Advances in Analog Circuits PDF

ADVANCES IN
ANALOG CIRCUITS
Edited by Esteban Tlelo-Cuautle
Advances in Analog Circuits
Edited by Esteban Tlelo-Cuautle
Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia
Copyright © 2011 InTech

All chapters are Open Access articles distributed under the Creative Commons
Non Commercial Share Alike Attribution 3.0 license, which permits to copy,
distribute, transmit, and adapt the work in any medium, so long as the original
work is properly cited. After this work has been published by InTech, authors
have the right to republish it, in whole or part, in any publication of which they
are the author, and to make other personal use of the work. Any republication,
referencing or personal use of the work must explicitly identify the original source.
Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted
for the accuracy of information contained in the published articles. The publisher
assumes no responsibility for any damage or injury to persons or property arising out
of the use of any materials, instructions, methods or ideas contained in the book.
Publishing Process Manager Katarina Lovrecic

Technical Editor Teodora Smiljanic
Cover Designer Martina Sirotic
Image Copyright Vilmos Varga, 2010. Used under license from Shutterstock.com
First published February, 2011

Printed in India
A free online edition of this book is available at www.intechopen.com

Additional hard copies can be obtained from [email protected]
Advances in Analog Circuits, Edited by Esteban Tlelo-Cuautle

p. cm.
ISBN 978-953-307-323-1
free online editions of InTech
Books and Journals can be found at
www.intechopen.com
Contents
Preface IX
Part 1 Circuit Design 1
Chapter 1 Analog CMOS Design Automation

Methodologies for Low-Power Applications 3
Alessandro Girardi and Lucas C. Severo
Chapter 2 A New Approach to Biasing Design of Analog Circuits 15

Reza Hashemian
Chapter 3 New Port Modeling

and Local Biasing of Analog Circuits 47
Reza Hashemian
Chapter 4 Behavioral Modeling of Mixed-Mode Integrated Circuits 85

Esteban Tlelo-Cuautle, Elyoenai Martínez-Romero,
Carlos Sánchez-López, Francisco V. Fernández,
Sheldon X.-D. Tan, Peng Li and Mourad Fakhfakh
Part 2 Design Issues 109
Chapter 5 Parallel Preconditioned Hierarchical Harmonic

Balance for Analog and RF Circuit Simulation 111
Peng Li and Wei Dong
Chapter 6 Lifetime Yield Optimization of Analog Circuits Considering

Process Variations and Parameter Degradations 131
Xin Pan and Helmut Graeb
Chapter 7 Linear Analog Circuits Problems

by Means of Interval Analysis Techniques 147
Zygmunt Garczarczyk
Chapter 8 Analog Design Issues

for Mixed-Signal CMOS Integrated Circuits 165
Gabriella Trucco and Valentino Liberali
VI Contents
Chapter 9 Tunable Analog and Reconfigurable

Digital Circuits with Nanoscale DG-MOSFETs 181
Savas Kaya, Hesham F. A. Hamed and Soumyasanta Laha
Chapter 10 Statistical Analog Circuit Simulation:

Motivation and Implementation 207
David C. Potts
Chapter 11 Advanced Statistical Methodologies

for Tolerance Analysis in Analog Circuit Design 227
Bruno Apolloni, Simone Bassis, Angelo Ciccazzo,
Angelo Marotta, Salvatore Rinaudo and Orazio Muscato
Part 3 Applications 245
Chapter 12 Analog-aware Schematic Synthesis 247

Yuping Wu
Chapter 13 An SQP and Branch-and-Bound Based Approach

for Discrete Sizing of Analog Circuits 297
Michael Pehl and Helmut Graeb
Chapter 14 Analog Circuit for Motion Detection Applied

to Target Tracking System 317
Kimihiro Nishio
Chapter 15 Analog Circuits Implementing a Critical Temperature

Sensor Based on Excitable Neuron Models 327
Gessyca M. and Tovar Nunez
Chapter 16 Evolvable Metaheuristics on Circuit Design 347

Felipe Padilla, Aurora Torres, Julio Ponce,
María Dolores Torres, Sylvie Ratté and Eunice Ponce-de-León
Preface
Analog circuit design imposes many issues and challenges to guarantee the develop-
ment of successful applications. For instance, the accomplishment of target specifica-
tions requires the highest experience of analog designers along with their creativity
and ingenuity to deal with trade-offs and to discover the obscure interactions among
design parameters. From this point of view, analog circuit design is considered a kind
of art.
To enhance analog circuit’s performances, a designer very often applies rules of thumb,
making almost impossible the development of systematic or generic design receipts,
in part because there exist a very huge plethora of circuit topologies and each one
requires different design strategies. Fortunately, researchers around the world share
acquired experience and insights to introduce advances in analog circuit design, mod-
eling, simulation and optimization. That way, this book summarizes recent advances
in analog circuits, covering a wide range of topics from circuit theory to multidisci-
plinary applications. The key contribution of each chapter focus on recent advances in
analog circuits, open issues and new challenges to accomplish academic or industrial
target specifications.
Electronic design automation of analog circuits is presented in Chapter 1 for low-pow-

er applications. The appropriate biasing is covered in Chapter 2, and it is extended to
new port modeling and local biasing of analog components in Chapter 3. In Chapter 4
a behavioral modeling approach at a level of abstraction higher than the transistor one
for mixed-mode circuits, is introduced. The following chapters introduce novel design
issues. For instance, Chapter 5 presents a hierarchical harmonic balance for analog and
RF circuit simulation. This chapter is much focused on design automation challenges,
as well as Chapter 6 to yield optimization of analog circuits considering process varia-
tions and parameter degradation. Chapter 7 shows the application of interval analysis
techniques for linear analog circuit’s problems.
When analog circuits co-exist with digital ones they process mixed-signals, as shown
in Chapter 8 for CMOS integrated circuits. Chapter 9 presents the application of nano-
scale DG-MOSFETs for tunable analog and reconfigurable digital circuits. Nanotech-
nology needs the application of statistical simulation, as shown in Chapter 10. Chapter
11 presents advanced statistical methodologies for tolerance analysis.
Analog circuits applications like the analog-aware circuit schematic synthesis, is shown
in Chapter 12. Sizing is a very complex topic and an SQP and branch-and-bound based
X Preface
approach is presented in Chapter 13. Chapters 14 and 15 introduce analog circuits for
motion detection applied to target tracking system, and for implementing a critical
temperature sensor based on excitable neuron models, respectively. Chapter 16 finally
discusses evolvable metaheuristics on circuit design.
Enjoy the book!
Esteban Tlelo-Cuautle
INAOE
Department of Electronics
Mexico
Part 1
Circuit Design
1
Analog CMOS Design Automation

Methodologies for Low-Power Applications
Alessandro Girardi and Lucas C. Severo
Federal University of Pampa - UNIPAMPA
Brazil
1. Introduction
The design automation of analog CMOS integrated circuits (ICs) is a demanding task in
microelectronics industry, because of the crescent necessity for low-power design and reduced
time-to-market. Nowadays, most analog sizing designs are done manually - with some aid
of simulation tools and equation-based models - and the quality of the resulting circuit is
dependent on the expertise of the designer. A system-on-chip (SOC) design has analog and
digital parts, each one designed with different methodologies and tools. The analog design
time must be compatible with the highly automated digital design time, which employs
advanced design automation tools (Gielen & Rutenbar, 2000).
The automation of fundamental analog design steps is extremely relevant for the success of
a project. The transistor sizing stage is, perhaps, the most difficult to automate due to the
large and highly non-linear design space. This stage is time consuming and might induce
significant delays relating to time-to-marketing. Nowadays, there is no analog circuit sizing
tools fully automatic searching the entire design space and taking advantage of state-of-the-art
fabrication technologies. Also, layout generation of analog blocks is error-prone and time
demanding.
An analog integrated circuit design is composed by transistors with different gate widths
and lengths, requiring complex techniques of layout generation to minimize variations and
improve matching. A traditional analog design methodology includes poor automated
calculations with electrical models based on first order equations, several iterations of spice
simulations and analysis, and full-custom layout generation. The experience of the designer
is fundamental for the quality of the resulting design and for the amount of time spent.
In general, the entire design space is rarely explored, mainly in transistor weak and moderate
inversion regions, which are the most appropriated for power-constrained applications.
The design space for the automatic synthesis of analog CMOS integrated circuits is highly
nonlinear. There are tens of free variables in the design of a typical analog integrated block
(such as an operational transconductance amplifier), related to gate dimensions (W and
L), bias currents or inversion levels. As the relation between transistor sizes and circuit
specifications (design objectives) is sometimes conflicting, the problem of finding an optimum
solution point is difficult to be exactly solvable. Some works have been done in this theme
describing the development of tools for analog design automation (ADA), using different
meta-heuristics and algorithms (Liu et al., 2009) (Vytyaz et al., 2009). The goal is always
the automation of time-consuming tasks and complex searches in highly non-linear design
4 Advances in Analog Circuitsi
spaces (Xu et al., 2009) (de Smedt & Gielen, 2003) (Hershenson et al., 2001). Basically
all of them can be categorized as equation-based or simulation-based automatic designs.
In the equation-based design strategy, analytical equations are used for modeling device
electrical characteristics, such as drain current, inversion level or small-signal parameters.
These models are often simplified or manipulated in order to fit certain limitations imposed
by optimization heuristics. The simulation-based strategy is based on results of electrical
simulations of the circuit to extract device parameters and design characteristics. The
simulation can be automated and performed several times until reaching the design objective.
Both strategies have demonstrated limitations but, together with powerful optimization
meta-heuristics, they are very promising for finding near-optimum design solutions in an
acceptable computational time. The goal of this text is to compare two different techniques
for automatic sizing of analog integrated amplifiers. The first one exploits the analytical
gm/ID methodology, in which the transconductance (gm) to drain current (ID ) ratio of the
transistors are free variables and gate width and length are defined in terms of the technology
independent gm/ID versus ID /(W/L) curve; and the second one is numeric, based on
an automated sequence of simulations of a spice netlist with W and L as free variables.
We employed Genetic Algorithms (GA) as optimization heuristics. Both methodologies
were implemented for sizing a power-constrained design of a two-stage Miller operational
transconductance amplifier for three different gain-bandwidth requirements.
2. Operational amplifier sizing optimization

The design of analog integrated circuits requires extensive design practice with a given
technology to correctly size transistors in order to achieve the required performance.
Analytical knowledge-based equations describe the relations between the transistors (design
parameters), design specifications (e.g. slew-rate grater or equal 10V/μs) and design
objectives (such as minimum power, area, noise, etc, or a combination thereof). These
equations are topology-specific and can be used within an automatic synthesis methodology,
which must perform the resolution of a system of non-linear equations. This system usually
has more independent variables than equations, returning a wide solution space. As a design
example using the two design methodologies here described, we used a two-stage CMOS
Miller operational transconductance amplifier (OTA). The circuit schematic of this amplifier
is shown in fig. 1. The Miller OTA is composed by an input differential pair and a current
mirror with active load in the first stage. The second stage is composed by an inverter
amplifier. Between the first and second stages is connected a compensation capacitor for
stability purposes. Chosen the analog IC cell topology, the initial task of the optimization
is to define search variables, specifications, and constraints in an appropriate manner. The
free variables can be the channel lengths and widths of MOS transistors, transistor inversion
levels, bias currents, capacitor values, etc.
As design specifications, we can include slew rate (SR), low frequency voltage gain (AV0 ),
gain bandwidth product (GBW), phase margin (PM), input common mode range (ICMR),
power dissipation and silicon area (Allen & Holberg, 2002). The slew rate (SR) is calculated
using the following equation:
I7
SR = (1)
Cf
Analog CMOS Design Automation Methodologies for Low-Power Applications 5
Here, I7 is the drain current of T7 and C f is the compensation capacitance. The low-frequency
voltage gain of this amplifier is the product of first gain stage and the second gain stage and
is given by
gm1 gm5
Av0 = · (2)
gds2 + gds4 gds5 + gds6
where gm is the gate transconductance and gds is the output conductance of MOSFETs
transistors. The Gain Bandwidth Product (GBW) is calculated using the transconductance
gm1 and the capacitance C f :
gm1
GBW = (3)
Cf
The minimum and maximum values for the input common-mode range (ICMR) are evaluated
using the large signal model, given by eq. 4 and 5, respectively.

+ I7
ICMR = VDD − − |VT2 | − VDS7(sat) (4)
β2

− I7
ICMR = VSS + + VT4 − VT2 (5)
β4
Here, VT is the threshold voltage, VDS is the voltage between the drain and source terminals
and β is a factor which depends on transistor size, carrier mobility (μ0 ), gate oxide thickness
(Tox ) and silicon oxide permittivity (ox ), given by
ox W
β = μ0 · · (6)
Tox L
The circuit power dissipation is given by the product between the supply voltage and total
current consumption.
Pdiss = (VDD − VSS ) · IDD (7)

The area occupied by the circuit is also an important specification. It cannot be exactly
calculated in the design sizing stage because it depends on the layout strategy to be used
in the physical synthesis design stage. However, an approximation considering gate area as
the main parameter can give a good indication of the circuit total area.
k
A gate = ∑ Wi · Li + AC f
(8)
i =1
Here, k is the number of transistors in the circuit. We also include the area occupied by the
compensation capacitor (AC f ), which is proportional to its capacitance value (in general, it is
implemented with double poly in CMOS technology).
The optimization strategy relies on minimizing a cost function, given as
n m
fc = ∑ αi p̂i (X ) + ∑ β j ĉ j (X ) (9)
i =1 j =1
where αi is the weighting coefficient for performance parameter p̂i ( X ), which is a normalized
function of the vector of independent design parameters X (free variables). This function
allows the designer to set the relative importance of competing performance parameters,
such as, for example, a weighted relation between power and area. The parameter ĉ j ( X ) is
a constraint normalized function, which limits the design space to feasible solutions of design
specifications. The coefficient β j indicates how closely the specification must be pursued.
The constraint function, for specification of a minimum, has the following form:
⎧c
⎨ jre f if c jre f > a · c jre f or c jre f < c j ( X ),
ĉ j ( X ) = c j ( X ) (10)
⎩0 if c ≤ c ( X ) ≤ a · c .
jre f j jre f
So, once the constraint value is achieved, it does not contribute for the increasing of the
cost function value. The constant a means a percentage of the constraint overvalue that
is considered accepted and it is necessary for avoiding an overestimation of a determined
parameter during the optimal point search procedure. For a specification of a maximum,
the constraint function has the inverse form. If c j ( X ) is inside a given specification, ĉ j ( X )
is set to zero. The cost function is computed in every iteration in the optimization loop. The
correct design space exploration is directly related to the cost function formulation (Koza et al.,
1997)(Alpaydin et al., 2003).
Fig. 1. Schematics of a two-stage Miller OTA.

The genetic algorithm, used in this work, is a heuristic for non-linear optimization based on
the analogy with biologic evolution theories (Venkataraman, 2001). It is a non-deterministic
algorithm and it works with a variety of solutions (population), simultaneously. The
population is a set of possible solutions for the problem. The size of the population is defined
in order to maintain an acceptable diversity considering an efficient optimization time. Each
possible solution of population is denominated a chromosome, which is a chain of characters
(gens) that represent the circuit variables. This representation can be in binary number, float
or others. The quality of the solution is defined by an evaluation function (cost function).
The algorithm receives an initial population, created randomly, some recombination and
mutation operators and the MOSFET technology model parameters. The population is
evaluated using a conventional SPICE electrical simulator. Based on valuation and roulette
method the parent chromosomes are selected for generating new chromosomes. The new
chromosomes are created including recombination and mutation - analogy with biology. In
the recombination, the chromosomes of two parents are divided and the union of the parts
produces a recombination. By the other side, mutation is a random error that happens
in a chromosome. The probability of mutation is defined by the user and it is compared
with a random value. If this random value is smaller than the probability value then a
gene on chromosome is randomly changed. In the case of analog design, it means that a
random variation is created over a certain design parameter. The next step is the exclusion
of parents and evaluation of new chromosomes, using again the electrical simulator and a
cost function. Based on these values, new chromosomes are introduced in the population. At
the end of each iteration, the stopping condition is tested and, if true, then the optimization
is finished. Otherwise, new parents are selected and the process is repeated. The stopping
condition can be the number of generations (iterations), minimal variation between variables
or cost function, or others. In GA, the number of individuals in the population is very
relevant, because it deals with several solutions simultaneously. Larger population increases
the diversity of solutions but also increases the optimization time. Then, the number of
population individuals must be chosen according to criteria of assuring solution diversity
but maintaining a practical optimization time. The implementation of GA used in this work
was GAOT (Genetic Algorithms Optimization Toolbox) for Matlab™(Houck et al., 1996).
3. Simulation-based methodology
The simulation-based strategy for automatic sizing of analog circuits is based on the results
obtained by electrical simulations of the target circuit. Several runs of simulations must be
performed, each one with different values for the circuit free variables. Variable perturbation
is defined by the optimization meta-heuristic and the convergence for an optimal solution
point depends on the correct search of the design space.
The sizing tool receives design specifications and technology model as parameters. Design
specifications are the required values of circuit specifications. These values are used as
objective and constraints in the optimization flow. The technology parameters and device
models are used for the electrical circuit simulation of MOS transistors. Knowing the input
values, the solution (population) is generated using an initialization function in the genetic
algorithm. This function generates a population of possible solutions for the circuit. In the
initialization function the initial solutions are generated randomly and evaluated by means
of electrical simulations. The solution evaluation function analyses the constraints and the
specification of the circuit to be optimized, as, for example, power dissipation, circuit area,
noise or others. The design flow of simulation-based strategy using Genetic Algorithms is
shown in fig. 2. The next step is to select solutions (parents) for generating a new set of
solutions using the techniques of crossover and mutation previously described. The new
solutions are evaluated using the electrical simulation and the evaluation function. After each
iteration, new solutions are inserted in the population and the old members (old solutions)
are excluded. The end of the optimization process happens when a stop condition is satisfied.
The stop condition can be a maximum number of population generations (iterations) or the
minimum variation of the cost function value (evaluation function).
Fig. 2. Simulation-based design flow using genetic algorithms.
4. gm/ID methodology
In the design procedure herein described, a methodology called gm/ID is used for the circuit
performance evaluation. This methodology considers the relationship between the ratio
of the transconductance gm over DC drain current ID and the normalized drain current
In = ID /(W/L) as a fundamental design parameter (Silveira et al., 1996), such as the
curve shown in fig. 3. The gm/ID characteristic is directly related to the performance
of the transistors, gives a clear indication of the device operation region and provides a
way for straightforward estimation of transistors dimensions. The main advantage of this
method is that the gm/ID xIn curve is unique for a given technology, reducing the number
of electrical parameters related to the fabrication process. Additionally, its analytical form
covers all transistor operation regimes, from weak to moderate to strong inversion. The
gm/ID xIn curve can be automatically evaluated by electrical simulation or by measurement
data. The analog circuit modeling for using with genetic algorithms is straightforward. Fig.
4 shows the proposed optimization design flow. The user enters the design specifications,
technology parameters and configures the cost function according to the required design
objectives and specifications. The optimization loop performs perturbations on the design
variables, whose amplitude is defined by the algorithm. These variables are defined by the
user, and are always related to the transistor geometry, large and small-signal parameters,
such as W, L, ID , gm and gm/ID . Following, the design properties evaluation is performed
by the calculation of the circuit characteristics such as voltage gain, cut-off frequency, phase
margin, dissipated power, input common-mode range, etc. This is done using circuit-specific
analytical equations, the gm/ID versus In curve and a transistor model for calculation of
transconductances, drain-source saturation voltages and currents. If the circuit is feasible,
i.e., transistor sizes are within an allowed range, the cost function can be evaluated and the
solution is accepted if the cost decreased. The final solution returns the devices dimensions.
Fig. 3. gm/ID x ID /(W/L) curves for 0.35μm CMOS technology.
Fig. 4. Design flow for the gm/ID design methodology.

Fig. 5. Cost function evolution.
5. Design example
In order to compare both previously described automatic synthesis strategies, three
corner designs were implemented for a Miller OTA, for three different specifications of
gain-bandwidth product (GBW): 0.1, 1 and 10MHz. The slew-rate, directly proportional to
GBW, was also defined as 0.1, 1 and 10V/μs. These designs are named Design 1, Design 2 and
Design 3, respectively. The other design constraints were held unchanged for the three designs
and are shown in table 1. The design objective is to minimize power consumption and area,
i.e., minimize I1 and I2 currents according to the schematics of fig. 1, since supply voltage is
constant, keeping gate dimensions as smaller as possible. The cost function equation has the
same format as shown in eq. 11. Here, the performance parameter is given by
Pdiss A gate
p̂( X ) = + (11)
Pdiss(re f ) A gate(re f )
where Pdiss and A gate are the DC power consumption and gate area, respectively - estimated
for each iteration - and Pdiss(re f ) and A gate(re f ) are reference values for normalization purposes.
Design constraints include minimum gain-bandwidth product (GBW), minimum voltage DC
gain (Av0 ), minimum phase margin (PM), minimum slew rate (SR) and the minimum and
maximum input common mode range (ICMR+ and ICMR− ).
Both design strategies implemented used the same set of design constraints. Also, as a
topology characteristic of Miller amplifier of fig. 1, some transistors need to be matched,
such as the input differential pair M1-M2 and the current mirrors M3-M4 and M7-M8
(multiplication factor of 1), diminishing the number of design free variables. The AMS CMOS
0.35μm was the target fabrication technology. Transistor lengths were limited in the range
between 0.35μm and 10μm and the widths between 1μm and 500μm for avoiding infeasible
solutions. The value of Cout was fixed in 10pF and VDD and VSS in 1.65V and -1.65V,
respectively. Next subsections describe the optimization setup for both methodologies and
the comparison of results.
5.1 Methodology 1: Simulation-based

In the simulation-based (SB) methodology with genetic algorithms, the design space
exploration was performed with a population of 1000 individuals. The specifications were
estimated by SPICE electrical simulations using the ACM transistor compact model (Cunha
et al., 1998), guaranteeing the exploration of weak, moderate and strong inversion regions.
Different types of SPICE analysis need to be generated for complete performance estimation.
For estimation of low frequency voltage gain, GBW and phase margin, the AC analysis is
executed, generating the Bode Diagram. For the ICMR evaluation a DC analysis is necessary.
For slew rate, DC currents and large and small signal parameters estimation it is used the
operation point (OP) analysis. Design specifications are calculated based on the simulation
results. In this design, 11 design free variables were selected, including the transistor
dimensions (W and L) and the bias current Ibias . These variables suffer a perturbation by
the algorithm at each iteration and the values are updated in the circuit netlist. Fig. 6 shows
the evolution of GBW, phase margin, low-frequency voltage gain and slew-rate for Design 3
in relation to the iteration number using SB methodology.
(a) Gain-bandwidth product (b) Phase margin
(c) Low-frequency voltage gain (d) Slew-rate
Fig. 6. Evolution of 4 design specifications for Design 3 with Simulation-Based methodology.

5.2 Methodology 2: gm/ID

In this design strategy, the independent variables are the gm/ID relationships and channel
lengths of each transistor. All design equations are put in terms of these parameters.
The drain current for these transistors can be calculated with the information about the
transconductance-to-current ratio:
gmi
IDi = (12)
gm
ID i
With the ACM transistor model we can estimate the Early voltage according to the transistor
length. The free variables subjected to perturbations by the genetic algorithm are: L1 = L2 ,
L3 = L4 , L5 , L6 , L7 = L8 , ( gm/ID )1 = ( gm/ID )2 , ( gm/ID )3 = ( gm/ID )4 , ( gm/ID )5 ,
( gm/ID )6 , ( gm/ID )7 , and the dependent parameters are W1 = W2 , W3 = W4 , W5 , W6 ,
W7 = W8 , C f and bias current. The range of gm/ID is well known from device physics
and behaves smoothly over a wide range of transistor biases, which is advantageous for
the search robustness. Moreover, the design space is limited by values of gm/ID between
zero and 28V −1 , which is the theoretical maximum gm/ID of bulk MOS transistors. Design
objectives and design specifications are evaluated in terms of free variables ( gm/ID )i and Li .
The same occurs with the dependent variables such as Wi and IDi . So, the transistor width can
be calculated as:
I Di · L i
Wi = (13)
Ini
where Ini is the normalized current of the ith device, given by the gm/ID xIn curve. The design
characteristics calculation is straightforward. The low-frequency gain, for example, is given
by
gm VA1 · VA3 gm VA5 · VA6
Av = · · · (14)
ID 1 VA1 + VA3 ID 5 VA5 + VA6
VA is the Early Voltage, directly dependent on gate length.
5.3 Comparison results

Table 1 shows the results of the performance obtained for designs 1, 2 and 3 using both
described methodologies. Table 2 shows the transistor sizes, inversion levels and the values
obtained for the bias current and compensation capacitor. Although each methodology used
a totally different approach for finding an optimum design, they achieved similar results.
In Design 1, with a target GBW of 100kHz, the gm/ID methodology provided a power
consumption of 3.52μW, against 4.48μW achieved by the simulation-based methodology. The
values of gm/ID of the input differential pair (M1 and M2) achieved similar values in both
methodologies, located in the weak inversion region. The same is valid for Designs 2 and 3,
with GBW in 1MHz and 10MHz, respectively, in which the input pair biasing was also located
in moderate or weak inversion. In Design 2, the SB methodology achieved the best result, with
power consumption of 47.8μW. In Design 3, however, the gm/ID approach achieved a power
consumption of about a third from that obtained by the SB methodology, at the expense of
larger gate area.
Av0 GBW PM [°] SR ICMR [V] Pdiss A gate

[dB] [MHz] [V/μs] [μW] [μm2 ]
+ -
Design 1
Spec. 70.0 0.1 60 0.1 -0.70 0.70 min. min.
gm/ID meth. 73.5 0.1 63 0.1 -1.64 1.32 3.52 740.8
SB meth. 73.4 0.1 61 0.1 -1.65 1.32 4.48 4420.0
Design 2
Spec. 70.0 1.0 60 1.0 -0.70 0.70 min. min
gm/ID meth. 70.1 1.0 61 1.0 -1.62 1.35 58.2 502.3
SB meth. 70.0 1.0 60 1.1 -1.65 1.34 47.8 5200.0
Design 3
Spec. 70.0 10.0 60 10.0 -0.70 0.70 min. min
gm/ID meth. 76.0 10.0 98 10.0 -1.64 1.31 296 6678.2
SB meth. 72.8 11.0 60 10.0 -1.59 1.44 852 2370.0
Table 1. Miller OTA synthesis results using gm/ID and simulation-based (SB) design
methodologies.
Parameter Design 1 Design 2 Design 3
gm/ID SB meth. gm/ID SB meth. gm/ID SB meth.
meth. meth. meth.
(W/L) M1,M2 31.1/4.9 97.0/6.0 113.0/2.2 296.0/3.1 126.0/0.9 217.0/0.4
(W/L) M3,M4 3.9/4.7 208.0/5.4 154.5/4.6 463.0/3.4 97.9/0.7 208.0/5.0
(W/L) M5 35.2/3.8 600.0/1.6 143.8/0.5 673.0/0.4 335.1/0.4 306.0/0.4
(W/L) M6 35.9/4.3 5.6/1.0 59.0/3.8 3.8/4.1 25.1/1.8 5.0/1.0
(W/L) M7,M8 3.7/4.9 3.2/4.8 1.0/0.6 1.0/5.0 4.4/2.3 1.0/3.4
( gm/ID ) M1,M2 25.1 28.8 23.5 27.3 17.4 26.9
( gm/ID ) M5 20.4 28.7 25.5 28.5 21.9 18.8
( gm/ID ) M6 14.8 8.34 7.8 2.18 2.9 0.48
Ibias [μA] 0.27 0.23 3.10 2.42 30.26 24.6
C f [pF] 2.71 2.20 2.91 2.20 3.02 2.20
Table 2. Miller OTA transistor sizes synthesized with gm/ID and simulation-based (SB)
automatic design methodologies. (gm/ID values are in V −1 and W and L are in μm.)
6. Conclusion
There are several techniques for automating analog integrated circuit design. The automation
has advantages over manual design, exploiting more effectively the design space and
searching for close to optimum solutions. However, circuit modeling and cost function
formulation have great impact on the final optimization solution. This work presented the
implementation of two different automatic design methodologies for sizing a two-stage Miller
OTA: analytical gm/ID methodology and numerical simulation-based methodology with
Genetic Algorithms. Considering exactly the same conditions for both methodologies - same
technology parameters, design objectives and constraints -, three power-constrained corner
designs were executed for three values of GBW: 0.1, 1 and 10MHz. As the optimization results
showed, both design methodologies achieved similar results, exploring weak, moderate
and strong inversion regions. The slightly differences in the results demonstrate that both
methodologies, even though using distinct design strategies, are adequate for the automatic
design of OTAs, with advantages over manual design. Genetic algorithms are very suitable for
analog design automation by the fact that the convergence of the final solution is not directly
dependent on the initial solution, and it is not necessary a deep knowledge by the human
designer about the circuit characteristics. However, it is very important to determine the size
of population (number of individuals) because it is directly related to the quality and to the
amount of time expended by the optimization process.
7. References
Allen, P. E. & Holberg, D. R. (2002). CMOS Analog Circuit Design, 2nd edn, Oxford University
Press, Oxford.
Alpaydin, G., Balkir, S. & Dundar, G. (2003). An evolutionary approach to automatic synthesis
of high-performance analog integrated circuits, IEEE Transactions on Evolutionary
Computation 7(3): 240–252.
Cunha, A. I. A., Schneider, M. C. & Galup-Montoro, C. (1998). An MOS transistor model for
analog circuit design, IEEE Journal of Solid-State Circuits 33(10): 1510–1519.
de Smedt, B. & Gielen, G. G. E. (2003). Watson: Design space boundary exploration and model
generation for analog and rf ic design, IEEE Transactions on Computer-Aided Design of
Integrated Circuits and Systems 22(2): 213–224.
Gielen, G. & Rutenbar, R. A. (2000). Computer-aided design of analog and mixed-signal
integrated circuits, Proceedings of the IEEE 88: 1825–1852.
Hershenson, M. D. M., Boyd, S. P. & Lee, T. H. (2001). Optimal design of a CMOS op-amp
via geometric programming, IEEE Transactions on Computer-Aided Design of Integrated
Circuits and Systems 20(1): 1–21.
Houck, C. R., Joines, J. A. & Kay, M. G. (1996). A genetic algorithm for function optimization:
A matlab implementation, Technical report, North Carolina State University.
Koza, J. R., III, F. H. B., Andre, D., Keane, M. A. & Dunlap, F. (1997). Automated synthesis
of analog electrical circuits by means of genetic programming, IEEE Transactions on
Evolutionary Computation 1(2): 109–128.
Liu, B., Fernandez, F. V., Gielen, G., Castro-Lopez, R. & Roca, E. (2009). A memetic
approach to the automatic design of high-performance analog integrated circuits,
ACM Transactions on Design Automation of Electronic Systems 14.
Silveira, F., Flandre, D. & Jespers, P. G. A. (1996). A gm/ID based methodology fo the design
of CMOS analog circuits and its application to the synthesis of a silicon-on-insulator
micropower OTA, IEEE Journal of Solid-State Circuits 31(9): 1314–1319.
Venkataraman, P. (2001). Applied Optimization with MATLAB Programming, Wiley-Interscience.
Vytyaz, I., Lee, D. C., Hanumolu, P. K., Moon, U.-K. & Mayaram, K. (2009). Automated design
and optimization of low-noise oscillators, Transactions on Computer-Aided Design of
Xu, Y., Hsiung, K.-L., Li, X., Pileggi, L. T. & Boyd, S. P. (2009). Regular analog/rf
integrated circuits design using optimization with recourse including ellipsoidal
uncertainty, IEEE Transactiions on Computer-Aided Design of Integrated Circuits and
Systems 28(5): 623–637.
2
A New Approach to
Biasing Design of Analog Circuits
Reza Hashemian
Northern Illinois University
United States
1. Introduction
A new approach for biasing analog circuits is introduced in this chapter. This approach is an
attempt to address some of the biasing complexities that exist today in biasing large analog
circuits. There are three steps involved in this methodology. First, in circuit analysis, the
methodology separates nonlinear components (transistors), particularly drivers, from the
rest of the circuit. Second, it uses local biasing introduced in the previous chapter to bias the
transistors individually and to the specs provided for the design. Finally, the method
presents a new way to change the local biasing into normal (global) circuit biasing with
choices of DC supplies at right locations in the circuit. It is the last step that will be our main
topic of discussion in this chapter. Here we see how we can remove all sources related to the
local biasing and replace them with normal circuit supplies without altering the design
specifications. These circuit supplies can be voltage sources, current sources or mirrors. In
case the supplies are already specified and in place, this method can still maintain the
design specs by re-evaluating some of the power-conducting components in the circuit.
Power-conducting components are those circuit components, such as resistors, that conduct
DC power (current) from the power supplies to the circuit drivers (transistors), for biasing
purposes.
Limitations in local Biasing - We fully discussed local biasing, its properties and applications
in the previous chapter. Despite all the advantages that local biasing offers one problem still
remains unresolved and that is: how to deal with so many DC sources generated due to
local biasing, known as distributed supplies? To see the problem, just take a single bipolar
transistor: it normally needs four (voltage and current) sources to get locally biased;
however, with coupling capacitors used this number reduces to two current sources and
two capacitors (taking care of the voltage drops). Similarly, we may need to use four sources
to locally bias a MOS transistor. Again, with coupling capacitors this number can get as low
as one source-drain current source. The problem, however, is that for the gate, and possibly
the substrate, the coupling capacitors need to have charging paths (a resistive path to a DC
supply). One way to handle the case and bring the number of DC supplies down to a
minimum of one or two is to use source transformation and replacement techniques, such as
voltage dividers, Δ-Y transformation, and current sources/mirrors. Nevertheless, the sheer
number of such sources in a fairly complex circuit can get so high that unless we find a
shortcut to the final solution the validity of local biasing as an effective methodology is
undermined.
16 Advances in Analog Circuits
A new strategy - We are introducing a different strategy for biasing analog circuits in this
chapter. The core of this strategy lies on the fact that in an analog circuit design environment
we only need to anchor down certain critical biasing specs and not all. By critical biasing specs
we mean those operating conditions that are essential in achieving the design criteria, such
as gains, undistorted output signals or power consumption. Other design criteria usually
adhere to these critical specs and adapt to the situation fairly well enforced by the critical
specs. The fact that DC supplies are present in a circuit only to bias the nonlinear
components reveals the fact that for each biasing (critical) spec we need to provide a path to
a DC power supply, controlled by the spec. With this in mind, the proposed strategy makes
a one-to-one correspondence between the circuit biasing requirements (specs) and those DC
(voltage or current) supplies needed to support these requirements. Hence, we need at least
as many path to DC power supplies as we have biasing specs in a circuit. Consequently, the
first task in this strategy is to pair each biasing spec with a biasing supply (voltage, current
or a power-conducting component). Second, the method must be capable of replacing
“distributed supplies” -- if a local biasing strategy is already in use -- with normal circuit
supplies, such as VCC and VDD. The idea here is to keep the main properties of local biasing –
translated into the critical biasing specs -- while removing local biasing sources to be
replaced with the normal biasing supplies.
The main advantages in employing this strategy are: i) to pin down the operating conditions
for the critical transistors while replacing the local biasing sources with a much fewer
designated DC supplies, ii) to minimize design efforts to fulfill only critical specs, hence
speeding up the process, and iii) the possibility to perform biasing entirely linearly. The last
point is particularly important and makes biasing almost a one-step process.
This chapter introduces two new circuit elements, fixator and norator, that are the center
pieces in our biasing design strategy. Fixators and norators come in pairs as effective tools to
perform a targeted biasing. It is shown that these pairs are very instrumental in matching
biasing critical specs with DC power resources. The method simply associates a designated
supply source (or a power-conducting component) with an arbitrary biasing spec. Fixator-
norator pairs cause local biasing sources (distributed supplies) to be entirely replaced with
normal circuit supplies designated by the designer. It is shown that the pair, when used
properly and in combination, will adhere to Kirchhoff laws as well.
Important properties of fixator-norator pairs are introduced in this chapter, and the
relationships between a fixator-norator pair and other circuit components (such as resistors,
voltage sources, and current sources) are discussed. Rules and regulations corresponding to
the use of fixator-norator pairs in a circuit are investigated. Being special circuit
components, fixators and norators must be used so that KVL and KCL are not violated in a
circuit. However, it is important to note that the use of fixator-norator pairs is only
temporary in this methodology; i.e., the pairs are removed as soon as the final circuit biasing
is established and the DC power is provided for the circuit. This is important in a sense that
ideal controlled sources, with very high gains, can be used to mimic fixator-norator pairs
without any restrictions. Because fixators can model fixed-biased ports, these devices can
also model nonlinear components for specified biasing situations. These nonlinear
components can be p-n junction diodes (as single port devices), bipolar transistors (as two
port devices), and MOS transistors (as three port devices).
An algorithm that explains the biasing design procedure of analog circuits is also introduced
in this chapter. This algorithm classifies circuit design procedure into two areas: the
performance (AC) design and the biasing design. The performance design (gain, bandwidth,
A New Approach to Biasing Design of Analog Circuits 17
SNR, power, distortion, and so on) is done first. Here is where the circuit topology and the
major circuit components are determined to achieve the design goals. In the performance
design the circuit is treated entirely linear, where the transistors are replaced with their
linear models at specifies operating points. Upon finishing the performance design the
circuit biasing design begins by providing a set of critical biasing specs. It is in this stage that
the linear models of the transistors (used in the performance design) are replaced with the
fixator models. Next, the designer needs to accommodate for the norators that must pair
with the fixators. He/she has variety of choices to place the pairing norators in the circuit;
having in mind that they are place holders for the power supplies, current sources/mirrors,
or power-conducting components. When finished, the circuit is ready for simulation, while
still linear. The results from the DC solution contain the voltage and current values for each
norator; where, each in turn can be replaced with an appropriate component. This completes
the DC design procedure.
1.1 Circuit biasing

Biasing is a major step in designing analog circuits [1 – 3]. In large and complex circuits
biasing has always been a great challenge for designers. The challenge is normally in two
areas. First, to get the number of iterations minimized and make the convergence possible
and fast; second, to move to the right regions of operations for active components
(transistors) so that acceptable performance is attained and the output signals are far from
being distorted or clipped during the AC operation. Both problems grow in complexity as
the number of transistors increases, design requirements become tighter, or more efficient
designs are in demand. One difficulty in the traditional approach appears to be the lack of
separation between linear and nonlinear components, as well as between the nonlinear
components themselves during the biasing process. Typical biasing techniques deal with the
entire circuit as a whole with no classification or circuit partitioning; hence, the complexity
quickly increases as the circuit grows. In case of analog ICs, where almost all circuit
components are nonlinear, distinction between linear and nonlinear components becomes
meaningless. Instead we can categorize components into two categories: i) drivers, and ii)
supporting components. In conventional methods used for the analysis and simulation of
analog circuits all nonlinear components, regardless of their categories and functionalities,
are included in a global biasing (DC) analysis. Whereas in more advanced methods we can
distinguish between the drivers and those supporting components, such as current
sources/sinks, current mirrors, and active loads [4 - 6]. Drivers typically reside along the
signal path directly shaping the output waveforms. They are strongly influencing the design
specifications, and are more sensitive to signal conditionings. Consequently, drivers must be
biased with more care and precision compared to the supporting components in a circuit.
2. Nullators, fixators and norators

We first need to define terms that are used in this chapter. In addition, all our discussions
here apply to DC power unless stated otherwise.
2.1 Port parameter fixing

The methodology introduced here is based on assigning specific operating points to
nonlinear ports (in diodes and transistors) during the biasing design of an analog circuit.
These operating points are considered the critical specs for the design. Once the critical Q-
points are assigned to the ports of the transistors the methodology holds them fixed during
the entire design period. Now, the question is how to keep a Q-point fixed while other
variables (voltages and currents) in the circuit are changing? As we will see, the answer to
this question lies in the use of fixator-norator pairs. A fixator is an expanded version of a
nullator.
A nullator is a two-terminal element with both its current and voltage equal to zero. A
norator is a two-terminal element with unspecified current and voltage [7 – 12].
Consider two networks N1 and N2 connected through a port j(Vj, Ij), as shown in Fig.1(a).
Nullify port j(Vj, Ij) from both sides by augmenting the port with voltage and current
sources that have the same port values, Vj, Ij, as discussed in the previous chapter. As a
result a new null port k(Vk, Ik) is created in the process, as shown in Fig.1(b). Now, because
port k is a null port (Vk = 0 and Ik = 0) we can split the two networks from port k and attach
each with a nullator, as depicted in Fig.2. Apparently, the operation has not changed any
current or voltage inside N1 or N2. In addition, it has fixed the port operating point (Ij and
Vj) so that any internal changes inside N2 (or N1) do not change the port’s Q-point. This
simply means that we can replace port j by a fixator.
Vj Vj
Ij Ij Ik Ij
N1 Vj N2 N1 Vj Ij Vk Ij Vj N2
(a) (b)
Fig. 1. Port nullification procedure
Vj Vj
Ij Ik Ik Ij
N1 Vj Ij Vk Vk Ij Vj N2
Fig. 2. Two networks N1 and N2 disjointed at port k(vk, ik) and each terminated by a nullator.
Fixator: A two-terminal component1 in a circuit is called a fixator if both the voltage across
the component and the current through the component represent independent sources [4].
Figures 3(a) and 3(b) represent two types of fixators and Fig. 3(c) is a symbol representing a
fixator. Note that a nullator is a special case of a fixator represented by Fx(0, 0), where both
1 A component here can represent any size two-terminal network.

the device voltage and current are zero. Also, note the difference between the two fixators
Fx(Vj, Ij) and Fx(Ij, Vj); in Fx(Vj, Ij) the voltage source Vj provides (or consumes) power and
the current source Ij is inactive2; whereas, in Fx(Ij, Vj) the current source Ij provides (or
consumes) power and the voltage source Vj is inactive. Note also the similarity between a
fixator and an H-model, discussed in the previous chapter. Both fixator and H-model model
a port, representing the existing situation of the port. The major difference, however, is that
in a fixator the equivalent impedance Req in the H-model is replaced with a nullator,
stamping on the port variables. This is because in an H-model the current going through the
Req is also zero making the voltage zero, as well. However, the replacement of Req with a
nullator removes the dynamics of the terminal and fixes the port values, Ij and Vj, for the
entire operation of the circuit; whereas in the case of Req the H-model behaves normally as
the Thevenin or Norton equivalent circuits behave. In fact, we can think of a fixator as a
snapshot of a port’s behavior, whereas an H-model represents the entire dynamics of the port
during the circuit operation. For example, take the case of two networks N1 and N2
connected through a port j, as in Fig.1(a); we can replace N1 by its H-model or alternatively
we can replace it with a fixator Fx(Vj, Ij), as shown in Fig. 4. In the later case we are bounded
with fixed values of Vj and Ij for the port; hence, the idea of fixing the design specs is born!
To further expand the idea, we need to look for a different role for a fixator. Notice that in
Fig. 4 we replaced the linear circuit N1 (or its H-model) with a fixator Fx(Vj, -Ij). Now we
can do the opposite; a fixator can replace a nonlinear component (or port) N2 in a circuit.
This is stated in Property 1.
Property 1: A two-terminal component, linear or nonlinear, in a circuit that is biased by a
current I and exhibits a terminal voltage V can be replaced with a fixator Fx(I, V) without
causing any change in the currents and voltages within the rest of the circuit.
One important conclusion from Property 1 is that, fixators are not only helping to fix the
design specs for biasing purposes, they also linearize a circuit by replacing all the nonlinear
components with fixators that are constructed from linear components. In addition, fixators
Vj Vj
Ij Ij
Fx(Vj, I j) Fx (Ij , Vj )
(a) (b) (c)
Fig. 3. (a) Voltage Fixator; (b) current Fixator; (c) Symbol representing a Fixator.
2
A source is inactive if it neither produces power or consumes power; hence, in an inactive
source either voltage or current is zero.
Ij
Vj N2
Fx (Vj , -I j)
Fig. 4. A Fixator replaced for the biasing circuit N1.
add to the stability of the design by performing a controlled approach to the design criteria.
For example, if for a certain specified biasing situation the circuit behaves unstably, one can
simply search for a more stable situation by slightly modifying the Q-points of certain
transistors. This can be done by modifying their corresponding fixators without really
touching any other parts in the circuit, or leaving the linearity conditions in the circuit.
In using fixators for port specification and stability, we realize that for each fixator used we
need to have one norator in the circuit to pair it with. As it turns out, fixator-norator pairs
provide an effective tool for us to perform the biasing strategy we are looking for in this
chapter. Here we show that the pair is the foundation for biasing circuits according to
biasing design specifications. The method shows how, through the use of fixator-norator
pairs, we can solve the problem of distributed supplies, generated because of local biasing. It
actually shows how a pair can be used to couple a biasing spec with a supporting supply
source; and in case the supply source is already specified in the design, the match is done
with a power-conducting component. Note that a fixator provides a solution and a pairing
norator finds, through the analysis, the resource needed for the solution. Hence, when used
in combination, the pair will adhere to Kirchhoff’s laws. In short, when a biasing criterion
requires inclusion in a design, a fixator keeps this criterion fixed while a norator provides,
allocated in an arbitrary location, the sourcing needed for the requirement. This is, of course,
only possible if the fixator can control the norator and, conversely, the fixator must also be
sensitive to the changes in the norator. Again, in case a designated DC supply is already in
place for the design, the norator can be placed in a location designated for a power-
conducting component, say a resistor, and then find its value through the analysis.
There is a different interpretation of fixator-norator pairs that is worth discussing. In
general, each circuit component is identified by its two variables, voltage and current. From
the two usually only one variable is specified, such as the voltage in a voltage source or the
current in a current source; alternatively the two may be related such as ohms law in a
resistor. This indicates that from the two variables one must be found through the circuit
laws, KVL and KCL. What makes fixators and norators different is that, in a fixator both
component variables are specified but in a norator neither is specified. Hence, none of them
can live alone in a circuit; whereas, when they pair they complement each other; i.e. overall,
the two carry two specified variables and two are left for the circuit to find. This description
of fixator-norator pairs suggests that the pair are no longer limited to DC operations and
they can be used in any circuit operation including linear and AC circuits. What it means is
that, in any type of circuit (linear or nonlinear) with any operation (DC or AC) one can set
(fix) some circuit variables in exchange for some component values. To think of it
differently, we can argue that fixator-norator pairs change a circuit analysis procedure to a
design procedure that guaranties certain design specifications, if obtainable. This is because
in circuit analysis we are given all component values and resources needed to analyze a
circuit; whereas, in a design procedure there are some component values or resources to be
determined in exchange for achieving some design specs.
Example 1: To show how the process works, we start with a simple diode circuit depicted in
Fig. 5 with an unspecified supply voltage V1. Suppose the design requirement in this
example is to find the value for V1 so that the diode current reaches 1mA. Figure 6 shows
the circuit arrangement for this design using a fixator-norator pair to satisfy the design
criteria. As shown, the added fixator -- a current source ID = 1 mA in parallel with a nullator
-- forces the assigned current through the diode. Now, because the voltage across the
current source is kept zero, the added fixator has no effect on the overall operation of the
circuit. In addition, a norator is substituted for the unknown supply voltage V1. Next, we
simulate the circuit and get a voltage of V1 = 2.2 V across the norator with a current I1 = 1.2
mA through it. This suggests that although we have aimed for the voltage source V1 to
replace the norator, we have in fact two more choices to make: i) replace the norator with a
current source I1 = 1.2 mA, or ii) replace the norator with a resistor R1 = -V1/I1 = -2.2/1.2 = -
1.8 KΩ. However, the last choice of a negative (active) resistance is not definitely acceptable
for this design.
1
1KΩ 2
300Ω
3
V1 5KΩ D
Fig. 5. A diode circuit with an unspecified supply voltage V1
1
1KΩ 2
300Ω
3
4
1mA
V1 5KΩ D
Fig. 6. The diode circuit arrangement using a nullor pair to satisfy the design criteria
ID = 1 mA
Note that after the supply V1 = 2.2V (or the current source I1 = 1.2 mA) is replaced with the
norator, the fixator-norator pair are removed from the circuit without inflecting any changes
to the circuit operation, i.e., still the current through the diode remains ID = 1 mA. Note that
in the case of replacing the norator with a current source I1 = 1.2 mA, the circuit operation is
not changed but the circuit structure (topology) can get modified. For instance, the 1 KΩ
resistor in series with the source becomes redundant and could be removed.
Now we are going to examine a third alternative. Let us assume that the voltage supply in
the original circuit, Fig.5, is already assigned for V1 = 2.5 V, but it is still necessary to have ID
= 1 mA, as a design requirement. This is the case that we need to decide on the value of a
“power-conducting” component. To proceed, let us assume the resistor R2 is the “power-
conducting” component that we need to adjust. We replace R2 with a norator, Fig.7, and
simulate the circuit. As usual, we replacing the norator with a very high gain controlled
source (VCVS), which is controlled by the fixator. From the simulated results we get a
voltage of V2 = 1.0 V across the norator and a current of I2 = 0.485 mA through it. This
simply means that the choice is to replace the norator with a resistor R2 = V2/I2 = 2.09 KΩ.
1
1KΩ 2
300Ω
3
4
1mA
V1 2.5 V R2
D
Fig. 7. The diode circuit arrangement using a nullor pair to satisfy the design criteria ID = 1 mA
In general, in a circuit a norator with computed voltage V1 and current I1 can be replaced
with i) a voltage source of V1 volts, ii) a current source of I1 amps, or iii) a component, such
as a resistor R = V1/I1.
Before we continue further we must realize that although our main use of fixator-norator
pairs here is for biasing purposes their application goes beyond this. The following simple
example goes one step further.
Example 2: Take the case of the diode circuit discussed in Example 1 (Fig. 5). There are two
design criteria to fulfill for this example: i) the power supply is specified with V1 = 3.3 V,
and the supply current is also fixed at I1 = 1.5 mA; ii) the diode current still remains fixed at
ID = 1 mA. Now, because we have two criteria to meet we must use two fixators, Fx(0, I1)
and Fx(0, ID), to keep the specified values fixed during the circuit biasing. The two fixators
need to match with two norators to make two fixator-norators pairs. Within several choices
we have we select two resistors R2 and R3 as “power-conducting” resistors to be
recalculated. Hence, we replace them with two norators, as depicted in Fig. 8. Now, we need
to decide which fixator is pairing which norator, as we have two choices to select; either (I1
with R2, ID with R3) or (I1 with R3, ID with R2). As it turns out, both choices work fine, except
the choice (I1 with R2, ID with R3) is preferred because it converges faster.
1KΩ R3
ID = 1mA
I1 = 1.5mA R2
D
V 1 = 3.3V
Fig. 8. The diode circuit arrangement using two nullor pairs to satisfy the design criteria of
I1 = 1.5 mA and ID = 1 mA.
After simulating the circuit with the fixator-norator pairs we can find all the current and
voltages for the circuit components including the two norators. With VR2 and IR2 found for
the norator R2, and VR3 and IR3 found for the norator R3 we get the actual resistor values as:
R2 = VR 2 / I R 2 = 1.8 / 0.5 = 3.6 KΩ
and
R3 = VR 3 / I R 3 = 1.08 / 1.0 = 1.08 K Ω
2.2 Rules governing fixators and norators in a circuit

Following the introducing of fixators and norators two major issues come up. First, how
shall we deal with fixators and norators in a circuit that contains other circuit components so
that the KVL and KCL are not violated? Second, for n fixators and n norators in a circuit,
how can we pair them for an effective performance? We discuss the first issue as the
properties of fixator-norator pairs, and leave the other issue for a later investigation. As we
already know fixators must pair with norators in order to have computational stability in a
circuit. We should also remember that a fixator represents a current source as well as a
voltage source combined; hence, it must adhere to both rules governing voltage sources and
current sources. For instance, a current source in series with a fixator may violate the KCL,
and a voltage source in parallel with a fixator may violate the KVL. In general, a cutset of
fixators with or without current sources may violate the KCL and a loop of fixators with or
without voltage sources may also violate the KVL. On the other hand, norators can be
considered a current source, a voltage source or a resistive component. As such they can
form a cutset with other current sources, and they can make loops with other voltage
sources with no restrictions. However, the problem with norators is independency, and it
becomes a serious issue when multiple numbers of norators are used in a circuit. For
example, two norators in series or in parallel do not violate the Kirchhoff’s laws but one
loses its independency. In general, a loop of all norators does not violate the KVL but we can
always remove (open) one from the loop without changing the circuit results. Similarly, a
node or cutset of all norators does not violate the KCL, but we can always short circuit one
norator in the group without changing the circuit performance. Other properties of fixator-
norator pairs are as follows [13]:
• The power consumed in a fixator Fx(V, I) is P = V*I; and the power is delivered by only
one of the sources, V (for Fx(V, I) ) or I (for Fx(I, V) ).
• A resistance R in series with a fixator Fx(V, I) is absorbed by the fixator and the fixator
becomes Fx(V1, I), where V1 = V + R*I. A resistance R in parallel with a fixator Fx(V, I) is
absorbed by the fixator and the fixator becomes Fx(V, I1) ; where I1 = I + V/R.
• A current source IS in parallel with a fixator Fx(V, I) is absorbed by the fixator and the
fixator becomes Fx(V, I1) , where I1 = I + IS.
• A voltage source VS in series with a fixator Fx(V, I) is absorbed by the fixator and the
fixator becomes Fx(V1, I) , where V1 = V + VS.
• Connecting a fixator Fx(V, 0) across a port with the port voltage V does not affect the
operation of the circuit; it only fixes the port voltage.
• Connecting a fixator Fx(0, I) in series with any component in a circuit with current I
does not affect the operation of the circuit; it only fixes the current going through that
component.
• In general, any two-terminal element in series with a fixator losses it’s current to the
fixator; and any two-terminal element in parallel with a fixator losses its voltage to the
fixator.
• A current source in series with a norator absorbs the norator; and a voltage source in
parallel with a norator absorbs the norator. In addition, a current source in parallel with
a norator is absorbed by the norator; and a voltage source in series with a norator is
absorbed by the norator.
• A resistance in series or in parallel with a norator is absorbed by the norator.
• A norator in series with a fixator Fx(V, I) becomes a current source I; and a norator in
parallel with a fixator Fx(V, I) becomes a voltage source V.
3. Circuit solutions containing fixator-norator pairs

3.1 Selective biasing
Selective biasing is a procedure that fixes part of or the entire operating regions of a
nonlinear component (say a transistor) during the circuit operation. To fix a biasing current,
I, in a port we can use a fixator Fx(0, I). Similarly, to fix a biasing voltage, V, across a port we
can use a fixator Fx(V, 0). However, as we discussed earlier, the use of fixators alone is not
permissible in a circuit; we must pair each with a norator. On the other hand, both fixators
Fx(0, I) and Fx(V, 0) carry zero power; hence, they alone cannot provide the biasing power
to the serving component they are attached to. This simply means that for each fixator that is
used to anchor certain biasing value in a circuit we need to provide the supplying power
and direct it to the component. Our solution is either i) find a location for the supply power
(voltage or current) and have the circuit find its magnitude, or ii) route the required power
from an existing power supply through a power-conducting component. As it turns out the
norators paring with the fixators can do both, provided that the pair are mutually sensitive,
i.e., change in one causes the other to change accordingly.
3.2 Sensitivity in fixator-norator pairs

In a circuit, each fixator can only work with a norator in a pair. A norator can be a source of
power, a consumer of power or a power-conducting component. This means a norator must
share power with a port that is anchored by a fixator. However, to satisfy this property the
following condition must hold. A fixator paring with a norator must be “sensitive” to the changes
happening in the norator and vice versa. This simply means that between a fixator and its
pairing norator there must be a feedback. We can think of a norator as a placeholder for a
DC supply or a power conductor in the circuit that must somehow “reach” to the
corresponding fixator. In a way, when we replace a transistor port with its fixator model, we
are getting a ticket, in exchange, to assign a DC source in the circuit wherever we like. This
is true provided that the DC source is “reachable” by the fixator.
Apparently, considering this property the choice of a norator pairing a fixator is not unique.
In a connected circuit a (voltage or current) change within a component normally causes
(voltage or current) changes throughout the circuit, although there are exceptions,
particularly in cases of controlled sources without feedback. Therefore, in pairing a fixator
with a norator we may have multiple numbers of choices to make; only avoiding those with
zero feedback. This brings us to another issue, mentioned earlier, that can be stated as
follows: for n fixators and n norators in a circuit how can we pair them for an effective design
performance? This is certainly a challenging problem and we do not intend to make a
comprehensive study on the subject here. What we would like to address is to find an
acceptable relationship between a fixator and a norator in a pair so that it helps to speed up
the biasing process in a circuit. The core issue in this relationship is the “sensitivity” issue
[14, 15].
Simulating fixator-norator pairs - Before we continue further on the sensitivity issue we need
to know how we can analyze or design a circuit that has fixator-norator pairs. Or simply,
how can we simulate a circuit that contains nullator-norator pairs? As far as we know the
existing circuit simulators, such as SPICE, do not have the means to directly handle the cases
[16, 17, 18]. Traditionally, transistors and high gain operational amplifiers have been used
for the purpose, and have done the job fairly successfully within acceptable accuracies [7, 9,
12]. However, in our case the situation is different. The fixator-norator pairs are only used
symbolically in a circuit in order to establish the design criteria we have adopted. They are
acting as catalyst and will be removed after the biasing is established in the circuit. Hence,
we can assume the pairs to be ideal in order to provide the component values accurately.
Within circuit components acceptable by a circuit simulator such as SPICE, controlled
sources with very high gains are the ideal candidates for the job. Now, the question is what
type of controlled sources must be used to simulate fixator-norator pairs? Evidently, if a
fixator is used to fix a specified current in a circuit component, the source replacing the
corresponding norator must be controlled by the voltage across the fixator. Similarly, if a
fixator is used to fix a specified voltage in the circuit, the source replacing the corresponding
norator must be controlled by the current through the fixator. Finally, the choice of the
controlled source itself can be arbitrary. For example, if the job is to find the supply voltage
VCC in response to a fixed current IB in the circuit then the controlled source is a voltage
controlled voltage source (VCVS). On the other hand, if in the previous case the supply
voltage VCC is already specified but we need to know how much current, IC, is conducted
from VCC, then we can use a voltage controlled current source (VCCS) to manage to find IC,
instead.
3.3 Paring fixators and norators in a circuit

As mentioned earlier, one of the conditions to pair a fixator with a norator is to have
feedback from the norator to the fixator. The purpose of this feedback is to harness the
growth of the voltage or current in the pairing norator. In fact, because we are simulating a
fixator-norator pair with a very high gain controlled source, the lack of feedback between
them can cause serious instability and cause blow up values; i.e., it can generate a very high
(negative or positive) voltage or current at the norator location or elsewhere in the circuit.
The only way to control this growth is to establish feedback between the two in the pair. The
following two examples show this feedback effects in dealing with fixator-norator pairs. A
detailed analysis on the subject is also given in the Appendix.
Example 3: - To see the feedback effect between a norator and its pairing fixator, let us
consider the biasing circuit of a simple common emitter BJT amplifier with feedback, shown
in Fig 9(a). In this example we assume the transistor operates linearly in its active region, so
that we can linearize the biasing circuit accordingly, as shown in Fig. 9(b). Table I provides
the component values for the linearized amplifier.
RC
RC RB Rf
Rf V1 V2
V CC
VCC
RB VBE
VBB
Q1 βI B RO
V BB IB
RBE
(a) (b)
Fig. 9. (a) The biasing circuit of a common emitter BJT amplifier with feedback; (b) linearized
biasing circuit for the amplifier;
VCC V VBB V VBE V RB KΩ RBE KΩ RO KΩ β

5 0.83 0.64 16.7 2 50 120
Table I. Component Values for the Linearized Amplifier
Now, in our first step we assume RC = 2 KΩ and do two experiments with this amplifier. In
the first experiment we remove the feedback resistance Rf from the circuit (no feedback),
and in the second experiment we assign Rf = 200 KΩ. Table II provide the simulation results
for the two experiments.
Rf KΩ V1 V V2 V IB μA
Open 0.66 2.42 10.36
200 0.668 1.526 9. 9
Table II. Simulation Results for the Linearized Amplifier
In the next step we take the case with feedback (Rf = 200 KΩ) and try to find the power-
conducting resistor RC for a fixed IB = 9.9 μA. Figure 10 shows the circuit constructed for this
situation. As shown the fixator Fx(VBE, IB) is paired with the norator RC. The simulation
results for this case provides VRC = 3.474104 V, and IRC = 1.737051 mA, where VRC and IRC
are the voltage across and the current through the norator RC. This brings us to RC = VRC /
IRC = 2 KΩ, as we expected.
Now we remove the feedback and repeat the circuit simulation with a fixed IB = 10.36 μA,
that is slightly different from the previous value. This time the results from the simulation
become surprisingly different. We get VRC = 53.3 V, and IRC = 0.2762 mA, which are
obviously not correct and unstable. Again, the reason for this instability and defective result
is due to the lack of feedback between the norator RC and the fixator Fx(VBE, IB). That is,
changes in the current through RC and the voltage across it is not “sensed” by the
controlling fixator Fx(VBE, IB).
RC
RB Rf
V1 V2
VCC
Fx(V BE , IB )
VBB
βIB RO
RBE
Fig. 10. The common emitter amplifier circuit with fixator-norator pair
Example 4: Consider a two stage BJT amplifier shown in Fig. 11(a). First we run the SPICE
simulation on the circuit with the component values as specified. The results, displayed
below, show the operating conditions for the two transistors.
VBE1 = 5.790227e-01
VCE1 = 7.225302e-01
VBE2 = 6.434079e-01
VCE2 = 2.382333e+00
IB1 = 4.405489e-07
WinSpice 1 ->
Next, we make the following changes in the circuit. i) Keep IB1 = 4.405489e-07 fixed, as it
resulted from the simulation. This is done by adding a fixator Fx(0, IB1) to the base of Q1. ii)
Remove RC2 = 5 KΩ and replace it with a pairing norator RC2, as depicted in Fig. 11(b). Next,
we simulate the new circuit with SPICE, and the following is the simulation results listed.
VBE1 = 5.790105e-01
VCE1 = 7.229068e-01
VBE2 = 6.434051e-01
VCE2 = 2.547247e+00
VRC2 = 2.013071e+00
IC2 = 3.867745e-04
RC2 = VRC2/ IC2 = 5.204765e+03
WinSpice 2 ->
VBB = 5 V VCC
V CC = 5 V
VBB = 5 V V CC
100 KΩ RC2
413 KΩ
413 KΩ 100 KΩ 5 KΩ 100 KΩ RC2
413 KΩ Rf1 Rf 2
Q2 Q2 Q2
Fx(0, I B1) Fx(0, I B1)
Q1 Q1 Q1
100 KΩ 10 KΩ 1 KΩ 100 KΩ 10 KΩ 1 KΩ 100 KΩ 10 KΩ 1 KΩ
(a) (b) (c)

Fig. 11. (a) Two stage BJT amplifier; (b) amplifier circuit with fixator-norator pair; (c)
amplifier circuit with feedback.
Note that the results in this case are just slightly different from that of the original circuit
(Fig. 11(a)), with difference of about 4%. Now, if we change the base current IB1 by a tiny
amount of 0.5 PPM (part per million) the responses take unrealistic values, as displayed in
the following SPICE responses. For example, the negative resistance RC2 cannot be correct.
This is of course expected because there is almost no feedback from the norator to the
fixator.
VBE1 = 5.789974e-01
VCE1 = 7.619999e-01
VBE2 = 6.398944e-01
VCE2 = 2.206873e+01
IB1 = 4.405491e-07
RC2 = -3.11725e+04
WinSpice 3 ->
In another try we modify the circuit by incorporating feedback into the circuit; one from the
output to the second stage and one from the second stage to the first stage, so that changes in
the norator RC2 reach the fixator Fx(0, IB1), as depicted in Fig. 11(c). The following SPICE
simulation shows the results after the base current IB1 is changed by 100 PPM. The results are
shown to be more reasonable, this time. For example, we notice that the power-conducting
resistance RC2 replacing the norator, is RC2 = 4.73 KΩ, changed only by about 5%. Again, due to
the feedback from the norator to the fixator, the circuit stability is back to normal now.
VCE2 = 5.802151e-01
VBE2 = 7.020994e-01
VCE1 = 6.432040e-01
VBE1 = 2.509425e+00
VRC2 = 2.054483e+00
IC2 = 4.343896e-04
RC2 = 4.729587e+03
WinSpice 4 ->
4. Component modeling with fixator

As stated in Property 1, a fixator can model a two-terminal device for a fixed biasing
condition (snapshot). For example, for a diode biased at (ID, VD) the fixator that replaces it is
Fx(ID, VD), where for positive ID and VD, the diode consumes power. However, because the
device is not locally biased (as discussed in the previous chapter) it must get power from the
supplies in the circuit, i.e., global biasing. Property 1 can also be extended to include devices
with multiple ports such as bipolar and MOS transistors. Here, for a fix component biasing
the original component can be removed from the circuit and be replaced with fixators that
mimic the same biasing; hence, imposing no change to the rest of the circuit. In general,
there are two types of fixator modeling for nonlinear devices. In the first type, called
complete modeling, the component is entirely removed from the circuit and replaced with one
or more fixators that represent the component with their intended biasing. In the second
method, called partial modeling, the component remains in the circuit but one or more
fixators keep its biasing fixed at the specified values. We will discuss each type separately.
4.1 Complete modeling of devices

As stated in Property 1 a two-terminal device (or network) can be modeled by a single
fixator. Likewise, for a multiple port device or network we can model each port separately
with a fixator [19]. Hence, an n-port device can be removed from a circuit and replaced by n
fixators with the same biasing currents and voltages without inflicting any changes within
the rest of the circuit. For example an MOS device can be completely modeled by using three
fixators. Figure 12 shows the complete fixator-models for nMOS and pMOS transistors,
neglecting the substrate effects. Similarly, Fig. 13 depicts the complete fixator-models for
npn and pnp transistors. Again, the models represent the devices with the same voltages
Fig. 12. Fixator models of nMOS and pMOS transistors when globally biased for VGS (VSG),
VDS (VSD), ID, and VBS (VSB). Both symbolic and expanded versions are shown.
Fig. 13. Fixator models of npn and pnp transistors when globally biased for VBE (VEB), VCE
(VEC), and IC.
and currents that they need to get biased to the specified Q-points. Note that two changes
are taking place in the circuit after the modeling is done: i) the resulted circuit becomes
linear, and ii) the circuit is DC-freezed at fixed biasing conditions. What it means is that,
addition (or removal) of any source or signal to the circuit may change signal conditions
within the circuit but no change in inflicted on the modeled transistors. Hence, circuits with
fixator-modeled components are not prepared for AC analysis.
4.2 Partial modeling of devices

In partial modeling the device remains biased in the circuit. In addition one or more fixators
are used to freeze one or more device (port) variables at given Q-points. We have already
used partial modeling in previous examples; for instance, in Example 4 we have freezed the
base current IB1 of Q1 during the entire biasing process. The advantage here is that we can
limit the number of fixators to the number of biasing specs provided for the design. Also, a
limited number of fixators makes it easier to match the number of fixators with that of
norators in the circuit. This helps to speed up the biasing procedure in a large circuit.
Another advantage in using partial modeling is that, in partial modeling the fixators are
only responsible to provide some critical biasing requirements and the rest are left to the
actual device, placed in the circuit, to adjust. For example, in a bipolar transistor only base
current IB and the collector-emitter voltage VCE might be considered critical; because with IB
given the transistor will decide on the value of VBE. Similarly, with the gain factor β known
the collector current IC is automatically established through the device characteristics.
However, the disadvantage here is that the circuit remains nonlinear.
In contrast with partial modeling, in complete modeling the transistors are totally absent
from the circuit and have been replaced with the fixators. This means the fixators are fully in
charge to accurately place the Q-points on the characteristic curves. This produces an extra
work for the designer, who, prior to the actual design, needs to run the transistors
individually and record the port values for the Q-points he/she has in mind. Then he/she
needs to place the port values into the fixators and exchange the fixators with the
corresponding transistors for the actual design.
The third option is to have a mixture of the two; i.e., some transistors get complete modeling
by fixators, while others are partially modeled. However, we are not allowed to have partial
modeling on a port of a transistor and apply complete modeling on another port of the same
transistor for obvious reasons.
Example 5: The objective in this example is to design a cascade CMOS amplifier, shown in
Fig. 14(a). The transistor sizes and the critical specs given for the design are listed in Table
III.
Devices W/L μm VGS V VDS V

M1 150/5 -2.0 -4.4
M2 50/5 1.4 2.4
Table III. The design Critical specs for the amplifier
VDD = 5 V
VDD = 5 V
Fx(VSG 1, 0)
R2 1 3 R2
M1 V out
Fx(VSD1, ID1)
AC vin Vout
AC 2 Fx(VDS2, ID2)
M2
VB 3
VI Fx(VGS 2, 0)
1 VI
4 4
R1 VB
2 R1
(a) (b)
Fig. 14. (a) A cascade CMOS amplifier; (b) the amplifier with complete fixator modeling of
the transistors.
To demonstrate different schemes, we are going to design the amplifier once using complete
modeling of both devices using fixators, and next we will use mixture of complete and
partial modeling.
Complete modeling – To perform the design by complete device modeling we first remove the
MOS transistors from the circuit and replace them with the fixator models shown in Fig. 12.
Note that the fixators carry the critical specs given in Table III. They also include the drain
currents ID1 = 289 μA and ID2 =30 μA that are computed when the transistors are
individually simulated using the design specs (refer to “Complete modeling of devices”).
Figure 14(b) shows the amplifier after the fixators have replaced the transistors. Note that
the circuit is linearized after the transistors are replaced with fixator-norator pairs. Another
important observation is the equality of the number of norators -- representing the
unspecified component values -- and fixators -- representing the design specs. After pairing
the fixators with the norators (identified by the same numbers in the figure) we represent
each pair by a high gain controlled source for simulation purposes. Table IV shows the
design values resulted from the SPICE simulation.
R1 R2 VGG VB
KΩ KΩ V V
1.9 66.3 3.0 2.0
Table IV. The Amplifier design Values for the Norators
Mixture modeling – In this design procedure we use the mixture of complete and partial
modeling devices by fixators. As displayed in Fig. 15(a) the transistor M1 is partially
modeled whereas the transistor M2 is complete modeled. Note that the number of fixator-
norator pairs is reduced to three but the circuit remains nonlinear. Similar to the previous
case, the fixators carry the critical specs for both transistors plus the drain currents ID1 and
ID2 for both transistors, as given in Table V. After pairing the fixators with the norators and
following the same routine as explained in the previous case we get the circuit simulated by
SPICE. The results from the simulation provide the component values as listed in Table VI.
W/L VGS VDS ID

Devices
μm V V μA
M1 150/5 - -4.37 289
M2 50/5 1.37 2.4 24.7
Table V. The design specs for the amplifier
R1 KΩ R2 KΩ VB V
2.0 80.0 2.0
Table VI. The Amplifier design Values for the Norators
VDD = 5 V V DD = 5 V
M1 80 KΩ
R2
AC vin M1
Vout
AC V out
vin
VI = 3 V
Fx(VDS 2, ID2) M2
VI = 3 V
Fx(VGS 2, 0) VB = 2 V
Fx(V SD1, I D1)
2 KΩ
VB
R1
(a) (b)
Fig. 15. (a) mixture of complete and partial modeling in the cascade CMOS amplifier; (b) the
amplifier with biasing design completed.
Finally, a complete design of the cascade amplifier is depicted in Fig. 15(b). Figure 16 shows
the transient response of the amplifier with a full output swing with negligible distortion.
Discussion - This study still needs to address two questions. First, what is the solution if the
DC supplies (mainly the voltage sources) so obtained are beyond the conventional and
standard values – such as 12V, 5V, 3.3V…? In the case of smaller voltage values techniques
such as voltage dividers can help to generate the right choices. For larger values, however,
the solution may get more complecated. An adjustment in the “power-conducting” resistors
is one possible solution. Because of the linearity involved, scaling is another simple tool to
adjust the circuit supplies to match the conventional supply values. The second question is:
Fig. 16. The transient response of the amplifier for a full output swing that displays
negligible distortion.
how to deal with the cases in which the number of fixators and norators are not equal?
Typically the number of fixators exceeds the number of norators. For example, in a three
stage amplifier with three driving transistors, we might need to have as many as six fixators;
whereas one power supply VCC or VDD, can be represented by only one norator. The good
news is that there are other components in the circuit that can be represented by norators. In
general, norators can represent three types of components, i) voltage sources, ii) current
sources/mirrors, or iii) power conducting devices, which are represented by resistors in
lumped analog circuits, and in the case of integrated circuits they can also be represented by
active loads. A second approach to achieve equality between the number of fixators and
norators is to limit the number of fixators to the number of critical biasing specs in a circuit.
In this approach we can identify the biasing design specs first; then classify the nonlinear
ports as critical and non-critical, where the critical ports carry the design specs. In the
second step, fixators are assigned only to those critical ports, which is necessary to keep
those design specs protected (fixed) during the biasing procedure. We will be covering this
subject in the next section in more detail.
4.3 Singularity and circuit divergence

Before leaving our discussion on the subject, there are issues that must be dealt with
regarding fixator-norator pairs. First, as mentioned earlier, the equality between the number
of fixators and norators is necessary to solve the circuit equations but it is not sufficient. The
problem is related to the independency of the circuit (KCL and KVL) equations. There is
always the possibility of inequality that may occur between the number of independent
fixators and nullators, even though they may have originally been set equal. The problem is
often caused by violating the rules related to fixators or nullators as discussed in Section 3.
Both fixators and norators are relatively new elements in circuit theory; and the rules of
engagement in KVL and KCL for these components are different from those of conventional
elements, such as resistors, voltage sources, and current sources. The following example
explains a similar case.
Example 6: Consider a simple nMOS circuit shown in Fig. 17(a). With the circuit values
specified the (SPICE) circuit simulator produces the biasing specs that are listed in Table VII.
Further test shows that these biasing values well respond to the AC operation. Next, we
keep the voltages VGS and VDS as two critical biasing values and fix them by using two
fixators, as depicted in Fig. 17(b). Next we need to assign two independent norators to
match the fixators. We first select two resistors RD and RS to be reevaluated for the given
design specs (VGS and VDS). To do this, we place the two norators in RD and RS locations.
After simulating the circuit with fixator-norator pairs, we get the resistors calculated as: RS =
997.6009 Ω, and RD = 9997.974 Ω, which are almost exactly as originally assigned for the
circuit.
W/L μm VGS V VDS V ID μA

50/5 1.966961 2.436567 233
Table VII. The Biasing specs for the NMOS Circuit

Next, we still keep VGS and VDS the same two critical biasing values and represent them by
the same two fixators, except, this time, we change the location of one norator switching
from RS to the supply voltage VDD, as shown in Fig. 17(c). We definitely have not violated
the KCL by creating a node of two norators but when we run the circuit we get unacceptable
responses. The SPICE simulation results produce: RD = -11411.8 Ω, and VDD = 10.42594 mV,
which both values are invalid! This is again because the two norators are in series and this
leave the voltage of the node common between the norators floating.
RD
VDD
RD
10 KΩ RD 5V
5V 200 KΩ M1
V DD M1
200 KΩ
200 KΩ M1 VGG RG
RG Fx(VDS , ID )
VGG 2.2 V
RG 2.2 V Fx(VDS, ID )
V GG Fx(VGS , 0)
2.2 V RS Fx(VGS , 0)
1 KΩ RS 1 KΩ RS
(a) (b) (c)

Fig. 17. (a) A simple nMOS circuit; (b) biasing design of the circuit with two fixator-norator
pairs; (c) the same as (b), except the norators form an illegal common node.
5. Circuit design for biasing

Design of high performance analog circuits can be a complex and often multi stage process –
noise, distortion, gain, bandwidth, biasing and so on. One approach to simplify the design
and cut loops and feedbacks between the stages is to use as much orthogonality as possible
[3]. This orthogonality is practiced in this chapter, between the circuit performances and the
biasing of the nonlinear components, or simply between AC and DC circuit designs. The
first task is to design for the circuit performances, mainly noise, signal power, and
bandwidth [3]. The biasing design typically comes last, except for possible circuit
modification that may require us to go back to the performance design, repeatedly. We only
deal with the biasing situation in this chapter. A full discussion on the performance design
and other related circuit design issues can be found in the literature [3].
Our approach to designing analog circuit biasing starts with a circuit topology (structure)
that is suitable for the design. There is, of course, no restriction on this topology and
structural modifications are acceptable during the design, as long as the final structure can
fulfill the design criteria. In case the circuit structure for the performance design is different
from that of the biasing design -- such as those with coupling or bypass capacitors -- we
restrict ourselves only to the bias (DC) handling structure. Our next move is to select regions
of operations for the transistors that fulfill the design requirements. This step may need
some individual testing of the transistors to make sure of their behavior in the circuit. In the
third step, and because the operating points for the transistors are specified, the components
can be replaced with their small signal linear models; and here is where the performance
(AC) design can start and continue until the design criteria are met. Following the
performance design we need to bias the components in the circuit so that each one operates
at the regions (Q-points) specified by the circuit performances. Algorithm 1 provides a
systematic procedure to do the circuit biasing using fixator-norator pairs.
5.1 Algorithm 1:
Preparation - Given the design specification, we begin with the performance design by
selecting a working circuit topology. We then choose the desired operating points for the
drivers3 that best meet the design requirements. Then we replace all the transistors with
their small signal linear models, to make the circuit entirely linear and ready for the AC
design. Note that as long as the linear models, representing locally biased devices, are not
altered the circuit topology as well as the component values (including the W/L ratios in
MOS transistors) can be changed for an optimal performance of the circuit. Finally, upon the
completion of the performance (AC) design, we can start the biasing design as follows:
1. Assign one fixator, carrying the biasing spec, to each “critical” transistor port. Also
assign one norator to a location in the circuit that is a candidate for i) a DC supply
voltage, b) a DC supply current, or iii) a power-conducting component such as a
resistor. Note: be sure that the number of fixators and norators match.
2. Pair each fixator with a norator in the circuit. This step is rather critical and needs to be
handled with care (see Sensitivity in fixator-norator pairs in Section 3). In general, any
pair must work (although may not be optimal), except for the cases where a fixator is
not sensitive to the changes in the norator.
3 In amplifiers drivers are the circuit transistors that are along the signal path and are directly involved
in circuit performance. Other non-driver transistors may exist in the circuit, such as those used in active
loads or current mirrors.
3. Assign one controlled source with high gain to each pair of fixator-norator so that the
fixator controls the source at the norator location. It is permissible to assume an ideal
controlled source with very high gain; this is because these controlled sources will
disappear afterwards, leaving the actual DC supplies or power-conducting components
in place. A controlled source can be one of the four types: VCVS, VCCS, CCVS, or
CCCS. The choice depends on the individual situation as follows:
a. For a fixator keeping a specified current fixed the controlled source is either VCVS,
or VCCS.
b. For a fixator keeping a specified voltage fixed the controlled source is either CCVS,
or CCCS.
c. For a norator holding the place for a voltage supply the best choice is either a
VCVS, or CCVS.
d. For a norator holding the place for a current (mirror) supply the best choice is
either a VCCS, or CCCS.
e. For a norator holding the place for a power-conducting component any of the four
will work.
4. Solve the linear circuit equations as prepared. The DC solution (simulation) provides
the currents and voltages for the circuit components including those of the norators that
are represented by the controlled sources.
5. Remove all the controlled sources from the circuit and replace each with an appropriate
voltage supply, Vj, a current supply, Ij, or a resistor Rj = Vj,/Ij; where Vj and Ij are the
voltage and current found for that controlled source (norator).
This concludes the biasing design algorithm.
6. Design examples
The following examples provide a systematic procedure for biasing design of analog circuits
using the new approach, given in Algorithm 1.
Example 7: This example presents a negative feedback BJT amplifier; fully explained in
reference [3]. Figure 18 shows a simplified AC schematic of the amplifier after it has gone
through the performance design in three areas: noise reduction, clipping/distortion
reduction, and high loop-gain-poles-product4. To perform the biasing of the circuit we need
to first specify the values of the DC supplies and their locations in the circuit. Next, we need
to select the operating points for the transistors so that they can fulfill the design specs. For
the actual power supplies, we choose two DC sources of 4V and - 4V, as assigned in the
reference [3]. Next we need to select DC power-conducting components that provide biasing
power to the drivers. However, there are certain performance design criteria that must be
given priority in this selection so that the biasing is smoothly aligned with the rest of the
design. These major performance design criteria are as follows:
• The emitters of Q1 and Q2 must be driven by a high impedance current source, Ie.
• The base of Q2 must be driven by a low impedance voltage source, Vb2.
• The collector of Q1 can be driven directly by VCC.
4 For details please refer to Chapter 10 in [3].

Fig. 18. A three stage amplifier topology after going through the performance, AC, design [3].
• The collector of both Q2 and Q3 must be driven by high impedance current sources IS2
and IS3, to maximize the gain.
• The base current of Q1 can be provided through a feedback resistor Rf5.
For this particular design we choose the collector-emitter voltages of two transistors Q2 and
Q3 (vce2 and vce3) as the “critical” design values. The collector-emitter voltage of Q1 (Vce1) is
considered “non-critical” because it is directly connected to VCC. Also all three collector
currents ic1, ic2, and ic3 are considered “critical” for this design. Table VIII, columns 1 and 2,
provides all five critical values for the selected operating points; also all five fixators that
keep these critical values fixed during the design are listed. Column 3 shows the matching
norators that are later replaced with computed components: a voltage source, three current
sources and one feedback resistor (DC power-conducting component). Figure 19 is extracted
from Fig. 18 after the fixator-norator pairs, specified in Table VIII, are added to the circuit.
Fixator
Critical specs Norator representations
representations
IC1 = 0.1 mA Fx(0, 0.1mA) RF
VCE2 = 0.67 V Fx(0.67V, 0) VB2
IC2 = 0.5 mA Fx(0, 0.5mA) IE
VCE3 = 2.2 V Fx(2.2V, 0) IS3
IC3 = 3.6 mA Fx(0, 3.6mA) IS2
Table VIII. Bias design specs and fixator-norators.
Below is a piece of the WinSPICE program code simulating the DC biasing of the amplifier.
Note that each fixator-norator pair is simulated by a very high gain controlled source
(namely VCVS, CCVS, VCCS, CCCS, and VCCS in sequence).
ic1 2 a DC 1.0e-04
e1 4 51 a 2 1000MEG
vce2 c 7 DC 0.67
hb2 Vb2 0 vce2 1000MEG
ic2 3 c DC 0.5m
ge 7 11 3 c 1000MEG
vce3 e 0 DC 2.2
fc3 21 4 vce3 1000MEG
ic3 4 e DC 3.6m
gc2 12 3 4 e 1000MEG
5 The resistance Rf is in the bias loop and part of a required AC filter as well, see [3].
Fig. 19. The three stage amplifier with fixator-norator pairs indicating the biasing design
specs.
The results from the WinSPICE simulation are shown below and listed in Table IX.
TEMP=27 deg C
DC analysis ... 100%
(v(4)-v(5))/vf#branch = 1.528640e+06
vb2 = 6.770538e-01
ve#branch = 6.068945e-04
vs3#branch = 3.601024e-03
vs2#branch = 5.229127e-04
WinSpice 6 ->
RF = 1.53 MEGΩ
VB2 = 0.677 V
IE = 0.607 mA
IS3 = 3.601 mA
IS2 = 0.523 mA
Table IX. Component Values for the Specified Biasing.
Finally, we remove the controlled sources (representing the fixator-norator pairs) from the
circuit and replace each with the computed voltage source, current sources, and one
feedback resistance. The final amplifier so designed is depicted in Fig. 206. As expected, the
resulted DC sourcing matches with those in [3].
6 For simplicity the current sources are presented in their ideal form in Fig. 12. A detailed current
sourcing and mirroring can be found in [3].
Fig. 20. The three stage amplifier with complete biasing.

Example 8: The purpose here is to complete the biasing design of a two stage CMOS differential
amplifier shown in Fig. 21. The design criteria set for this amplifier requires that both the input
offset voltage VG and the output offset voltage VO, remain stable at 0.5V. Hence, we have two
design criteria to fulfill and need two fixators to fix VIN = 0.5 V and VOUT = 0.5 V. The circuit
with fixator (or rather nullator)-norator pairs is shown in Fig. 22. Next, because the supply
voltage VDD is already specified for the design at VDD = 1V, we need to focus on finding the
two current sources (mirrors), as power-conducting components. So we can replace the current
sources with two norators and simulate the circuit (Fig. 22). The SPICE simulation finds the
currents flowing through the norators as I1 = 1.26 μA and I2 = 21 μA. This means we can
replace the norators with two current sources at the designated locations, as they were before.
VDD= 1 V
I1 I2
M1 M2 VO
VG
M3
500 KΩ 500 KΩ
Fig. 21. A two stage CMOS differential amplifier.

VDD= 1V
I1 I2
M1 M2 VO= 0.5V
VG= 0.5V
M3
500 KΩ 500 KΩ
Fig. 22. Design stages of a CMOS differential amplifier

Note, in this example, that the choice of two current sources replacing the norators is only
one option. Here the source resistance for each current source happens to be infinity, but
this is not a requirement. In fact, any component, or combination of components as a two-
terminal circuit, is permissible to replace the norator, say norator I1, provided that the DC
current through the two-terminal amounts to the current I1, and the voltage across the two-
terminal is the same as that obtained for the norator I1, in the circuit simulation. For
instance, let us take the following case: let us assume that in doing the AC performance
design of the amplifier, we have come up with a resistance of RI1 necessary to place it at the
location of the current source I1. Now, to make this resistance also available for the DC
biasing, all we need to do is to add RI1 in parallel with the current source I1. The only
correction we need to make is to reduce the current in the source from I1 to I’1; where I’1 = I1
– IRI1, and IRI1 is the DC current that is conducted through the resistance RI1. In short, the
overall branch current must stay fixed at I1. The significance of this issue is in providing link
between design of DC and AC in analog circuits. It simply opens a new procedure in the
design where both DC and AC design are pursued in combinations, but they may differ in
some component values. This is more apparent in design of integrated circuits, where the
roles of active loads and current mirrors are different from DC biasing to AC signal loading.
However, this is a topic of further investigation.
Let us get back into our design. Now that we have substituted for the norators, the design is
complete, after removing the fixator-norator pairs from the circuit. Next, to perform the
transient operation, we apply 0.5V DC supply to the gate of M2 and run the amplifier with
an input signal Vi = 500 + 5*sinωt mV applied to the gate in M1. As shown in Fig. 23, the
generated output voltage Vout,pp = 0.8 V still remains undistorted. Note that the output offset
voltage stays at 0.5V, as expected.
Example 9: The purpose of this example is to complete the design of a CMOS differential
amplifier with a buffer stage. Figure 24(a) depicts the circuit configuration. As shown, the
performance design of the amplifier is completed giving the transistor sizes listed in
Table X.
Fig. 23. The undistorted output waveform for the CMOS differential amplifier
VDD = 2.5V VDD = 2.5V
100 KΩ 100 KΩ 100 KΩ 100 KΩ
M3 M3
M1 M2 M1 M2 Vout
Vout
Fx(0 , 0)
M4 Id M4
Is Id
Vb Vb
Fx(0 , 20 μA)
-VSS = -2.5V
- VSS = -2.5V
(a) (b)
Fig. 24. (a) A CMOS differential amplifier with buffer stage; (b) biasing design procedure for
the amplifier
To complete the biasing design we need to do the following: i) specify the biasing voltage Vb
so that we can get a current sink of IS = 20 μA, and ii) specify the current mirror ID in the
buffer stage so that the output offset voltage Vout = 0. Figure 24(b) shows the biasing design
procedure, where two fixator-norator pairs are used for IS and Vout, and Vb and Id. Again,
because of the two fixator-norator pairs used in this example the problem is to find the best
pairing situation among the four so that it provides the fastest and most accurate solution.
Within the two existing choices it turns out that the fixator Fx(0, 20mA) and the norator Vb
make a good match; likewise, Fx(0, 0) and the norator Id also produce good results. Again,
the fixator-norator pairs are replaced with two high gain controlled sources, prepared for
circuit simulation. Following the SPICE simulation of the circuit the two unknown values
are computed as: Vb = -1.56V, and Id = 48 μA. Next, the amplifier circuit is completed by
making Vb = -1.56V, and Id = 48 μA in Fig. 24(a). Because the two voltage supplies VDD =
2.5V and -VSS = -2.5V are available in this design we can simply generate Vb = -1.56V
through a voltage referencing (divider) circuit; and for Id = 48 μA a current mirror circuit can
be put in place. This completes the biasing design of the amplifier.
M1 M2 M3 M4
W/L - μm W/L - μm W/L - μm W/L - μm
20/2 20/2 200/2 40/2
Table X. The CMOS Transistor Sizes
7
6.1 Some challenges and potential impacts of the proposed methodology
We believe the proposed methodology can have a profound impact on the research and
development of techniques for designing analog circuits. It provides circuit designers a
collection of choices and short cuts to create better designs in shorter time periods. The
design tools and procedures introduced in this and a previous chapter are new and
expandable. The proposed tools can be interpreted as the beginning of a new methodology
in analog circuit designs. Through this methodology, one can see the challenges that exist
for more direct, faster and cost effective designs of otherwise complex analog circuits. What
it brings to a designer is simplicity, time and management. It brings simplicity because no
matter how complex the circuit might be, it can be partitioned and linearized. The designer
can save time because by linearization he/she has entirely removed the nonlinear iterations
from the analysis. The designer is in full control of the management of the design because
he/she is not faced with a complex network of mixed linear and nonlinear components, but
individual transistors to assume the right operating points for. By a mixture of global and
local biasing (see the previous chapter) a skilled designer can maneuver around and find a
selective path for gradually applying DC supplies in the circuit, aiming at a smooth and fast
converging biasing. Finally, because of the exact and selective environment that is provided
by this methodology, the designer is capable of accurately calculating for possible
distortions, noise, bandwidth, power and other design attributes. Last but not least, this
study introduces new missions and roles for some virtual components: nullator, fixator and
norators, that have not been practiced in the past.
Here are some of the evidences for the challenges discussed:
• No matter how complex, the nonlinearity is entirely removed and replaced with the
linearized equivalent circuits for biasing.
• If selected, each transistor (nonlinear component) is individually biased to the selective
and desirable operating points without affecting the rest of the circuit.
• Local biasing minimizes the DC power consumption in the circuit. In general, the
methodology can be used to monitor the DC power consumption in a circuit and direct
it so that one can reduce the power effectively.
• Through the use of fixator-norator pairs a circuit designer can specify and fix the design
criteria (pertinent to the biasing) all throughout the design. The pair also serves to
locate and find values for voltage/current supplies or components that conduct the DC
power.
• Although fixator-norator pairs, as non realistic circuit components, are used in the
biasing design, they only act as a catalyst and removed after the proper components are
substituted.
A mixture of the traditional and the new method is also possible for the design; which is in
fact recommended for circuit modification and debugging.
7 This discussion was suggested by one of the reviewers.

7. Appendix
Feedback effect in fixator-norator pairs: - In pairing fixators with norators in a circuit, one of the
essential conditions is to have mutual feedback between the two. In one direction, it is the
fixator that generates the current and voltage values of the pairing norator; but in the other
direction it is the feedback from the norator to the fixator that controls the event and puts
harness into the growth of the voltage or the current in the pairing norator. The following
analysis is an attempt to show this effect through an example by using feedback theory.
Analysis - To see the feedback effect between a norator and its pairing fixator, let us consider
the biasing circuit of a simple common emitter BJT amplifier with feedback, shown in Fig
A1(a). With the assumption that the transistor operates close to its linear regions on the
characteristic curves we can linearize the biasing circuit according to Fig. A1(b). Next, we
can even simplify the circuit more as represented in Fig. A1(c); where we can easily find the
circuit values as
VBB VBE
I1 = + ,
RB RBE
V1 = RBE I B + VBE ,
RB RBE
Rin = ,
RB + RBE
β
Gm = , (1)
RBE
VCC
I2 = ,
RC
ICE = GmVBE , and
RC RO
Rout =
RC + RO
RC Rf
RC V1 V2
RB Rf
Rf
V CC
VCC
I1
RB VBE RIN
VBB Gm V1 ICE I2
Q1 βI B ROUT
V BB IB RO
RBE
(a) (b) (c)
Fig. A1. (a) The biasing circuit of a common emitter BJT amplifier with feedback; (b)
linearized biasing circuit for the amplifier; (c) reduced equivalent circuit.
Now, we can start writing the node equations for the circuit (Fig. A1(c)), and after solving
the equations we get
I 2 = (Gout Gin / G f + Gout + Gin + Gm ) V1 − (Gout / G f + 1) I 1 − ICE
Gi = 1 / Ri for all i. (2)
We substitute from Eqs. (1) into Eq. (2), and after proper simplification we get
VB = RGC VCC − (1 − RGBE (Gout Gin / G f + β + 1))VBE + RGB (Gout / G f + 1) VBB (3)
Where
VB = RBE I B , R = 1 / G ,
and
G = Gout Gin / G f + Gout + Gin + Gm . (4)
The assumption is that the supply voltage VBB is already given and stays constant; also VBE
stays constant. Suppose the design requires having IB stay fixed at its specified value. Then
according to Eq. (3) the amount of feedback voltage that VCC can contribute to the base
voltage of the transistor is.
VB' = RGC VCC (5)
Equation (5) provides the feedback effect from VCC (the norator) to the transistor base where
the fixator is located. Now, to complete the loop we need to get the feed forward effect, i.e.,
how the fixator in the transistor base generates VCC. As mentioned earlier, for simulation
purposes we can use a very high gain controlled source (VCVS, in this case) to handle the
case. Hence, for a gain of Av we can write the relationship as
VCC = AvVB (6)
This is how the norator voltage (VCC) is generated due to the variation across the fixator IB,
i.e. VB. Now, to get the feedback part strait we first substitute for R from Eq. (4) into Eq. (5).
Next, we simplify Eq. (5), for very high feedback resistance Rf, to get
ROUT RIN
VB' = VCC = FVCC (7)
R f RC
The variable F is the feedback coefficient. From the feedback control systems we know that,
for high gain AV, where F*Av >> 1, the closed loop gain AC can be approximated as
1 R f RC
AC = = (8)
F ROUT RIN
As Eq. (8) indicates, AC will be limited for limited values of the feedback resistance Rf. On
the other hand, if Rf grows high the system become more unstable; eventually with broken
feedback a fixator fails to generate the required DC supply (VCC) as a substitution for the
pairing norator.
8. Acknowledgment
The author would like to thank Ms. Golnaz Hashemian for her valuable suggestions and
editing the chapter.
9. References
[1] A.S. Sedra, and K.C. Smith, Microelectronic Circuit 6th ed. Oxford University Press, 2010.
[2] R.C.Jaeger, and T.N. Blalock, Microelectronic Circuit Design 4th ed. Mc Graw-Hill Higher
Education, 2010.
[3] C. J. Verhoeven, Arie van Staveren, G. L. E. Monna, M. H. L. Kouwenhoven, E. Yildiz,
Structured Electronic Design: Negative-Feedback Amplifiers, Kluwer Academic
Publishers, 2003.
[4] R. Hashemian, “Local Biasing and the Use of Nullator-Norator Pairs in Analog Circuits
Designs,” VLSI Design, vol. 2010, Article ID 297083, 12 pages, 2010.
doi:10.1155/2010/297083.
http://www.hindawi.com/journals/vlsi/2010/297083.html
[5] ___, "Analog Circuit Design with Linearized DC Biasing ", Proceedings of the 2006 IEEE
Intern. Conf. on Electro/Information Technology, Michigan State University;
Lancing, MI, May 7– 10, 2006.
[6] ___, " Designing Analog Circuits with Reduced Biasing Powe", Proceedings of the 13th
IEEE International Conference on Electronics, Circuits and Systems, Nice, France
Dec. 10– 13, 2006
[7] R. Kumar, and R. Senani, “Bibliography on Nullor and Their Applications in Circuit
Analysis, Synthesis and Design”, Analog Integrated Circuit and Signal Processing,
Kluwer Academic Pub, 2002.
[8] H. Schmid, “Approximating the universal active element”, IEEE Trans. on Cir. and Sys.
II, Volume 47, Issue 11, Nov 2000, pp 1160 – 1169.
[9] E. Tlelo-Cuautle, M.A. Duarte-Villasenor, C.A. Reyes-Garcia, M. Fakhfakh, M. Loulou, C.
Sanchez-Lopez, and G. Reyes-Salgado, “Designing VFs by applying genetic
algorithms from nullator-based descriptions”, ECCTD 2007, 18th European
Conference on Circuit Theory and Design, Volume, Issue , 27-30 Aug. 2007, pp 555
– 558.
[10] E. Teleo-Cuautle, L.A. Sarmiento-Reyes, “Biasing analog circuits using the nullor
concept”, Southwest Symp. on Mixed-Signal Design, 2000.
[11] D.G. Haigh, and P.M. Radmore, “Admittance Matrix Models for the Nullor Using Limit
Variables and Their Application to Circuit Design”, IEEE Transactions on Circuits
and Systems I: Regular Papers, Volume 53, Issue 10, Oct. 2006, pp 2214 – 2223.
[12] Claudio Beccari "Transmission zeros", Departimento di Electronica, Turin Institute of
Technology, Turino, Italy; December 6, 2001.
[13] D.G. Haigh, T.J.W. Clarke, and P.M. Radmore, “Symbolic Framework for Linear Active
Circuits Based on Port Equivalence Using Limit Variables”, IEEE Transactions on
Circuits and Systems I: Regular Papers, Volume 53, Issue 9, Sept. 2006, pp 2011 –
2024.
[14] T.L. Pillage, R.A. Rohrer, C. Visweswariah, Electronic Circuit & System Simulation
Methods, McGraw-Hill, Inc., 1995.
[15] J. Vlach and K. Singhal, computer methods for circuit analysis and design, Van Nostrand
Reinhold Electrical/Computer Science and Engineering Series, 1983.
[16] L.W. Nagel, "SPICE2, A computer program to simulate semiconductor circuits," Univ. of
California, Berkeley, CA, Memorandum no. ERL-M520, 1975.
[17] Mike Smith, "WinSpice3 User’s Manual, v1.05.08",
http://www.ousetech.co.uk/winspice2/, May 2006.
[18] R. Jacob. Baker, CMOS, Circuit Design, Layout, and Simulation, 2nd ed. IEEE Press, Wiley
Interscience, 2008, pp. 613 – 823.
[19] R. Hashemian, “Source Allocation Based on Design Criteria in Analog Circuits”,
Proceedings of the 2010 IEEE International Midwest Symposium On Circuits And
Systems, Seattle, WA, August 1 - 4, 2010.
3
New Port Modeling and

Local Biasing of Analog Circuits
Reza Hashemian
Northern Illinois University
United States
1. Introduction
In today’s high-speed technology, analog and mixed signal integrated circuit technology has
an important and decisive place in communication and signal processing. In particular with
CMOS technology rapidly embracing the field, analog circuit design has become more
challenging than ever [1–8]. Other developments in the technology such as lower supply
voltages, low-power consumption, performance complexity, and high transistor counts have
substantially increased the demand for new design methodologies and techniques.
A major difficulty in dealing with analog circuits is the DC biasing – getting desirable
operating points with quick convergence; and the problem is getting worse with the
advancement of the technology which is due to increase in size and circuit complexity. The
analysis may even lead to multiple DC operating points, or instability in the operating
points caused by positive feedbacks [9, 10]. In SPICE circuit simulator [3, 4], for examples,
methods such as Newton-Raphson iteration techniques are employed to deal with
nonlinearities; the major difficulty sometime is to get the circuit to converge within a limited
number of iterations. Schemes such as adding minimum conductance (GMIN), shunt
resistors, changing the tolerance values for the results, and supply stepping are typically
adopted in the simulator to make the convergence possible.
There are several causes for these problems. A major difficulty arises from the fact that, in
traditional methods, an analog circuit is usually analyzed and simulated as a whole – with
the linear and nonlinear components all together. Usually a poor selection of initial
conditions or adopting large and unregulated steps of iterations cause instability or it may
even cause the circuit to diverge. Another difficulty can result from a fixed circuit topology
with fixed DC supplies throughout the biasing procedure. With such a pre-setting
conditions the operating points are naturally found through long and timely iterations. All
this adds up to the design burden and timely process. We need a more guided design
procedure; a procedure that helps a designer to go through a top-down and piece-by-piece
design strategy.
The objective in this chapter is to introduce such a guided design procedure for biasing. The
purpose is to use a “divide and concur” strategy for a better handling the case. This strategy
separates linear and nonlinear portions of an analog circuit, and takes more control of the
nonlinear portions. This separation of portions (components) within the circuit is
accomplished by introducing a new port modeling that nullifies the ports of nonlinear
devices. This in turn leads to a new biasing technique for nonlinear components. The result
is to replace the regular DC supplies with alternative supplies that are directly attached to
the nonlinear devices. It is shown that a unique and very powerful additivity property takes
charge in performing this component biasing operation. Another useful property using this
strategy is the removal of nonlinearity in the biasing design. This is done because being
locally biased the nonlinear components can be replaced with their linear models operating
at those Q-points; hence making the biasing design of the circuit entirely linear. However,
one major drawback that exists in using local biasing is the sheer number of DC supplies
needed in local biasing. There are source transformations methods that help to reduce these
supplies and possibly end up with the regular circuit supplies. As discussed in the next
chapter, one direct and simple technique is introduced that removes the distributed local
biasing sources all together and replaces them with the regular supplies, such as VDD or VCC,
in a single step. Finally, because the proposed strategy offers a complete isolation of
individual nonlinear devices (transistors), it makes it possible to modify, adjust and tune the
circuit locally without disturbing the rest of the circuit.
Another important outcome of this methodology is that it provides an ability to control and
reduce power consumption in a circuit. It is shown that by local biasing nonlinear devices
we actually reduce the DC power to its minimum – just enough to get the devices biased. In
other words, by locally biasing we are totally cutting off the DC power from entering the
linear elements in the circuit.
1.1 Thevenin and Norton equivalent circuits

Thevenin and Norton equivalent circuits, also known as Thevenin and Norton models, are
two known conventional models that explain terminal behavior of linear circuits. They are
crucial for circuit analysis, replacing a terminal port with a source and an impedance [11].
Both Thevenin and Norton models are very useful circuit simplification techniques that are
often used in order to concentrate only on the terminal behavior of certain linear portion of a
circuit that normally supplies power or signal to the rest of the circuit. In general, both
models are used in different circuit analysis and applications such as, in source
transformation, DC analysis, Transform (frequency or phasor and s-domain) analysis [1, 6].
Here we limit our discussion to DC analysis only. Figure 1(a) represents a two terminal
linear resistive circuit, N, with both independent and internally-dependent sources; and
Figs. 1(b) and (c) are the Thevenin and Norton models of N, respectively. Where, VTh
represents the open-circuit voltage and IN is the short-circuit port current in the original
circuit. For Req, we can either remove all independent sources from the circuit and calculate
the port resistance, or alternatively get Req from Eq.(1).
VTh
Req = (1)
IN
Example 1: Figure 2(a) shows a simplified small signal equivalent circuit of a single stage BJT
amplifier with the virtual biasing supplies included. The Thevenin model for the amplifier
looking from the output port is given in Fig. 2(b). Figure 2(c) shows the port’s characteristic
curve (line), indicating the circuit linearity. The figure also shows how we can move from
the Thevenin model, specified by point T(2.5V, 0), to the Norton model, given as point N(0,
1.25mA) on the characteristic line.
New Port Modeling and Local Biasing of Analog Circuits 49
N x Req x
x
A resistive circuit
with independent VTh IN Req
& self -dep.
sources y y y
(a) (b) (c)
Fig.
1. (a) A two terminal linear resistive circuit; (b) Thevenin, and (c) Norton equivalent circuit.
Ib x Ix
25 K Ω 2 KΩ
N 1.25 mA
2 KΩ x
124*Ib
0.5 V
2.5 V
200 Ω 5 V
y
y T vx
0,0 2.5 V
(a) (b) (c )
Fig. 2. (a) A simplified small signal equivalent circuit of a single stage BJT amplifier; (b) the
Thevenin equivalent circuit; (c) the port’s characteristic curve, indicating the linearity.
However, despite their simplicity, there is a rigidity involved in port representation by
either the Thevenin or Norton equivalent circuits. As indicated in Fig. 2(c), Thevenin or
Norton model occupy only one point on the characteristic line, where the line meets one of
the axis. This characteristic line also serves as a load line in some biasing situations, where it
identifies the port’s operating point (Q-point) when the two characteristic curves from both
sides of the port cross. The limitation for Thevenin or Norton model is that it represents only
the “sourcing” network with no information given about the “target” network, unless the
two are connected and the analysis is done with the combined circuit. This of course fits
with most circuit applications where all we need is a simplified two terminal linear circuit
that gets connected with the target circuit for the rest of the process; but again, we perform
the analysis only when the two are combined. The circuit complexity created this way may
not be so evident for a single port connection, but for multiple ports the complexity may get
quite significant. There are other cases where circuits in both sides of a port need to get
engaged in some (sources or components) exchanges; hence a more dynamic port modeling
may be needed. Examples can be found in source transformation, noise-source modeling,
and power transport cases. Port nullification is another example that uses Hybrid modeling,
as discussed next.
2. Hybrid equivalent circuit

A Hybrid equivalent circuit, or simply an H~-model, of a two-terminal network is a
generalized version of Thevenin or Norton equivalent circuit; for resistive circuits it consists
of a voltage source, a current source and an equivalent resistance, Req, which is identical th
that in the Thevenin or Norton model. Apparently here one source, VH or IH, can be selected
arbitrarily and the other source is found through Eq(2).
VH
IH = IN − or VH = VTh − I H Req (2)
Req
Note that, like the Thevenin or Norton models, here only two measurements are needed to
get all H~-model parameters. For example, for a selective value of IH and two measurements
of VTh and IN, Eqs. (1) and (2) can be used to obtain Req and VH for the model. Now, consider
two networks N1 and N2 connected through port j(Vj, Ij), as shown in Fig. 3. There are two
types of H~-models for the linear two terminal network N1. Type 1 H~-model is shown in
Fig. 4(a). To find this model first open circuite the port where Ij = 0. By referring to Fig. 4(a)
and considering Eq.(2) we get
Vj = VH + I H Req = VTh (3)
Next, short circuit the port terminals to get Vj = 0, and find
I j = I H + VH / Req = I N (4)
In Type 2 H~-model, however, the sources remain the same as in Type 1, but instead of
calculating the equivalent resistance Req we let N1 remain unaltered except all its DC power
supplies are removed, as shown in Fig. 4(b). The term ”DC power removed” means that all
Ij
N1
A res is tiv e c irc uit
Vj N2
with independent
& s elf -dep.
s ourc es
Fig. 3. Two networks N1 and N2 connected through a port j(Vj, Ij).
Ix
VH VH IN
Ij N1 Ij
IH H
Req N2 N2
IH Vj IH Vj
No DC vx
Power 0, 0 VH V Th
(a) (b) (c)
Fig. 4. A two-terminal Hybrid equivalent circuit for N1; (a) Type 1 representation; (b) Type 2
representation; (c) the location on the port’s characteristic curve.
independent DC supplies are removed from N1, including charges on the capacitors and
currents through the inductors. Type 2 H~-model is useful in a number of applications, such
as moving the DC sources in a circuit to its port terminals without disturbing the internal
structure (topology) of the network.
Note that, because of having two sources instead of one, an H~-model represents an axis of
freedom that acts as a tool in dynamic modeling of a port. As indicated in Fig. 4(c), an H~-
model covers a full and continuous range of equivalent circuits for a two-terminal network.
It is evident from Eq. (2) and Fig. 4(c) that both the Thevenin and Norton models are two
special cases of an H~-model.
Example 2: Figure 5(a) shows the same circuit given in Example 1 (Fig. 2(a)), except this time
the x-y port is connected to a load RL. Here we would like to have: i) an H~-model for the
two terminal circuit, on the left of x-y, so that the power consumption on both sides of the
port are equal; and ii) modify the H~-model in part i) so that the power consumption in the
two terminal circuit (the left of x-y) becomes zero.
Ib x VH
x Ij
25 K Ω
2 KΩ
124*Ib
0. 5 K Ω 0.5 KΩ
0.5 V RL 2 KΩ
IH Vj
RL
200 Ω 5 V
y y
(b)
(a)
Fig. 5. (a) A simplified small signal equivalent circuit of a single stage BJT amplifier with
load; (b) an H~-model of the amplifier.
Solution: We first find an H~-model representation for the two-terminal circuit as depicted
in Fig. 5(b), with the source values, VH and IH, unspecified. Second, to make the power
consumption on both sides of port j equal we need to have
RL I 2j = Req ( I H − I j )2
By using Eq.(2), and knowing that VTh = 2.5V and Req = 2KΩ we get
I j = 1 mA , I H = 1.5 mA , VH = − 0.5 V , and the power consumed for each side is
W j = 0.5 mW .
For part ii), because the situation for the load RL is not changed we still have I j = 1 mA ,
Vj = 0.5 V , and W j = 0.5 mW . Now, to make the power consumption to the left of X – Y
zero we must have Req ( I H − I j )2 = 0 ; or simply I H = I j = 1 mA , and as a result
VH = Vj = 0.5 V . This concludes the solution with the fact that in the part ii) the total power
consumption is reduced to half, i.e., from 1.0 mW to 0.5 mW.
2.1 Universality
Universality is an important property of an H~-model. H~-models can be accurately applied
to all possible cases of linear two-terminal networks, regardless of the port impedances;
whereas both Thevenin and Norton equivalent circuits lose their sensitivity in some specific
cases where port impedances take extreme low or extreme high values. For example,
consider measuring the Thevenin (open circuit) voltage of a two terminal network N1 that
has the equivalent resistance of Req = 2 MΩ. Suppose the measuring voltmeter has the input
impedance of RM = 20 MΩ and the measured open circuit voltage displayed is VM = 3V.
Apparently selecting VTh = VM = 3V as the Thevenin voltage for the port carries an error of
10%. Whereas, an H~-model with VH = VM = 3V and IH = IM = 136nA represents an exact H~-
model for the port. Note that there is no need for any extra measurement to find IM, because
we can simply get it from IM = VM/RM.
3. Input-referred noise using hybrid models

H~-model representation can be very helpful in noise analysis, particularly in the input-
referred noise calculations [12]. It simplifies and produces uniformity in noise analysis by
using only one noise model for all possible cases, dealing with different values of the source
impedance RS and the amplifier input impedance Rin.
Let us consider an amplifier with a gain factor of G and input impedance Rin, shown in Fig.
6(a). Because noise is more conveniently measurable at the output port of a circuit we can
represent the output noise of the amplifier in its power spectrum density, denoted by V2o,n(f) in
V2/Hz. However, to specify a measured output noise we need to have a frequency band.
For simplicity, suppose the measurement frequency bandwidth is B = fH – fL Hz; where fH
and fL are the high and low frequency of the spectrum, respectively. With relatively constant
(within -3 dB) gain factor within the bandwidth the measured output noise can be found as:
Vo2,n ,rms = BVo2,n ( f ) (5)
On the other hand, depending on the type of input signal to the amplifier, the gain factor G
can be considered as a voltage gain A or as a trans-impedance RM depaeding on the input
voltage or current representation, respectively. Next, to calculate the input-referred noise of
the amplifier1 we need to attenuate the output noise by the gain factor G to bring it into the
input loop of the amplifier. The question is how this input-referred noise must be
represented when transferred into the input loop: as a voltage source, a current source, or in
combination of the two? It of course depends on the values of the two parameters: the
source impedance RS and the amplifier input impedance Rin [12]. Note that our objective
here is to find the input-referred noise of the amplifier that corresponds to the measured
noise at the open circuit output port. Hence, the assumption is that the thermal noises
associated with RS, Rin and the amplifier output impedance, among others are all included in
the process, and there is no need to separately calculate and add up to the input-referred
noise. However, exception might arise for a case where the source input impedance is not
included in the output noise measurement. In such a case, because of linearity, the thermal
noise of RS must be added to the input-referred noise to get the final response. In our
analysis, however, we assume the inclusive case, i.e., the entire amplifier noise, including
that of RS, is all measured at the amplifier output port.
1Input-referred noise is a virtual input noise that creates Vo,n,rms at the output, in case the amplifier is
noise free.
3.2 Input-referred noise computation

We first consider the case where the input-referred noise is represented either as a voltage
source or as a current source. The two choices are depicted in Figs. 6(b) and (c), and the
values of the input-referred noises are expressed in Eqs. (6) and (7), respectively. To simplify
this representation, again, we assume the thermal noise from RS, as well as other noise
components, to be included in Vi ,n , rms or I i ,n , rms .
Rin + RS
Vi ,n , rms = Vo ,n.rms (6)
ARin
Rin + RS
I i ,n , rms = Vo ,n.rms (7)
ARin RS
Vo,n, rms Vi, n,rms

RS RS
Iin Iin Iin
A A A
AC V in V Out AC Vin VOut AC V in V Out
RM RM RS RM
Ii ,n,rms
(a) (b) (c)
Fig. 6. (a) An amplifier with a gain factor of G (A or RM), and input impedance Rin, and the
measured output noise Vo ,n.rms ; (b) the input-referred noise as a voltage source; (c) the input-
referred noise as a current source.
However, in a special case where RS or Rin gets an extreme (low or high) value the situation
may become different so that Eq.(6) or Eq.(7) may not produce the correct response as
discussed below.
1. For a very low value of RS the input-referred noise is represented by a voltage source
(Fig. 6(b)) calculated by using Eq. (6) as
Vo ,n.rms
Vi ,n ,rms = (8)
A
For the case when both RS and Rin are very small we get the ratio α = RS/Rin and from
Eq. (6) we can get
1+α
Vi ,n ,rms = Vo ,n.rms (9)
A
2. For very high value of RS the input-referred noise is represented by a current source
(Fig. 6(c)) calculated by using Eq. (7) as
Vo ,n.rms
I i ,n , rms = (10)
ARin
For the case when Rin is very small the gain facto G can be represented by the trans-
impedance RM; the input-referred noise is obtained as
Vo ,n.rms
I i ,n , rms = (11)
RM
3. For the case when both RS and Rin are very large and they approach infinity there is an
ambiguity in the circuit and a rational solution cannot be pursued. This is because we
are basically pushing current through an open circuit! However, for large but limited
values of RS and Rin, either Eqs. (6) or (7) can provide the input-referred noise. For
example, we can use Eq. (9) to get Vi ,n ,rms .
3.2 Use of H~-models in noise computation

The problem with the foregoing procedure is that in each case we need to know the range of
values of RS and Rin in order to decide on the circuit topology; hence, decide on the right
type of the input-referred noise source. This definitely makes the analysis rather impractical.
It is only in an H~-model representation that all cases discussed above can be combined and
integrated into one. An H~-model can simply provide a universal and accurate model for the
noise calculation, regardless of the value of RS or Rin. Figure 7 shows an H~-model
representation of the input-referred noise for the selected amplifier. As shown, we can use
both types of input-referred noise sources in Fig. 7 to calculate the output noise, as shown
below.
Vh,n, rms
RS I in
A
Ih,n, rms
AC Vin VOut
RM
Fig. 7. Use of H~-modeling for computation of input-referred noise.
ARin 2 ARS Rin 2 (12)

Vo2,n ,rms = Vh2, n ,rms ( ) + I h2,n ,rms ( )
RS + Rin RS + Rin
Equation (12) can be written as
2 2 2 2
VTh , n = Vh , n , rms + I h , n , rms RS (13)
Where VTh ,n is the Thevenin noise voltage at the input loop, and is given by
RS + Rin
VTh , n = Vo ,n ,rms (14)
ARin
A comparison between Eq. (13) and Eq. (2) reveals that Eq. (13) is, indeed, the result of H~-
modeling of the input-referred noise; except that the representation here is in terms of noise
power rather than the noise voltage or current values.
Vh ,n ,rms and I h ,n ,rms can be found using Eqs. (13) and (14) with RS = 0 and RS = ∞ ,
respectively. This results in
Vo ,n , rms|Rs = 0 Vo ,n , rms|Rs =∞
Vh ,n ,rms = and I h ,n ,rms = (15)
A ARin
Here Vo ,n ,rms Rs =∞ stand for the output noises obtained when the amplifier input port is open
circuited; similarly, Vo ,n ,rms Rs = 0 stand for the output noises obtained when the amplifier
input port is short circuited. We are now ready to show that for all the cases discussed
earlier (with different values of RS and Rin) the proposed H~-model can be exclusively used
to calculate the input-referred noise. For example, for RS = 0 we get from Eq. (13) that
V
VTh , n = Vh ,n ,rms = Vi , n , rms , and from Eq. (14) we get Vi ,n ,rms = o ,n.rms which is the same as
A
Eq.(8). For RS very large by combining Eqs. (13) and (14) we get
R + Rin V
VTh , n = Vo ,n ,rms S = I h ,n ,rms RS = I i , n ,rms RS , which simply results in I i ,n , rms = o ,n.rms ,
ARin ARin
which is the same as given in Eq. (10).
Example: 3 - Consider an amplifier with a voltage gain of A = 40 dB, source impedance RS = 2
KΩ and the input impedance Rin = 8 KΩ. The output noise is measured for two cases of RS
and RS = ∞ and for a bandwidth of 300 MHz. For RS we measure Vo ,n ,rms|Rs = 0 = 200 μV,
and for RS = ∞ we measure Vo ,n ,rms|Rs =∞ = 400 μV. Calculate i) the hybrid noise voltage and
current for the input-referred noise Vh ,n ,rms and I h ,n ,rms ; ii) VTh , n , iii) and the overall output
noise Vo ,n ,rms .
Solution – The amplifier gain is A = 100 V/V. From Eq. (13) we get
Vh ,n ,rms = 200/100 = 2 μV, and I h ,n ,rms = 400/(100*8) = 0.5 nA.

2
From Eq. (13) VTh , n = 4.0e-12 + 0.5e-18 * 4.0e+06 = 6.0e-12.
Which results in VTh ,n = 2.45 μV.

Next, from Eq. (14) we get Vo ,n ,rms = 2.45 * 100 * 8/10 = 200 μV.
4. Nullified Hybrid equivalent circuit

A Nullified Hybrid equivalent circuit, called H-model, is an especial case of an H~-model;
where, the values of the voltage and current sources in the model are identical to the
corresponding port voltage and current values. What this means is that the sources in an H-
model are representing the biasing situation of the corresponding port. For example, take
the case of Fig. 3, where the network N1 provides the voltage Vj and the current Ij to bias the
network N2. The two models for this example are shown in Figs. 8(a) and 8(b). Note that
Figs. 8(a) and 8(b) are identical to Figs. 4(a) and 4(b) except here the model-sources represent
the port values. Note also from Fig. 8 that, as a result of H-modeling another port, k(Vk, Ik),
is created across N1, where both Vk and Ik are zero. Port k(Vk, Ik) is called a “null” port and
the process of creating it is called “port nullification”, as will be discussed shortly.
Vj Vj
Ik Ij N1 Ik Ij
Req Vk Ij Vj N2 Vk Ij Vj N2
No DC
Power
N’2 N’1 N’2

(a) (b)
Fig. 8. An H-model for a two terminal N1; (a) Type 1 representation; (b) Type 2
representation.
Theorem 1 introduces an important property of an H-model dealing with power
distribution in a network [13]. It adds an extra dimension to the power analysis and power
segmentation in a network.
Theorem 1: Consider a network N2 connected to another network N1 through a port j(Vj, Ij),
as in Fig. 3. Replacing N1 with its Type 1 or Type 2 H-model reduces the power
consumption in N’1 to zero, while the power consumption in N2 remains unchanged.
Proof: Consider the H~-model in Fig. 4(a) or 4(b). Both sources, IH and VH, provide power to
networks N1 and N2. The power delivered to N2 is fixed and it amounts to P2 = Vj * Ij;
whereas in Type 1 H~-model the power consumed for N1 (Fig. 4(a)) is P1 = Req(IH – Ij)2.
Hence, the power P1 in N1 becomes zero if IH = Ij which also results in VH = Vj. For Type 2 H-
model however, notice from Fig. 8(b) that N’1 has no DC supply to get power from, plus its
port is also nullified. Therefore, all currents and voltages inside N’1 must be zero, resulting
in zero power consumption.
Port Nullification: Consider a network N2 connected to another network N1 through a port
j(Vj, Ij) as shown in Fig. 3. One way to nullify Port j is to augment the port from both sides
(N1 and N2) by current sources Ij and voltage sources Vj as depicted in Fig 9. The result is the
creation of another port k(Vk, Ik) that, by definition, is a null port, i.e., both Ik and Vk are
zero.
Vj Vj
Ij Ik Ij
N1 Vj Ij Vk Ij Vj N2
N’2
Fig. 9. A simple port nullification procedure with no change imposed on N1 or N2.
However, there is an alternative method to create a null port when two networks N1 and N2
are connected through a port j(Vj, Ij), shown in Fig. (3). Here we can simply replace N1 with
its H-model (Type 1 or Type 2) and create the null port k(vk, ik), as depicted in Fig. 8. Note
that as a result of port nullification procedure, shown in Figs. 8 and 9, an extended network,
N’2, is created that contains N2 plus the sources belonging to the H-model. Similarly,
another network N’1 is also created, on the left hand side, when the H-model loses its
sources. As we can see it later, these extended networks are of particular importance in
circuit biasing.
Note that the characteristic curves of ports j and k are identical except for shifts of v and i,
coordinate axis, from the origin to the Qj(Vj, Ij) point. This makes the operating point Qj(Vj,
Ij) to fall on the origin, creating a new operating point Qk(0, 0) for the port k, shown in Fig.
10. This simply means that, for any pair of networks, N1 and N2, connected through a port j
it is always possible to nullify the port and change N1 and N2 to N’1 and N’2, where N’1 and
N’2 are identical to N1 and N2, except the v and i coordinate axis are move to the port’s
operating point. This is stated in Property 1.
ij ik
vk
Ij Q
vj
Vj
Fig. 10. The i-v coordinate axis moved from (0, 0) for the j port to a new position, Qj(Vj, Ij),
for the k port.
Property 1: Consider two networks N1 and N2 connected through a port j, as in Fig.3. If port j
is null then the i-v characteristic curve of the port, looking through either network, passes
through the origin and the origin is the operating point of that port. In case port j is not null
it is always possible to nullify the port to get the corresponding networks N’1 and N’2 with a
null port k, as shown in Fig.8.
Example 4: Consider the circuit of Fig. 11(a), where two sections of a circuit are connected
through a port j(Vj, Ij). Let the MOS diode be characterized by i = K (V-1)2 mA for V > 1V,
and let K = 0.5 mA/V2. The analysis shows that port j is not a null port because Ij = 1 mA
and Vj = 3 V. Next, we augment port j of N2 by two current and voltage sources Ij = 1 and Vj
= 3 V and then remove the supply sources of 5 V and 1 mA from N1. As a result a new null
port k(Vk, Ik) is created, as shown in Fig. 11(b). Note that although the i–v characteristic
curve of port j (associated with both networks) does not pass through the origin that of port
k does (property 1). In addition the Q-point of port k is located at the origin, as expected.
Note that i) the network N’1, on the left hand side, is still linear, and ii) the new port k has an
i–v characteristic curve that passes through the origin, and the origin is also the Q-point for
the port. This simply means that the Thevenin equivalent circuit of N’1, looking from port k,
must be a resistance with no source attached to it.
VDD
1 mA
4 KΩ 1 mA 4 KΩ 1 mA
2 KΩ 3V
2 KΩ Ij Ik Ij
5V
8 KΩ Vj 8 KΩ Vk 1 mA Vj
N1 N2 N’1 N’2
(a) (b)
Fig. 11. (a) Example of two networks N1 and N2 separated by a port j; (b) creation of a null
port k in an H-modeling representation.
5. H-modeling in multi-port networks

H-model is also capable of representing a multi-port network; and this representation is of
Type 2, introduced in Section 4. Consider a linear network N1 connected to another network
N2 through n-ports j(Vj, Ij), for j = 1, 2, …, and n, as shown in Fig. 12. Similar to a two
terminal network, the Type 2 H-model representation of N1 is obtained by removing all
independent sources2 from N1, and instead augmenting the ports with voltage and current
sources that match the corresponding port values, as depicted in Fig. 13.
Note that, similar to a single port network, the H-model procedure described above creates
n null ports k(Vk, Ik), for k = 01, 02, …, and 0n. Also note from Fig.13 that, as a result of the
H-modeling, two networks N’1 and N’2 are created that are connected together through n
null ports. Property 2 is similar to Property 1 that holds for n-port networks.
Property 2: Consider two networks N1 and N2 connected through n ports j, for j = 1, 2, …,
and n. Replace N1 with its Type 2 H-model representation to create n null ports k, for k = 01,
02, …, and 0n, as shown in Fig.13. Then for any of n nullified port the i-v characteristic curve
passes through the origin and the origin is the operating point of that port.
In another interpretation, Property 2 clearly states that port nullification through the H-
modeling does not change the ports’ i-v characteristic curves; it only moves the v and i
coordinate axis so that the ports’ operating points fall on the origins, for all n ports.
Similarly, Theorem 1 also applies to n-port networks, as stated in the following corollary.
Corollary 1: Consider a network N1 connected to another network N2 through n ports j(Vj, Ij),
for j = 1, 2, …, and n. Replacing N1 with its (Type 2) H-model reduces the power
consumption in N’1 to zero.
The proof of Corollary 1 is similar to that of Theorem 1 in that we only need to note that N’1
has no source to get power from, and that all its n ports are nullified and cannot deliver
power to N’1. Corollary 1 has several applications in power analysis of analog circuits. One
2 Again, N1 does not have dependent source that is controlled from outside of N1.
I1
V1
I2
V2
N1 N2
.
.
.
In
Vn
Fig. 12. Multi-port networks N1 and N2 connected through n ports.

V1
I01 I1
V01 I1 V1
V2
I02 I2
V02 I2 V2
N1 N2
No
DC supply .
.
.
Vn
I0n In
V0n In Vn
N’2
Fig. 13. H-model representation of the n-port network N1.
application is to verify the power consumption in different parts of a network without

disturbing the rest of the circuit. For instance, to calculate the power used in an amplifier
core, minus the losses in the DC suppliers and the power supporting circuit elements, we
can do as follows: replace the DC supply sections of the circuit with their H-models and
then calculate the total power consumed in the circuit. This is equal to the power consumed
in the amplifier core. This is in fact true for any type of power consumption including AC
power. For example, to calculate the power consumed in a circuit alone, minus the input
sources, we can represent the input sources by their H-models and calculate the total power
in the circuit. Another important application of Corollary 1 is in low power designs of
analog circuits. Here we can start designing a circuit, say an amplifier, with minimum DC
power consumption, i.e., just enough to bias the transistors in the circuit. However, the
circuit so obtained may not be very practical, after all. This is because there might be too
many DC sources, known as “distributed supplies”, being added to the circuit as a result of
the H-modeling. Nevertheless, this is a good starting point for an efficient design for power
consumption. The question asked is: how to remove the “distributed supplies” in the circuit
and replace them with typical circuit supplies, but still keep the DC power consumption
minimized? One simple solution to deal with the distributed supplies is to move them to
their destination one at a time, having in mind to keep the power consumption minimized.
This process definitely takes time and programming it may need a major effort. A more
strait forward methodology for DC supply allocation in analog circuits has been recently
developed [14] that makes this journey much simpler. The next chapter discusses this new
methodology in more details.
5.1 Coupling capacitors in H-modeling

Another useful property of H-model is that from two sources used in the model only one
souse provides power to the circuit and the other source is inactive (sitting idle with zero
voltage or current). For example, in the H-modeling shown in Figs. 8 and 13 the current
sources Ij provide power to N2, but the voltage sources Vj are only to provide voltage drops
necessary to create the null ports k, for k = 01, 02, …, and 0n, without delivering (or
consuming) any power to the circuit. It is also possible to reverse the situation and have the
voltage sources provide power and the current sources sitting inactive. Figure 14 shows
such a modeling for a single port network that is identical to Fig. 8(b) except here the
positions of the model-sources have been swapped. This is summarized in Property 3.
Vj
Ik Ij
N1 Vk Ij Vj N2
No
Source
N’2
Fig. 14. An alternative H-modeling representation
Property 3: Consider two networks N1 and N2 connected together through one or multiple
ports j(Vj, Ij), for all j, as shown in Figs. 3 and 12. Next, replace N1 with its H-model such as
those in Figs. 8, 13 and 14. Then there is only one active model-source, Ij or Vj, for each port
delivering power to N2 and the other model-source is inactive.
According to Property 3 only half of the sources used in H-models are active sources and the
other half are inactive; they are there to establish the voltage or current requirement for the
null ports. This brings up an alternative representation for an H-model. In this
representation we can replace an inactive source with a storage element such as capacitor or
inductor. Forexample, Figs. 8(b) and 13 are two circuit examples where the voltage sources
are inactive. Apparently replacing these voltage sources with capacitors that are charged to
the same voltages must satisfy the H-modeling: hence, making no changes in the voltages
and currents within N1 or N2, as depicted in Fig. 15. In fact, these capacitors play similar
roles as the coupling capacitors in ordinary amplifiers. Traditionally, coupling capacitors
are used in amplifier designs to confine the DC power within the stages of the amplifier, or
to block the DC from entering the input source or the load. The same role is played here;
except here the choice is broader. In general a circuit can arbitrarily be partitioned into two
blocks, N1 and N2 connected through n ports, where one block, say N2, receives the DC
power it needs to bias the (nonlinear) components and the other one does not need it. For
example, take again the case of Fig. 13; assume N2 is the collection of all the nonlinear
components (transistors) and N1 represents the rest of the circuit. This simply means that the
DC supplies are limited to directly bias nonlinear components in N2 and nothing else.
Figure 15 shows how the voltage sources in local biasing in Fig. 13 are replaced with
coupling capacitors; and these capacitors are going to get charged at the beginning of the
C1
I01 I1
V01 I1 V1
C2
I02 I2
V02 I2 V2
N1 N2
No
DC supply .
.
.
Cn
I0n In
V0n In Vn
N’2
Fig. 15. H-model representation of an n-port network using coupling capacitors.
circuit operation, known as the transient response. It is during this period that the capacitors
are charged to the same voltages as those voltage sources, Vj, provided that each capacitor
has a (resistive) charging path, providing an RC time constant.
6. Component biasing
One of the applications of H-modeling, leading to port nullification, is in biasing of
nonlinear components, individually or in clusters. This is known as component biasing. Take
the case of Fig. 3 or Fig. 12 and assume N2 consists of one or more nonlinear components
connected to the rest of the circuit, N1. This simply means that N1 is biasing all the
components accumulated in N2, and it establishes operating points for the ports at Qj(Vj, Ij),
for j = 1, 2, …, n. Now, compare Fig. 3 with Fig. 8(b), or Fig. 12 with Fig. 13; in both cases no
change in the biasing of the components inside N2 takes place i.e. the ports are still operating
at Qj(Vj, Ij) points. The difference, however, is that in the former circuits (Figs. 3 and 12) the
components in N2 are globally biased through N1, whereas in the later cases (Figs. 8(b) and
13) the ports are directly biased through the H-model sources, leaving N1 with no DC
supply. This brings us to introduce a new biasing scheme, known as local biasing. We can
simply show that component biasing is the combination of local biasing applied to all ports
of a nonlinear component (transistor). Next we introduce local biasing and its applications.
6.1 Local biasing

A port is locally biased if it is augmented with a voltage source and a current source so that
they exactly provide the voltage and current the port needs to operate at its desired Q-point.
Apparently the port receives its biasing power exclusively from one of those DC (voltage or
current) supplies and that DC supply is fully dedicated to the port.
A component is individually biased (called component biasing) if all its ports are locally
biased. Likewise, an m-port network consisting of multiple components is locally biased if
all its ports are locally biased.
Property 4: A nullified port is locally biased.
The proof of Property 4 is quite evident because when a port is nullified the exchange of DC
power through the port becomes zero and that is exactly what local biasing is all about.
However, in local biasing the exchange of power between two sides of the nullified port is
zero only at the designated operating point. The port behaves quite normal and like when it
is globally biased, when a signal is applied to the port. In other words, local biasing only
shifts the port’s i-v coordinate axis to the operating point.
Local biasing Using Coupling Capacitors: As discussed in Section 5, coupling capacitors can be
used in place of voltage sources in H-modeling, as shown in Fig. 15. Because of the identity
between the two concepts the same rules apply to local biasing ports as well. Now we must
realize that although both local biasing solutions (one with two sources and one with a
current source and a coupling capacitance) serve the same purpose of confining the DC
power within the nonlinear components, they do not perform identically; and they are not
interchangeable in some cases. Here are the major differences between the two. As we
discussed earlier, a locally biased port j with both sources being present create a null port k;
and as long as k stays null it guaranties that port j operates at Qj(Vj, Ij), as shown in Figs. 8,
13, and 14. However, any new DC supply in the circuit that effects port k causes port j to
shift from Qj(Vj, Ij) accordingly. Hence, local biasing, with both sources present, is
transparent to any signal (DC and AC) in the circuit; the same it is in a normal biasing
situation. This, for example, helps in amplifier designs where the frequency band includes
DC. However, this is not the case when coupling capacitors are used in local biasing. Once
the port’s operating point is established in the coupling capacitor case it remains unaltered,
no matter how much DC supply we bring to the main circuit. In fact, here, it is the current
source across the port that provides the biasing condition for the port and as long as it
remains constant at Ij the operating point stays unaltered at Qj(Vj, Ij). That is why in a
capacitor coupling case we lose the low frequency bandwidth to a non-zero value of fL,
depending on the RC time constants; C being the coupling capacitor. The following property
is valid for both types of local biasing.
Property 5: Consider a linear circuit N connected to one or more nonlinear components
through p ports. Suppose the DC supplies in N bias the p ports to their Q-points Qj(Vj, Ij),
for j = 1, 2, …, and p. Now, if we remove all DC supplies from N and instead locally bias all
p ports to their assigned operation points Qj(Vj, Ij) then we observe no change happening in
the AC performance of the entire circuit, i.e., the gains, input and output impedances,
frequency responses, and signal distortion remain unaltered. The exception is in the case
when coupling capacitors are used. The later causes the low frequency response of the
amplifier to change from DC to a higher frequency fL.
The proof of Property 5 is quite evident. For the case of local biasing using two DC sources
for each port, the sources are transparent to the AC signals and they can simply be removed
for AC analysis (including DC signal). For the case of local biasing with coupling capacitors
the capacitors bypass AC signals except for the frequencies below the low cut-off frequency
fL of the circuit.
Example 5: Consider designing a two stage BJT amplifier with feedback. The circuit structure
(topology) is shown in Fig. 16, and the design specifications are given in Table I. The
VCC
RC
IS vout
Q2
Q1
RF
iin RB RE
AC
Fig. 16. A two stage BJT amplifier with feedback

Av dB Ai dB Rin KΩ Rout KΩ fL Hz fH KHz

44 46 1.2 65 30 300
Table I. The design specs for the amplifier
Transistors are two npn, 2N3904, and their selected operating points during the AC
operation are listed in Table II. We first start with the AC performance design. This is done
by replacing the transistors with their small signal linear models at the designated Q-points.
Next, in a routine linear analysis, the circuit components are found so that the design meets
the given criteria. Table III lists the resistor values resulting from the AC design.
Trans. VBE V IB μA VCE V IC mA

Q1 0.57 0.32 1.5 0.025
Q2 0.7 25.0 4.2 4.0
Table II. The selected operating points for the transistors in the amplifier
RB KΩ RC KΩ RE Ω RF KΩ
100 1 200 40
Table III. the resistor values resulting from the AC design of the amplifier
V CC
vout
I C2
V C2
VCC
IC1
RC Q2
V CC
VB 1 IB1
Q1
RF
VE 2
RB RE IE2
AC
N2
N’1 N’2
Fig. 17. Separation of linear and nonlinear sections in the two stage BJT amplifier using H-
modeling
Our next step is to bias the transistors through local biasing. To do this we first separate the
nonlinear components from the rest of the circuit. Next, we remove the unknown DC
supplies (VDD and IS) from the circuit, and instead locally bias the transistors to their desired
Q-points, as shown in Fig. 17. Notice how the circuit is partitioned into two sub-networks
N’1 and N’2; where N’1consistes of the linear (resistive) components of the amplifier with
zero DC power, and N’2 contains the locally biased transistors. Our formal amplifier design
is over by now and the circuit should work perfectly fine. However, there is still one
practical problem left that must be taken care of; which is to reduce the number of DC
supplies and possibly allocate only one or two normal DC voltage supplies at the designated
locations. We leave this to the next chapter where the problem is tackled and a systematic
solution for DC power management and supply allocation is provided for analog circuits.
Instead, here we will continue to focus on local biasing. Because there is a low cut off
frequency specified for this design we have a choice to use coupling capacitors for the local
biasing. Figure 18(a) shows the amplifier locally biased with coupling capacitors. Note that
the capacitor values are selected based on the low cutoff frequency response, fL = 30 Hz,
specified for the amplifier. Figure 18(b) shows the output voltage swing in full range with
negligible distortion, and Fig. 18(c) shows the output frequency spectrum of the amplifier,
which tightly meets the design criteria. Theoretically our design objective for this amplifier
is accomplished at this point; however, one may argue about the practicality aspects of this
design with four current sources distributed within the circuit. In case of integrated circuits
this may be acceptable because the current sources can be replaced with active loads,
current mirrors and current sinks. For our design, as a lumped amplifier circuit, this may
create problems. One simple solution is to replace the current sources with resistors that
provide the same DC currents to the devices. But the problem with these resistors is that
when added to the circuit they may, to some extent, change the AC performances of the
amplifier, such as the gains. In some cases the changes might be negligible. In tighter design
however we can repeat the AC design; this time analyze the linear circuit with the resistors
included.
VCC
IC2 1.0 uF
IS vout
IB 1
Q2 RC
150 nF
Q1
1.0 uF
RF
I E2
i in RB RE
AC
Fig. 18. (a) – The amplifier, locally biased with coupling capacitors substituted for the
voltage sources.
(b) (c)
Fig. 18. (b) The output voltage swing of the amplifier; (c) the output frequency spectrum of
the amplifier
6.2 Local biasing vs. normal biasing

Discussion - Note from Fig. 18(a) that, any change in the biasing resources, such as making
changes in VCC or in the DC current sources, shifts the operating points of the transistors;
but the effect is negligible to the rest of the circuit. This is due to the coupling capacitors. For
example, the feedback circuit (RB, RE and RF), a very sensitive part of the amplifier, is not
affected much by these changes. This is sometimes considered a serious deviation from the
design purposes. For instance, suppose one of the purposes of the design is to provide
feedback for the biasing to help to stabilize the transistor’s operating points against shifts in
the operating points during the amplifier operation. The purpose is definitely defeated by
using this type of local biasing. This is because, in using local biasing the rest of the circuit
becomes DC isolated except for the nonlinear block (N2); hence, no DC power runs outside
the block to make the feedback effective. In short, if negative feedback exists in the circuit to
help stabilize the transistors biasing then local biasing with coupling capacitors does not
help the situation and cuts off the feedback for DC. The good news is that we can do local
biasing with a mixture of both methods. That is, we can leave the sensitive areas, which
need DC feedback, with normal locally biasing, i.e. including voltage sources instead of
coupling capacitor, and do the rest with the coupling capacitors. This way we are benefiting
from both worlds, i.e. i) getting stability through the feedback, and ii) having the coupling
capacitors to fix the operating regions when needed.
To summarize, we have introduced two types of local biasing for nonlinear components:
one with both voltage and current sources present for each component, and one with
coupling capacitors replacing the voltage courses. A third type is to use the mixture of both
as appropriate. Here are some similarities and differences between the two types. For
certain operating point assigned to a transistor in a circuit both types of local biasing
provide voltage and current required to bias the transistor. In a way, this local biasing
causes the v and i coordinate axis to move to the Q-point. The difference between the two
types, however, is that in the coupling capacitor case the Q-point is fixed on the origin as
long as the local biasing current stays fixed; whereas, in the former (with the voltage source)
case the biasing behaves exactly like ordinary biasing except initially the Q-point is located
at the origin, but it can move with adding extra DC sources to the circuit.
6.3 Local biasing of devices

Because local biasing deals with nonlinear devices, one way to efficiently analyze or design
an analog circuit is to bias the transistors individually before placing them in the circuit.
This allows the circuit to go directly for AC analysis. Within the three major semiconductor
devices p-n junction diodes are one-port devices and can be locally biased. Bipolar-junction
transistors are two-port devices, but they can also be modeled with two one-port devices in
case Ebers-Moll or the transport large signal model [11, 16] is used to replace them. Figure
19 shows an npn and a pnp transistor locally biased with their symbolic representation also
shown.
E C
IB V CE
B
IC
VEB VBE
IC
B
VEC IB
C E
(a) (b) (c)
Fig. 19. Locally biased bipolar transistors; (a) an npn; (b) a pnp; and (c) the symbol for a
locally biased BJT
MOS transistors, on the other hand, are considered three-port devices with only four sources
needed to represent locally biasing of the device. This is because for the drain-source we
need both ID and VDS sources to nullify the port; whereas for the gate-source and the
substrate-source we only need VGS and VBS to nullify the ports, respectively. Figure 20
illustrates both an nMOS and a pMOS being locally biased; however, for simplicity purposes
we may normally drop the substrate effect, VBS, and consider the device as a two port
(drain-source and gate-source) component.
S D
VGS VDS
VBS
G
B ID
ID VSG B
G VSB
VSD
D S (b) (c)
(a)
Fig. 20. Locally biased MOS transistors; (a) an NMOS transistor; (b) a PMOS transistor; and
(c) the symbol for a locally biased transistor
C
V CE
V BE
IC
B
IB
E
(a) (b)
Fig. 21. (a) Locally biased bipolar transistor using coupling capacitors; and (b) the
representing symbol
6.4 Source reduction in the local biased transistors

Coupling capacitors described earlier can also be used in device biasing to replace the
voltage sources. Again, this is because only one of the two sources is needed to provide DC
power to the port, and the other source is inactive. In case of an inactive voltage source we
can replace it with a coupling capacitor. However, there are two related issues that must be
addressed when coupling capacitors are used in amplifiers. First, a coupling capacitor must
have a (resistive) path to DC supplies to get charged during the operation. Second, as we
know, coupling capacitors have direct effect on the low frequency response of the amplifier;
therefore, they must be selected so that the low frequency response criterion of the amplifier
is not violated. Figure 21 is an example of coupling capacitor used in local biasing of a BJT.
Example 6: Figure 22(a) shows a single stage nMOS amplifier. To apply local biasing we
remove all DC supplies from the circuit and locally bias the transistor. In this design the
transistor is locally biased with capacitor coupling, shown in Fig. 22(b). However, the
coupling capacitors need to be charged to the level needed for local biasing; hence, they
need charging path. The capacitor CDS has its charging path to the current source ID, but the
capacitor CGS lacks such a path and we need to create one through an extra resistance RF =
10MEGΩ. Although RF is not part of the original amplifier, it is large enough to neglect its
effect on the amplifier operations. The next step is to decide on the capacitors values. Each
capacitor creates a pole (also zero) for the output transfer function, and for the correct
frequency response the poles must lie below the low cutoff frequency of the amplifier, set at
fL = 100 Hz. We first assume CGS = 0.5 nF and CDS = 100 nF and compute the poles
individually. For the gate capacitor, roughly speaking, we have
1 1
f GS = = = 32 Hz
2π ( RF + RG )CGS 2π * 10.2 e + 06 * 0.5 e − 09
And for the drain capacitor we get
1 1
f DS = = = 53 Hz
2π RDC DS 2π * 30 e + 03 * 100 e − 09
Both pole locations are below fL = 100 Hz and hence are accepted for our design. This will
conclude the design. The circuit of Fig. 22(b) is simulated by SPICE and the results for both,
the transient response and the frequency response, are provided in Figs. 23 (a) and (b),
respectively. Note from Fig. 23(a) that it takes about 4 msec for both CGD and CDS to charge
to the level needed for local biasing. Also note that, since all biasing is accomplished by
current sources we do not need to specify the DC supply value VDD, unless certain voltage
swing for the output waveform is needed.
VDD = 5 V
VDD
80 μA
ID
30 KΩ
10 MΩ
CDS
200 KΩ V Out 100 nF VOut

200 KΩ CGS
30 KΩ
AC 1.5 KΩ 0.5 nF
AC
80 μA
ID
VGG = 1.6 V 1.5 KΩ
(a) (b)
Fig. 22. (a) Single stage NMOS amplifier; and (b) locally biased transistor with coupling
capacitors
(a) (b)
Fig. 23. The SPICE simulation results; (a) the transient waveforms; and (b) the amplifier
frequency response
Before we leave our discussion about the coupling capacitors here we need to closely look at
their effect on the AC operation of the circuit. As we add each capacitor to a circuit3 we
basically add one pole, and possibly one zero, to the transfer function; and in the case of
large number of capacitors they may initiate circuit instability and oscillation. A rule of
thumb that often applies here is that, if an amplifier has feedback and it exceeds more than
two stages, the extra number of coupling capacitors for local biasing should be avoided.
3 The assumption is that the capacitors are independent, i.e., they can arbitrarily assume any voltage
across.
7. Additivity in local biasing

Additivity is a major property of linear circuits; it permits superposition as a convenient tool
for breaking the circuit response to multiple stimuli into the sum of the circuit responses to
individual stimulus, one at a time. As we know additivity does not apply to nonlinear
circuits, but as we will prove, it works in circuits with nonlinear components provided that
the circuit situation (response) is restored before applying the next stimulus (DC supplies).
Theorem 2 describes the procedure for two terminal networks.
Theorem 2 – Additivity: Consider a network N2 connected to another network N1 through a
port j(V, I), as shown in Fig. 24(a), and with its characteristic curve shown in Fig. 24(b). Let
N1 contain n DC supplies. Further, assume we are dealing with simple (non-multiple)
operating points4 in this case. Group the sources arbitrarily into p mutually exclusive
groups5. Perform p number of biasing cycles to the circuit; each time applying only one
group of supplies and remove the rest. Then, the final operating point of the port due to all
n supplies can be determined by adding the (voltage and current) values associated with all
p number of operating points in the p biasing cycles provided that the port is nullified
(called partial local biasing) before the next operation is performed.
(a) (b)
Fig. 24. (a) Two terminal networks connected; and (b) the port’s operating point on the
characteristic curve
Proof: Suppose a network N1 with n DC supplies is connected to another network N2
through a port j(V, I) (Fig. 20(a)), and suppose Q(V, I) is the operating point of the port
looking to N2, as shown in Fig. 20(b). Now, split n supplies into p groups of mutually
exclusive supplies n1, n2, …, and np. First keep the group of n1 supplies in N1 and remove
the rest (Fig. 25(a)). Suppose for this case the operating point moves to a new point, Q1(V1,
I1), on the characteristic curve, as depicted in Fig. 25(d). Next do the followings: i) augment
port j1(V1, I1) with current I1 and voltage V1 supplies, and ii) remove n1 sources from N1. This
creates a nullified port j2(V2, I2) next to j1(V1, I1). Now we have completed a partial local
biasing, which causes the v and i coordinate axis to move from (0, 0) to Q1, and make it the
4 In a case of multiple operating points we may end up with more than one operating point for a single
set of supply sources.
5 It is also permissible to have a supply used in more than one group. In this case the supply is
partitioned and each part is exclusively used in one group.

new origin. Next, add the group of n2 supplies to N1, as shown in Fig. 25(b). This causes the
operating point to move from Q1 (the new origin) to Q2(V2, I2), as indicated in Fig. 25(d).
Likewise, augment port j2 with current I2 and voltage V2 supplies and then remove n2 group
of supplies from N1 to create a nullified Port j3(V3, I3). Again, the last operation causes the v
and i coordinate axis to move from Q1 to the new location, Q2, (Fig. 25(d)). Similarly,
introduce n3 group of supplies to N1 (Fig. 25(c)) and move the operating point to a new
point Q3(V3, I3) on the characteristic curve. Without loss of generality we can now assume
that the sources in N1 are exhausted at this point. Then Q3 and Q must be the same point on
the characteristic curve. This is because the process, just explained, is not different from
applying all n supplies to the circuit in p steps of n1, n2, …, and np groups, but this time
without removing any of them. This simply means that V = V1 + V2 + V3, and I = I1 + I2 + I3,
as we can see in Fig. 25(d). This proves the theorem.
V1 I3 V2 V1
I1 N1 I2 N1
N1
V1 N2 V 2 I1 N2 V3 N2
n1 DC n2 DC n3 DC
I2 I1
s upplies s upplies s upplies
(c)
(a) (b)
Fig. 25. The process of additivity in local biasing; (a) network with the first group of
supplies; (b) and (c) Sequences of adding groups of supplies one at a time and accumulating
the biasing results
i1 i2 i3
I3
Q3 v3
I2 v2
Q2
I1 Q1 v1
V1 V2 V3
Fig. 25. (d) Progressive move of the operating point on the characteristic curve as the
supplies are adding.
The method just described allows us to progressively bias a complex circuit in a step-by-step
procedure. By using additivity property we can break down the DC supplies into p separate
groups of supplies so that each time we only apply one group. At the end it is the sum of
partial results that determines the final operating points of the transistors. This separation of
multi-step biasing procedure, called progressive biasing, has only been possible by using local
biasing methodology. In a way, local biasing keeps (stores) the progression of the biasing
status in the circuit in order to accumulate and direct the biasing to its destination. It can be
thought of as a ladder procedure: in each step of the ladder one group of the circuit supplies
are replaced with local biasing supplies so that the Q-points of the transistors stay
unchanged on the characteristic curves, but all the coordinate axis move to the Q-points,
making them new origins for the next step. This continues until the circuit supplies are
exhausted. It is this additivity property that makes superposition, a valuable tool, available
for nonlinear circuits. It is through this superposition that we can break down the
complexity of biasing for large circuits and manage a smooth biasing convergence.
Another notable point regarding this step-by-step biasing procedure is that we can arrive at
a final Q-point in a port from different directions, depending on the sequence of the supply
groups we select to apply. And in these options we might be able to select the quickest one
or the one that assures convergence. On the same line, following the procedure stated in
Theorem 2 we might arrive at different Q-points when we approach from different
directions. This is the case when we are dealing with multiple Q-points; and the described
procedure can provide an alternative technique for searching for multiple operating points
in a nonlinear circuit [9].
Although Theorem 2 is given for two terminal networks it can easily be extended to include
multiple-port networks, as stated in Corollary 2.
Corollary 2: Consider a network N2 connected to another network N1 through m ports. Let
N1 contain n number of DC supplies used for biasing N2. Further, assume all the operation
points for the m ports are simple (non-multiple) Q-points. Next, group the sources
arbitrarily into p mutually exclusive groups. It then follows that for each port the final
operating point Q(V, I), due to all n DC supplies, can be found by adding the Q-point
(voltage and current) values, Q(Vj, Ij), for all p number of group of supplies, provided that
the following condition holds:
The port is nullified by being locally biased after each group of supplies is applied; making
the Q-point a new origin for the port’s characteristic curve.
Hence we can write:
p p
V = ∑ Vj and I = ∑ I j (16)
j =1 j =1
The proof of Corollary 2 is similar to that of Theorem 2 in the sense that in each sequence of
applying a group of DC supplies to the circuit we can extend the procedure to include all m
ports. However, we must remember that in each step the nullification of ports must be total
and simultaneous. That is, for each application of a certain group of DC supplies we need to
find the corresponding H-model of all m ports of N1. This process does two things:
i. it generates m null ports -- one for each port --, and
ii. it finds the v and i values of the partial Q points for all ports at the same instance.
Again, we must emphasize that this additivity procedure is applied to circuits with simple
operating points; where, for each port, any route taken ends up at a fixed location (Q-point)
on the characteristic curve. For circuits with multiple Q-points the procedure works as well,
except we may reach to different Q-points when we follow different sequences of supply
groups.
This additivity property provides a new and remarkable methodology for the analysis and
simulation of nonlinear circuit with multiple nonlinear components. Another unique feature
of the additivity property is that it provides a simple mechanism through which we can
arbitrarily and gradually (ladder type) replace the normal supplies in an amplifier with
supplies that locally bias the individual transistors. Conversely, in a design procedure, we
can start with local biasing the transistors to get them to the desired Q-points; then move
and combine the distributed supplies (by techniques such as source transformation) to
merge into normal circuit supplies such as VDD, VCC.
Algorithm 1 provides a circuit analysis procedure based of the progressive biasing stated in
Corollary 2.
Algorithm 1:
1. Given a nonlinear circuit, first identify all nonlinear devices and put them into one
nonlinear network N2 with m ports, j(vj, ij), connected to the rest of the circuit as a linear
network N1.
2. Select a grouping scheme for the DC supplies and put them into an arbitrary sequence
that best performs the biasing of m ports in N2. The sequence selected should possibly
guaranty a quick convergence. This is a crucial step and needs design experience to
achieve a good result.
3. Keep the first group of supplies in N1 and remove the rest. Assume this group of
supplies makes N2 to operate at Q1 (V1, I1) on the characteristic curve (for simplicity the
algorithm is given for one port but it is extendable to all m ports, as well). Next,
augment the port with I1 and V1 sources for local biasing, and remove the first group of
supplies from N1. This will create a nullified port.
4. Include the second group of supplies into N1 and remove the rest. This will cause the
operating point to move from Q1 (V1, I1) (now the origin) to Q2 (V2, I2), which is the new
operating point.
5. Continue with step 4 until all groups of supplies are sequentially applied.
6. The biasing of the transistors is complete and the entire circuit -- N1 plus N2 -- is ready
for the application of the AC signals. The output signal, in this case, is only AC without
being mixed with any DC component.
The following two examples are going through the progressive biasing procedure by using
Algorithm 1.
Example 7: Figure 26(a) shows part of the circuit of the MC1553, a three stage BJT amplifier
with feedback [11]. The circuit apparently works (biases) with a single supply of VCC = 9V.
To produce a progressive biasing for the amplifier we have spit the 9 volt supply into four
separate unspecified supplies VBB, VCC, VDD, and VEE, shown in Fig. 26(b). Note that we have
also replaced the transistors with their locally biased counterparts; where each transistor has
its own voltage sources VBE and VCE, and current sources IB and IC used for the local biasing,
as depicted in Fig. 19. Next, we are going to make three groups of supplies: (VCC = 5V and
VDD = 9V), (VBB = 7V and VEE = 9V) and (VBB = 2V and VCC = 4V), and then apply Corollary 2
for a progressive biasing procedure. Table IV is the result of this biasing procedure. Column
2 in the table displays the biasing results (Q-points) of the transistors when the original
circuit of Fig. 26(a) is used. Columns 3, 4 and 5 are the results of the progressive biasing
sequentially applying the groups of supplies as indicated. As shown, column 5, which is the
accumulation of all the three steps, is identical to column 2, as expected. Another interesting
observation from Table IV is that, although the transistors may go into different modes of
operations in the progressive biasing – such as saturation or cut off, for example – the results
are coming out correctly at the end.
VCC = 9 V VBB VCC V DD VEE
26.8 K Ω 9 KΩ 600 Ω 26.8 KΩ 9 KΩ 600 Ω

5 KΩ 5 KΩ
vout vout
Q3 Q3
Q2 Q2
Q1 Q1
1.0 μF 1.0 μF
640 Ω 640 Ω
2. 71 K Ω 100 Ω 100 Ω 2.71 K Ω 100 Ω 100 Ω

AC AC
(a) (b)
Fig. 26. (a) Part of the circuit of the MC1553, a three stage BJT amplifier with feedback; and
(b) a progressive biasing of the amplifier using additivity property along with the local
biasing.
Items All four VCC = 5V + VBB = 7V + VBB = 2V

supplies 9V VDD = 9V +VEE = 9V +VCC = 4V
IB1 6.97246e-06 -9.53064e-13 1.845914e-06 6.972464e-06

IB2 1.08180e-05 4.782478e-04 2.838049e-04 1.081804e-05
IB3 3.39962e-05 1.741652e-13 1.20893e-11 3.399617e-05
IC1 9.13995e-04 1.662226e-12 1.950662e-04 9.139951e-04
IC2 1.52165e-03 1.788923e-03 1.786860e-03 1.521652e-03
IC3 5.64041e-03 -1.43753e-13 2.230231e-11 5.640411e-03
VBE1 6.634702e-01 2.281979e-09 6.231302e-01 6.634702e-01
VCE1 5.279963e-01 6.957695e-01 6.728130e-01 5.279964e-01
VBE2 6.766817e-01 6.957695e-01 6.901600e-01 6.766817e-01
VCE2 1.221757e+00 5.538356e-02 6.570034e-02 1.221757e+00
VBE3 7.109050e-01 5.538356e-02 6.335615e-02 7.109050e-01
VCE3 5.104901e+00 7.513105e-11 8.997656e+00 5.104902e+00
Table IV. the results of progressive biasing of the amplifier using additivity property with
local biasing
Example 8 –BJT Circuit with multiple Operating Points: The circuit shown in Fig. 27 has
multiple operating points, and for a similar circuit Goldgeisser and Green [9] have reported
nine operating points. We originally simulate the circuit with all three external supplies,
12V, 10V and 2V simultaneously applied. When using WinSpice3 [4] it takes 163 iterations to
converge to stable operating points for all transistors. In an effort to minimize the effect of
other convergence factors both source stepping and the shunt convergence aids are disabled
in this program – performed by enabling OPTIONS: ITL6=1 and MINCONVSHUNT=0.
Table V shows the circuit node voltages obtained.
Next, we use local biasing methodology. First, we identify all four BJTs in the circuit and
put them into a multi-port nonlinear block N2. Next, we separate the DC supplies into two
groups: i) the 12V and 10V supplies, and ii) the 2V supply. In step 1 we keep the 12V and
10V supplies in the circuit and remove the 2V supply, and simulate the circuit using Spice3
with applying the same conditions (OPTIONS) we did originally. Here we notice that the
circuit converges fairly quickly into a set of operating points. In the second step we remove
the supplies from the circuit and instead locally bias the transistors to the same operating
points reached. Then we add the 2V supply to the circuit and simulate the circuit again. The
circuit converges this time to a new set of operating points after a few more iterations. It is
observed, as expected, that these new operating points are the same as those originally
obtained, i.e., located at the same Q-points on the devices’ characteristic cures.
Fig. 27. A BJT circuit with multiple operational points
V(2) V(3) V(4) V(5) V(8)

10.425 0.171 0.5216 10.428 1.0967
Table V. The circuit node voltages for figure 27 external biasing
In comparing the two methods of biasing for this example, we notice that while 163
iterations was necessary for the biasing to converge in the original biasing scheme it only
took 10 iterations for both steps in the local biasing scheme to converge. This is an
outstanding achievement owing to the additivity property of local biasing, which is very
essential in the analysis and simulation of complex circuits. One may argue that a similar
power supply “stepping” is also provided in the conventional analog circuit simulators such
as SPICE. However, the differences are quite evident. The proposed method gives choices to
skilled designers to select their own DC supply grouping and the sequences they choose to
apply. For example, in applying the local biasing scheme in this example if we apply the
“2V” supply first and then the “12V and 10V” the number of iterations would substantially
increase. The second, and the most important property of the new methodology is its
additivity property; where the sequences of steps in biasing are “stored” in local biasing as it
adds up to the final result (again, like stepping on a ladder).
Figure 28 depicts the circuit when it is locally biased, and as we notice the entire external DC
supplies are removed leaving each transistor with its own biasing. Table VI provides the
augmented voltage and current supplies used during the two steps of the local biasing. Note
that the values in the column 5 are the sum of the corresponding values in the columns 3
and 4, which is due to the additivity property.
Fig. 28. The BJT circuit with locally biased transistors
BJT BJT
12V and 10V 2V All Supplies
Ports
VBE1 0.667 -0871 -0.204
Q1 VCE1 4.17 5.35 9.52
IB1 1.18e-05 -1.18e-05 -1.61e-12
IC1 1.63e-03 -1.63e-03 1.24e-11
VBE2 0.437 0.248 0.685
Q2 VCE2 9.66 -9.61 5.16e-02
IB2 7.09e-09 9.79e-05 9.79e-05
IC2 2.48e-07 2.50e-03 2.50e-03
VBE3 0.589 -0.014 0.575
Q3 VCE3 9.40 0.51 9.91
IB3 8.80e-07 -3.27e-07 5.53e-07
IC3 8.80e-05 -3.69e-05 5.11e-05
VBE4 -0.728 0.377 -0.351
Q4 VCE4 9.58 0.32 9.90
IB4 -1.71e-12 1.09e-14 -1.70e-12
IC4 1.26e-11 3.46e-13 1.30e-11
Table VI. the voltage and current supplies used for local biasing
Progressive biasing, employed in the last two examples, has other applications in the
analysis and design of analog circuits. One application in circuit design is in setting the
operating regions of the transistors based on the design specs. In this situation the
transistors are initially locally biased to their assigned Q-points. What is then left to
complete the DC circuit design portion is to move the generated local biasing sources to the
locations designated for the circuit power supplies. The other application of local biasing is
in circuit diagnosis and modification; where local treatments of a malfunctioning circuit can
solve the problem rather then doing a complete redesign. Both applications are briefly
explained next.
7.1 Using local biasing in circuits design

The methodology just described allows a circuit designer to locally bias the individual
nonlinear devices in a circuit as desired; for the rest he/she will be dealing with the linear
portion of the circuit. In other words, in this methodology the biasing become local and
isolated from the rest of the circuit, while the AC signals remains global and not mixed with
the biasing. This allows the designer to concentrate on the AC signal operations and design
the circuit based on its best performance.
Another advantage in using local biasing for design purposes is the convenience it provides
for the designer to play with the operating regions of individual transistors until he/she is
satisfied. Whereas, in the traditional (global) biasing the DC and AC signals are mixed;
making the design complex. In a way, by local biasing we are creating an orthogonality
between AC and DC design and operation of circuits. Here, only the nonlinear devices are
individually biased to meet the designated operating regions. Whereas, in the AC design the
rest of the circuit contribute to the signal performance, with the small signal linear models of
the devices included in the circuit.
In addition local biasing minimizes DC power consumptions in the circuit; hence by starting
the circuit design with local biasing we in fact have started the design with minimum
power. For any modification such as source transformations to another destination in the
circuit we can monitor the power variations and go for optimal power as the design
progresses.
Design Procedure: We are now ready to implement the proposed methodology in design
stages. Algorithm 2 provides a stepwise procedure to design an analog circuit using the new
methodology. Although given for an amplifier the procedure is equally applicable to any
other analog circuit.
Algorithm 2:
1. To design an amplifier for a given topology and design specs first select the desired
operating regions for the devices (diodes and transistors) so that the devices can best
respond to the design specs. Locally bias the devices by augmenting current and
voltage sources to each individual6 device to meet the DC design specs.
2. Replace the locally biased devices with their small signal linear models and proceed
designing the linear amplifier for its AC performance. It is important to note that as
long as the linear models, representing the locally biased devices, are not altered the
circuit topology, as well as the component values (including the W/L ratios in MOS
transistors) can be changed for optimal performance of the amplifier.
6 It is possible to combine multiple (transistor) devices in an m-port network and locally bias the m-port
network instead.
3. After the AC performance design is completed satisfactorily go back and replace the
linear models of the transistors with their corresponding locally biased devices.
Theoretically, both DC and AC design of the amplifier is over by now, except for the
existence of the distributed DC sources.
4. Use source transformation techniques combined with current sourcing and mirroring
techniques to move and reduce the current and voltage sources used for the local
biasing, in such a way that the result could end up with one or a few supplies -- VDD
and VSS – in the circuit.
The following example provides the design of an amplifier using the proposed methodology
described in Algorithm 2.
Example 9 –Three-stage CMOS Op-Amp: Consider designing a three-stage operational
amplifier with circuit configuration shown in Fig. 29. For simplicity the current mirrors are
substituted by ideal current sources. The transistors’ biasing currents ID1 = 21.6 μA, ID2 = 21.6
μA, ID3 = 110 μA, and ID4 = 2.63 mA are provided as design specs; which are based on the
power expectation for each amplifier stage. Also the design is targeted for a maximum
output voltage swing of 7 V peak to peak. In addition, base on the design specs we expect to
get about 5 mW of output power to the load.
Fig. 29. Basic configuration for a three stage operational amplifier.

The next step in the design is to size the transistors. The channel lengths are assumed fixed
for L = 2 μm; the transistor widths, based on the drain currents, are calculated and shown in
Table VII. Now we locally bias the transistors so that the operating points are located far
enough into the linear portion of the characteristic curves, in the saturation regions. It is
reasonable to assume that the operating points of M3 and M4 to be two critical design specs.
This is because M3 and M4 represent output stages and need to exhibit maximum voltage
swings with high currents. The next step is to locally bias the transistors in the circuit and
then remove all external DC supplies, as depicted in Fig.30. Table VII shows the biasing
design specs for the transistors (see Fig.20).
Next, we can proceed with the design, taking the followings specs into consideration:
• For a maximum of 7 V peak to peak output voltage swing (M4) we need the DC power
supplies VDD = VSS = 5 V.
• The selection of the operating currents for the transistors is based on the power
expectation for each stage. For example, in the buffer stage, the device current ID4 = 2.63
mA is selected to deliver about 5 mW power to the load. Likewise, given the current
gain for the buffer stage AI3 = 24 A/A we can calculate the drain current for M3 as ID3 =
2.63 / 24 = 0.11 mA.
• The selection of VGS for M4 is important in pushing the operating region of the buffer
transistor far enough into the linear saturation region and to produce Voutp-p = 7 V
without distortion.
• Other design parameters such as the resistor values are also calculated for the targeted
performance of the amplifier. For this design we find RM1 = 51 KΩ, RM2 = 51 KΩ, and
RM4 = 4.5 KΩ to best fit the specs.
Locally biasing Sources W/L μm VDS VGS VSB ID

pMOS-1 15/2 -5.55 -1.65 -3.35 -21.6 μA
pMOS-2 15/2 -5.55 -1.65 -3.35 -21.6 μA
nMOS-3 30/2 2.71 1.10 0.00 110.0 μA
pMOS-4 500/2 -4.74 -2.03 -5.26 -2.63 mA
Table VII. Transistor Sizes and DC sources for local biasings of Transistors
Fig. 30. The Op-Amp configuration with locally biased devices

The initial stage of the design of the amplifier including the component biasing is over now.
In the next stage we need to replace the transistors with their small signal models to perform
the performance design such as the gains, bandwidth, and so on. However, because our
main intention at this point is the biasing design we ignore details on the performance
design. Here we are allowed to modify the component values (except for the transistors’
model values which are anchored by the local biasing) until the desired responses are
obtained and the design criteria are met. Following the performance design we need to
replace the linear transistor models with their locally biased transistors, as shown in Figure
30. Note that no external DC supply other than those included in the local biasing is needed
to run the amplifier. Figure 31 shows the WinSpice3 simulation results for the amplifier with
the local biasing. Both the transient responses (the output signals before and after the buffer
stage) and the frequency responses are provided. Note that all node signals in the transient
responses lack any DC component, due to local biasing; hence no need for coupling
capacitors or to stop offset voltages.
Fig. 31. The transient and frequency responses of the Op-Amp with locally biased
configuration
Finally, for practical reasons we need to replace the local biasing supplies with limited
external supplies located at the designated locations in the amplifier. Application of certain
procedures (not explained here) has results in having three current sources I1 = 43 μA, I2 =
68 μA, and I3 = 1.12 mA plus two voltage sources VDD = 5 V and VSS = 5 V, as originally
shown in Fig. 29. These sources are replacing the local biasing sources in the amplifier.
7.2 Circuit diagnosis and partial local biasing

By partial local biasing (PLB) we mean to perform local biasing on a device (or a port)
without disturbing any other part in the circuit, even without changing the regular DC
supplies in the circuit. Hence, PLB allows a designer to diagnose an analog circuit and
locally tune it by changing the biasing conditions of one or more components in the circuit
without changing the operating points of other components. PLB is different from local
biasing in which, local biasing makes the entire circuit DC-static (zero DC power) except for
the locally biased devices; whereas in PLB the DC supplies remain intact within the circuit,
except that the operating points of the ports, selected for modification, can be changed
through PLB. This modification is done by augmenting those ports with a combination of
voltage and current sources that have values equal to the differences between the old and
the new Q-points of those ports.
PLB has two main properties; it is local and it is not destructive. It is local because it only
affects the component under test. Second, because of the additivity property of local biasing
and due to being local, PLB can be progressive in steps of one or more components at a time.
For example, if in a circuit modification the biasing conditions of several components need
to be changed, we can change one device at a time and look for the responses as we progress
[15]. One application of PLB is in circuit diagnosis and repair. If the problem relates to a
faulty transistor, for example, we can take it out and replace it with a new one. We can also
replace it with a different type of transistor, such as changing BJTs to MOS transistors, in a
circuit. Another application of PLB is in partially testing a complex circuit looking for the
troubled places. For example, consider the circuit in Fig. 32(a), where the MOS transistor M
is malfunctioning because its output port is at Q(V, I), which is at the wrong place on the
characteristic curve (Fig. 32(b)). To correct the situation we need to move the operating point
to the right on the characteristic curve, positioning it at Q1(V+δV, I+δI), as indicated in Fig.
32(b). We use PLB by augmenting the transistor with one voltage and one current source
that has values δV and δI, respectively. This causes the OP to move from Q to Q1 without
affecting the rest of the circuit, as depicted in Fig. 32(c). Later, we may need to move the
sources, δV and δI, and integrate them with the rest of the DC supplies in the circuit by
using techniques such as source transformations. Of course, we need to be careful in this
source transformation so that the other operating points, for other transistors, are not
disturbed.
δV
N N δI
M M
(a) (b) (c)
Fig. 32. Partial local biasing of an MOS in a circuit; (a) the original circuit with distorted
output; (b) the device characteristic curve; and (c) .corrected operating point through partial
local biasing.
The following example further explains the procedure.
Example 10: In this example we are considering a two stage MOS amplifier with feedback, as
shown in Fig. 33. Initially both transistors, M1 and M2, are assumed identical with W/L =
50/5 μm. The amplifier works fine with this configuration without distortion. However, in
an attempt to improve the output power of the amplifier we modify it by changing the size
of M2 from W/L = 50/5 to W/L = 100/5, doubling the transistor channel length. The change
disturbs the biasing situation in the amplifier and distorts the output response, as shown in
Fig. 34. Next we apply the PLB on M2 to correct its biasing situation. It turns out that locally
adding an extra current ID2 = 560 μA to the drain current of M2 would correct its operating
point. Both output waveforms, one before the biasing correction and one after, are shown in
Fig. 34. Note that the gross distortion observed in the output waveform of the original
amplifier has disappeared from the output waveform of the modified amplifier. We also
notice a better gain for the second stage of the amplifier, which is mainly due to a better and
flatter operating region created for M2 transistor.
VDD = 5V
240 KΩ 10 KΩ 4 KΩ
vout
50/5
M2
50/5
20 nF M1
80 KΩ
300 KΩ
AC 2 KΩ
Fig. 33. Two stage MOS amplifier with feedback with the output distorted for W/L = 100/5
Output response
after biasing
correction
Output response
before biasing
correction
Fig. 34. The output response of the amplifier before and after bias correction.
8. Chapter summary
A new modeling technique, called H~-modeling, is introduced for one and multiple port
networks. It is shown that H~-models are more dynamics compare to Thevenin or Norton
equivalent circuits, and they have the ability to more accurately describe the port behavior.
The properties of this model, particularly in calculating the input-referred noise, is
discussed. A special type of H~-model, called nullified H~-model, or simply H-model, is also
introduced; and many properties of H-modeling including power management in the circuit
is investigated. It is shown that H-models are not limited to single port networks but cover
multi-ports, as well. A major property of H-modeling is in local biasing of transistors. It
separates nonlinear components from the linear portion of the circuit for faster and more
efficient circuit biasing. Here a designer can take advantage of H-modeling and bias
individual transistors (or in combinations) with no need to perform the the normal circuit
biasing. Because of the distributed supplies, created due to local biasing, the method is
extended to include coupling capacitors for biasing purposes as well. The fact that local
biasing helps to do a mixture of regular but progressive biasing in complex circuits is
discussed. Here, local biasing keeps (stores) the status of partial biasing in any stage of a
gradual and step-wise biasing procedure, i.e., it allows the global biasing to keep
progression toward the completion of the biasing. Next, partial-local biasing is introduced,
which helps to modify and locally correct the biasing of a circuit. This is important in
debugging, modifying and repairing complex analog circuits.
9. Acknowledgment
The author would like to thank Ms. Leyla Hashemian for her valuable suggestions and
editing the chapter.
10. References
[1] T.L. Pillage, R.A. Rohrer, and C. Visweswariah, “Electronic Circuit & System Simulation
Methods,” New York, McGraw-Hill, 1995.
[2] J. Vlach and K. Singhal, computer methods for circuit analysis and design, Van Nostrand
Reinhold Electrical/Computer Science and Engineering Series, 1983.
[3] L.W. Nagel, "SPICE2, A computer program to simulate semiconductor circuits," Univ. of
California, Berkeley, CA, Memorandum no. ERL-M520, 1975.
[4] Mike Smith, "WinSpice3 User’s Manual, v1.05.08",
http://www.ousetech.co.uk/winspice2/, May 2006.
[5] C.W. Ho, A.E. Ruehli, and P.A.Brennan, "The modified nodal approach to network
analysis," IEEE Trans. Circuits Syst., vol. CAS-22, no.6, pp.504-509, June 1975.
[6] C. A. Desoer and E. S. Kuh, Basic Circuit Theory. New York: McGraw Hill, 1969.
[7] Y. Inouea, "Dc analysis of nonlinear circuits using solution-tracing circuits," Trans. IEICE
(A). vol. J74 A, pp. 1647-1655, 1991.
[8] ___, "A practical algorithm for dc operating-point analysis of large scale circuits," Trans.
IEICE (A), vol. J77-A, pp. 388-398, 1994.
[9] L. B. Goldgeisser and M. M. Green "A Method for Automatically Finding Multiple
Operating Points in Nonlinear Circuits," IEEE Trans. Circuits Syst. I, vol. 52, no. 4,
pp. 776-784, April. 2005.
[10] R. C. Melville, L. Trajkovic, S.C. Fang, and L. T. Watson, "Artificial parameter homotopy
methods for the dc OP problem,” IEEE Trans. Computer-Aided Design, vol. 12, no.
6, pp. 861-877, Jun. 1993.
[11] A.S. Sedra, and K.C. Smith, Microelectronic Circuit 6th ed. Oxford University Press, 2010.
[12] R. Jacob. Baker, CMOS, Circuit Design, Layout, and Simulation, 2nd ed. IEEE Press, Wiley
Interscience, 2008, pp. 613 – 823.
[13] R. Hashemian, "Designing Analog Circuits with Reduced Biasing Power", to be

published in the Proceedings of the 13th IEEE International Conf. on Electronics,
Circuits and Sys., Nice, France Dec. 10– 13, 2006.
[14] ___, “Local Biasing and the Use of Nullator-Norator Pairs in Analog Circuits Designs,”
VLSI Design, vol. 2010, Article ID 297083, 12 pages, 2010. doi:10.1155/2010/297083.
http://www.hindawi.com/journals/vlsi/2010/297083.html
[15] ___, “Partial Local Biasing, A New Method to Modify/Tune Amplifiers for a Desirable
Performance”, 2007 IEEE International Conference on Electro/Information
Technology, IIT, Chicago, May 17 – 19, 2007.
[16] R.C.Jaeger, and T.N. Blalock, Microelectronic Circuit Design 4th ed. Mc Graw-Hill Higher
Education, 2010.
4
Behavioral Modeling of Mixed-Mode

Integrated Circuits
Esteban Tlelo-Cuautle1 , Elyoenai Martínez-Romero2 , Carlos

Sánchez-López3 , Francisco V. Fernández4 , Sheldon X.-D. Tan5 ,
Peng Li6 and Mourad Fakhfakh7
1,2 INAOE
3 UAT,
3,4 IMSE. CSIC and University of Sevilla
5 University of California at Riverside,
6 Texas A&M University,
7 University of Sfax
1,2,3 México
4 Spain
5,6 USA
7 Tunisia
1. Introduction
Modeling is a preliminary work or construction that serves as a plan from which a final
product can be made. Modeling at the transistor level of abstraction in the integrated circuit
(IC) industry has roots in the primitives found in the popular simulation program with
integrated circuit emphasis (SPICE). Although the SPICE models have evolved to increased
accuracy, improvements in simulation speed have been small without going to higher levels
of abstraction, rules and guidelines to enhance the design of modern analog integrated
circuits (Alvarado et al., 2010; Beelen et al., 2010; Fakhfakh et al., 2010; McAndrew, 2010;
Muñoz-Pacheco & Tlelo-Cuautle, 2009; S. Steinhorst & L. Hedrich, 2010).
Behavioral modeling is performed according to the kind of application, for example not only
transistors models can be refined to work at radio frequency (RF) and microwave applications
(Gaoua et al., 2010), but also integrated resistors can be refined to include parasitic effects
(McAndrew, 2010). Additionally, transistors and parasitic elements can be modeled into
hardware description languages (Alvarado et al., 2010), so that the development time of
integrated circuits may be shrinked and the models can be tested before they are included
into commercial simulators, namely SPICE and ELDO.
An important issue is the application of symbolic analysis to generate analytical expressions
to describe the behavior of devices and circuits (Beelen et al., 2010; Tan & Shi, 2004). More
recently, McConaghy & Gielen (2009) introduced a template-free symbolic performance
modeling of analog circuits, mainly focused on operational transconductance amplifier
(OTA) based circuits. The application of symbolic analysis has also shown its usefulness in
parasitic-aware optimization and retargeting of analog layouts (Lihong et al., 2008). In fact,
the circuit design cycle covers different stages which can be performed in a hierarchical way,
from the specifications down to the layout, and from the extraction of layout-parasitics up to
the simulation of the whole circuit or system. In all cases, a refinement of the model is very
much needed at low- and high-level of abstraction (Ruiz-Amaya et al., 2005; Vasilevski et al.,
2009).
In some cases, symbolic analysis is combined with numerical simulation to perform
semi-symbolic behavioral modeling (Balik, 2009). Other important issues in behavioral
modeling of analog circuits is the generation of noise expressions (Martinez-Romero et al.,
2010), and the determination of dominant circuit-elements for the design of low-voltage
amplifiers (Tlelo-Cuautle, Martinez-Romero, Sánchez-López & X.-D. Tan, 2010).
Although many novel approaches for symbolic behavioral model generation have been
introduced for analog circuits, as recently reported in (Fakhfakh et al., 2010), yet the generation
of compact analytical expressions is an open problem. Some recent research has been
oriented to apply model order reduction (MOR) techniques (Qin et al., 2005; Shi et al.,
2006; Sommer et al., 2008; Tan & He, 2007), to capture the dominant behavior, but as
already mentioned in (Shi et al., 2006), a reduced symbolic expression is very difficult to
generate with MOR techniques. In this manner, this book chapter highlights some recent
developments in applying symbolic analysis to generate behavioral models of mixed-mode
integrated circuits (Bhadri et al., 2005; Krishna et al., 2007; McConaghy & Gielen, 2009;
Sánchez-López, Fernández & Tlelo-Cuautle, 2010; Sánchez-López & Tlelo-Cuautle, 2009; Tan
& Shi, 2004; Tlelo-Cuautle et al., 2009; Tlelo-Cuautle, Sánchez-López, Martinez-Romero & Tan,
2010; Tlelo-Cuautle, Sánchez-López & Moro-Frias, 2010).
In the following sections, we show the generation of behavioral models of mixed-mode
devices and circuits. This process is performed by using the nullor element to describe
the topology of the active devices and by applying symbolic nodal analysis to compute the
analytical expressions of the devices and circuits. Furthermore, to show the usefulness of the
generated symbolic behavioral models, they are used in the design process of an oscillator, for
which some insigths are derived in order to determine the circuit-element values and to speed
up circuit simulation. The chapter finishes by discussing some issues related to the application
of MOR techniques to approximate the dominant behavior of mixed-mode circuits, and the
generation of symbolic models including noise and distortion behavior.
2. Mixed-mode devices
In the analog domain, the input and output transfer relationships can be expressed by two
kinds of signals: voltage and current. When the signals are voltages, the device or circuit
is working in voltage-mode. This is the case of operational amplifier based circuits. On the
other hand, when the signals are currents, the device or circuit is working in current-mode.
However, when the device or circuit drives both voltage and current signals, it is working in
mixed-mode.
The first active device allowing the transfer of voltage and current was introduced in 1968
(Smith & Sedra, 1968), it was named current conveyor. Nowadays, the current conveyor
has evolved into three generations with direct and inverting characteristics (Tlelo-Cuautle,
Sánchez-López & Moro-Frias, 2010). All kinds of current conveyors work in mixed-mode
and basically they are composed of unity gain cells (Soliman, 2009; Tlelo-Cuautle,
Duarte-Villaseñor & Guerra-Gómez, 2008), which can be superimposed (Tlelo-Cuautle,
Behavioral Modeling of Mixed-Mode Integrated Circuits 87
Moro-Frias & Duarte-Villaseñor, 2008) to generate different kinds of active devices (Biolek
et al., 2008), all of them useful for analog signal processing applications. Among the unity gain
cells, the voltage mirror (Tlelo-Cuautle, Duarte-Villaseñor & Guerra-Gómez, 2008) and current
mirror can be modeled by using nullators and norators (Tlelo-Cuautle, Sánchez-López,
Martinez-Romero & Tan, 2010), but also they have the pathological representation introduced
in (Saad & Soliman, 2010), and they can be used to model the behavior of active devices with
inverting characteristics.
Although the current conveyor is a mixed-mode device, it can be used to implement
voltage-mode circuits such as active filters (Chen, 2010; Maheshwari et al., 2010). Some
mixed-mode integrated circuits implemented with other active devices can be found in
(and A. Bentrcia and S.M. Al-Shahrani, 2004; Bhadri et al., 2005; Soliman, 2007), and one
approximation to generate their behavioral models is given in (Krishna et al., 2007). The
modeling of all kinds of active devices by using controlled-sources can be found in (Biolek
et al., 2008). However, that models may generate systems of equations bigger than by using
nullors. For instance, in Fig. 1 are shown the models of the operational amplifier, OTA and
negative-type second generation current conveyor (CCII- (Tlelo-Cuautle, Sánchez-López &
Moro-Frias, 2010)), using nullors.
- -
+ + Y
gm
+ -
+ Z
X
(a) (b) (c)
Fig. 1. Modeling the (a) operational amplifier (opamp), (b) operational transconductance
amplifier (OTA), and (c) negative-type second generation current conveyor (CCII-) using
nullors
From the properties of the nullator whose voltage and current are zero (Sánchez-López,
Fernández & Tlelo-Cuautle, 2010), and for the norator whose voltage and current are arbitrary,
the active devices shown in Fig. 1 have the following relationships:
• From Fig. 1(a), the voltage and current at the input port of the opamp are zero due to the
properties of the nullator. At the output port, the voltage and current can be infinity due
to the property of the norator. Then, the ideal behavior of the opamp is well described by
using one nullator and one norator.
• From Fig. 1(b), the voltage across the conductance gm is just the differential voltage at the
input port because the voltage across each nullator is zero. Further, the current through gm
is the one leaving the output port of the OTA, i.e. i o = gm (v+ − v− ), where v+ − v− is the
differential voltage at the input port of the OTA.
• From Fig. 1(c), the property of the nullator generates iY = 0 and v X = vY , while the
property of the norator allows i Z = −i X . These three equations describe the ideal behavior
of the CCII-.
Among the mixed-mode active devices, the positive-type second generation current conveyor
(CCII+) is very versatile because if it is connected with a voltage follower, they describe
the current-feedback operational amplifier (CFOA). Both the CCII+ and CFOA are useful
to realize linear and nonlinear circuits (Sánchez-López, Trejo-Guerra, Muñoz-Pacheco &
Tlelo-Cuautle, 2010; Trejo-Guerra et al., 2010). Other useful mixed-mode active devices
are the transimpedance amplifier (van der Horst et al., 2010), operational transresistance
amplifier (OTRA) and current operational amplifier (COA) (Sánchez-López, Fernández &
Tlelo-Cuautle, 2010). In the following section we show how to generate the fully-symbolic
behavioral model of amplifiers and oscillators by including parasitic effects of the active
devices. For instance, when the analog circuits are modeled using nullors, their input-output
relationships can be generated by applying the symbolic nodal analysis (NA) method given
in (Sánchez-López et al., 2008; Sánchez-López & Tlelo-Cuautle, 2009; Tlelo-Cuautle et al.,
2009; Tlelo-Cuautle, Sánchez-López & Moro-Frias, 2010). The models used are very useful
for low frequency behavior, but for high frequency behavior yet one needs to investigate how
to approximate the gain, poles and zeros, noise and distortion. These aspects are discussed in
the following sections.
3. Behavioral modeling of analog circuits using pathological elements

Behavioral modeling has shown its advantages for successful development of analog
electronic design automation (EDA) tools due to various types of systems that can be
represented by means of an abstract model (Muñoz-Pacheco & Tlelo-Cuautle, 2009). The
abstraction levels indicate the degree of detail specified about how a function is to be
implemented. Therefore, behavioral models try to capture as much circuit functionality
as possible with far less implementation details than the device-level description of the
circuit. Some recent developments related to symbolic behavioral modeling can be found
in (Fakhfakh et al., 2010).
The generation of behavioral models is very useful to perform different design tasks, such
as synthesis (Saad & Soliman, 2008) and sizing (Diaz-Madrid et al., 2008). The applications
to analog design also include behavioral modeling of power (Suissa et al., 2010), carbon
nanotube field-effect-transistors (Chek et al., 2010), statistical modeling (Li et al., 2010),
efficient RF/microwave transistor modeling (Gaoua et al., 2010), etc. In all cases, the goal is not
only to capture the dominant behavior (Beelen et al., 2010), but also to generate refined models
to enhance high-level simulation (Alvarado et al., 2010; Vasilevski et al., 2009). The refinement
helps to approximate the behavior of circuits with strong nonlinearities (McAndrew, 2010; S.
Steinhorst & L. Hedrich, 2010), and to improve timing analysis (Hao & Shi, 2009), for instance.
The application of symbolic behavioral modeling approaches allows to perform sensitivity
analysis (Shi & Meng, 2009), which can be very useful to determine design-limits in designing
nonlinear circuits. For example, to determine the tuning range of mixed-mode quadrature
oscillators (Ansari & Maheshwari, 2009), the phase margin of opamps (Pugliese et al.,
2010), to identify the dominant circuit-elements in low-voltage amplifiers (Tlelo-Cuautle,
Martinez-Romero, Sánchez-López & X.-D. Tan, 2010), and to identify the noisy elements at
the transistor level of design (Martinez-Romero et al., 2010).
From the advantages infered above, we present the symbolic behavioral modeling of analog
circuits using the pathological elements: nullators and norators. The other two pathological
elements: voltage mirrors and current mirrors can be described as already shown in (Saad &
Soliman, 2010; Tlelo-Cuautle, Sánchez-López, Martinez-Romero & Tan, 2010).
Some examples for the generation of behavioral models for mixed-mode devices and circuits
are introduced in (Fakhfakh et al., 2010; Sánchez-López, Fernández & Tlelo-Cuautle, 2010;
Tlelo-Cuautle, Sánchez-López, Martinez-Romero & Tan, 2010; Tlelo-Cuautle, Sánchez-López
& Moro-Frias, 2010). In this subsection we show the model generation for simple
low-voltage amplifiers using the pathological elements nullators and norators (Tlelo-Cuautle,
Martinez-Romero, Sánchez-López & X.-D. Tan, 2010), because they are quite useful to perform
symbolic analysis by applying only nodal analysis (NA). Furthermore, to generate a symbolic
behavioral model we should replace every transistor and every non-NA-compatible circuit
element with their nullor-equivalent, as already shown in (Tlelo-Cuautle et al., 2009). Here,
we summarize the NA formulation (i = Yv) of analog circuits modeled with nullors.
1. Describe the interconnection relationships of norators (Pj ), nullators (O j ), and admittances
by generating tables including names and nodes.
2. Calculate indexes associated to set row and column to group grounded and floating
admittances:
• ROW: Contains all nodes ordered by applying the norator property which nodes (m, n )
are virtually short-circuited. These indexes are used to fill vector i and the admittance
matrix Y.
• COL: Contains all nodes ordered by applying the nullator property which nodes (m, n )
are virtually short-circuited. These indexes are used to fill vector v and the admittance
matrix Y.
• Admittances: They are grouped into two tables: Table A includes all nodes (ordered),
and in each node is the sum of all admittances connected to it. Table B includes all
floating admittances and its nodes (m, n ).
3. Use sets ROW and COL to fill vectors i and v, respectively. To fill Y: if in Table A a node
is included in ROW and COL, introduce that admittance(s) in Y at position (ROW index,
COL index). For each admittance in Table B, search node m in ROW and n in COL (do the
same but search n in ROW and m in COL), if both nodes exist the admittance is introduced
in Y at position (ROW index, COL index), and it is negative.
The solution of the NA formulation can be obtained by applying determinant decision
diagrams (DDD) (Fakhfakh et al., 2010; Tan & Shi, 2004).
Now we are able to generate the symbolic behavioral model of low-voltage amplifiers. Let’s
consider the common source amplifier with an active load shown in Fig. 2(a). Our goal is to
obtain its behavioral model expressed as a fully symbolic transfer function (Tlelo-Cuautle,
Martinez-Romero, Sánchez-López & X.-D. Tan, 2010). The first step consists to obtain its
nullor equivalent, which is shown in Fig. 2(b), where the input signal is the current source
emulating the voltage vin . As it can be seen, the input voltage from Fig. 2(a) was converted
into a current source using one nullator, one norator and one unity-resistor, making it an
NA-compatible element, i.e. an element which can be stamped directly into the nodal
analysis formulation, and also it does not increase the order of the system of equations, as
already shown in (Fakhfakh et al., 2010; Sánchez-López, Fernández & Tlelo-Cuautle, 2010;
Tlelo-Cuautle et al., 2009; Tlelo-Cuautle, Martinez-Romero, Sánchez-López & X.-D. Tan, 2010;
Tlelo-Cuautle, Sánchez-López, Martinez-Romero & Tan, 2010; Tlelo-Cuautle, Sánchez-López
& Moro-Frias, 2010).
The interconnection relationships of the nullators and norators is shown in Table 1, from which
the sets COL and ROW are generated as: COL ={(1,2,3),(4,5)}, and ROW = {(1),(3,4,5)}. This
means that the order of the admittance matrix is 2×2. The admittances are listed as shown in
Table 2, where only one admittance is floating. The formulation of the system of equations is
given by (1), and the solution for the behavioral model, i.e. the voltage transfer function, is
given by (2).

1 0 v1,2,3 vin
= (1)
gm1 − sCgd1 s ( Cgd1 + Cgs2 ) + go1 + go2 + gm2 v4,5 0
gm2
4 go2
Cgs2
O3 P3
5
1 2 Cgd1
O1
O2 P2
P1 go1
1 3
Vin Cgs1
gm1
(a) (b)
Fig. 2. (a) Low voltage amplifier with active load, and (b) Nullor equivalent.
Nullator Nodes Norator Nodes
O1 1,2 P1 2,0
O2 2,3 P2 3,5
O3 4,5 P3 4,5
Table 1. Data structure of nullators and norators from Fig. 2(b).
Table A Table B
Node Grounded Admittances Nodes Floating Admittances
1 1 2,5 sCgd1
2 sCgs1+sCgd1
3 gm1
4 gm2
5 go1 + go2 + sCgd1 + sCgs2
Table 2. Data structure of admittances from Fig. 2(b).
vo gm1 − sCgd1
=− (2)
vin s(Cgd1 + Cgs2 ) + go1 + go2 + gm2
Another example is taken from (Sanchez-Sinencio, 2009), the three stages uncompensated
low-voltage amplifier shown in Fig. 3, which nullor equivalent is given in Fig. 4. To
formulate the admittance matrix, we follow the steps provided above so that the sets COL
and ROW are (Tlelo-Cuautle, Martinez-Romero, Sánchez-López & X.-D. Tan, 2010): COL =
{(1,3,4), (2,8,9), (5,6,7), (10,13), (11), (15,16), (18)}, and ROW = {(1), (2), (4,5,6), (7,9,10), (11,12),
(13,14,15), (17,18)}. The admittance matrix is of order 7×7, and it is shown by (3). Following
the steps provided at the beginning of this section, the symbolic behavioral model, i.e. the
transfer function is given by (4). As one sees, the symbolic expression is very large, and it was
generated by using simple nullor equivalentes for the MOSFETs, i.e. every MOSFET from
Fig. 3 was modeled only with a nullor and its transconductance (some MOSFETs include
the output conductance to minimize error according to (Tlelo-Cuautle, Martinez-Romero,
Sánchez-López & X.-D. Tan, 2010)). Furthermore, were the parasitic capacitors of every
MOSFET be used, the expression in (4) becomes huge. A further step should be performed
to simplify large symbolic expressions which can also be done by applying model order
reduction approaches as shown in the following section.
M3 M4
M6
I ref
M1 M2 M8
Vin1 Vin2
Cp2 Vout
Cp1
M7
M5 CL
Mb M9
Fig. 3. Three stages uncompensated low voltage amplifier.
gm3 gm4 gm6
6 7 go4 13 go6
P3 O3 O5 P5 O8 P8
10
1 3 5
2 8 15
O1 O6
O7 P7 O10 P10
O2 P2
P1 P6 go2 go8
1 4 V 1 9 Cp1 16
Vin1 in2 Cp2
gm1 gm2 gm8
18
11
O4 P4 O9 P9 O11 P11
12 go5 14 17
go7 CL
go9
gm5 gm7 gm9
Fig. 4. Nullor equivalent from Fig. 3.

⎡ ⎤
1 0 0 0 0 0 0
⎢ 0 ⎥
⎢ 1 0 0 0 0 0 ⎥
⎢ ⎥
⎢ gm1 0 gm3 0 −gm1 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 gm2 gm4 go2 + go4 + sCp1 −go2 − gm2 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ −gm1 −gm2 0 −go2 gm1 + gm2 + go2 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎣ 0 0 0 gm6 0 go6 + go7 + sCp2 0 ⎦
0 0 0 0 0 −gm8 go8 + go9 + gm8 + sCL
(3)
vo gm8 ggm6 gm1 ( gm3 go2 + gm3 gm2 + gm4 go2 + gm4 gm2 )
=− (4)
vin D (s)
D(s)=gm2 gm3 go4 go6 go9+gm1 gm3 go2 go7 go9+gm1 gm3 go2 go7 gm8+ ( gm1 gm3 go2 go6 CL+
gm2 gm3 go4 Cp2 go8+ gm2 gm3 go4 Cp2 go9+gm2 gm3 go4 go6 CL+gm1 gm3 Cp1 go7 go9+
gm1 gm3 Cp1 go7 gm8+go2 gm3 go4 Cp2 go9+go2 gm3 go4 Cp2 go8+go2 gm3 Cp1 go6 go9+
go2 gm3 Cp1 go6 gm8+go2 gm3 go4 go6 CL+go2 gm3 go4 go7 CL+gm4 gm1 go2 Cp2 go9+
gm2 gm3 Cp1 go7 go8+gm2 gm3 Cp1 go7 go9+gm2 gm3 Cp1 go7 gm8+gm4 gm1 go2 go6 CL+
gm4 gm1 go2 go7 CL+gm4 gm1 go2 Cp2 go8+go2 gm3 go4 Cp2 gm8+go2 gm3 Cp1 go6 go8+
go2 gm3 Cp1 go7 go8+go2 gm3 Cp1 go7 go9+go2 gm3 Cp1 go7 gm8+gm1 gm3 go4 Cp2 go8+
gm1 gm3 go4 Cp2 go9+gm1 gm3 go2 go7 CL+gm1 gm3 go2 Cp2 go8+gm1 gm3 go2 Cp2 go9+
gm4 gm1 go2 Cp2 gm8+gm1 gm3 go4 Cp2 gm8+gm1 gm3 Cp1 go6 go8+gm1 gm3 Cp1 go6 go9+
gm1 gm3 Cp1 go6 gm8+gm1 gm3 Cp1 go7 go8+gm2 gm3 Cp1 go6 go9+gm2 gm3 Cp1 go6 gm8+
gm1 gm3 go2 Cp2 gm8+gm1 gm3 go4 go6 CL+gm1 gm3 go4 go7 CL+gm2 gm3 go4 go7 CL+
gm2 gm3 go4 Cp2 gm8+gm2 gm3 Cp1 go6 go8 ) s+gm2 gm3 go4 go6 go8+gm2 gm3 go4 go6 gm8+
go2 gm3 go4 go6 gm8+ ( gm2 gm3 Cp1 Cp2 CL+gm1 gm3 Cp1 Cp2 CL+go2 gm3 Cp1 Cp2 CL ) s3
+gm1 gm3 go4 go6 gm8+ gm1 gm3 go4 go7 go8+gm1 gm3 go4 go7 go9+gm1 gm3 go4 go7 gm8+
gm4 gm1 go2 go6 gm8+gm2 gm3 go4 go7 go8+gm2 gm3 go4 go7 go9+gm2 gm3 go4 go7 gm8+
gm4 gm1 go2 go7 go9+gm4 gm1 go2 go7 gm8+ ( gm1 gm3 Cp1 Cp2 go8+gm2 gm3 Cp1 Cp2 go9+
gm1 gm3 Cp1 go7 CL+gm1 gm3 Cp1 Cp2 go9+go2 gm3 Cp1 Cp2 gm8+gm1 gm3 Cp1 Cp2 gm8+
go2 gm3 Cp1 Cp2 go9+gm2 gm3 go4 Cp2 CL+go2 gm3 Cp1 Cp2 go8+gm2 gm3 Cp1 Cp2 go8+
gm2 gm3 Cp1 Cp2 gm8+go2 gm3 go4 Cp2 CL+go2 gm3 Cp1 go6 CL+gm2 gm3 Cp1 go6 CL+
gm1 gm3 go2 Cp2 CL+gm1 gm3 go4 Cp2 CL+go2 gm3 Cp1 go7 CL+gm2 gm3 Cp1 go7 CL+
gm4 gm1 go2 Cp2 CL+gm1 gm3 Cp1 go6 CL ) s2 +gm4 gm1 go2 go6 go9+go2 gm3 go4 go6 go8+
go2 gm3 go4 go7 gm8+go2 gm3 go4 go6 go9+go2 gm3 go4 go7 go8+go2 gm3 go4 go7 go9+
gm4 gm1 go2 go7 go8+gm1 gm3 go2 go6 go8+gm1 gm3 go2 go6 go9+gm1 gm3 go2 go6 gm8+
gm1 gm3 go2 go7 go8+gm1 gm3 go4 go6 go8+gm1 gm3 go4 go6 go9+gm4 gm1 go2 go6 go8
If the low voltage amplifier is designed with standard CMOS integrated circuit technology, its
gain performance comparison with respect to its behavioral model given by (4) is shown in
Fig. 5. To minimize the error it is necessary to include more symbolic elements, as shown in
the following section. However, the symbolic expression becomes huge originating a trade-off
between the size of the exact symbolic behavioral model and the allowed error compared with
HSPICE simulation.
Fig. 5. Comparison between HSPICE and (4).
4. Simplification approaches
To simplify the symbolic expression given in (4), several approaches can be found in
(Fakhfakh et al., 2010). Those approaches combine numerical and symbolic techniques to
reduce the analytical expression. For instance, the expression reduction can be performed
by the application of three complementary methods: simplification before generation (SBG)
techniques (negligible elements are pruned from the circuit, graph or matrix associated to
the circuit equation formulation); simplification during generation (SDG) techniques (only
the significant parts of the symbolic expressions are generated); and simplification after
generation (SAG) techniques (least significant terms are pruned from the symbolic expression
resulting from the previous approximate analysis steps).
Both, SBG and SDG approaches are usually tied to the kind of analysis method used. In
this way, some SBG methods operate at the matrices resulting from analysis methods like
nodal analysis. The approaches in (Hsu & Sechen, 1994; Sommer et al., 1993) eliminate device
parameters from each cofactor of the nodal matrix if the error induced in the cofactor is below
a given error threshold. Concurrently with the device parameter elimination, this technique
tries to reduce determinant dimension by factoring out rows and columns with only one
nonzero entry and performs row and column operations to reduce the number of symbols
or nonzero entries. Other methods by (Guerra et al., 1998; Yu & Sechen, 1996) operate at
the graph level; usually, the voltage and current graphs, as the two-graph method has been
demonstrated to be the most efficient symbolic analysis method (Wambacq et al., 1996). In
this case, graph branches are removed or its terminal nodes are contracted if their contribution
(appropriately) measured to the transfer function is sufficiently small. In all cases, an adequate
error mechanism is needed to control which matrix entries can be deleted or graph branches
can be deleted and graph nodes contracted without exceeding some prescribed maximum
magnitude/phase errors. Most approaches (Hsu & Sechen, 1994; Sommer et al., 1993; Yu
& Sechen, 1996) perform the evaluation of the contributions to the network function of the
elimination of matrix entries or the successive node contractions and branch removals at a set
of frequency samples within the range being considered. An obvious trade-off between the
number of frequency samples (directly related to computational time) and the possibility of
exceeding the maximum errors between frequency samples exist. An exception is the efficient
approach in (Fernández et al., 1998; Guerra et al., 1998; Rodriguez-Garcia et al., 1999), that
selects a small set of frequency samples, uses interval analysis techniques to detect if the
error is exceeded in some intermediate frequency and new frequency samples are added
accordingly.
SDG techniques generate symbolic terms in decreasing order of magnitude until the number
of terms is enough to model the behavior of the circuit with a given accuracy. Term
generation algorithms in decreasing order of magnitude were originally developed for the
two-graph approach. The modeling of analysis problems in terms of matroids has allowed
the term generation with many other methods but the two-graph (voltage and current
graph) method still remains as the most efficient one. Valid symbolic terms corresponds to
ordered enumeration of common spanning trees to both graphs, that in terms of matroids it
corresponds to the ordered enumeration of bases common to two graphic matroids. If the
terms must belong to a given power of the complex frequency s of functions like (2) and (4),
then the bases must be also common to a partition matroid, determined by spanning trees
that have a fixed number of frequency-dependent elements. Although there are efficient
algorithms for the ordered enumeration of bases common to two matroids, the ordered
enumeration of bases common to three matroids is in general a NP-complete problem. The
possible alternatives are linked to the error control mechanism used:
• Enumerate bases common to the partition matroid and one of the graphic matroids, and
for each one, check if it is also a base of the other graphic matroid (Wambacq et al., 1995;
Yu & Sechen, 1995). The generation algorithm is most efficient known but many generated
bases may not be common to the three matroids. In this case, error mechanisms that control
the error in each coefficient of the transfer function can be used, for instance, by means of
a sensitivity driven mechanism (Daems et al., 1999).
• Enumerate bases common to the two graphic matroids (therefore, admittance of frequency
dependent elements must be evaluated at a given frequency) and for each one, check if
it contains the required number of frequency-dependent elements. The frequency can be
selected via a sensitivity-driven mechanism that increases the probability of generation of
terms with the desired number of frequency-dependent elements (Wambacq et al., 1998).
The same error control mechanism than in the previous case can be used.
• Enumerate bases common to the two graphic matroids for a given frequency. One
possibility is to enumerate bases at several frequencies and merge the results (Yu &
Sechen, 1997). Another possibility to avoid the use of an excessive number of frequencies,
generation of unnecessary terms and possible error excesses between frequency samples is
to use a similar sampling approach and error control mechanism to the SBG case: a reduced
number of samples, detection of error excesses by interval analysis and step-by-step
addition of sampling frequencies (Guerra et al., 1998).
Special attention deserves the approximation of symbolic poles and zeros of transfer functions.
Extraction of poles and zeros from symbolic transfer functions is subject to strong limitations
for two reasons:
1. The maximum polynomial order that can be extracted analytically is limited to four (in
practice, for symbolic roots it is limited to two);
2. Approximation of the transfer function under magnitude and phase error control
mechanisms does not map to controlled pole and zero errors.
For these reasons, more powerful approaches specifically devoted to symbolic pole and
zero extraction have been developed. One of these techniques (Henning, 2002) applies
a simplification before generation technique based on the approximation of the nodal
admittance matrix for a selected eigenvalue by ranking the eigenvalue shifts induced
by different device parameter eliminations and performing the least significant device
prunings while some error criterion in the eigenvalue shift is met. The eigenvalue shift
is obtained from a linear prediction formula derived from a Taylor series approximation
of the generalized eigenvalue problem, similar to the sensitivity analysis above, yielding a
ranking of candidate parameter eliminations. The approach in (Guerra et al., 2002) exploits
the Haley’s modification-decomposition method (Haley, 1991) to transform the generalized
eigenvalue problem into a standard eigenvalue problem. This new formulation can use
the efficient QR algorithm to numerically track pole and zero errors, and it contains a
time-constant matrix whose entries can be calculated symbolically very efficiently. This opens
the possibility to apply simplification before generation techniques at the matrix level (by
selecting only the appropriate entries of matrix T, entries that correspond to analysis of simple,
purely resistive circuits), simplification before generation techniques at the circuit level (by
eliminating negligible devices of the resistive circuit associate to each entry of interest) and
simplification during generation techniques at the circuit level (by applying conventional SDG
techniques to simplified, purely resistive circuits).
Model order reduction (MOR) technique is another simplification approach (Qin et al., 2005;
Shi et al., 2006; Tan & He, 2007). Besides, the technique based on the asymptotic waveform
evaluation (AWE) approach (Qin et al., 2005; Tan & He, 2007), can be applied to reduce the
order of the behavioral model either symbolic (Shi et al., 2006), or numerically (Sommer et al.,
2008). In the rest of this section we show the drawbacks when performing a fully-symbolic
AWE approach. Let’s us consider the circuit in Fig. 3, by replacing each MOSFET with
its nullor equivalent including two parasitic capacitors (connected between gate-source and
gate-drain), the nullor circuit is shown in Fig. 6. Compared to the nullor circuit in Fig. 4, in
this case there are many floating admittances, so that the system of equations grows and so
the size of the symbolic behavioral model.
For this example, the ROW and COL sets are: ROW = (1), (2), (4,5,6), (7,9,10), (11,12), (13,14,15),
(17,18), and COL = (1,3,4), (2,8,9), (5,6,7), (10,13), (11), (15,16), (18). Twelve admittances are
floating ones, so that the formulation includes the generation of Table A and Table B, as it was
done for the low voltage amplifier with active load described above.
The generation of the fully symbolic transfer function from Fig. 6 leads to a fifth order
denominator. For instance, when the uncompensated amplifier is designed with standard
CMOS integrated circuit technology, and by replacing every symbol-circuit-element with its
numerical value computed from an HSPICE simulation, the rational "s-domain" fifth order
function is given by (5).
1536(2.38E8s5 + 2.75E20s4 − 4.4E30s3 − 4.19E41s2 − 5.69E51s − 1.94858E61)

H (s) = (5)
7.67E18s5 + 1.32E29s4 + 3.5E38s3 + 1.51E41s2 + 3.29E54s + 1.29637E61
Besides, usually a second order polynomial is very sufficient to approach the behavior of an
amplifier in analog design. In this manner, AWE is very useful to generate a reduced order
behavioral model through Padé approximation. The main operations can be found in (Qin
gm3 gm4 gm 6
go 3
6 Cgs 3 Cgs 4 go 4 Cgs 6 13 go 6
7
Cgd 4 Cgd 6
5 10
1 3 Cgd 1 Cgd 2 Cgd 8
2 8
go2 Cp2 go 8
1 4 go 1 V 1 9 Cp1 16
Vin1 in2
Cgs 1 Cgs 2 Cgs 8
gm 1 gm2 gm8
15
11
18
Cgd 5 Cgd 7 Cgd 9
12 go 5 14 17
go 7 CL
go 9
gm5 gm 7 gm9
Fig. 6. Nullor equivalent from Fig. 3 including parasitic capacitors to all MOSFETs.
et al., 2005; Tan & He, 2007). Basically, from the computation of the symbolic transfer function
(H(s)), one evaluates derivatives to compute moments with respect to the variable "s". The
iterative formula is given by (6).
1 dk H (s)
| mk = (6)
k! dsk s =0
Afterwards, Padé approximation generates an expression with reduced order of the form:
P (s) a0 + a1 s + a2 s2 + . . . + a p s p
H p,q (s) = = (7)
Q(s) 1 + b1 s + b2 s 2 + . . . + b q s q
The coefficients a p and bq can be obtained by solving two system of equations, for numerator
and denominator, respectively. As already described in (Tan & He, 2007). As one can infer,
doing this work fully-symbolically to generate a fully-symbolic behavioral model instead
of the rational expression in (5), is very time-consuming and it requieres a lot of memory.
For instance, the fully symbolic moment m0 is given by a very large expression where the
numerator is espressed as:
-gm8 gm6 gm1 ( go3 gm2+go2 gm4+go3 go2+gm3 gm2+gm2 gm4+gm3 go2+go5 gm4 )
And the denominator is expressed by:

go1 gm3 go2 go7 gm8+go8 go6 go1 go2 go3+go1 gm2 go4 go6 go9+go1 go5 go4 go6 gm8+go1 gm4 go2 go7 go8+
go1 gm2 go4 go6 go8+go1 go5 go4 go6 go9+go1 go5 go2 go6 go9+go1 gm3 go4 go7 gm8+go1 go2 go4 go6 go9+
go1 go3 go2 go6 gm8+go8 go6 go1 go2 gm3+go8 go7 go1 go2 gm3+go9 go7 go1 go2 gm3+ go1 gm4 go2 go6 go8+
go1 gm3 go2 go6 gm8+go1 gm4 go2 go7 go9+go1 go5 go2 go6 go8+go9 go6 go1 go2 go3+go1 go5 go2 go6 gm8+
go1 go3 go4 go7 gm8+go1 go5 go4 go7 gm8+go9 go6 go1 go2 gm3+go1 gm2 go4 go6 gm8+go1 gm2 go4 go7 go8+
go1 go3 go4 go6 gm8+go1 gm2 go4 go7 gm8+go8 go7 go1 go2 go3+ go1 go2 go4 go7 gm8+go1 go3 go2 go7 gm8+
go1 gm4 go2 go7 gm8+go1 gm4 go2 go6 gm8+go1 go5 go2 go7 go8+go1 gm3 go4 go6 go8+go9 go7 go1 go2 go3+
go1 gm4 go2 go6 go9+go1 gm3 go4 go6 go9+go1 gm3 go4 go7 go8+go1 go2 go4 go6 gm8+go3 gm1 go2 go6 go8+
go3 gm1 go2 go7 go8+go3 gm1 go2 go7 go9+ go3 gm1 go2 go7 gm8+go3 gm1 go2 go6 go9+go3 gm1 go2 go6 gm8+
go3 gm1 go4 go6 gm8+go3 go2 go4 go7 go9+gm3 go5 go4 go7 gm8+go3 go5 go2 go6 go8+go3 go5 go2 go7 go8+
go3 go5 go2 go7 go9+go3 go5 go2 go7 gm8+go3 go5 go2 go6 go9+go3 go5 go2 go6 gm8+gm3 go5 go2 go7 go9+
gm3 go5 go4 go6 gm8+ go3 gm2 go4 go6 gm8+go3 go5 go4 go6 gm8+go3 gm1 go4 go6 go9+go3 gm1 go4 go6 go8+
go3 gm1 go4 go7 go8+go3 gm1 go4 go7 go9+go3 gm2 go4 go6 go9+go3 gm2 go4 go6 go8+go3 gm2 go4 go7 go8+
go3 gm2 go4 go7 go9+go3 go2 go4 go7 gm8+gm3 gm1 go2 go6 go8+gm3 gm1 go2 go7 go8+gm3 gm1 go2 go7 go9+
gm3 gm1 go2 go7 gm8+gm3 gm1 go2 go6 go9+gm3 gm1 go2 go6 gm8+go3 go5 go4 go6 go9+go3 go5 go4 go6 go8+
go3 go5 go4 go7 go8+go3 go5 go4 go7 go9+gm3 gm1 go4 go6 gm8+gm3 gm2 go4 go6 go9+gm3 gm2 go4 go6 go8+
gm3 gm2 go4 go7 go8+gm3 gm2 go4 go7 go9+gm3 gm2 go4 go6 gm8+gm3 go5 go2 go6 go8+ gm3 go5 go2 go7 go8+
gm3 go5 go2 go7 gm8+gm3 go5 go2 go6 go9+gm3 go5 go2 go6 gm8+go3 gm1 go4 go7 gm8+go3 go2 go4 go6 gm8+
gm3 go2 go4 go6 go9+gm3 go2 go4 go6 go8+gm3 go2 go4 go7 go8+gm3 go2 go4 go7 go9+gm4 gm1 go2 go6 go8+
gm4 gm1 go2 go7 go8+gm4 gm1 go2 go7 go9+gm4 gm1 go2 go7 gm8+ gm4 gm1 go2 go6 go9+gm4 gm1 go2 go6 gm8+
gm3 gm2 go4 go7 gm8+go1 go5 go2 go7 go9+go1 gm3 go4 go7 go9+go1 go5 go2 go7 gm8+go1 go2 go4 go7 go9+
go1 go2 go4 go6 go8+go1 gm2 go4 go7 go9+go1 gm3 go4 go6 gm8+go1 go2 go4 go7 go8+go1 go5 go4 go7 go8+
gm3 gm1 go4 go7 gm8+gm3 go2 go4 go7 gm8+ gm3 gm1 go4 go7 go9+gm3 gm1 go4 go6 go9+gm3 gm1 go4 go6 go8+
gm3 gm1 go4 go7 go8+go3 gm2 go4 go7 gm8+go3 go2 go4 go6 go9+go3 go2 go4 go6 go8+go3 go2 go4 go7 go8+
go3 go5 go4 go7 gm8+gm3 go5 go4 go6 go9+gm3 go5 go4 go6 go8+gm3 go5 go4 go7 go8+gm3 go5 go4 go7 go9+
gm3 go2 go4 go6 gm8+ go1 go5 go4 go6 go8+go1 go3 go4 go6 go9+go1 go3 go4 go6 go8+go1 go3 go4 go7 go8+
go1 go3 go4 go7 go9+go1 go5 go4 go7 go9
It can be clearly infered that the size of the expressions for the next moments by applying (6)
may grow exponentially, and when the Padé approximation is done by generating (7), much
more memory will be needed. Fortunately, in (Fakhfakh et al., 2010) there is a simplification
approach for analog integrated circuits designed with MOSFETs, so that the behavioral model
for the uncompensated low voltage amplifier can be reduced to a third order described by:
− gm1 gm6 gm8

H (s) = (8)
CL C p1 C p2 s3 + C p1 C p2 s2 + ( c p1 ( go6 + go7 ) + C p2 ( go2 + go4 )) gm8 s + gm8 ( go2 + go4 )( go6 + go7 )
From this example, it can be appreciated that a combination of symbolic and numerical model
order reduction approaches can be very useful to generate simplified behavioral models of
analog integrated circuits.
5. Behavioral modeling of sinusoidal oscillators

This section provides an overview of the usefulness of generating symbolic behavioral models
in the design of sinusoidal oscillators implemented with mixed-mode devices, such as the
operational transresistance amplifier (OTRA). This device can also be designed with standard
CMOS integrated circuit technology, so that an accurate simplified symbolic behavioral model
can be difficult to generate. The following section approaches the behavior of the OTRA with a
simple transfer function. This analytical expression can be used in a higher abstraction level,
e.g. into Verilog-A as already shown in (Tlelo-Cuautle, Duarte-Villaseñor, Garcia-Ortega &
Sánchez-López, 2007; Tlelo-Cuautle, Sánchez-López, Fakhfakh & Loulou, 2007), to accelerate
design development time.
The OTRA is an important building block in mixed-mode analog integrated circuit design.
One reason is that circuit designers have been focusing their attention on analog signal
processing applications extended to high-frequency by using current-mode techniques
(Hwang et al., 2009). Although the OTRA is commercially available in bipolar technology,
it does not provide a virtual ground at the input terminals and only allows the input current
to flow in one direction 1 .
Recent realizations have been suggested to design OTRA based circuits in multiple-mode
(Lee, 2010), e.g. voltage, current, transconductance, and transresistance modes. Among the
applications of OTRAs one can find implementations such as: instrumentation amplifiers,
integrators, continuous-time filters, immitance simulators, waveform generators, bistable
multivibrators, oscillators and amplification of signals from current-source transducers.
These applications using OTRAs overcome the finite gain-bandwidth product associated
to conventional opamps. Additionally, both the inputs and outputs of the OTRA are low
impedance terminals, that way, all parasitic capacitors will have little effect and the time
response limitations incurred by parasitic capacitors can be minimized (Chen et al., 2001).
On the other hand, since the OTRA is a high gain current-input voltage-output device, it can
be considered as a current-to-voltage converter and its behavior can be modeled by using a
current-controlled voltage source (CCVS). This CCVS can be modeled using nullors, so that
once again, as already shown in the previous section, we are able to apply the symbolic NA
method to generate symbolic behavioral models of OTRA based circuits.
In this section, a new nullor-based model for the OTRA, which is composed by four nullors
and three grounded resistors is introduced. In this manner, the symbolic NA method can
easily be applied to compute small-signals characteristics of OTRAs-based analog circuits.
The nullor-based model not only reduces the admittance matrix size, if it is compared with
the element stamp method, but also analog circuits with both inputs currents flowing toward
the OTRA can easily be analyzed.
The OTRA is a three-terminal analog building block, where its input-output terminals are
characterized by low impedances, and its behavior can be described as already shown in
(Hwang et al., 2009). Since external negative feedback is required for OTRA based analog
circuits, it is better to design an OTRA with high DC open-loop gain.
To show the usefulness of the nullor equivalent of the OTRA, lets us consider the
OTRAs-based oscillator shown in Fig. 7. The system of equation by applying the symbolic NA
method is given by (9). The evaluation of the determinant of the admittance matrix generates
the characteristic equation given by (10).
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
− sC1 0 1 0 v3,9 0
⎢ 1
− R1
1 ⎥
1 0 ⎥ ⎢ v6,13 ⎥ ⎢ ⎥
⎢ Rm1 ⎥= ⎢ 0 ⎥
⎢ ⎥ ⎢⎣ ⎦ ⎣ ⎦ (9)
⎣ − 1
R2 − sC2 − 1
Rs 0 1 ⎦ v 7,8 0
0 1 − 1 0 1 v11,12 0
R m2 R4
1 1 1 1 1 1 1 1 1 1
s2 + ( + + ( − ))s + + ( + − ) (10)
Rm1 C1 Rm2 C2 C2 Rs R4 R1 R2 C1 C2 Rm1 C1 C2 Rm2 Rs R4
Since the gain of the OTRA is finite the two-pole behavioral model can be described by (11).
Where ω p1 and ω p2 are the angular frequencies of the first and second pole and Rm0 is the DC
gain of the OTRA.
1 National Semiconductors Corp., Designing with a new super fast dual norton amplifier. Linear
Applications Data Book, 1981.
National Semiconductors Corp., The LM 3900: a new current differencing quad of the input amplifiers.
Linear Applications Data Book, 1986.
Rm0
Rm (s) = (11)
(1 + s )(1 + s )
ω p1 ω p2
For middle frequency applications, the transfer of the OTRA denoted by Rm (s) can be
expressed by (12). Therefore, since | s = jω | < < ω p2 (10) is approached by (13).
1 1
Rm (s) = ; Cm = (12)
sCm (1 + ω p2 )
s Rm0 ω p1
1 1 1 1
s2 + ( − )s + (13)
C2 + Cm2 Rs R4 R1 R2 (C1 + Cm1 )(C2 + Cm2 )
That way, the condition and frequency of oscillation are given by (14).
1
R3 = R4 , ωo =
(14)
R1 R2 (C1 + Cm1 )(C2 + Cm2 )
R2 R3
C1 C2
3
R2 R3
C1 C2 1 1 1 1
3 4 6 1 4 6
1 8 12
7 11
Rm1(s) Rm2(s)
2
+ 5
+ 9 13
R4 2 5
R1
Rm1(s) R4 Rm2(s)
R1
(a) (b)
Fig. 7. (a) OTRAs-based oscillator taken from (Salama & Soliman, 2000), and (b) nullor
equivalent.
The oscillator in Fig. 7 was designed and simulated using HSPICE to verify its behavior at
several frequencies. By choosing R1 =R2 =2kΩ, R3 =R4 =10kΩ, the value of the frequencies of
oscillation are shown in Fig. 8 as:
f 1 =2.65MHz (Dashed-line) with C1 =C2 =24pF,

f 2 =6.29MHz with C1 =C2 =6.46pF (Dotted-line), and
f 3 =12MHz (Solid line) with C1 =C2 =0.1pF,
The parasitic capacitances were calculated by applying (12) so that they were approximated to
Cm1 =Cm2 =6.46pF. On the other hand, the calculation of the frequencies of oscillation for very
high frequency applications, needs to be performed by applying the approximation given in
(15). In this manner, we obtain f 1 =2.61MHz, f 2 =6.16MHz and f 3 =12.1MHz which are in good
agreement with the simulated results.
ω p2
Rm (s) = (15)
s 2 Cm
From Fig. 8 one can observe that the maximum frequency of oscillation ( f 3 =12MHz) is limited
by Cm1,2 =CZ2 according to (12), with Rm0 =R Z2 , where R Z2 and CZ2 are the parasitic resistance
and capacitance associated to the Z terminal of the commercially available AD844AN.
Fig. 8. Time responses of the OTRA-based sinusoidal oscillator
6. Noise and distortion behavioral modeling

Symbolic analysis has been demonstrated its usefulness in computing second order effects
such as noise and distortion, some works can be found in (Fakhfakh et al., 2010). For the noise
behavioral modeling approach, the nature of the equations in noise analysis allows applying
determinant decision diagrams (DDD)s (Shi & Tan, 2000), to improve the calculation of noise
expressions (Martinez-Romero et al., 2010).
The generation of behavioral expressions for noise figure, input and output noise for analog
circuits is presented in (Tlelo-Cuautle & Sánchez-López, 2004). In this section we show the
results of the calculation of the output noise for the three stages uncompensated low voltage
amplifier shown in Fig. 3. As already shown in (Martinez-Romero et al., 2010), the behavioral
noise expression is compared with Hspice simulations using the level zero Spice 2 models.
The equivalent circuit for noise analysis of the three stages amplifier is shown in Fig. 9.
The details on the formulation of the system of equations is provided in (Martinez-Romero
et al., 2010). It generates an admittance matrix of order 5×5, which can be easily solved
by applying DDDs. The symbolic behavioral expression of the output noise is also given
in (Martinez-Romero et al., 2010), while the comparison between HSPICE and the evaluation
of the symbolic expression is given in Fig. 10. The error can be minimized by addition of other
noisy elements but the symbolic expression can increase. Other examples related to symbolic
noise behavioral modeling are provided in (Fakhfakh et al., 2010).
gm 3 gm 4 gm6
2
In M 3 go4
2
3 4
I n M4
9 go 6 I 2n M 6
8
2
go 2
go 1 go 8
1 I 2n M 1 5 11 12
I 2n M 2 I 2n M 8
gm1 gm 2 gm 8
14
7 2 10 2 13 2
I n M5 go 7 I n M7 In M 9
go 9
gm5 gm 7 gm 9
Fig. 9. Nullor equivalent from Fig. 3.
Fig. 10. Symbolic and HSPICE noise responses for the uncompensated amplifier.
For the symbolic distortion behavioral modeling, an analysis approach is presented by
(Floberg, 1997), it deals with bipolar transistor circuits. However, due to the difficulty
to generate analytical expressions in circuits with hard-distortion and at high-frequencies
(Wambacq et al., 1999), the distortion analysis is generally performed for weakly nonlinear
circuits (Li & Pileggi, 2005). In this case, the application of symbolic analysis for behavioral
model generation is suitable for the distortion analysis in single-, two- and three-stage
amplifiers (Hernes & Sansen, 2005).
Combinations of symbolic methods with numerical analysis for nonlinear circuits are
presented in (Daems et al., 2002; Manthe et al., 2003). Besides, recent developments are
oriented to nonlinear Model Order Reduction of Analog/RF Circuits.
For instance, abstracting transistor-level circuit details that include important weakly
nonlinear effects into a compact macromodel can be instrumental in assisting the analysis
and design of analog and RF circuits. For many such applications, while circuit blocks often
exhibit weak nonlinearities, the design specification for linearity is often extremely important
and very stringent. Hence, it is important to be able to model distortions in a compact and
accurate way.
Along this line, a number of research attempts have emerged in the literature. Symbolic
modeling of weakly nonlinear circuits has been used to build system-level models (Wambacq
et al., 2000; Wambacq & Sansen, 1998), by using the notation of Volterra series. Neural network
and time series models have also been proposed for nonlinear modeling (Root et al., 2003).
Nonlinear reduction techniques have been studied, which may target only strongly nonlinear
behaviors, or may include both weakly and nonlinear aspects (Dong & Roychowdhury, 2003;
Rewienski & White, 2001).
For many applications where weakly nonlinear distortions are important aspects of design
specifications, Volterra series provides a good choice for system description. In (Phillips, 2000;
Roychowdhury, 1999), the projection-based nonlinear model order-reduction frameworks for
weakly nonlinear systems were first developed by extending moment-matching projection
techniques used for interconnect modeling. Here the basic idea is to view a weakly nonlinear
system as a set of interconnected linear networks and then each of such linear circuits is
reduced via model order reduction.
While the concept of projection-based model order reduction is highly relevant for nonlinear
distortion modeling, it is worthy noting that the reduced model compactness is critical
for effective nonlinear model reduction. Without proper handling, resulting size of a
projection-based nonlinear reduced model tends to grow rapidly.
To this end, the most general matrix-form nonlinear transfer functions, or in other words,
frequency-domain Volterra kernels, are used as a starting point for nonlinear model order
reduction (Li & Pileggi, 2003; 2005). In this so-called NORM algorithm, to disclose the problem
structure of nonlinear model order reduction, moments of nonlinear transfer functions and
associated Krylov subspaces have been derived in the matrix form. With this, relationships
between Krylov subspace projection and nonlinear transfer function moment matching are
understood. Using this as a basis, the model size is further optimized for a targeted number
of matched moments, leading to significant improvements on the model compactness. The
reduced order model structure can be tailored by controlling the moment matching orders for
different orders of nonlinearity in a coherent fashion. Furthermore, it is shown that multipoint
expansions for projection-based nonlinear model order reduction is advantageous in terms of
model compactness at the expanse of additional computational cost (Li & Pileggi, 2003; 2005).
Under the same based framework, weakly nonlinear distortions of time-varying RF circuits
can be also captured.
7. Conclusion
We have presented the symbolic behavioral model generation of mixed-mode circuits. It
was shown that the use of the nullor properties allows us not only to describe the dominant
behavior of active devices, but also to add or remove parasitic elements in order to generate
simplified analytical expressions. Furthermore, the nullor equivalent of a mixed-mode circuit
is suitable to formulate a compact system of equations by applying nodal analysis.
Several examples were presented to demonstrate the usefulness of the nullor models
to generate symbolic behavioral expressions of mixed-mode circuits and of sinusoidal
oscillators.
A new nullor-based model for the mixed-mode device named operational transresistance
amplifier (OTRA) was introduced and it was used to compute small-signal characteristics
of a sinusoidal oscillator oriented to circuit design.
Some open research problems in the generation of symbolic behavioral models were listed
along the chapter. That problems may be solved by using the properties of the nullor and
nodal analysis, and by applying model order reduction (MOR) techniques. Finally, to generate
a simplified behavioral model, it could be much better to combine numerical and symbolic
approaches and to develop new MOR approaches to deal with analog VLSI circuits.
8. Acknowledgment
This work is partially supported by CONACyT through the grant for the sabbatical stay
of the first author at University of California at Riverside, during 2009-2010. The authors
acknowledge the support from UC-MEXUS-CONACYT collaboration grant CN-09-310; by
Promep México under the project UATLX-PTC-088, and by Consejeria de Innovacion Ciencia
y Empresa, Junta de Andalucia, Spain, under the project number TIC-2532. The third author
thanks the support of the JAE-Doc program of CSIC, co-funded by FSE.
9. References
Alvarado, J., Iniguez, B., Estrada, M., Flandre, D. & Cerdeira, A. (2010). Implementation of the
symmetric doped double-gate MOSFET model in Verilog-A for circuit simulation,
International Journal of Numerical Modelling: Electronic Networks, Devices and Fields
23(2): 88–106.
and A. Bentrcia and S.M. Al-Shahrani, M. A. (2004). A novel mixed-mode
current-conveyor-based filter, International Journal of Electronics 91(3): 191–197.
Ansari, M. & Maheshwari, S. (2009). Electronically tunable MOSFET-C mixed-mode
quadrature oscillator, International Multimedia, Signal Processing and Communication
Technologies, pp. 158–160.
Balik, F. (2009). A semi - symbolic method of electronic circuit design by pole and zero
distribution optimization using time - constants approximation, IEEE International
Conference on Electronics, Circuits and Systems, pp. 848–851.
Beelen, T., ter Maten, E., Sihaloho, H. & van Eijndhoven, S. (2010). Behavioral modeling
of the dominant dynamics in input-output transfer of linear(ized) circuits, Procedia
Computer Science 1(1): 347–355.
Bhadri, P., Srinivasan, R., Mal, P., Beyette, F. & Carter, H. (2005). Mixed mode integrated
circuits, IEEE Potentials 24(1): 6–11.
Biolek, D., Senani, R., Biolkova, V. & Kolka, Z. (2008). Active elements for analog signal
processing: Classification, review, and new proposals, Radioengineering 17(4): 15–32.
Chek, D., Tan, M., Ahmadi, M., Ismail, R. & Arora, V. (2010). Analytical modeling of high
performance single-walled carbon nanotube field-effect-transistor, Microelectronics
Journal 41(9): 579–584.
Chen, H. (2010). Single CCII-based voltage-mode universal filter, Analog Integrated Circuits
and Signal Processing 62(2): 1573–1979.
Chen, J., Tsao, H. & Liu, S. (2001). Voltage-mode MOSFET-C filters using operational
transresistance amplifier (OTRA’s) with reduced parasitic capacitance effect, IEE
Proc. Circuits Devices Syst. 148(5): 242–249.
Daems, W., Gielen, G. & Sansen, W. (2002). A fitting approach to generate symbolic
expressions for linear and nonlinear analog circuit performance characteristics,
Design, Automation and Test in Europe, pp. 268–273.
Daems, W., Verhaegen, W., Wambacq, P., Gielen, G. & Sansen, W. (1999). Evaluation of error
control strategies for the linear symbolic analysis of analog integrated circuits, IEEE
Trans. Circuits and Systems I 46(5): 594–606.
Diaz-Madrid, J., Hinojosa, J. & Doménech-Asensi, G. (2008). Fuzzy logic technique for
accurate analog circuits macromodel sizing, International Journal of Circuit Theory and
Applications 38(3): 307–319.
Dong, N. & Roychowdhury, J. (2003). Piecewise polynomial nonlinear model reduction, Proc.
ACM/IEEE Design Automation Conf., pp. 484–489.
Fakhfakh, M., Tlelo-Cuautle, E. & Fernández, F. (2010). Design of Analog Circuits through
Symbolic Analysis, Bentham Sciences Publishers Ltd.
Fernández, F., Guerra, O., Rodriguez-Garcia, J. & Rodriguez-Vázquez, A. (1998). Symbolic
analysis of analog integrated circuits: the numerical reference generation problem,
IEEE Trans. on Circuits and Systems - II 45(10): 1351–1361.
Floberg, H. (1997). Symbolic analysis in analog integrated circuit design, Kluwer Academic
Publishers.
Gaoua, S., S.ăAsadi, M.C.E.ăYagoub & F.A.ăMohammadi (2010). CAD tools for efficient
RF/microwave transistor modeling and circuit design, Analog Integrated Circuits and
Signal Processing 63(1): 59–70.
Guerra, O., Rodriguez-Garcia, J., Fernández, F. & Rodriguez-Vázquez, A. (2002). A symbolic
pole/zero extraction methodology based on analysis of circuit time-constants, Analog
Integrated Circuits and Signal Processing 31: 101–118.
Guerra, O., Rodriguez-Garcia, J., Roca, E., Fernández, F. & Rodriguez-Vázquez, A. (1998). A
simplification before and during generation methodology for symbolic large-circuit
analysis, IEEE Int. Conf. on Electronics, Circuits and Systems, pp. 81–84.
Haley, S. (1991). Modification-decomposition transformation in analog design, Int. Journal of
Computer Aided VLSI Design pp. 427–428.
Hao, Z. & Shi, G. (2009). Symbolic techniques for statistical timing analysis of rcl mesh
networks with resistor loops, IEEE International Symposium on Integrated Circuits,
pp. 470–473.
Henning, E. (2002). Matrix approximation techniques for symbolic extraction of poles and
zeros, Analog Integrated Circuits and Signal Processing 31: 81–100.
Hernes, L. & Sansen, W. (2005). Distortion in single-, two- and three-stage amplifiers, IEEE
Trans. on Circuits and Systems I 52(5): 846–856.
Hsu, J. & Sechen, C. (1994). DC small signal symbolic analysis of large analog integrated
circuits, IEEE Trans. on Circuits and Systems, Part I 41(12): 817–828.
Hwang, Y., Wu, D., Chen, J., Shih, C. & Chou, W. (2009). Design of current-mode
MOSFET-C filters using OTRAs, International Journal of Circuit Theory and Applications
37(3): 397–411.
Krishna, S., Baghini, M. & Mukherjee, J. (2007). Behavioral models of basic mixed-mode
circuits: practical issues and application, Proceedings of European conference on circuit
theory and design, Spain, pp. 854–857.
Lee, C. (2010). Multiple-mode OTA-C universal biquad filters, Circuits, Systems and Signal
Processing 29(2): 263–274.
Li, P. & Pileggi, L. (2003). NORM: compact model order reduction of weakly nonlinear
systems, Proc. of 40th IEEE/ACM Design Automation Conference, pp. 472–477.
Li, P. & Pileggi, L. (2005). Compact reduced-order modeling of weakly nonlinear analog and
RF circuits, IEEE Trans. on Computer-Aided Design of Integrated Circuits and Systems
23(2): 184–203.
Li, X., McAndrew, C., Wu, W., Chaudhry, S., Victory, J. & Gildenblat, G. (2010). Statistical
modeling with the PSP MOSFET model, IEEE Trans. on Computer-Aided Design of
Lihong, Z., Jangkrajarng, N., Bhattacharya, S. & Shi, C. (2008). Parasitic-aware optimization
and retargeting of analog layouts: A symbolic-template approach, IEEE Tran. on
Computer-Aided Design of Integrated Circuits and Systems 27(5): 791–802.
Maheshwari, S., Mohan, J. & Chauhan, D. (2010). Voltage-mode cascadable all-pass sections
with two grounded passive components and one active element, IET Circuits, Devices
& Systems 4(2): 113–122.
Manthe, A., Zhao, L. & Shi, R. (2003). Symbolic analysis of analog circuits with hard
nonlinearity, IEEE/ACM Design Automation Conference, pp. 542–545.
Martinez-Romero, E., Tlelo-Cuautle, E., Sánchez-López, C. & X.-D. Tan, S. (2010). Symbolic
noise analysis of low voltage amplifiers by using nullors, SM2ACD, IEEE, Tunisia.
McAndrew, C. (2010). Integrated resistor modeling, in G. Gildenblat (ed.), Compact Modeling:
Principles, techniques and applications, Springer, Netherlands, pp. 271–297.
McConaghy, T. & Gielen, G. (2009). Template-free symbolic performance modeling of
analog circuits via canonical-form functions and genetic programming, IEEE Tran.
on Computer-Aided Design of Integrated Circuits and Systems 28(8): 1162–1175.
Muñoz-Pacheco, J. & Tlelo-Cuautle, E. (2009). Automatic synthesis of 2D-n-scrolls chaotic
systems by behavioral modeling, Journal of Applied Research and Technology 7(1): 5–14.
Phillips, J. (2000). Automated extraction of nonlinear circuit macromodels, Proc. Custom Integr.
Circuits Conf., pp. 451–454.
Pugliese, A., Amoroso, F., Cappuccino, G. & Cocorullo, G. (2010). Analysis of op-amp phase
margin impact on SC sigma-delta modulator performance, Microelectronics Journal
41(7): 440–446.
Qin, Z., Tan, S. & Cheng, C. (2005). Symbolic analysis and reduction of VLSI circuits, Springer.
Rewienski, M. & White, J. (2001). A trajectory piecewise-linear approach to model order
reduction and fast simulation of nonlinear circuits and micromachined devices, Proc.
IEEE/ACM Int. Conf. Computer-Aided Design, pp. 252–257.
Rodriguez-Garcia, J., Guerra, O., Roca, E., Fernández, F. & Rodriguez-Vázquez, A. (1999).
Error control in simplification before generation algorithms for symbolic analysis of
large analogue circuits, Electronic Letters 35(4): 260–261.
Root, D., Wood, J. & Tufillaro, N. (2003). New techniques for nonlinear behavioral modeling of
microwave/RF ICs from simulation and nonlinear microwave measurements, Proc.
ACM/IEEE Design Automation Conf., pp. 85–90.
Roychowdhury, J. (1999). Reduced-order modeling of time-varying systems, IEEE Trans. on
Circuits and Systems II 46(10): 1273–1288.
Ruiz-Amaya, J., de la Rosa, J., Fernández, F., Medeiro, F., del Rio, R., Pérez-Verdú,
B. & Rodriguez-Vázquez, A. (2005). High-level synthesis of switched-capacitor,
switched-current and continuous-time SD modulators using SIMULINK-based
time-domain behavioral models, IEEE Trans. on Circuits and Systems, Part I
52(9): 1795–1810.
S. Steinhorst & L. Hedrich (2010). Advanced methods for equivalence checking of analog
circuits with strong nonlinearities, Formal Methods in System Design 36(2): 131–147.
Saad, R. & Soliman, A. (2008). Use of mirror elements in the active device synthesis by
admittance matrix expansion, IEEE Tran. on Circuits and Systems I 55(9): 2726–2735.
Saad, R. & Soliman, A. (2010). A new approach for using the pathological mirror elements
in the ideal representation of active devices, International Journal of Circuit Theory and
Salama, K. & Soliman, A. (2000). Novel oscillators using the operational transresistance
amplifier, Microelectronics Journal 31(1): 39–47.
Sánchez-López, C., Fernández, F. & Tlelo-Cuautle, E. (2010). Generalized admittance matrix
models of OTRAs and COAs, Microelectronics Journal 41(8): 502–505.
Sánchez-López, C., Moro-Frias, D. & Tlelo-Cuautle, E. (2008). Improving the formulation
process of the system of equations of analog circuits, International Workshop on
Symbolic and Numerical Methods, Modeling and Applications to Circuit Design, Erfurt,
Germany, pp. 102–106.
Sánchez-López, C. & Tlelo-Cuautle, E. (2009). Symbolic behavioral model generation
of current-mode analog circuits, International Symposium on Circuits and Systems
(ISCAS), IEEE, Taiwan, pp. 2761–2764.
Sánchez-López, C., Trejo-Guerra, R., Muñoz-Pacheco, J. & Tlelo-Cuautle, E. (2010). N-scroll
chaotic attractors from saturated functions employing CCII+s, Nonlinear Dynamics
61(1-2): 331–341.
Sanchez-Sinencio, E. (2009). Low voltage amplifier design techniques, Tutorial in IEEE Midwest
Symposium on Circuits and Systems, Cancun, Mexico.
Shi, C. & Tan, S. (2000). Canonical symbolic analysis of large analog circuits with determinant
decision diagrams, IEEE Trans. on Computer-Aided Design of Integrated Circuits and
Systems 19(1): 1–18.
Shi, G., Hu, B. & Shi, C. (2006). On symbolic model order reduction, IEEE Trans. on
Computer-Aided Design of Integrated Circuits and Systems 25(7): 1257–1272.
Shi, G. & Meng, X. (2009). Variational analog integrated circuit design via symbolic sensitivity
analysis, IEEE International Symposium on Circuits and Systems, pp. 3002–3005.
Smith, K. & Sedra, A. (1968). Proceedings of the IEEE 56(The current conveyor: a new circuit
building block): 1368–1369.
Soliman, A. (2007). Voltage mode and current mode tow thomas bi-quadratic filters using
inverting CCII, International Journal of Circuit Theory and Applications 35(4): 463–467.
Soliman, A. (2009). Applications of voltage and current unity gain cells in nodal admittance
matrix expansion, IEEE Circuits and Systems Magazine 9(4): 29–42.
Sommer, R., Halfmann, T. & Broz, J. (2008). Automated behavioral modeling and analytical
model-order reduction by application of symbolic circuit analysis for multiphysical
systems, Simulation Modeling Practice and Theory 16(8): 1024–1039.
Sommer, R., Henning, E., Droge, G. & Horneber, E. (1993). Equation-based symbolic
approximation by matrix reduction with quantitative error prediction, Alta Frequenza
5(6): 317–325.
Suissa, A., Romain, O., Denoulet, J., Hachicha, K. & Garda, G. (2010). Empirical method based
on neural networks for analog power modeling, IEEE Transactions on Computer-Aided
Design of Integrated Circuits and Systems 29(5): 839–844.
Tan, S. & He, L. (2007). Advanced Model Order Reduction Techniques in VLSI Design, Cambridge
University Press.
Tan, S. & Shi, C. (2004). Efficient approximation of symbolic expressions for analog behavioral
modeling and analysis, IEEE Tran. on Computer-Aided Design of Integrated Circuits and
Systems 23(6): 907–918.
Tlelo-Cuautle, E., Duarte-Villaseñor, M., Garcia-Ortega, J. & Sánchez-López, C. (2007).
Designing SRCOs by combining SPICE and Verilog-A, International Journal of
Electronics 94(4): 373–379.
Tlelo-Cuautle, E., Duarte-Villaseñor, M. & Guerra-Gómez, I. (2008). Automatic synthesis of
VFs and VMs by applying genetic algorithms, Circuits, Systems and Signal Processing
27(3): 391–403.
Tlelo-Cuautle, E., Martinez-Romero, E., Sánchez-López, C. & X.-D. Tan, S. (2009). Symbolic
formulation method for mixed-mode analog circuits using nullors, International
Conference on Electronics, Circuits, and Systems (ICECS), IEEE, Mexico, pp. 856–859.
Tlelo-Cuautle, E., Martinez-Romero, E., Sánchez-López, C. & X.-D. Tan, S. (2010).
Symbolic behavioral modeling of low voltage amplifiers, International Conf. on
Electrical Engineering, Computing Science and Automatic Control (CCE), IEEE, Mexico,
pp. 510–514.
Tlelo-Cuautle, E., Moro-Frias, D. & Duarte-Villaseñor, M. (2008). Synthesis of CCII-s by
superimposing VFs and CFs through genetic operations, IEICE Electron. Express
5(11): 411–417.
Tlelo-Cuautle, E. & Sánchez-López, C. (2004). Symbolic computation of NF of transistor
circuits, IEICE Trans on Fundamentals of Electronics, Communications and Computer
Sciences E87(9): 2420–2425.
Tlelo-Cuautle, E., Sánchez-López, C., Fakhfakh, M. & Loulou, M. (2007). Designing SRCOs
by symbolic-behavioral-modeling of unity-gain cells, IEEE International Conference on
Electronics, Circuits and Systems, Marrakech, pp. 1035–1038.
Tlelo-Cuautle, E., Sánchez-López, C., Martinez-Romero, E. & Tan, S. (2010). Symbolic analysis
of analog circuits containing voltage mirrors and current mirrors, Analog Integrated
Circuits and Signal Processing 65(1): 89–95.
Tlelo-Cuautle, E., Sánchez-López, C. & Moro-Frias, D. (2010). Symbolic analysis of
(MO)(I)CCI(II)(III)-based analog circuits, International Journal of Circuit Theory and
Trejo-Guerra, R., Sánchez-López, C., Tlelo-Cuautle, E., Cruz-Hernández, C. &
Muñoz-Pacheco, J. (2010). Realization of multiscroll chaotic attractors by using
current-feedback operational amplifiers, Revista Mexicana de Fisica .
van der Horst, M., Linnenbank, A., Serdijn, W. & Long, J. (2010). Systematic design of a
transimpedance amplifier with specified electromagnetic out-of-band interference
behavior, IEEE Trans.on Circuits and SystemsŮI 57(3): 530–538.
Vasilevski, M., Aboushady, H. & Louerat, M. (2009). Automatic model refinement of Gm-C
integrators for high-level simulation of continuous-time sigma-delta modulators,
International Symposium on Circuits and Systems, IEEE, pp. 2769–2772.
Wambacq, P., Dobrovolny, P., Donnay, S., Engels, M. & Bolsens, I. (2000). Compact modeling
of nonlinear distortion in analog communication circuits, Proc. ACM/IEEE Design,
Automation Test Eur. Conf., pp. 350–354.
Wambacq, P., Dobrovolny, P., Gielen, G. & Sansen, W. (1998). Symbolic analysis of large
analog circuits using a sensitivity-driven enumeration of common spanning trees,
IEEE Trans. Circuits and Systems - II 45(10): 1342–1350.
Wambacq, P., Fernández, F., Gielen, G., Sansen, W. & Rodriguez-Vázquez, A. (1995).
Efficient symbolic computation of approximated small-signal characteristics of
analog integrated circuits, IEEE J. Solid-State Circuits 30(3): 327–330.
Wambacq, P., Fernández, F., Gielen, G., Sansen, W. & Rodriguez-Vázquez, A. (1996). A family
of matroid intersection algorithms for the computation of approximated symbolic
network functions, IEEE International Symposium on Circuits and Systems, pp. 806–809.
Wambacq, P., Gielen, G., Kinget, P. & Sansen, W. (1999). High-frequency distortion analysis of
analog integrated circuits, IEEE Trans. on Circuits and Systems II 46(3): 335–345.
Wambacq, P. & Sansen, W. (1998). Distortion Analysis of Analog Integrated Circuits, Norwell,
MA: Kluwer.
Yu, Q. & Sechen, C. (1995). Efficient approximation of symbolic network functions using
matroid intersection algorithms, IEEE International Symposium on Circuits and Systems,
pp. 2088–2091.
Yu, Q. & Sechen, C. (1996). A unified approach to the approximate symbolic analysis of large
analog integrated circuits, IEEE Trans. on Circuits and Systems I 43(8): 656–669.
Yu, Q. & Sechen, C. (1997). Efficient approximation of symbolic network functions using
matroid intersection algorithms, IEEE Trans. on Computer Aided Design 16: 1073–1081.
Part 2
Design Issues
5
Parallel Preconditioned Hierarchical Harmonic

Balance for Analog and RF Circuit Simulation
Peng Li1 and Wei Dong2
1 Department of Electrical and Computer Engineering, Texas A&M University
2 Texas Instruments
USA
1. Introduction
Circuit simulation is a fundamental enabler for the design of integrated circuits. As the design
complexity increases, there has been a long lasting interest in speeding up transient circuit
simulation using paralellization (Dong et al., 2008; Dong & Li, 2009b;c; Reichelt et al., 1993;
Wever et al., 1996; Ye et al., 2008).
On the other hand, Harmonic Balance (HB), as a general frequency-domain simulation
method, has been developed to directly compute the steady-state solutions of nonlinear
circuits with a periodic or quasi-periodic response (Kundert et al., 1990). While being
algorithmically efficient, densely coupling nonlinear equations in the HB problem formulation
still leads to computational challenges. As such, developing parallel harmonic balance
approaches is very meaningful.
Various parallel harmonic balance techniques have been proposed in the past, e.g. (Rhodes
& Perlman, 1997; Rhodes & Gerasoulis, 1999; Rhodes & Honkala, 1999; Rhodes & Gerasoulis,
2000). In (Rhodes & Perlman, 1997), a circuit is partitioned into linear and nonlinear portions
and the solution of the linear portion is parallelized; this approach is beneficial if the linear
portion of the circuit analysis dominates the overall runtime. This approach has been
extended in (Rhodes & Gerasoulis, 1999; 2000) by exposing potential parallelism in the form
of a directed acyclic graph. In (Rhodes & Honkala, 1999), an implementation of HB analysis
on shared memory multicomputers has been reported, where the parallel task allocation and
scheduling are applied to device model evaluation, matrix-vector products and the standard
block-diagonal (BD) preconditioner (Feldmann et al., 1996). In the literature, parallel matrix
computation and parallel fast fourier transform / inverse fast fourier transform (FFT/IFFT)
have also been exploited for harmonic balance. Some examples of the above ideas can be
found from (Basermann et al., 2005; Mayaram et al., 1990; Sosonkinaet al., 1998).
In this chapter, we present a parallel approach that focuses on a key component of modern
harmonic balance simulation engines, the preconditioner. The need in solving large practical
harmonic balance problems has promoted the use of efficient iterative numerical methods,
such as GMRES (Feldmann et al., 1996; Saad, 2003), and hence the preconditioning techniques
associated with iterative methods. Under such context, preconditioning is a key as it not only
determines the efficiency and robustness of the simulation, but also corresponds to a fairly
significant portion of the overall compute work. The presented work is based upon a custom
hierarchical harmonic balance preconditioner that is tailored to have improved efficiency and
robustness, and parallelizable by construction (Dong & Li, 2007a;b; 2009a; Li & Pileggi, 2004).
The latter stems from the fact that the top-level linearized HB problem is decomposed into a
series of smaller independent matrix problems across multiple levels, resulting a tree-like data
dependency structure. This naturally provides a coarse-grained parallelization opportunity as
demonstrated in this chapter.
In contrast to the widely used standard block-diagonal (BD) preconditioning (Feldmann et
al., 1996; Rhodes & Honkala, 1999), the presented approach has several advantages First,
purely from an algorithmic point of view, the hierarchical preconditioner possess noticeably
improved efficiency and robustness, especially for strongly nonlinear harmonic balance
problems (Dong & Li, 2007b; Li & Pileggi, 2004) . Second, from a computational point
of view, the use of the hierarchical preconditioner pushes more computational work onto
preconditioning, making an efficient parallel implementation of the preconditioner more
appealing. Finally, the tree-like data dependency of the presented preconditioner allows
for nature parallelization; in addition, freedoms exist in terms of how the overall workload
corresponding to this tree may be distributed across multiple processors or compute nodes
with a suitable granularity to suit a specific parallel computing platform.
The same core parallel preconditioning technique can be applied to not only standard
steady-state analysis of driven circuits, but also that of autonomous circuits such as
oscillators. Furthermore, it can be used as a basis for developing harmonic-balance
based envelope-following analysis, critical to communication applications. This leads to
a unifying parallel simulation framework targeting a range of steady-state and envelope
following analyses. This framework also admits traditional parallel ideas that are based
upon parallel evaluations of device models, parallel FFT/IFFT operations, and finer grained
matrix-vector products. We demonstrate favorable runtime speedups that result from
this algorithmic change, through the adoption of the presented preconditioner as well as
parallel implementation, on computer clusters using message-passing interface (MPI) (Dong
& Li, 2009a). Similar parallel runtime performances have been observed on multi-core
shared-memory platforms.
2. Harmonic balance
A circuit with n unknowns can be described using the standard modified nodal analysis
(MNA) formulation (Kundert et al., 1990)
d
h(t ) = q ( x (t)) + f ( x (t)) − u (t) = 0, (1)
dt
where x (t) ∈ n denotes the vector of n unknowns, q ( x (t)) ∈ n represents the vector of the
charges/fluxes contributed by dynamic elements, f ( x (t)) ∈ n represents the vector of the
currents contributed by static elements, and u (t) is the vector of the external input excitations.
If N harmonics are used to represent the steady-state circuit response in the frequency domain,
the HB system of the equations associated with Equation 1 can be formulated as
H ( X ) = ΩΓq (·)Γ −1 X + Γ f (·)Γ −1 X − U = 0, (2)
where X is the Fourier coefficient vector of circuit unknowns; Ω is a diagonal matrix

representing the frequency domain differentiation operator; Γ and Γ −1 are the N-point FFT
and IFFT (inverse FFT) matrices; q (·) and f (·) are the time-domain charge/flux and resistive
equations defined above; and U is the input excitation in the frequency domain. When
Parallel Preconditioned Hierarchical Harmonic Balance for Analog and RF Circuit Simulation 113
the double-sided FFT/IFFT are used, a total number of N = 2k + 1 frequency components

are included to represent each signal, where k is the number of positive frequencies being
considered.
It is customary to apply the Newton’s method to solve the nonlinear system in Equation 2.
At each Newton iteration, the Jacobian matrix J = ∂H/∂X needs to be computed, which is
written in the following matrix form (Feldmann et al., 1996; Kundert et al., 1990)
J = ΩΓCΓ −1 + ΓGΓ −1, (3)

∂q ∂f
where C = diag{ck = ∂x | x = x ( tk) } and G = diag{ gk = ∂x | x = x ( tk) } are block-diagonal
matrices with the diagonal blocks representing the linearizations of q (·) and f (·) at N sampled
time points t1 , t2 , · · · , t N . The above Jacobian matrix is rather dense. For large circuits,
storing the whole Jacobian matrix explicitly can be expensive. This promotes the use of
an iterative method, such as Generalized Minimal Residual (GMRES) method or its flexible
variant (FGMRES) (Saad, 1993; 2003). In this case, the Jacobian matrix needs only to be
constructed implicitly, leading to the notion of the matrix-free formulation. However, an
effective preconditoner shall be applied in order to ensure efficiency and convergence. To
this end, preconditioning becomes an essential component of large-scale harmonic balance
analysis.
The widely-used BD preconditioner discards the off-diagonal blocks in the Jacobian matrix
by averaging the circuit linearizations at all discretized time points and uses the resulting
block-diagonal approximation as a preconditioner (Feldmann et al., 1996). This relatively
straightforward approach is effective for mildly nonlinear circuits, where off-diagonal blocks
in the Jacobian matrix are not dominant. However, the performance of the BD preconditoner
deteriorates as circuit nonlinearities increase. In certain cases, divergence may be resulted for
strongly nonlinear circuits.
3. Parallel hierarchical preconditioning

A basic analysis flow for harmonic analysis is shown in Fig.1.
Clearly, at each Newton iteration, device model evaluation and the solution of a linearized
HB problem must be performed. Device model evaluation can be parallelized easily due its
apparent data-independent nature. For the latter, matrix-vector products and preconditioning
are the two key operations. The needed matrix-vector products associated with Jacobian
matrix J in Equation 3 are in the form
JX = Ω(Γ (C (Γ −1 X ))) + Γ ( G (Γ −1 X )), (4)
where G, C, Ω, Γ are defined in Section 2. Here, FFT/IFFT operations are applied

independently to different signals, and hence can be straightforwardly parallelized. For
preconditioning, we present a hierarchical scheme with improved efficiency and robustness,
which is also parallelizable by construction.
3.1 Hierarchical harmonic balance preconditioner

To construct a parallel preconditioner to solve the linearized problem JX = B defined by
Equation 4, we shall identify the parallelizable operations that are involved. To utilize, say m,
Fig. 1. A basic flow for HB analysis (from (Dong & Li, 2009a) ©[2009] IEEE ).
processing elements (PEs), we rewrite Equation 4 as

⎡ ⎤⎡ ⎤ ⎡ ⎤
J11 J12 · · · J1m X1 B1
⎢ J21 J22 · · · J2m ⎥ ⎢ X2 ⎥ ⎢ B2 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ . .. . . .. ⎥ ⎢ .. ⎥ = ⎢ . ⎥, (5)
⎣ .. . . . ⎦ ⎣ . ⎦ ⎣ .. ⎦
Jm1 Jm2 · · · Jmm Xm Bm
where Jacobian J is composed of m × m block entries; X and B are correspondingly partitioned

into m segments along the frequency boundaries. Further, J can be expressed in the form
⎛⎡ ⎤ ⎞
Ω1
⎜⎢ Ω2 ⎥ ⎟
⎜⎢ ⎥ ⎟
[ J ] m×m = ⎜⎢ .. ⎥ C c + Gc ⎟ , (6)
⎝⎣ . ⎦ ⎠
Ωm
where circulants Cc , Gc are correspondingly partitioned as

⎡ ⎤
Cc11 ··· Cc1m
⎢ .. ⎥
Cc = ΓCΓ −1 = ⎣ ... ..
. . ⎦
C ··· Ccmm
⎡ cm1 ⎤. (7)
Gc11 ··· Gc1m
⎢ .. ⎥
Gc = ΓGΓ −1 = ⎣ ... ..
. . ⎦
Gcm1 ··· Gcmm
A parallel preconditioner is essentially equivalent to a parallelizable approximation to J.

Assuming that the preconditioner is going to be parallelized using m PEs, we discard the
off-diagonal blocks of Equation 7, leading to m decoupled linearized problems of smaller

dimensions
⎧
⎪
⎪ J11 X1 = [ Ω1 Cc11 + Gc11 ] X1 = B1
⎪
⎨ J22 X2 = [ Ω2 Cc22 + Gc22 ] X2 = B2
.. . (8)
⎪
⎪ .
⎪
⎩
Jmm Xm = [ Ωm Ccmm + Gcmm ] Xm = Bm
By solving these decoupled linearized problems in a parallel way, a parallel preconditioner is

efficiently provided.
(a) Matrix view (b) Task dependence view
Fig. 2. Hierarchical harmonic balance preconditioner.

This basic idea of divide-and-conquer can be extended in a hierarchical fashion as shown in
Fig. 2. At the topmost level, to solve the top-level linearized HB problem, a preconditioner is
created by approximating the full Jacobian using a number (in this case two) of super diagonal
blocks. Note that the partitioning of the full Jacobian is along the frequency boundary. That is,
each matrix block corresponds to a selected set of frequency components of all circuit nodes in
the fashion of Equation 5. These super blocks can be large in size such that an iterative method
such as FGMRES is again applied to each such block with a preconditioner. These lower-level
preconditioners are created in the same fashion as that of the top-level problem by recursively
decomposing a large block into smaller ones until the block size is sufficiently small for direct
solve.
Another issue that deserves discussion is the storage of each subproblem in the preconditioner
hierarchy. Note that some of these submatrix problems are large. Therefore, it is desirable to
adopt the same implicit matrix-free presentation for subproblems. To achieve this, it is critical
to represent each linearized sub-HB problem using a sparse time-domain representation,
which has a decreasing time resolution towards the bottom of the hierarchy consistent with the
size of the problem. An elegant solution to this need has been presented in (Dong & Li, 2007b;
Li & Pileggi, 2004), where the top-level time-varying linearizations of device characteristics are
successively low-pass filtered to create time-domain waveforms with decreasing resolution
for the sub-HB problems. Interested readers are redirected to (Dong & Li, 2007b; Li & Pileggi,
2004) for an in-depth discussion.
3.2 Advantages of the hierarchical preconditioner

Purely from a numerical point of view, the hierarchical preconditioner is more advantageous
over the standard BD preconditioner. It provides a better approximation to the Jacobian, hence
leading to improved efficiency and robustness, especially for strongly nonlinear circuits.
Additionally, it is apparent from Fig. 2 that there exists inherent data independence in the
hierarchical preconditioner. All the subproblems at a particular level are fully independent,
allowing natural parallelization. The hierarchial nature of the preconditioner also provides
additional freedom and optimization in terms of parallelization granularity, and workload
distribution, and tradeoffs between parallel efficiency and numerical efficiency. For example,
the number of levels and the number of subproblems at each level can be tuned for the best
runtime performance and optimized to fit a specific a parallel hardware system with a certain
number of PEs. In addition, difference in processing power among the PE’s can be also
considered in workload partitioning, which is determined by the construction of the tree-like
hierarchical structure of the preconditioner.
4. Runtime complexity and parallel efficiency

Different configurations of the hierarchial preconditioner lead to varying runtime
complexities and parallel efficiencies. Understanding the tradeoffs involved is instrumental
for optimizing the overall efficiency of harmonic balance analysis.
Denote the number of harmonics by M, the number of circuit nodes by N, the number of
levels in the hierarchical preconditioner by K, the total number of sub-problems at level i by
Pi (P1 = 1 for the topmost level), and the maximum number of FGMRES iterations required to
reach the convergence for a sub-problem at level i by IF,i . We further define S F,i = Πik=1 IF,k ,
i = 1, · · · , K and S F,0 = 1.
The runtime cost in solving a sub-problem at the ith level can be broken into two parts: c1) the
cost incurred by the FGMRES algorithm; and c2) the cost due to the preconditioning. In the
serial implementation, the cost c1 at the topmost level is given by: αIF,1 MN + βIF,1 MN log M,
where α, β are certain constants. The first term in c1 corresponds to the cost incurred within
the FGMRES solver and it is assumed that a restarted (F)GMRES method is used. The second
term in c1 represents the cost of FFT/IFFT operations. At the topmost level, the cost c2
comes from solving P2 sub-problems at the second level IF,1 times, which is further equal
to the cost of solving all the sub-problems starting from the second level in the hierarchial
preconditioner. Adding everything together, the total computational complexity of the serial
hierarchically-preconditioned HB is
K −1
M
MN ∑ Pi S F,i−1 α + β log + γS F,K MN 1.1 , (9)
i =1
Pi
where the last term is due to the direct solve of the diagonal blocks of size N at the bottom of
the hierarchy. We have assumed that directly solving an N × N sparse matrix problem has a
cost of O( N 1.1 ).
For the parallel implementation, we assume that the workload is evenly split among m
PEs and the total inter-PE communication overhead is Tcomm, which is proportional to the
number of inter-PE communications. Correspondingly, the runtime cost for the parallel
implementation is

MN ∑iK=−11 Pi S F,i−1 α + β log M
Pi + γS F,K MN
1.1
+ Tcomm . (10)
m
It can be seen that minimizing the inter-PE communication overhead (Tcomm ) is important
in order to achieve a good parallel processing efficiency factor. The proposed hierarchical
preconditioner is parallelized by simultaneously computing large chunks of independent
computing tasks on multiple processing elements. The coarse-grain nature of our parallel
preconditioner reduces the relative contribution of the inter-PE communication overhead and
contributes to good parallel processing efficiency.
5. Workload distribution and parallel implementation

We discuss important considerations in distributing the work load across multiple processing
elements and parallel implementation.
5.1 Allocation of processing elements

We present a more detailed view of the tree-like task dependency of the hierarchical
preconditioner in Fig. 3.
Fig. 3. The task-dependency graph of the hierarchical preconditioner (from (Dong & Li,
2009a) ©[2009] IEEE ) .
5.1.1 Allocation of homogenous PE’s

For PE allocation, let us first consider the simple case where the PEs are identical in compute
power. Accordingly, each (sub)problem in the hierarchical preconditioner is split into N
equally-sized sub-problems at the next level and the resulting sub-problems are assigned to
different PE’s. We more formally consider the PE allocation problem as the one that assigns
a set of P PEs to a certain number of computing tasks so that the workload is balanced and
there is no deadlock. We use the breadth-first traversal of the task dependency tree to allocate
PEs, as shown in Algorithm 1.
The complete PE assignment is generated by calling Allocate(root, Pall ), where the root is the
node corresponding to the topmost linearized HB problem, which needs to be solved at each
Newton iteration. Pall is the full set of PEs. We show two examples of PE allocation in Fig.
4 for the cases of three and nine PEs, respectively. In the first case, three PEs are all utilized
at the topmost level. From the second level and downwards, a PE is only assigned to solve
a sub-matrix problem and its children problems. Similarly, in the latter case, the workload
at the topmost level is split between nine PEs. The difference from the previous case is that
there are less number of subproblems at the second level than that of available PEs. These
three subproblems are solved by three groups of PEs: {P1 , P2 , P3 }, {P4 , P5 , P6 } and {P7 , P8 , P9 },
respectively. On the third level, a PE is assigned to one child problem of the corresponding
parent problem at the second level.
Algorithm 1 Homogenous PE allocation

Inputs: a problem tree with root n; a set of P PEs with equal compute power;
Each problem is split into N sub-problems at the next level;
Allocate(n, P)
1: Assign all PEs from P to root node
2: If n does not have any child, return
3: Else
4: Partition P into N non-overlapping subsets, P1 , P2 , · · · , P N :
P
5: IF N == NP
6: P has P/N PEs (1 ≤ i ≤ N)

i
7: Elseif (P >N)

8: P i has NP
+ 1 PEs (1 ≤ i < N) and
P
P N has P − ( N + 1)( N − 1) PEs
9: Else
10: P i has one PE (1 ≤ i ≤ P) and others have no PE; return a warning message
11: For each child n i : Allocate(n i , P i ).
Fig. 4. Examples of homogenous PE allocation (from (Dong & Li, 2009a) ©[2009] IEEE ).
5.1.2 Deadlock avoidance

A critical issue in parallel processing is the avoidance of deadlocks. As described as follows,
deadlocks can be easily avoided in the PE assignment. In general, a deadlock is a situation
where two or more dependent operations wait for each other to finish in order to proceed. In
an MPI program, a deadlock may occur in a variety of situations (Vetter et al., 2000). Let us
consider Algorithm 1. PEs P1 and P2 are assigned to solve matrix problems M A and M B on
the same level. Naturally, P1 and P2 may be also assigned to solve the sub-problems of M A
and M B , respectively. Instead of this, if one assigns P1 to solve a sub-problem of M B and P2 a
sub-problem of M A , a deadlock may happen. To make progress on both solves, the two PEs
may need to send data to each other. When P1 and P2 simultaneously send the data and the
system does not have enough buffer space for both, a deadlock may occur. It would be even
worse if several pairs of such operations happen at the same time. The use of Algorithm 1
reduces the amount of inter-PE data transfer, therefore, avoids certain deadlock risks.
5.1.3 Allocation of heterogenous PE’s

It is possible that a parallel system consists of processing elements with varying compute
power. Heterogeneity among PEs can be considered in the allocation to further optimize the
performance. In this situation, subproblems with different sizes may be assigned to each PE.
We show a size-dependent allocation algorithm in Algorithm 2. For ease of presentation,
we have assumed that the runtime cost of linear matrix solves is linear in problem size. In
practice, more accurate runtime estimates can be adopted.
Algorithm 2 Size-dependent Heterogenous PE allocation

Inputs: a problem tree with root n; a set of P PEs; problem size S;
each problem is split into N sub-problems at the next level;
compute powers are represented using weights of PEs : w1 ≤ w2 ≤ · · · ≤ w P
Allocate(n, P, S)
1: Assign all PEs to root node
2: If n does not have any child, return
3: Else
4: Partition P into N non-overlapping subsets: P1 , P2 , · · · , P N ,
with the total subset weights w s,i , (1 ≤ i ≤ N ).
5: Minimize the differences between w s,i ’s.
6: Choose the size of the i-th child node n i as:
P
Si = S · w s,i / ∑ w j
j =1
7: For each n i : Allocate(n i , P i , Si ).
An illustrative example is shown in Fig. 5. Each problem is recursively split to three

sub-problems at the next level. The subproblems across the entire tree are denoted by
n i , (1 ≤ i ≤ 13). These problems are mapped onto nine PEs with compute power weights
w1 = 9, w2 = 8, w3 = 7, w4 = 6, w5 = 5, w6 = 4, w7 = 3, w8 = 2 and w9 = 1, respectively.
According to Algorithm 2, we first assign all PEs (P1 ∼ P9 ) to n1 , the top-level problem. At
the second level, we cluster the nine PEs to three groups and map a group to a sub-problem at
the second level. While doing this, we minimize differences in total compute power between
these three groups. We assign {P1 , P6 , P7 } to n2 , {P2 , P5 , P8 } to n3 , and {P3 , P4 , P9 } to n4 , as
shown in Fig. 5. The sum of compute power of all the PE’s is 45, while those allocated to n2 ,
n3 and n4 are 16, 15 and 14, respectively, resulting a close match. A similar strategy is applied
at the third-level of the hierarchical preconditioner as shown in Fig. 5.
Fig. 5. Example of size-dependent heterogenous PE allocation (from (Dong & Li, 2009a)
©[2009] IEEE ).
5.2 Parallel implementation

The proposed parallel preconditioner can be implemented in a relatively straightforward way
either on distributed platforms using MPI or on shared-memory platforms using pThreads
due to its coarse grain nature. Both implementations have been taken and comparisons were
made between the two. Similar parallel scaling characteristics for both implementations
have been observed, again, potentially due to the coarse grain nature of the proposed
preconditioner.
We focus on some detailed considerations for the MPI based implementation. On distributed
platforms, main parallel overheads come from inter-PE communications over the network.
Therefore, one main implementation objective is to reduce the communication overhead
among the networked workstations. For this purpose, non-blocking MPI routines are adopted
instead of their blocking counterparts to overlap computation and communication. This
strategy entails certain programming level optimizations.
As an example, consider the situation depicted in Fig. 5. The solutions of subproblems n5 ,
n6 and n7 computed by PEs P1 , P6 and P7 , respectively, need to be all sent to one PE, say P1 ,
which also works on a higher-level parent problem. Since multiple sub-problems are being
solved concurrently, P1 may not immediately respond to the requests from P6 (or P7 ). This
immediately incurs performance overhead if blocking operations are used.
Instead, one may adopt non-blocking operations, as shown in Fig. 6, where a single data
transfer is split into several segments. At a time, P6 (or P7 ) only prepares one segment of data
and sends a request to P1 . Then, the PE can prepare the next segment of data to be sent. As
such, the communication and computation can be partially overlapped.
Fig. 6. Alleviating communication overhead via non-blocking data transfers (from (Dong &
Li, 2009a) ©[2009] IEEE ).
Note that the popularity of recent multi-core processors has stimulated the development
of multithreading based parallel applications. Inter-PE communication overheads may be
reduced on shared-memory multi-core processors. This may be particularly beneficial for fine
grained parallel applications. In terms of parallel circuit simulation, for large circuits, issues
resulted from limited shared-memory resources must be carefully handled.
6. Parallel autonomous circuit and envelope-following analyses

Under the context of driven circuits, we have presented the hierarchical preconditioning
technique in previous sections. We further show that the same approach can be extended
to harmonic balance based autonomous circuit steady-state and envelope-following analyses.
6.1 Steady-state analysis of autonomous circuits

Several simulation techniques have been developed for the simulation of autonomous circuits
such as oscillators (Boianapally et al., 2005; Duan & Mayaram, 2005; Gourary et al., 1998;
Kundert et al., 1990; Ngoya et al., 1995). In the two-tier approach proposed in (Ngoya et al.,
1995), the concept of voltage probe is introduced to transform the original autonomous circuit
problem to a set of closely-related driven circuit problems for improved efficiency. As shown
in Fig. 7, based on some initial guesses of the probe voltage and the steady-state frequency, a
driven-circuit-like HB problem is formulated and solved at the second level (the lower tier).
Then, the obtained probe current is used to update the probe voltage and the steady-state
frequency at the top level (the upper tier). The process repeats until the probe current becomes
(approximately) zero.
Fig. 7. Parallel harmonic balance based autonomous circuit analysis (from (Dong & Li, 2009a)
©[2009] IEEE ).
It is shown as follows that the dominant cost of this two-tier approach comes from a series
of analysis problems whose structure resembles that of a driven harmonic balance problem,
making it possible to extend the aforementioned hierarchical preconditioner for analyzing
oscillators.
Fig. 8. Partitioning of the Jacobian of autonomous circuits (from (Dong & Li, 2009a) ©[2009]
IEEE ).
In the two-tier approach, the solution of the second-level HB problem dominates the overall
computational complexity. We discuss how these second level problems can be sped up by an
extended parallelizable hierarchical preconditioner. The linearized HB problem at the lower
tier corresponds to an extended Jacobian matrix

AnN ×nN BnN ×l
· X( nN +l )×1 = V( nN +l )×1, (11)
Cl ×nN Dl ×l
where n and N are the numbers of the circuit unknowns and harmonics, respectively,
and l (l << nN ) is the number of additionally appended variables corresponding to the
steady-state frequency and the probe voltage. It is not difficult to see that the structure
of matrix block AnN ×nN is identical to the Jacobian matrix of a driven circuit HB analysis.
Equation 11 is rewritten in the following partitioned form

AX1 + BX2 = V1
. (12)
CX1 + DX2 = V2
From the first equation in Equation 12, we express X1 in terms of X2 as
X1 = A−1 (V1 − BX2 ). (13)
Substituting Equation 13 into the second equation in Equation 12 gives
X2 = ( D − CA−1 B )−1 (V2 − CA−1 V1 ). (14)
The dominant computational cost for getting X2 comes from solving the two linearized matrix
problems associated with A−1 B and A−1 V1 . When X2 is available, X1 can be obtained by
solving the third matrix problem defined by A in Equation 13, as illustrated in Fig. 8.
Clearly, the matrix structure of these three problems is defined by matrix A, which has a
structure identical to the Jacobian of a driven circuit. The same hierarchical preconditioning
idea can be applied to accelerate the solutions of the three problems.
6.2 Envelope-following analysis

Envelope-following analysis is instrumental for many communication circuits. It is
specifically suitable for analyzing periodic or quasi-periodic circuit responses with slowly
varying amplitudes (Feldmann& Roychowdhury, 1996; Kundert et al., 1988; Rizzoli et al.,
1999; 2001; Silveira et al., 1991; White & Leeb, 1991). The principal idea of the HB-based
envelope-following analysis is to handle the slowly varying amplitude, called envelope, of
the fast carrier separately from the carrier itself, which requires the following mathematical
representation of each signal in the circuit
K
x (t) = ∑ Xk (t)e jkω0 t , N = 2K + 1, (15)
k =− K
where the envelope Xk (t) varies slowly with respect to the period of the carrier T0 = 2π/ω0 .
This signal representation is illustrated in Fig. 9.
As a result, the general circuit equation in Equation 1 can be cast to
K
d
h(t) = h(te , tc ) = ∑ [ jkω0 Qk (te ) + Q (te ) + Gk (te ) − Uk (te )] e jkω0 tc ,
dt k
(16)
k =− K
where different time variables te , tc are used for the envelope and the carrier. Correspondingly,
the Fourier coefficients shall satisfy
d
H ( X (te )) = ΩΓq (·)Γ −1 X (te ) + Γq (·)Γ −1 X (te ) + Γ f (·)Γ −1 X (te ) − U (te ) = 0, (17)
dte
which can be solved by using a numerical integration method. Applying Backward Euler (BE)
to discretize Equation 17 over a set of time points (t1 , t2 , · · · , tq , · · · ) leads to

Γq (·)Γ −1 X (tq ) − Γq (·)Γ −1 X (tq−1 ) /(tq − tq−1 )
(18)
+ ΩΓq (·)Γ −1 X (tq ) + Γ f (·)Γ −1 X (tq ) − U (tq ) = 0.
To solve this nonlinear problem using the Newton’s method, the Jacobian is needed
−1
Jenv = tΓCΓ
q − tq−1
+ ΩΓCΓ −1 + ΓGΓ −1 =
⎡ I1 ⎤
Ω1 + t q − t q − 1
⎢ ⎥ (19)
⎢ .. ⎥ · C c + Gc ,
⎣ . ⎦
Ωm + tq −Imtq−1
where the equation is partitioned into m blocks in a way similar to Equation 6; I1 , I2 , · · · , Im

are identity matrices with the same dimensions as the matrices Ω1 , Ω2 , · · · , Ωm , respectively;
Circulants Cc and Gc have the same forms as in Equation 7. Similar to the treatment taken
in Equation 8, a parallel preconditioner can be formed by discarding the off-block diagonal
entries of Equation 7, which leads to m decoupled linear problems of smaller dimensions
⎧
⎪
⎪ [(Ω1 + ( t −It1 ) )Cc11 + Gc11 ] X1 = B1
⎪
⎪
q q−1
⎪
⎨ [(Ω2 + ( t −It2 ) )Cc22 + Gc22 ] X2 = B2
q q−1
.. . (20)
⎪
⎪
⎪
⎪ .
⎪
⎩ [(Ω + Im
m (t −t )
)Ccmm + Gcmm ] Xm = Bm
q q−1
Fig. 9. Signal representations in envelope-following analysis (from (Dong & Li, 2009a)
©[2009] IEEE ).
To summarize, the mathematical structure of these sub-problems is identical to that of a

standard HB problem. The same matrix-free representation can be adopted to implicitly
form these matrices. A hierarchical preconditioner can be constructed by applying the above
decomposition recursively as before.
7. Illustrative examples
We demonstrate the presented approach using a C/C++ based implementation. The MPICH
library (Gropp & Lusk, 1996) has been used to distribute the workload over a set of networked
Linux workstations with a total number of nine CPUs. The FFTW package is used for
FFT/IFFT operations (Frigo & Johnson, 2005) and the FGMRES solver is provided through
the PETSC package (Balay et al., 1996). Most of the parallel simulation results are based upon
the MPI based implementation unless stated otherwise.
7.1 Simulation of driven circuits

A list of circuits in Table 1 are used in the experimental study. For the hierarchical
preconditioning technique, a three-level hierarchy is adopted, where the size of each
sub-problem is reduced by a factor of three at the next lower level.
Serial and parallel implementations of the block diagonal (BD) preconditioner (Feldmann
et al., 1996) and the hierarchical preconditioner are compared in Table 2. Here a parallel
implementation not only parallelizes the preconditioner, but also other parallelizable
components such as device model evaluation and matrix-vector products. The second and
third columns show the runtimes of harmonic balance simulations using the serial BD and
hierarchical preconditioner, respectively. The columns below ’T3(s)’, ’T5(s)’ and ’T4(s)’, ’T6(s)’
correspond to the runtimes of the parallel HB simulations using the BD preconditioner and
the hierarchical preconditioner, respectively. The columns below ’X1’-’X4’ indicate the parallel
runtime speedups over the serial counterparts. It is clear that the hierarchical preconditioner
Index Description of circuits Nodes Freqs Unknowns

1 frequency divider 17 100 3,383
2 DC-DC converter 8 150 2,392
3 diode rectifier 5 200 1,995
4 double-balanced mixer 27 188 10,125
5 low noise amplifier 43 61 5,203
6 LNA + mixer 69 86 11,799
7 RLC mesh circuit 1,735 10 32,965
8 digital counter 86 50 8,514
Table 1. Descriptions of the driven circuits (from (Dong & Li, 2009a) ©[2009] IEEE ).
speeds up harmonic balance simulation noticeably in the serial implementation. The

MPI-based parallel implementation brings in additional runtime speedups.
Serial Parallel 3-CPU Platform Parallel 9-CPU Platform
Index BD Hierarchical BD Hierarchical BD Hierarchical
T1(s) T2(s) T3(s) X1 T4(s) X2 T5(s) X3 T6(s) X4
1 354 167 189 1.87 92 1.82 89 3.97 44 3.79
2 737 152 391 1.88 83 1.83 187 3.94 40 3.80
3 192 39 105 1.82 22 1.77 52 3.69 11 3.54
4 55 15 31 1.77 9 1.67 14 3.93 4 3.75
5 1,105 127 570 1.93 69 1.84 295 3.74 36 3.53
6 139 39 80 1.73 23 1.67 38 3.66 11 3.55
7 286 69 154 1.85 38 1.80 76 3.76 19 3.62
8 2,028 783 1,038 1.95 413 1.89 512 3.96 204 3.83
Table 2. Comparison on serial and parallel implementations of the two preconditioners

(modified from (Dong & Li, 2009a) ©[2009] IEEE ).
To show the parallel runtime scaling of the hierarchical preconditioner, the runtime speedups
of the parallel preconditioner over its serial counterpart as a function of the number of
processors for three test circuits are shown in Fig. 10.
In Fig. 11, we compare the distributed-memory based implementation using MPI with the
shared-memory based implementation using multithreading (pThreads) for the frequency
divider and the DC-DC converter. Two implementations exhibit a similar scaling
characteristic. This is partially due to the fact the amount of inter-PE communication is
rather limited in the proposed hierarchal preconidtioner. As a result, the potentially greater
communication overhead of the distributed implementation has a limited impact on the
overall runtimes.
7.2 Parallel simulation of oscillators

A set of oscillators described in Table 3 are used to compare two implementations of the
two-tier method (Ngoya et al., 1995), one with the block-diagonal (BD) preconditioner, and
the other the hierarchial preconditioner.
The runtimes of the serial implementations of the two versions are listed in the columns
labeled as "Serial Platform" in Table 4. At the same time, the runtimes of the parallel
simulations with the BD and hierarchical preconditioners on the 3-CPU and 9-CPU platforms
are also shown in the table. The columns below ’X3’ and ’X5’ are the speedups of parallel
simulations with the BD preconditioner. And the columns below ’X4’ and ’X6’ are the
speedups of parallel simulations with the hierarchical preconditioner.
Fig. 10. The runtime speedups of harmonic balance simulation with hierarchical
preconditioning as a function of the number of processors (from (Dong & Li, 2009a) ©[2009]
IEEE ).
Fig. 11. Comparison of shared-memory and distributed-memory implementations of

hierarchical preconditioning (from (Dong & Li, 2009a) ©[2009] IEEE ).
Index Oscillator Nodes Freqs Unknowns

1 11 stages ring oscillator 13 50 1,289
2 13 stages ring oscillator 15 25 737
3 15 stages ring oscillator 17 20 665
4 LC oscillator 12 30 710
5 digital-controlled oscillator 152 10 2890
Table 3. Descriptions of the oscillators (from (Dong & Li, 2009a) ©[2009] IEEE ).
Serial Platform Parallel 3-CPU Platform Parallel 9-CPU Platform
Osc. Two-tier BD Two-tier Hier. BD Hier. BD Hier.
T1(s) N-Its T2(s) N-Its T3(s) X3 T4(s) X4 T5(s) X5 T6(s) X6
1 127 48 69 43 74 1.71 41 1.68 32 3.97 18 3.83
2 95 31 50 27 55 1.73 29 1.72 24 3.96 13 3.85
3 83 27 44 23 48 1.73 26 1.69 22 3.77 12 3.67
4 113 42 61 38 67 1.68 37 1.66 30 3.80 17 3.69
5 973 38 542 36 553 1.76 313 1.73 246 3.95 141 3.86
Table 4. Comparisons of the two preconditioners on oscillators (from (Dong & Li, 2009a)
©[2009] IEEE ).
On the 3-CPU platform, the average values below the columns ’X3’ and ’X4’ are 1.72x, 1.70x,
respectively; On the 9-CPU platform, these average values are 3.89x and 3.78x respectively. It
can be observed that the proposed parallel method brings favorable speedups over both its
serial implementation and the parallel counterpart with the BD preconditioner.
7.3 Parallel envelope-following analysis

A power amplifier and a double-balanced mixer are used to demonstrate the proposed ideas,
and the results are shown in Table 5. The runtimes are in seconds. As a reference, the runtimes
of the serial transient simulation, the serial envelope-following simulations with the BD and
the hierarchical preconditioners are listed in the columns below "Serial Platform", respectively.
The columns below ’X2’ and ’X3’ indicate the speedups of the envelope-following simulation
over the transient simulation. In the columns labeled as "3 CPUs" and "9 CPUs", the runtime
results of the parallel envelope-following simulations with the BD preconditioner and the
hierarchical preconditioner using three and nine CPUs are shown. The columns below
’X4’-’X7’ indicate the runtime speedups of the parallel envelope-following analyses over their
serial counterparts. The runtime benefits of the proposed parallel approach are clearly seen.
Serial Platform 3 CPUs 9 CPUs
CKT Trans. BD Hier. BD Hier. BD Hier.
T1 T2 X2 T3 X3 T4 X4 T5 X5 T6 X6 T7 X7
PA 831 76 10.9 26 32.0 44 1.73 16 1.64 19 4.01 7 3.72
Mixer 1,352 102 13.2 39 34.6 60 1.70 24 1.62 26 3.94 11 3.67
Table 5. Comparison of the two preconditioners on envelope-following simulation (from

(Dong & Li, 2009a) ©[2009] IEEE ).
8. Conclusions
We address the computational challenges associated with harmonic balance based analog
and RF simulation from two synergistic angles: hierarchical preconditioning and parallel
processing. From the first angle, we tackle a key computational component of modern
harmonic balance algorithms that rely on the matrix-free implicit formulation and
efficient iterative methods. The second angle is meaningful as parallel computing has
become increasingly pervasive and utilizing parallel computing power is an effective
means for improving the runtime efficiency of electronic design automation tools. The
presented hierarchical preconditioner is numerically robust and efficient, and parallizable
by construction. Favorable runtime performances of hierarchical preconditioning
have been demonstrated on distributed and shared memory computing platforms for
steady-state analysis of driven and automatous circuits as well as harmonic balance based
envelope-following analysis.
9. Acknowledgments
This material is based upon work supported by the National Science Foundation under Grant
No. 0747423, and SRC and Texas Analog Center for Excellence under contract 2008-HC-1836.
10. References
Balay, S.; Buschelman, K.; Gropp, W.; Kaushik, D.; Knepley, M.; McInnes, L.; Smith, B. &
Zhang, H. (2001). PETSc Web pages: URL: www.mcs.anl.gov/petsc, Argonne National
Laboratory.
Basermann, A.; Jaekel, U.; Nordhausen, M. & and Hachiya, K.(2005). Parallel iterative solvers
for sparse linear systems in circuit simulation Future Gener. Comput. Syst., Vol. 21, No.
8, (October 2005) pp. 1275-1284, ISSN 0167-739X.
Boianapally, K.; Mei, T. & Roychowdhury, J. (2005). A multi-harmonic probe technique for
computing oscillator steady states, Digest of Technical Papers, IEEE/ACM Int. Conf.
on CAD, pp. 610-613, ISBN 0-7803-9254-X, San Jose, CA, USA, November 2005,
IEEE/ACM.
Dong, W. & Li, P. (2007a) Accelerating harmonic balance simulation using efficient
parallelizable hierarchical preconditioning, Proceedings of IEEE/ACM Design
Automation Conference, pp. 436-439, ISBN 978-1-59593-627-1, San Diego, CA, USA,
June 2007, IEEE/ACM.
Dong, W.; & Li, P. (2007b). Hierarchical harmonic-balance methods for frequency-domain
analog-circuit analysis, IEEE Trans. on Computer-Aided Design of Integrated Circuits and
Systems, Vol. 26, No. 12, (December 2007) pp. 2089-2101, ISSN 0278-0070.
Dong, W., Li, P. & Ye, X. (2008) WavePipe: parallel transient simulation of analog and digital
circuits on multi-core shared-memory machines, Proceedings of IEEE/ACM Design
Automation Conference, pp. 238-243, ISBN 978-1-60558-115-6, Anaheim, CA, USA, June
2008, IEEE/ACM.
Dong, W. & Li, P. (2009a). A parallel harmonic balance approach to steady-state and
envelope-following simulation of driven and autonomous circuits, IEEE Trans. on
Computer-Aided Design of Integrated Circuits and Systems, Vol. 8, No. 4, (April 2009) pp.
490 - 501, ISSN 0278-0070.
Dong, W. & Li, P. (2009b) Parallelizable stable explicit numerical integration for efficient circuit
simulation, Proceedings of IEEE/ACM Design Automation Conference, pp. 382-385, ISBN
978-1-6055-8497-3, San Francisco, CA, USA, July 2009, IEEE/ACM.
Dong, W.; & Li, P. (2009c). Final-value ODEs: stable numerical integration and its application
to parallel circuit analysis, Digest of Technical Papers, IEEE/ACM Int. Conf. on CAD, pp.
73-78, ISBN 978-1-60558-800-1, San Jose, CA, USA, November 2009, IEEE/ACM.
Duan, X. & Mayaram, K. (2005). An efficient and robust method for ring-oscillator simulation
using the harmonic-balance method, IEEE Trans. on Computer-Aided Design of
Integrated Circuits and Systems, Vol. 24, No. 8, (August 2005) pp. 1225-1233, ISSN
0278-0070.
Feldmann, P.; Melville, R. & Long, D. (1996) Efficient frequency domain analysis of large
nonlinear analog circuits, Proceedings of IEEE Custom Integrated Circuts Conf., pp.
461-464, ISBN 0-7803-3117-6, San Diego, CA, USA, May 1996, IEEE.
Feldmann, P. & Roychowdhury, J. (1996). Computation of circuit waveform envelopes using
an efficient matrix-decomposed harmonic balance algorithm, Digest of Technical
Papers, IEEE/ACM Int. Conf. on CAD, pp. 295-300, ISBN 0-8186-7597-7, San Jose, CA,
USA, November 1996, IEEE/ACM.
Frigo, M. & Johnson, S. (2005). The design and implementation of FFTW3, Proceedings of the
IEEE, Vol. 93, No. 2, (January 2005) pp. 216-231, ISSN 0018-9219.
Gourary, M.; Ulyanov, S.; Zharov, M.; Rusakov, S.; Gullapalli, K.; & Mulvaney, B (1998).
Simulation of high-Q oscillators, Digest of Technical Papers, IEEE/ACM Int. Conf.
on CAD, pp. 162-169, ISBN 1-58113-008-2, San Jose, CA, USA, November 1998,
IEEE/ACM.
Gropp, W. & Lusk, E. (1996). User’s Guide for mpich, a Portable Implementation of
MPI,Mathematics and Computer Science Division, Argonne National Laboratory.
Kundert, K.; White, J. & Sangiovanni-Vincentelli, A. (1988). An envelope-following method
for the efficient transient simulation of switching power and filter circuits, Digest of
Technical Papers, IEEE/ACM Int. Conf. on CAD, pp. 446-449, ISBN 0-8186-0869-2, San
Jose, CA, USA, October 1988, IEEE/ACM.
Kundert, K.; White, J.; & Sangiovanni-Vincentelli, A. (1990). Steady-state Methods for Simulating
Analog and Microwave Circuits, Kluwer Academic Publisher, ISBN 978-0-7923-9069-5,
Boston, USA.
Li, P. & Pileggi, L.(2004). Efficient harmonic balance simulation using multi-level frequency
decomposition, Digest of Technical Papers, IEEE/ACM Int. Conf. on CAD, pp. 677-682,
ISBN 1092-3152, San Jose, CA, USA, November 2004, IEEE/ACM.
Mayaram, K.; Yang, P.; Burch, R. Chern, J.; Arledge, L. & Cox, P.(1990). A parallel
block-diagonal preconditioned conjugate-gradient solution algorithm for circuit and
device simulations, Digest of Technical Papers, IEEE/ACM Int. Conf. on CAD, pp.
446-449, ISBN 0-8186-2055-2, San Jose, CA, USA, November 1990, IEEE/ACM.
Ngoya, E.; Suarez, A.; Sommet, R. & Quere, R. (1995). Steady state analysis of free or
forced oscillators by harmonic balance and stability investigation of periodic and
quasi-periodic regimes, nt. J. Microw. Millim.-Wave Comput.-Aided Eng., Vol. 5 No. 3,
(March 1995) pp. 210-233, ISSN 1050-1827.
Reichelt, M.; Lumsdaine, A; & White, J. (1993). Accerlerated waveform methods for parallel
transient simulation of semiconductor devices, Digest of Technical Papers, IEEE/ACM
Int. Conf. on CAD, pp. 270–74, ISBN 0-8186-4490-7, San Jose, CA, USA, November
1993, IEEE/ACM.
Rhodes, D.; & Perlman, B. (1997). Parallel computation for microwave circuit simulation. IEEE
Trans. on Microwave Theory and Techniques, Vol. 45, No. 5, (May 1997) pp. 587-592, ISSN
0018-9480.
Rhodes, D. & Gerasoulis, A.(1999). Scalable parallelization of harmonic balance simulation,

Proceedings of IPPS/SPDP Workshops, pp. 1055-1064, ISBN 3-540-65831-9, San Juan,
Puerto Rico, USA, April 1999, Springer.
Karanko, V.; & Honkala, M.(2004). A parallel harmonic balance simulator for shared memory
multicomputers, Proceedings of 34th European Microwave Conference, pp. 849-851, ISBN
3-540-65831-9, Amsterdam, The Netherlands, October 2004, IEEE.
Rhodes, D. & Gerasoulis, A.(2000). A scheduling approach to parallel harmonic balance
simulation. Concurrency: Practice and Experience, Vol. 12, No. 2-3, (February-March
2000) pp. 175-187.
Rizzoli, V.; Neri, A. Mastri, F. & and Lipparini, A. (1999). A krylov-subspace technique for the
simulation of RF/microwave subsystems driven by digitally modulated carriers, Int.
J. RF Microwave Comput.-Aided Eng., Vol. 11, No. 7, (August 2005) pp. 490-505, ISSN
1531-1309.
Rizzoli, V.; Costanzo, A. & Mastri, F. (2001). Efficient krylov-subspace simulation of
autonomous RF/microwave circuits driven by digitally modulated carriers, IEEE
Microwave Wireless Comp. Lett., Vol. 11, No. 7, (July 2001) pp. 308-310, ISSN 1531-1309.
Saad, Y. (1993). A flexible inner-outer preconditioned GMRES algorithm, SIAM J.Sci.Comput.,
Vol. 14, No. 2, (March 1993) pp. 461-469, ISSN 1064-8275.
Lima, P.; Bonarini, A. & Mataric, M. (2003). Iterative Methods for Sparse Linear Systems, Society
for Industrial and Applied Mathematics, ISBN 0898715342, Philadelphia, PA, USA.
Silveira, L.; White, J. & Leeb, S. (1991). A modified envelope-following approach to clocked
analog circuit simulation, Digest of Technical Papers, IEEE/ACM Int. Conf. on CAD, pp.
20-23, ISBN 0-8186-2157-5, San Jose, CA, USA, November 1991, IEEE/ACM.
Sosonkina, M., Allison, D. & Watson, L. (1998). Scalable parallel implementations of the gmres
algorithm via householder reflections, Int. Conf. on Parallel Processing, pp. 396-404,
ISBN 0-8186-8650-2, Minneapolis, MN, USA, August 1998, IEEE.
Vetter, J.; & de Supinski, B. (2000) Dynamic software testing of MPI applications with Umpire,
Proceedings of the 2000 ACM/IEEE conference on Supercomputing, pp. 51-60, ISBN
0-7803-9802-5, Washington, DC, USA, November 2000, IEEE.
Wever, U. & Zheng, Q. (1996) Parallel Transient Analysis for Circuit Simulation, Proceedings of
the Twenty-Ninth Hawaii Int. Conf. on System Sciences, pp. 442-447, ISBN 0-8186-7324-9,
Wailea, HI, USA, January 1996, IEEE.
White, J. & Leeb, S. (1991). An envelope-following approach to switching power converter
simulation, IEEE Trans. on Power Electronics, Vol. 6, No. 2, (April 1991) pp. 303-307,
ISSN 0885-8993.
Ye, X.; Dong, W.; Li, P. & Nassif, S. (2008). MAPS: Multi-Algorithm Parallel circuit
Simulation, Digest of Technical Papers, IEEE/ACM Int. Conf. on CAD, pp. 73-78, ISBN
978-1-4244-2819-9, San Jose, CA, USA, November 2008, IEEE/ACM.
6
Lifetime Yield Optimization of Analog Circuits

Considering Process Variations and Parameter
Degradations
Xin Pan and Helmut Graeb
Institute for Electronic Design Automation, Technische Universitaet Muenchen, Munich
Germany
1. Introduction
As semiconductor technology continuously scales, the joint effects of manufacture process
variations and operational lifetime parameter degradations have been a major concern for
analog circuit designers since they affect the lifetime yield value, i.e., the percentage of the
products which can satisfy all of the pre-defined specifications during lifetime operation
(Alam, Kang, Paul & Roy, 2007), (Gielen et al., 2008).
The analysis and optimization of analog circuits considering process variations alone have
been in research for decades, and certain design centering algorithms and commercial
software are available to achieve a design for yield (more specifically, fresh yield) (Nassif,
2008), (Antreich et al., 1994). On the other hand, the modeling of device parameter
degradations such as Negative Bias Temperature Instability (NBTI) and Hot Carrier Injection
(HCI) has been so far focusing mainly on the nominal values without considering the
underlying variations during manufacture process (Jha et al., 2005), (Liu et al., 2006) and
(Martin-Martinez et al., 2009). A robust analog circuit design is thus needed tolerant of both
process variations and lifetime parameter degradations, maximizing the lifetime yield value.
Most of the past works quantify the influences of process variations and lifetime degradations
separately. However, since lifetime degradations will drift certain device parameters, e.g., Vth ,
from their fresh values during circuit lifetime operation, the distribution of the circuit-level
performance will also shift its position during lifetime, as can be seen in Figure 1, where
1000 Monte-Carlo simulations are run on a fresh and 5-year-old Miller OpAmp. Values of
Gain-Bandwidth Product(GBW) and Rising Slew Rate (SR) are shown, both moving towards
negative direction. In order to ensure a robust design, it is thus necessary to consider the joint
effects of process variations and lifetime parameter degradations during design phase, such
that certain weak points can be detected early, and additional safe margins can be assigned
properly.
In this chapter, a novel design methodology to analyze and optimize the lifetime yield value
of analog circuits based on the idea of lifetime worst-case distance is presented. It does
not involve Monte-Carlo simulations, and considers process variations and major parameter
degradation mechanisms such as NBTI and HCI.
The proposed work is based on the preliminary methods presented in (Pan & Graeb, 2009)
and (Pan & Graeb, 2010). The content is augmented such that the sizing constraints for both
fresh and aged circuits are considered, as well as the required additional area penalty for the
reliability optimized design is analyzed.
The rest of the chapter is organized as follows. Section 2 introduces the physical behavior of
two main degradation effects, namely, NBTI and HCI. Section 3 briefly reviews the current
methods in literature which study the joint effect of process variations and parameter lifetime
degradation. Section 4 gives basic definitions needed throughout the chapter. The definitions
of lifetime yield and lifetime worst-case distance are proposed in Section 5, which are the key
concepts of the chapter. Section 6 introduces the sizing constraints which must be considered
in the proposed method. The new reliability-aware design flow is proposed in Section 7.
Section 8 introduces a linear prediction model in time domain which is used to speed up the
analysis of lifetime worst-case distance values. Then experimental results are given in Section
9. Finally Section 10 concludes the chapter.
7.0E+06
6.5E+06 degradation
6.0E+06
GBW (HZ)
5.5E+06
5.0E+06
4.5E+06
4.0E+06
2 2.5 3 3.5 4 4.5 5 5.5
SR (V/uS)
Fresh 5 Years
Fig. 1. Degradation of performance distributions from fresh circuit to 5 years of a Miller

OpAmp by 1000 Monte-Carlo simulations.
2. Degradation physics
In this section, the physical characters of HCI and NBTI will be briefly introduced. For a
more complete discussion, please refer to (Hu et al., 1985), (Schroder & Babcock, 2003), (Alam,
Kufluoglu, Varghese & Mahapatra, 2007) and (Wang et al., 2007).
2.1 HCI
Figure 2(a) shows the simplified physical behavior of HCI effect on an NMOS transistor. HCI
effect is the result of injection of channel carriers from the conducting channel under the gate
into the gate dielectric. It happens near the drain area where the lateral electric field is high
and the channel carriers gain enough kinetic energy during the acceleration along the channel.
The hot channel carriers may hit an atom in the substrate, breaking a electron-hole pair or a
Si-H bond, and introducing interface traps and a substrate current.
Traditional modeling method of HCI is by analyzing the substrate current Isub (Hu et al., 1985).
The correlation is due to the fact that both hot-carriers and substrate current are driven by a
common factor-the maximum channel electric field Em at the drain end. Some recent research
(Wang et al., 2007) point out that, as technology scales, Isub will be dominated by various
leakage components such as gate leakage, junction current, etc. Authors in (Wang et al., 2007)
proposed the following reaction-diffusion based model for the degraded parameter ΔVth due
Considering Process Variations and Parameter Degradations 133
to HCI as:
q Eox ψ
ΔVth = K2 Qi exp exp − it tn (1)
Cox Eo2 qλEm
where Qi is the inversion charge, ψit is the trap generation energy and the time exponential
constant n is 0.45.
Gate Gate
Source Drain Source Drain
n+ n+ p+ p+
Insulator carrier injection Insulator H

H H
H
H H
hot electron Si Si Si Si Si Si
p- Substrate n- Substrate
(a) HCI of an nMOSFET (b) NBTI of a pMOSFET
Fig. 2. Physics of HCI and NBTI
2.2 NBTI
The physical behavior of NBTI on a PMOS transistor is shown in Figure 2(b). It is commonly
accepted that NBTI is the result of hole-assisted breaking of Si-H bonds at Si/SiO2 interface
(Alam, Kufluoglu, Varghese & Mahapatra, 2007) when a PMOS is biased in inversion using
the Reaction-Diffusion (R-D) model:
dNIT
= k F ( N0 − NIT ) − k R NH (0) NIT (2)
dt
where NIT is the fraction of Si-H bonds at the Si/SiO2 interface which breaks at time t, N0 is
the initial number of all Si-H bonds, and k F is the dissociation rate constant. The second term
in (2) describes the annealing process of the released H atoms. NH (0) is the H concentration
at the interface.
NBTI is getting more serious as technology scales, since the vertical oxide field is continuously
increasing to enhance transistor performance. Thus a hole in the channel can be easily
captured and a two-electron Si-H covalent bond at the Si/SiO2 interface can be weakened
by it. The weakened Si-H bonds break easily at certain high temperature. Atomic H’s are
released in short time, then they convert to and diffuse as molecular H2 in long time (>100 s)
(Alam, Kufluoglu, Varghese & Mahapatra, 2007).
NBTI effect will degrade certain transistor parameters, such as threshold voltage, drain
current, transconductance, etc. Threshold voltage degradation due to NBTI is given by (Yan
et al., 2009)
Vgs α Ea
Vth = A exp − tn (3)
tox kT
where K is Boltzmann’s constant, Ea is the activation energy, n = 0.25 for atomic H in short
time, and n = 0.16 for molecular H2 in long time as discussed above.
The intrinsic variations of NBTI are studied in (Rauch, 2002). The expression of variation in
Vth shift is
Ktox μ(Vth )
σ(Vth ) = (4)
AG
where Tox is effective gate oxide thickness, AG is its area and K is an empirical constant.
It is pointed out in (Schroder & Babcock, 2003) that, NBTI should not exhibit any gate length
dependence, since it does not depend on lateral electric fields. But NBTI is sometimes
enhanced with reduced gate length, which is not well understood yet. The closeness of the
source and drain maybe one of the reasons for that.
3. State of the art

It is only since very recent years that the joint effects of process variations and lifetime
parameter degradations are proposed in literature. They differ in the type of reliability effects
considered and the type of circuits studied.
For digital circuits, NBTI-aware statistical timing analysis considering process variations are
proposed in (Vaidyanathan, Oates, Xie & Wang, 2009), (Vaidyanathan, Oates & Xie, 2009),
(Wang et al., 2008) and (Lu et al., 2009). Authors in (Vaidyanathan, Oates, Xie & Wang,
2009) build up gate-level delay fall-out model by propagating the device parameter fall-out
model due to NBTI and process variations into the gate delay model. They consider in
addition the intrinsic variations of NBTI process in (Vaidyanathan, Oates & Xie, 2009). Using
variation-aware gate delay model, the timing behavior of a path is modeled in (Wang et al.,
2008). Authors in (Lu et al., 2009) apply the NBTI aging-aware statistical timing analysis into
circuit level. All of those methods rely on the analytical expression of performance features,
which is suitable for digital circuits but difficult in analog domain.
For analog circuits, authors in (Maricau & Gielen, 2009) use Monte-Carlo simulation loop
to obtain the degraded performance values for each fresh random sample at every lifetime
point. Then the most appropriate distribution function at each time is fitted, thus a failure
distribution throughout the lifetime can be found. It results in a high simulation effort and
difficulty for further optimization. They improve their method in (Maricau & Gielen, 2010)
using response surface model to speed up the simulations, where certain numbers of random
samples are still required to obtain the degraded distribution information.
None of the above mentioned methods proposes an analog circuit design flow considering the
joint effects of process variations and lifetime parameter degradations. A new methodology
concerning such joint effects is thus required to help designer during the design phase.
4. Definitions
We first consider the fresh circuit here, i.e., no degradation is occurred. We distinguish three
types of parameter vectors,
• design parameters d, for example transistor widths and lengths, which are optimization
parameters of the analog sizing process.
• statistical parameters s, for example, Vth , tox , Le f f , etc, that have variations during
manufacturing process. They are usually modeled by Gaussian, log-normal or uniform
distributions. Without loss of generality, those distributions can be transformed into a
Gaussian distribution (Eshbaugh, 1992) with mean vector s0 and covariance matrix C in
the following form: s ∼ N (s0 , C), with

1 β2 (s)
pdf N (s) = √ ns √ · exp − (5)
2π · detC 2
whose level contours are ellipsoids
β2 (s) = (s − s0 ) T · C−1 · (s − s0 ) (6)
• range parameters r, for example supply voltage and temperature.

It is obvious that the integration of (5), the multi-dimensional probability density function
(pdf) of a Gaussian distribution, over the entire statistical parameter space is 1. However, we
have to consider the specified performance features also.
The performance vector f results from the output of a numerical circuit simulation, for
example gain, bandwidth, slew rate:
d, s, r → f(d, s, r) (7)
Each element of f has a certain lower bound and/or upper bound. As a result, in the statistical
parameter space, certain part of the Gaussian pdf will be cut off, since a part of parameter
variations falls out of the acceptance region bordered by the performance specifications. In
other words, from (7) a corresponding set of statistical parameters can be found to make
performances fulfill their specifications for all range parameter vectors within their acceptance
region Tr . Thus a statistical parameter acceptance region As (d), the shape of which depends
on d, can be defined as
A s (d) = s| ∀ fl ≤ f(d, s, r) ≤ fu (8)
r∈ Tr
Fresh yield Y is the percentage of manufactured circuits that satisfy the performance
specification considering statistical parameter variations.
Y = prob{ ∀ fl ≤ f(d, s, r) ≤ fu }
r∈ Tr
= prob{s ∈ As (d)} (9)
As pointed out in (Graeb, 2007) and (Antreich et al., 1994), the worst-case distance, β w , can
be used as an indicator of circuit robustness. It corresponds to the ellipsoid among the level
contours of Gaussian pdf that just touches the performance boundary at worst-case point sw :
β2w = (sw − s0 ) T · C−1 · (sw − s0 ) (10)

sw has the highest probability density among all boundary parameters. In a practical analog
circuit yield analysis and design centering flow, it is found numerically by solving
sw = arg min{ β2 (s)| f (s) = f l,u } (11)

s
β w can be interpreted as β w -sigma circuit robustness. The resulting fresh yield with respect to
one performance specification f i can be approximated by
β w,i
1
√ e− 2 ξ · dξ
1 2
Yi = (12)
−∞ 2π
for a transformed standard Gaussian distribution.
5. Lifetime yield and lifetime worst-case distance

Extending to the definition of lifetime yield, we take the circuit operating time t into
consideration. For the three types of parameters defined previously, statistical parameters
will degrade from its fresh value during lifetime. Consequentially the performance value will
also drift. Thus both of them are functions of operating time.
At time t, we have s(t) ∼ N (s0 (t), C(t)), and
d, s(t), r → f(d, s(t), r) (13)
The statistical parameter acceptance region now becomes
As (d, t) = {s(t)| ∀ fl ≤ f(d, s(t), r) ≤ fu } (14)

r∈ Tr
The corresponding lifetime yield Y (t) at time t is the percentage of circuits which can
still fulfill the performance specifications after parameter degradation. Since the original
distribution around s0 will shift to a new distribution with new mean vector s0 (t) and
covariance matrix C(t), a certain percentage of the fresh circuits which satisfied the
specification will fall out of the acceptance region after time t (Figure 3).
Lifetime yield at time t can be defined as
Y (t) = prob{ ∀ fl ≤ f(d, s(t), r) ≤ fu }

r∈ Tr
= prob{s(t) ∈ As (d, t)} (15)
Comparing with (12), we can see that the fresh yield is just a special case for lifetime yield
when t is set to 0 for the fresh circuit without any degradation (denoted as t = t0 from now
on). The value of lifetime yield thus can reflect the influence of parameter degradations.
s2
Worst-case distance of
fresh circuit
Performance boundary
Worst-case
degr
adat distance of
ion
s0(t0) aged circuit
s0(t)
s1
Acceptance region
Fig. 3. Lifetime worst-case distances of fresh and aged circuits with corresponding ellipsoids
(in thick) during lifetime degradation for one performance specification.
Figure 3 shows lifetime worst-case distance of fresh and aged circuits in statistical parameter
space with corresponding ellipsoids (in thick) for one performance specification both at time
t0 and t. It is assumed from now on that worst-case distance degrades monotonically.
As can be seen, during lifetime degradation both s0 (t) and C(t) change their values, a part
of statistical parameters around worst-case parameter sw (t0 ) thus fall out of the acceptance
region As,i (d) for the ith performance feature in vector f. Worst-case distance decreases,
leading to a decreasing lifetime yield. Note that the mean vector s0 (t) still fulfills the
specification, but the parameters around sw (t0 ) are very sensitive to the degradation, since
they already locate on the boundary of As,i (d) before degradation occurs.
The value of lifetime yield Yi (t) with respect to one performance feature f i can be estimated
by:
β w,i (t)
1
√ e− 2 ξ · dξ
1 2
Yi (t) = (16)
−∞ 2π
where β w,i (t) is the corresponding lifetime worst-case distance
β2w,i (t) = (sw (t) − s0 (t)) T · C(t)−1 · (sw (t) − s0 (t)) (17)
Table 1 shows the correspondence between worse-case distance and yield.
βw -3 -2 -1 0 1 2 3
Yield 0.1% 2.2% 15.8% 50% 84.1% 97.7% 99.9%
Table 1. Worst-case distances and corresponding yield value

at any time t.
For example, β w,i (t0 ) = 3 refers to a 3-sigma design at t0 of a performance feature f i . The total
lifetime yield Y (t) is bounded by:
1 − ∑(1 − Yi (t)) Y (t) min Yi (t) (18)

i i
A smaller worst-case distance during lifetime leads to more significant yield loss, thus it is
important in our new design flow to analyze and optimize the worst-case distances and the
corresponding yield values during lifetime to ensure a robust design.
6. Sizing rules
As shown in (Massier et al., 2008), sizing rules of the analog circuits are the constraints that
must be satisfied during circuit sizing. They include, for example, geometry constraints (e.g.,
transistor width, length, area) and electrical constraints (e.g., transistor gate-source voltage
Vgs , drain-source voltage Vds ). They are used to ensure the proper functionalities of the
circuits, for example, preventing the transistors from entering the inappropriate operation
regions, or limiting the voltage difference of Vds in a transistor pair to a certain value, etc.
It is known that some of the sizing rules for the fresh circuit will not be fulfilled after the step of
lifetime yield optimization carried out on the aged circuit (Pan & Graeb, 2009). Which means,
even if the fresh yield happened to be high after the step of lifetime yield optimization, the
resulting circuit is very sensitive to the process variations at fresh time.
Considering such sizing rules for both fresh and aged circuits, we apply the fresh and aged
sizing rules checking during the lifetime yield optimization process, which will ensure the
functionality and robustness of both fresh and aged circuits.
7. New design flow

The proposed lifetime yield optimization flow uses a tool WiCkeD (Antreich et al., 2000) and
aging simulator RelXpert from Cadence with NBTI and HCI degradation engines. The lifetime
yield of the analog circuit is optimized by maximizing both the fresh worst-case distance
and lifetime worst-case distance values, considering the sizing constraints for both fresh and
degraded circuits. The result of the flow ensures a robust analog design tolerant of both
process variations and lifetime parameter degradations, at the cost of an additional layout
area.
7.1 Software
EDA tools that contain various degradation models are available today. One of the most
famous tools is the Berkeley Reliability Tools (BERT) (Tu et al., 1993), which is the origin of
the tool RelXpert by Cadence today. Our proposed new design flow does not rely on specific
tools or models. Here we take RelXpert as an exemplary degradation tool for demonstration
purpose.
A simplified working flow of RelXpert is shown in Figure 4. It can generate degraded
BSIM3/4 model cards for each transistor at a specified time t, taking the fresh circuit
netlist and Cadence’s AgeMOS model as input. The transistor degradation as well as the
degraded circuit netlist at time t are produced, ready for SPICE simulation to get a degraded
performance.
Fresh AgeMOS
Netlist model
RelXpert
Degraded
Netlist
Fig. 4. Simplified work flow of RelXpert from Cadence.

For design centering/yield optimization, we exemplarily use the design optimization
software WiCkeD (Antreich et al., 2000). Its yield analysis and optimization algorithms
are based on worst-case distances mentioned above. It can always produce an optimized
design parameter vector towards a maximized yield considering distributions of statistical
parameters.
7.2 New design flow

Considering both process variation and lifetime degradation, our new design flow to analyze
and optimize the lifetime yield is shown in Figure 5.
In the new design flow, the fresh circuit with initial design parameters d is first optimized
to obtain a maximum fresh yield, producing d for optimal yield. It involves internal loops
between circuit simulator and WiCkeD. During this step, the fresh sizing rules are checked.
Then the circuit lifetime yield optimization for a specified time point t is performed on
initial d
Fresh Yield Circuit Simulation

Fresh sizing rules
Optimization with
check
by WiCkeD Fresh Netlist
d for optimal yield
Fresh sizing rules

check Lifetime Yield Circuit Simulation
Optimization RelXpert with
Degraded sizing rules by WiCkeD Degraded Netlist
check
d for optimal yield and reliability
Fig. 5. The new design flow with reliability optimization.
the degraded netlist generated by RelXpert. Note that for each internal optimization loop,
an updated netlist from WiCkeD will be given to RelXpert to obtain a renewed version of
degraded netlist. During this step, both the fresh and degraded sizing rules are checked to
ensure the correct functionality of the circuit both at fresh time and after degradation. The
final obtained design parameters d for optimal yield and reliability are the resulting solution
of the design flow.
Fresh yield optimization step ensures that the smaller worst-case distances will be increased,
thus the fresh design is centered such that it is already less sensitive to parameter drift. This
provides a reasonable starting point for lifetime yield optimization, since the influence of
parameter degradation on the performance and yield is kept at minimum level.
After the lifetime yield optimization, optimized design parameters are obtained such that
any decreasing worst-case distances during lifetime are increased again as much as possible.
The design is centered now such that the most degradation-sensitive worst-case distance will
be kept maximum. The resulting design solution is thus optimal considering both process
variations and lifetime degradations.
8. Prediction: Speed up the β w (t) evaluation

In this section, a prediction model of lifetime worst-case distance in time domain is presented
to speed up the analysis of lifetime yield value. Only performance and statistical parameter
sensitivity analysis are needed, in comparison to the Monte-Carlo simulation method and
numerical optimization solutions. It is based on the linear performance model as follows. The
index i of ith performance in vector f is left out for simplicity. Without loss of generality, only
upper bound f u is considered hereafter.
8.1 Linear performance model

At any time t during the lifetime, the first-order Taylor expansion of performance f (t) with
respect to s(t) from worst-case point sw,u in s space is
f (s(t)) ≡ f (t) ≈ f (sw,u (t)) + ∇ f (sw,u (t)) T · (s(t) − sw,u (t)) (19)
By assuming a linear performance model, the sensitivity of performance over statistical
parameters keeps constant, i.e.,
∇ f (sw,u (t)) ≡ g (20)
is constant over the entire s space at any time. Thus the level contours of f in s space are
equidistant lines as illustrated in dashed lines in Figure 6. f (sw,u ) in (19) is the upper bound
value f u . So from (19) the linear performance model at t can be formulated as
f (t) ≈ f u + gT · (s(t) − sw,u (t)) (21)
s2
sw,u(t0)
g
sw,u(t)
degr
adat
ion
s0(t0) s0(t)
s1
Fig. 6. Linear performance model during lifetime degradation in statistical parameter space
(dashed lines are equidistant level contours of f , ellipsoids are level contours of statistical
parameters).
sw,u (t) is called worst-case statistical parameter vector at t. It is the statistical parameter vector
where the corresponding performance f reaches its boundary value f u at t. It corresponds
to the position in s space where the probability of occurrence reaches it s maximum in the
non-acceptance region (slashed area in Figure 6). A robust design indicates that such a
probability of occurrence should be kept minimum, i.e., sw,u should be positioned furthest
away from s0 (t) so that it is least sensitive to the s changes which may cause it fall into
non-acceptance region.
Since s(t) ∼ N (s0 (t), C(t)), the mean and the variance of the linearized performance model
can be formulated from (21) as
μ( f (t)) = f u + gT · (s0 (t) − sw,u (t)) (22)
σ2f (t) = g ·C·g ≡
T
σ2f (23)
where (23) is constant over time. Taking the process variation as second order effects on the
sensitivity towards degradation, C(t) is assumed to be constant, i.e., C(t) = C (Sobe et al.,
2009).
Considering parameter degradation from t0 to t, a first-order Taylor approximation of μ( f (t))

with respect to t from t0 can be expressed as
∂μ f
μ( f (t)) = μ( f (t0 )) + | t0 · ( t − t0 ) (24)
∂t
From (22) we have
μ( f (t0 )) = f u + gT · (s0 (t0 ) − sw,u (t0 )) (25)
and
∂μ f ∂s0 (t) ∂sw,u (t)
| t0 = g T · | t0 − | t0 (26)
∂t ∂t ∂t
It can be observed from (26) that the product
∂sw,u (t)
gT · | t0 (27)
∂t
∂s (t)
remains zero, since the two vectors g and w,u ∂t |t0 are orthogonal to each other. This is easy to
understand because during the degradation of s parameters, the worst-case point sw,u moves
along the performance boundary f u , as can be observed in Figure 6, while the performance
gradient g always points to the direction that is vertical to that boundary in the performance
model.
So (26) becomes
∂μ f ∂s (t)
| t = g T · 0 | t0 (28)
∂t 0 ∂t
and (24) can be further expressed as
∂s0 (t)
μ( f (t)) = f u + gT · (s0 (t0 ) − sw,u (t0 )) + gT · | t0 · ( t − t0 ) (29)
∂t
8.2 Prediction of β w,u (t)

To predict β w,u (t), a first-order Taylor expansion of β w,u (t) with respect to t from t0 is
dβ w,u (t)
β w,u (t) = β w,u (t0 ) + | t0 · ( t − t0 ) (30)
dt
dβ (t)
where the sensitivity part, w,u
dt |t0 can be derived using results from Section 8.1 as follows.
Since at the worst-case point sw,u (t), the corresponding level contour of s(t) is
β2w,u (t) = (sw,u (t) − s0 (t)) T · C−1 · (sw,u (t) − s0 (t)) (31)
It touches the performance boundary at sw,u (t), which means the orthogonal on (31) is parallel
to g:
C−1 · (sw,u (t) − s0 (t)) = λ · g (32)
Inserting (32) into (31) we have
β2w,u (t) = λ2 · gT · C · g (33)
By taking λ from (33) into (32) we obtain

β w,u (t)
(sw,u (t) − s0 (t)) = ·C·g (34)
gT · C · g
Then (34) is taken back into (22):

μ( f (t)) = f u − β w,u (t) · gT · C · g (35)
so that the worst-case distance at t can be expressed as

f u − μ( f (t))
β w,u (t) = (36)
gT · C · g
Then from (36) and (29) the worst-case distance degradation rate can be formulated as
dβ w,u (t) 1 ∂s (t)
| t0 = − · g T · 0 | t0 (37)
dt σf ∂t
which differs from (5) in (Sobe et al., 2009). From (37) it is clear that the evaluation of the
worst-case distance degradation rate for a performance upper bound involves only multiple
sensitivity evaluations which can be carried out efficiently. Especially in our case, both σ f and
g remain constant, requiring an one-time evaluation only. The sensitivity of s0 (t) over t is
calculated by finite-difference approximation. The values of s0 (t) at respective time points are
obtained from exemplary aging simulator in our experiment described in Section 7, then the
corresponding sensitivity and the worst-case distance degradation rate can be evaluated.
Thus, by taking (37) back into (30), the values of β w,u (t) at time t can be predicted
efficiently without searching for the worst-case statistical parameters sw,u (t) through iterative
optimization method.
9. Experimental results
Vdd
MP3 MP4 MP5
Ibias Cmiller
Vout
Vin- Vin+
MP1 MP2
MN3
MN1 MN2
Vss
Fig. 7. Circuit topology of Miller OpAmp used in the experiment.

The circuit structure of the two stage Miller OpAmp used in the experiment is shown in Figure
7. The first stage is the differential stage, with the input differential pair, consisting of PMOS
MP1 and MP2, and its active load, a current mirror consisting of NMOS MN1 and MN2. The
second stage is a CMOS inverter with an NMOS MN3 as driver and a PMOS MP5 as its active
load.
It is clear from the circuit structure that certain sizing constraints on transistors concerning the
node voltages impose certain stress levels of each transistor.
9.1 Results of the new design flow
yield-optimal reliability-optimal
fresh 10 years 10 years fresh
Gain≥80dB 4.0 3.9 3.9 3.9
Slew Rate ≥ 3V/μs 4.2 1.9 3.4 5.8
GBW ≥ 2MHz 5.8 5.7 5.8 5.9
Phase Margin≤ 120deg 5.2 4.3 5.1 5.9
Power≤ 2mW 5.9 5.8 6.2 6.6
CMRR≥80dB 3.4 2.2 3.3 4.2
Relative Area 100% 107%
Lifetime Yield 99.96% 94.50 % 99.93% 99.99%
Table 2. Experimental results of the new design flow with reliability optimization.
We apply the new design flow in Figure 5 to the Miller OpAmp as introduced above. One
of the stop criteria of the tool WiCkeD during fresh or lifetime yield optimization process,
the maximum yield difference between two consecutive iterations, is set to 0.1%. That is,
the fresh or lifetime yield optimization stops if the improvement of the yield value between
two consecutive iterations is smaller than 0.1%. A 180nm technology is used with a supply
voltage of 1.7V. The circuit is degraded to time t=10 years with example AgeMOS degradation
model parameters inside RelXpert. The covariance matrix of statistical parameters is assumed
to be constant over time. Table 2 shows the simulation results. Six of the performances are
considered here, namely, DC Gain, Rising Slew Rate (SR), Gain-Bandwidth Product(GBW),
Phase Margin, Power and Common-Mode Rejection Ratio (CMRR).
From result of fresh yield optimization we can see that the fresh circuit design is centered with
99.96% fresh yield, the corresponding design parameters are initial d at t0 . After degradation
to 10 years with the same design parameters, all of the performances and worst-case distances
will degrade, as well as the lifetime yield, which is only 94.50% now. Then a design centering
on the degraded circuit is performed during lifetime yield optimization step. The result
shows that the degraded circuit will have a lifetime yield of 99.93% with increased worst-case
distances. Thus a design solution d for optimal yield and reliability is found.
Verification result on last column shows that with this optimized design, fresh circuit at t0 will
be centered to a better position in terms of both fresh yield and lifetime yield. The fresh yield
is 99.99%, and almost all of the worst-case distances here are much bigger compared to the
fresh design where no degradation is considered.
For the price we pay for the more robust circuit, the approximated total area of the circuit
layout is evaluated. For the area of a transistor, it is simply the product of the width and the
length. For the area of the Miller capacitor, it is transformed into the corresponding area by a
constant. The results in Table 2 show that 7% more relative layout area is needed for the more
robust circuit.
9.2 Results of the prediction model

To verify the prediction model of (30), the lifetime worst-case distance values obtained from
the tool WiCkeD and the prediction model are compared for two performances, SR and
CMRR. The comparison results and relative errors at different t’s are plotted in Figure 8 and
Figure 9. It is clear from the results that the prediction model can track the β w (t) degradation
with an acceptable error. For the simulation time, it takes five minutes on average for WiCkeD
to evaluate one β w (t) for one performance at t, while using the prediction model it takes only
about forty seconds. A clear speedup about eight times is observed.
Lifetime worst-case distance
2.8 0.08
2.6 0.07
0.06
Relative error
2.4
0.05
of SR
2.2
0.04
2.0
0.03
1.8 Accurate 0.02
1.6 Prediction 0.01
1.4 0.00
2 4 6 8 2 4 6 8
Year Year
(a) Comparison of the accurate and (b) Relative error of the prediction
predicted values of β w (t) at different t
Fig. 8. Prediction results on SR
3.4 0.12
Lifetime worst-case
distance of CMRR
3.2 Accurate 0.10

Relative error
3.0 Prediction 0.08

2.8 0.06
2.6
0.04
2.4
2.2 0.02
2.0 0.00
2 4 6 8 2 4 6 8
Year Year
(a) Comparison of the accurate and (b) Relative error of the prediction
predicted values of β w (t) at different t
Fig. 9. Prediction results on CMRR
10. Conclusion
As semiconductor technology continuously scales, the joint effects of manufacturing process
variations and parameter lifetime degradations have been a major concern for analog circuit
designers, since the deviation of performance values from the nominal ones will impact both
the fresh yield and lifetime yield.
In this chapter, a new analog design flow with reliability optimization is presented. The effect
of both process-induced parameter variation and time-dependent parameter degradation
can be analyzed automatically. The remaining lifetime yield of the designed circuit can
be predicted and optimized early in the design phase. After lifetime yield optimization,
simulation results show that a more reliable design is achieved, tolerant of both process
variation and lifetime degradation.
A prediction model for the lifetime worst-case distances is proposed to speed up the analysis
of lifetime worst-case distance values. The experimental results show that the model can
effectively evaluate during design phase the remaining lifetime yield of the circuits after
degradation occurs in their lifetime.
11. References
Alam, M. A., Kufluoglu, H., Varghese, D. & Mahapatra, S. (2007). A comprehensive model for
PMOS NBTI degradation: Recent progress, Microelectronics Reliability 47(6): 853–862.
Alam, M., Kang, K., Paul, B. & Roy, K. (2007). Reliability-and process-variation aware design
of VLSI circuits, Proceedings of the 14th IEEE International Symposium on the Physical
and Failure Analysis of Integrated Circuits (IPFA), Bangalore, India, pp. 17–25.
Antreich, K., Eckmueller, J., Graeb, H., Pronath, M., Schenkel, F., Schwencker, R. & Zizala, S.
(2000). WiCkeD: Analog circuit synthesis incorporating mismatch, Proceedings of the
IEEE Custom Integrated Circuits Conference (CICC), Orlando, USA, pp. 511–514.
Antreich, K., Graeb, H. & Wieser, C. (1994). Circuit analysis and optimization driven by
worst-case distances, IEEE Transactions on Computer-Aided Design of Integrated Circuits
and Systems 13(1): 57–71.
Eshbaugh, K. (1992). Generation of correlated parameters for statistical circuit simulation,
IEEE Transactions on Computer-Aided Design 11(10): 1198–1206.
Gielen, G., De Wit, P., Maricau, E., Loeckx, J., Martín-Martínez, J., Kaczer, B., Groeseneken,
G., Rodríguez, R. & Nafría, M. (2008). Emerging yield and reliability challenges
in nanometer CMOS technologies, Proceedings of the Design, Automation and Test in
Europe (DATE), Munich, Germany, pp. 1322–1327.
Graeb, H. (2007). Analog Design Centering and Sizing, Springer, The Netherlands.
Hu, C., Tam, S., Hsu, F., Ko, P., Chan, T. & Terrill, K. (1985). Hot-electron-induced MOSFET
degradation– model, monitor, and improvement, IEEE Journal of Solid-State Circuits
20(1): 295–305.
Jha, N., Reddy, P., Sharma, D. & Rao, V. (2005). NBTI degradation and its impact for analog
circuit reliability, IEEE Transactions on Electron Devices 52(12): 2609–2615.
Liu, Z., McGaughy, B. & Ma, J. (2006). Design tools for reliability analysis, Proceedings of the
ACM/IEEE Design Automation Conference (DAC), San Francisco, USA, pp. 182–187.
Lu, Y., Shang, L., Zhou, H., Zhu, H., Yang, F. & Zeng, X. (2009). Statistical reliability
analysis under process variation and aging effects, Proceedings of the ACM/IEEE
Design Automation Conference (DAC), San Francisco, USA, pp. 514–519.
Maricau, E. & Gielen, G. (2009). Efficient reliability simulation of analog ICs including
variability and time-varying stress, Proceedings of the Design, Automation and Test in
Europe (DATE), Nice, France, pp. 1238–1241.
Maricau, E. & Gielen, G. (2010). Variability-aware reliability simulation of mixed-signal ICs
with quasi-linear complexity, Proceedings of the Design, Automation and Test in Europe
(DATE), Dresden, Germany, pp. 1094–1099.
Martin-Martinez, J., Rodríguez, R., Nafria, M. & Aymerich, X. (2009). Time-dependent
variability related to BTI effects in MOSFETs: Impact on CMOS differential
amplifiers, IEEE Transactions on Device and Materials Reliability 9(2): 305–310.
Massier, T., Graeb, H. & Schlichtmann, U. (2008). The sizing rules method for CMOS
and bipolar analog integrated circuit synthesis, IEEE Transactions on Computer-Aided
Design of Integrated Circuits and Systems 27(12): 2209–2222.
Nassif, S. R. (2008). Process variability at the 65nm node and beyond, Proceedings of the IEEE
Custom Integrated Circuits Conference (CICC), San Jose, USA, pp. 1–8.
Pan, X. & Graeb, H. (2009). Degradation-aware analog design flow for lifetime yield analysis
and optimization, Proceedings of the 16th IEEE International Conference on Electronics,
Circuits and Systems (ICECS), Yasmine Hammamet, Tunisia, pp. 667–670.
Pan, X. & Graeb, H. (2010). Reliability analysis of analog circuits by lifetime yield prediction
using worst-case distance degradation rate, Proceedings of the 11th IEEE International
Symposium on Quality Electronic Design (ISQED), San Jose, USA, pp. 861–865.
Rauch, S. E. (2002). The statistics of NBTI-induced VT and β mismatch shifts in pMOSFETs,
IEEE Transactions on Device and Materials Reliability 2(4): 89–93.
Schroder, D. K. & Babcock, J. A. (2003). Negative bias temperature instability: Road to cross
in deep submicron silicon semiconductor manufacturing, Jounal of Applied Physics
94(1): 1–18.
Sobe, U., Rooch, K., Ripp, A. & Pronath, M. (2009). Robust analog design for automotive
applications by design centering with safe operating areas, IEEE Transactions on
Semiconductor Manufacturing 22(2): 217–224.
Tu, R., Rosenbaum, E., Chan, W., Li, C., Minami, E., Quader, K., Ko, P. & Hu, C.
(1993). Berkeley reliability tools-BERT, IEEE Transactions on Computer-Aided Design
of Integrated Circuits and Systems 12(10): 1524–1534.
Vaidyanathan, B., Oates, A. & Xie, Y. (2009). Intrinsic NBTI-variability aware statistical
pipeline performance assessment and tuning, Proceedings of the IEEE/ACM
International Conference on Computer-Aided Design, San Jose, USA, pp. 164–171.
Vaidyanathan, B., Oates, A., Xie, Y. & Wang, Y. (2009). NBTI-aware statistical circuit delay
assessment, Proceedings of the 10th International Symposium on Quality Electronic Design
(ISQED), San Jose, USA, pp. 13–18.
Wang, W., Reddy, V., Krishnan, A., Vattikonda, R., Krishnan, S. & Cao, Y. (2007). Compact
modeling and simulation of circuit reliability for 65-nm CMOS technology, IEEE
Transactions on Device and Materials Reliability 7(4): 509–517.
Wang, W., Reddy, V., Yang, B., Balakrishnan, V., Krishnan, S. & Cao, Y. (2008). Statistical
prediction of circuit aging under process variations, Proceedings of the IEEE Custom
Integrated Circuits Conference (CICC), San Jose, USA, pp. 13–16.
Yan, B., Qin, J., Dai, J., Fan, Q. & Bernstein, J. (2009). Reliability simulation and circuit-failure
analysis in analog and mixed-signal applications, IEEE Transactions on Device and
Materials Reliability 9(3): 339–347.
7
Linear Analog Circuits Problems by

Means of Interval Analysis Techniques
Zygmunt Garczarczyk
Silesian University of Technology, Gliwice
Poland
1. Introduction
Inevitable fluctuations in the manufacturing processes and environmental operating
conditions of linear analog circuits cause circuit parameters to vary about their nominal
target values. The mathematical model of an engineering system evaluated by a transfer
function (e.g. of an active and even passive circuit) never describes exactly the system’s
behavior. The changes in the performance of linear circuit due to the variations in circuit
parameters are of great practical importance in engineering analysis and design. The
tolerance problem for linear analog circuit have been extensively studied and many results
have been published, e.g. (Antreich et al., 1994; Spence & Soin, 1997). Because of
uncertainties, the values of the parameters of a given circuit may be treated as belonging to
some intervals. In recent years, interval analysis becomes powerful tool for tolerance
computations of some design problems (Kolev et al., 1988; Femia & Spagnuolo, 1999).
Some results have been reported using algorithms for linear interval equations for
solving tolerance problems (Tian et al., 1996; Garczarczyk, 1999; Shi et al., 1999; Tian & Shi,
2000).
The structure of the chapter is the following: section 2 explains an interval analysis
techniques for linear analog tolerance problem. In that approach we are interested in
calculation tolerances (the range of values) for real and imaginary part of transfer function
with respect to change of one parameter of the circuit. Section 3 deals with the problem of
computing the frequency response of an uncertain transfer function whose numerator and
denominator are interval polynomials. Studying a solution set of corresponding 2×2 linear
interval equation one can obtain bounds on the frequency response. Using Kharitonov
polynomials family and complex interval division it’s also possible to evaluate the bounds.
In this section we compare results obtained by applying presented approaches. Numerical
studies are also reported in order to illustrate presented methods.
2. Evaluation of linear circuits tolerances

The objective of this section is to develop the interval analysis techniques for linear analog
circuit tolerance problem. In that approach we can compute effectively tolerances for real
and imaginary parts of the transfer function with respect to change of one parameter of a
circuit.
2.1 Bilinear and biquadratic form of a circuit function

The functional dependence of circuit performance on the designable parameters is known
implicitly through the circuit transfer function. If the dependence on the R, L, C elements
and on the controlled sources is investigated, the transfer function is a quotient of two linear
polynomials, i.e., a bilinear relation, is arrived at. We have the following well-known result:
L(s , x ) A( s) + xB(s )
F(s , x ) = = (1)
M(s , x ) C(s) + xD(s)
In the above equation the symbol x denotes dependence on the network element parameter
(R or L or C or gain of the controlled source). A(s), B(s), C(s) and D(s) are functions of the
complex frequency s. They depend on kind of transfer function and on the structure of a
circuit examined. A similar biquadratic relation was derived for the dependence on the ideal
transformer ratio n, on the ideal gyrator resistance r and on the conversion factor k of the
ideal negative impedance converter (Geher, 1971). The transfer function has the following
form:
L(s , x ) A( s ) + xB(s ) + x 2 C(s )

F(s , x ) = = (2)
M(s , x ) D(s ) + xE( s ) + x 2 G(s )
A(s), B(s), etc. are depending on the type of the transfer function and the topology of the
circuit. For some fixed frequency transfer function can be represented by its real and
imaginary part, i.e.
L 1 (ω, x ) L (ω, x )
F( x ) = F( jω, x ) = +j 2 (3)
M 1 (ω, x ) M 1 (ω, x )
Here L1(ω,x), L2(ω,x), M1(ω,x) denote polynomials in x of second order and fourth order
(maximally) for bilinear and biquadratic transfer functions, respectively. We are interested
in calculation tolerance (the range of values) for real and imaginary part of the transfer
function caused by some parameter x ranging in known interval, i.e. x ∈ x = [ x, x ] .
This one-parameter tolerance problem can be solved by means of the well-known circle
diagram method for bilinear transfer function, unfortunately biquadratic transfer function is
more difficult problem. Here we propose a unified approach to tolerance problem for
bilinear and biquadratic transfer function based on the range evaluation of a rational
function by means of interval analysis techniques.
2.2 Range values of a rational function

Let L(x) be a polynomial of degree n and M(x) a polynomial of degree m so that
f(x) = L(x)/M(x) is a rational function. We want to expand f(x) into its Taylor series
k
f( x) = ∑ c (x − x )
i =0
i 0
i
(4)
For computing the first k Taylor coefficients of f(x) at some point x0 where M(x0) ≠ 0, we
start by developing the polynomial L(x) into its Taylor series about the point x0
Linear Analog Circuits Problems by Means of Interval Analysis Techniques 149
n
L( x ) = ∑ a (x − x )
i =0
i 0
i
(5)
Similarly, let
m
M( x ) = ∑ b (x − x )
i =0
i 0
i
(6)
Note that max(m,n) = 2 or 4.

Coefficients ai and bi are obtained directly as
ai=L(i)(x0)/i! , bi=M(i)(x0)/i! , (7)

i = 1,2,...,m(n)
More effectively we can compute them by using the extended Horner scheme (Elden &
Wittmeyer-Koch, 1990).
It was derived in (Garczarczyk, 1995) that one can compute the values of the first k Taylor
coefficients of a rational function by solving a (k + 1)×(k + 1) lower triangular Toeplitz
system of the form:
⎡b0 0 ⎤ ⎡c0 ⎤ ⎡a 0 ⎤
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ b1 b0 ⎥ ⎢ c1 ⎥ ⎢ a 1 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢b2 b1 b0 ⎥ ⎢c 2 ⎥ ⎢a 2 ⎥
⎢ ⋅ ⎥⎢ ⋅ ⎥ = ⎢ ⋅ ⎥ (8)
⋅ ⋅ ⋅
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ ⋅ ⋅ ⋅ ⋅ ⎥⎢ ⋅ ⎥ ⎢ ⋅ ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ ⋅ ⋅ ⋅ ⋅ ⎥⎢ ⋅ ⎥ ⎢ ⋅ ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣b k b2 b1 b 0 ⎦ ⎣c k ⎦ ⎣a k ⎦
Note that for the case k > m(n), the lower triangular Toeplitz system is lower banded.
To compute the values of the Taylor coefficients of a rational function the main work is to
solve the lower triangular Toeplitz system (8). Special structure of Toeplitz systems leads to
the variety of solving algorithms, so they belong to more elaborated linear systems. Because
system (8) is lower triangular for a small k, we can use the usual forward substitution method
for its solving. For large k more efficient method is a variant of Trench algorithm for Toeplitz
band matrices (Trench, 1985). Inversion of a nonsingular Toeplitz matrix of the form
⎡b0 0⎤
⎢ ⎥
⎢ b1 b0 ⎥
⎢ ⎥
⎢b2 b1 b0 ⎥
(9)
T=⎢ ⋅ ⋅ ⋅ ⋅ ⎥
⎢ ⎥
⎢ ⋅ ⋅ ⋅ ⋅ ⎥
⎢ ⎥
⎢ ⋅ ⋅ ⋅ ⋅ ⎥
⎢ ⎥
⎣b k b2 b1 b0 ⎦
band or not may be computed following:

Let (without loss of generality) b0 = 1, then
T-1 = [hrs]kr,s=0 (10)

is the matrix given by
hrs = -ψr-s-1 , r = 0,1,...,k , s = 0,1,...,r (11)

with ψj = 0 if j < -1, ψ-1 = -1, and
j −1
ψ j = b j+ 1 − ∑b
s =0
j −s ψ s , 0≤j≤k–1. (12)
Note that matrix T-1 is also lower triangle Toeplitz matrix and is uniquely determined by its
first column (h00,...,hk0)t = (-ψ-1,-ψ0,...,-ψk-1)t. The solution
[c0,c1,...,ck]t = T -1 [a0,a1,...,ak]t (13)

of (8) can be calculated by using the fast Fourier transform.
For any function f(x) which has an interval arithmetic evaluation the range of values of f
over the interval x
R(f,x) := {f(x)⏐ x ∈ x} (14)

is contained in the interval arithmetic evaluation f(x), i.e.
R(f,x) ⊆ f(x) (15)

Additionally, it is strongly dependent on the arithmetic expression which is used for the
interval evaluation of the function (Neumeier, 1990; Moore et al., 2009).
Exact Taylor expansion for a rational function f(x) is following
f(x) = p(x) + r(x) (16)

where
k k
⎛r⎞
p( x) = ∑
i =0
α i x i , with α i = ∑ c ⎜⎜⎝ i ⎟⎟⎠(−x )
r =i
r 0
r −i
(17)
and
r(x) = f(k+1)(x0 + ξ(x - x0))(x - xk+1 )/(k + 1)! (18)

ξ ∈ [0,1] , x0 ∈ x (e.g. x0 = m(x)).
If f(x) : D ⊆ R → R is k + 1 times continuously differentiable, then for all x ⊆ D it’s fulfilled
(Garczarczyk, 1993):
(inclusion)
R(f,x) ⊆ V(f,x) := R(p,x) + f(k+1)(x)w(x)k+1/(k + 1)! (19)

(distance)
q(R(f,x),V(f,x)) ≤ γw(x)k+1 , γ≥0, (20)

where R(p,x) is the exact range of the polynomial p(x) over x, and
q(R,V) = max(|R - V|,| R - V |) means distance between intervals R = [ R , R ] and V = [ V , V ].
Relation (19) gives the way of range values evaluation: we need to calculate the range of
polynomial and the range of remainder term. It’s seen from (20) that the overestimation of
R(f,x) by V(f,x) decreases with a power k + 1 of w(x) (width of x), so if f(k+1)(x) is bounded we
can omit the remainder term in V(f,x) and then
R(f,x) ≈ R(p,x) (21)
2.3 Bernstein polynomials

Estimates for the maximum, resp. the minimum, of the polynomial over x are obtained by
computing Bernstein coefficients.
For some order v of Bernstein polynomial we have (Ratschek & Rokne, 1984)
min Bj ≤ min p(x) ≤ max p(x) ≤ max Bj , (22)

0≤j≤v,x∈x,
where v ≥ k and
⎛ j⎞
j k
⎜⎜ ⎟⎟
⎛t⎞ ⎝s⎠ ,
Bj = ∑∑ ⎜⎜ ⎟⎟ α t x t −s w( x )s
s
t =s ⎝ ⎠
⎛ v⎞
j = 0,1,...,v (23)
s =0 ⎜⎜ ⎟⎟
⎝s⎠
The coefficients Bj are computed using a following finite difference table
B0 B1 B2 " " " B ν−2 B ν −1 Bν

ΔB 0 " " " " " ΔB ν−1
Δ2 B 0 " Δ2 B ν−2 (24)
% $
Δν B 0
The initial slanted entries are generated basing on coefficients of polynomial p(x) following
k
⎛l⎞
Δr B 0 = A r ∑ ⎜⎜⎝ r ⎟⎟⎠α x
l =r
l
l −r
, (25)
k
⎛l⎞
∑ ⎜⎜⎝ r ⎟⎟⎠α x
l −r
Δr B ν−r = A r l (26)
l =r
−1
r⎛ν⎞
where A r = w( x ) ⎜⎜ ⎟⎟ , r = 0,1,...,ν, x = [ x , x] .
⎝r⎠
The top row of table contains the desired Bernstein coefficients. Finite differences are
computed following
Δr B j = Δr − 1B j + 1 − Δr − 1B j , r > 0, j = 0,1,...,ν. (27)

For example
ΔB 0 = B 1 − B 0 ⇒ B 1 = ΔB 0 + B 0 (28)
and
ΔB ν − 1 = B ν − B ν − 1 ⇒ B ν − 1 = B ν − ΔB ν − 1 . (29)
Relations (24) – (29) lead directly to the following scheme of computing of Bernstein
coefficients
B0 → (+) B1 → (+) B2 " " Bν − 2 (+) ← Bν −1 (+) ← Bν

(+) (+) (−) (−)
↑ ↑ ↑ ↑
ΔB 0 → ( + ) ΔB 1 " " ΔB ν − 2 ( + ) ← ΔB ν − 1
(+) (−) (30)
↑ ↑
Δ2 B 0 " " Δ2 B ν − 2
%$
It’s seen we can develop the algorithm of a parallel computation of Bernstein coefficients
starting from slanted entries. We note that since αl = 0 for l > k there is no need to compute
entries Δr B j for r > k ; a triangle table turns into trapezium one. In the trapezium table a
bottom row has all entries equal, i.e.
Δs B 0 = Δs B 1 = " = Δs B ν − s , ν > s . (31)
Realisation of scheme (30) leads to the three cases of parallel computation slightly different
according to the value of ν (Garczarczyk, 2002).
2.3 Numerical examples

To illustrate the basic ideas of our approach two examples are considered. The first example
refers to the bilinear transfer function and the second to the biquadratic one. Taylor
coefficients ai and bi i = 0,1,...,k, k = 2 or 4, were computed by means of extended Horner
scheme. For example, polynomial L(x) was developed by the algorithm written in Pascal-
like code as:
for i = 0,1,...,n
ai = coefficient(L(x));
for k = 0,1,...n
for i = n-1, n-2,...,k
ai = ai+1x0 + ai;
In both examples Toeplitz system (8) is banded and was solved using algorithm based on
Trench’s concept (10) – (13).
EXAMPLE 1. Consider a second-order low-pass filter section of Fig.1, originally proposed
by Sallen and Key.
C1
G1 G2
U1 A U2
C2
Fig. 1. Second-order low-pass filter section

Bilinear transfer function considered here is following
G G
x 1 2
U2 C 1C 2
F(s , x ) = =
U 1 s 2 + ( G 2 (1 − x) + G 1 + G 2 )s + G 1 G 2
C2 C1 C 1C 2
where x = A.
Assuming G1 = G2 = 1 and C1 = C2 = 1 for fixed frequency we obtain
( 1 − ω2 )x ωx − 3ω
F( x) = F( jω, x) = +j
M( ω , x ) M( ω , x )
where M(ω,x) = 1+7ω2+ω4-6ω2x+ω2x2 .

We have applied relation (21) for Taylor expansion of degree k = 5 and Bernstein coefficients
of degree v = 10 were used. For x ∈ x = A0[1-ε, 1+ε] with A0 = 1, ε = 0.01 we obtained results
presented in the Table 1. In the second column there are values of the ranges for real and
imaginary part of the transfer function, the third column contains their nominal values.
ω x∈x X = A0
0.2 [0.878692,0.899103] + j[-0.372772,-0.367971] 0.888889 - j0.370370
2.0 [-0.127097,-0.122930] + j[-0.168624,-0.164735] -0.125005 - j0.166667
20.0 [-0.024009,-0.023489] + j[-0.002395,-0.002367] -0.023749 - j0.002381
Table 1. Range values of transfer function of Sallen-Key low-pass section
EXAMPLE 2. Consider the gyrator circuit with feedback shown in Fig.2.
u1 u2
Fig. 2. Gyrator circuit

Let Y = sC1 and Z = 1/sC2. Biquadratic transfer function is of the form
U2 xsC 2
F(s , x ) = = 1+ 2 2
U1 x s C 1C 2 + 1
where x = r is the gyration resistance. This circuit appriopriately loaded can realize a
transfer function of phase equalizer.
For fixed frequency we have
U2 ωC 2 x
F( x ) = F( jω, x ) = = 1− j
U1 M(ω, x)
where M(ω,x) = ω2C1C2x2 – 1.

It was assumed for simplicity C1 = C2 = 1. For x ∈ x = r0[1-ε, 1+ε] with r0 = 2, ε = 0.05 we
have obtained following results
ω x∈x x = r0
0.1 1 + j[0.197086, 0.219623] 1 + j0.208333
1.0 1 – j[0.615803, 0.727913] 1 – j0.666667
10.0 1 – j[0.047712, 0.052745] 1 – j0.050125
Table 2. Range values of transfer function for circuit with gyrator
Degrees of Taylor and Bernstein coefficients were analogous to previous example.
3. Frequency response envelopes of interval systems

The computation of the frequency responses of uncertain transfer functions plays a major
role in the application of frequency domain methods for the analysis and design of robust
systems. There is a rich resource of prior works on this subject, e.g. (Bartlett et al., 1993;
Chen & Hwang, 1998a, 1998b; Tan & Atherton, 2000; Hwang & Yang, 2002; Tan, 2002;
Nataraj & Barve, 2003).
In this section we consider continuous-time systems characterized by rational transfer
functions. Motivated by the above we incorporate uncertainties into the transfer function.
We assume that the system’s performance is governed by the interval transfer function
N( s ) a 0 + a 1 s + " + a m s m
K( s ) = = (32)
D(s) b 0 + b 1s + " b n s n
where coefficients of numerator and denominator are not known exactly, but are given in
prescribed real intervals
ai ≤ ai ≤ ai , i = 0," , m
(33)
bj ≤ bj ≤ bj , j = 0 , " , n.
A problem of major importance and significance is to be able to determine the envelopes of

the amplitude and phase of K(jω) of the above family of transfer functions. Phase and
amplitude bounds have a simple geometric interpretation: they represent envelopes of the
Nyquist plot.
The objective of this section is to develop the interval analysis techniques to the problem
presented above. Focusing on this specific class of uncertain systems we compare two
approaches to computation of Nyquist plot collections.
3.1 Linear interval equations approach

In this section we collect some known results on the linear interval equations and their use
to the problem explained in the previous section. This approach was explicitly presented in
(Garczarczyk, 1999).
Let G(s) be the inverse of interval transfer function K(s). Introducing input signal x(jω) and
output signal y(jω) the input-output relationship for linear continuous-time system, can be
written as
x 1 (ω) + jx 2 (ω) = (Re{ G( jω, p)} + j Im{ G( jω, p)})( y 1 (ω) + jy 2 (ω)) (34)
where
x 1 (ω) = Re{x( jω)}, x 2 (ω) = Im{x( jω)},

and y 1 (ω) = Re{y( jω)},y 2 (ω) = Im{y( jω)}.
Assuming x1(ω)=1, x2(ω)=0 (sinusoidal input x(t) = cos(ωt) is applied) we can rewrite eq.(34)
as the system of two linear equations
⎡Re{ G( jω)} − Im{ G( jω)} ⎤ ⎡ y 1 (ω)⎤ ⎡1⎤

⎢Im{ G( jω)} = . (35)
⎣ Re{ G( jω)} ⎥⎦ ⎢⎣ y 2 ( ω)⎥⎦ ⎢⎣0 ⎥⎦
For a fixed frequency, we obtain following equation
⎡[a , b] −[ c , d]⎤ ⎡ y 1 ⎤ ⎡1⎤

⎢[ c , d ] [a , b] ⎥ ⎢ y ⎥ = ⎢0 ⎥ (36)
⎣ ⎦⎣ 2 ⎦ ⎣ ⎦
Here the ranges of values of Re{G( jω)} and Im{G( jω)} are represented by intervals [a, b]
and [c, d], respectively.
Equation (36) forms a system of linear interval equations. It can be denoted as
Ay = b (37)
Such a system represents a family of ordinary linear systems which can be obtained from it
by fixing coefficients values in the prescribed intervals. Every of these systems, under the
assumption that each A∈A is nonsingular, has a unique solution, and all these solutions
constitute a so-called solution set S.
The solution set of eq. (37) can be expressed as
S = {y : Ay = b ,A ∈ A ,b ∈ b} (38)
It forms some two-dimensional region of output values of a system in the sinusoidal steady-
state.
If interval matrix A is regular i.e. if det A≠0 for each A∈A, the solution set of a linear
interval equation is described by Oettli and Prager in their famous equivalence (Oettli &
Prager, 1964; Neumeier, 1990)
y ∈ S ⇔ Ay − b ≤ Δ y + δ (39)
where A=m(A), b=m(b) and Δ=w(A/2), δ=w(b)/2.

Applying Oettli-Prager formula to the equation (36) we obtain following inequality
⎡m 1 − m 2 ⎤ ⎡ y 1 ⎤ ⎡ 1 ⎤ ⎡ρ 1 ρ2 ⎤ ⎡y1 ⎤
⎢m − ≤ , (40)
⎣ 2 m 1 ⎥⎦ ⎢⎣ y 2 ⎥⎦ ⎢⎣0 ⎥⎦ ⎢⎣ρ 2 ρ 1 ⎥⎦ ⎢⎣ y 2 ⎥⎦
where m1 = (a+b)/2, m2 = (c+d)/2

and ρ1 = (b-a)/2, ρ2 = (d-c)/2.
Computation of the regions of values of y1 and y2 for which inequality (40) is true gives us
the full information about changes of frequency response caused by variations some of
system parameters. To obtain this information we solve inequality (40) for whole complex
plane. In Fig. 3 region of solutions (region of uncertainty) in the fourth quadrant is
represented by the tetragon ABCD. The straight lines l1 and l2 are here defined following
c d
l 1 :y 2 = − y1 , l 2 :y 2 = − y1 , (41)
b a
y2 1 1
b a y1
B
C
A
l1
D
1
− l2
d
1
−
c
Fig. 3. Region of uncertainty in the fourth quadrant

Calculation coordinates of the points of intersections in each quadrant leads to the bounds of
a frequency response.
At the border of two quadrants structure for the solution set is quite different. In Fig.4 is
shown a region at the border of III and IV quadrants, i.e. if m1=0 (a=-b) and m2>0.
The straight lines l1 and l2 are following
c c
l 1 :y 2 = − y 1 , l 2 :y 2 = y 1 (42)
b b
y2 1
1
b b y1
E 1
d A
B
D
1
c
F
C
l1
l2
Fig. 4. Structure of the solution set at the border of two quadrants
3.2 Kharitonov polynomials method

Problem of evaluating the frequency response envelopes can be treated as the task of finding
the maximum and minimum of P( jω) and of Arg [P(jω)] of a family of polynomials
P( s ) = α 0 + α 1 s + α 2 s 2 + " + α k s k
(43)
α i ≤ α i ≤ α i , i = 0, " , k
The value set of a polynomial with uncertain coefficients at a frequency ω denote the region
in the complex plane occupied by all the values of the polynomial over all allowable
coefficients values.
From (43) we have
P( jω) = Re{ P( jω)} + jJm{P( jω)} (44)
Formula (44) defines for every ω ∈ R, a linear transformation from the (k+1)-dimensional
real coefficient set to the complex plane. Assuming that the intervals of the coefficients are
independent, the (k+1)-dimensional interval vector (box) is mapped into a complex
rectangular interval (rectangle with edges parallel to the axes of the complex plane).
It has been observed in ( Dasgupta, 1988) that the corners of that rectangular interval clearly
correspond to the four Kharitonov polynomials (Kharitonov, 1979)
P1 ( jω) = α 0 + α 1s + α 2 s 2 + α 3 s 3 + " s = jω
P2 ( jω) = α 0 + α 1s + α 2 s 2 + α 3 s 3 + " s = jω
(45)
P3 ( jω) = α 0 + α 1 s + α 2 s 2 + α 3 s 3 + " s = jω
P4 ( jω) = α 0 + α 1s + α 2 s 2 + α 3 s 3 + " s = jω
From (45) it’s seen that the value sets of N(s) and D(s) are the members of the set of complex
rectangular intervals (is denoted here by R(C)).
They have the form
N( jω) = N = N 1 + jN 2 = [ n 1 , n 1 ] + j[ n 2 , n 2 ] , (46)
and
D( jω) = D = D 1 + jD 2 = [d 1 , d 1 ] + j[d 2 , d 2 ] (47)
To calculate value set of interval transfer function we need to divide those two complex
intervals. Complex interval operations should deliver the closest inclusion of the set of all
possible values, i.e.
{ a : b a ∈ N , b ∈ D} ⊆ N : D (48)
For rectangular complex arithmetic addition, subtraction and multiplication are optimal,
whereas division is not. We apply here an improved version of division (in the sense of
inclusion), namely (Rokne & Lancaster, 1971; Petkovic & Petkovic, 1998)
1
N : D = N⋅ (49)
D
where
1 ⎧⎪ ⎧1 ⎫ ⎫⎪
= inf ⎨X ∈ R(C ) ⎨ b ∈ D ⎬ ⊆ X ⎬ . (50)
D ⎪⎩ ⎩b ⎭ ⎪⎭
Relation (50) is illustrated in Fig. 5. for the interval D from the first quadrant.
Im
Re
E F
G H
Fig. 5. Optimal rectangular enclosure

Optimal enclosure has the form of rectangle EFGH. Curvilinear hatched region which was
generated by conformal mapping corresponds to the exact range of D −1 . The shape of the
exact region and adequate enclosure depend on the position of interval D on the complex
plane.
3.3 Numerical studies

To compare properties of presented approaches two examples are considered. The first
example refers to the transfer function of the form (32), the second one to the case
represented in the relation (50).
EXAMPLE 3. Let us consider T-bridged circuit depicted in Fig. 3. The frequency response is
represented by the transmittance (Chen, 2009)
U2 R 1 C 1 R 2 C 2 s 2 + (R 1 C 1 + R 2 C 2 )s + 1
K( s ) = =
U 1 R 1 C 1 R 2 C 2 s 2 + (R 1 C 1 + R 2 C 2 + R 2 C 1 )s + 1
R1 R2
U1 U2
C
Fig. 6. Bridget–T circuit

Let assume R1C1 = R2C2 = RC = [1-ε, 1+ε], ε = 0.05.
Then the interval transmittance is done as
K( s ) =
U2
=
[0.9025, 1.1025]s 2 + [1.9, 2.1]s + 1 .
U 1 [0.9025, 1.1025]s 2 + [2.85, 3.15]s + 1
0.2
0.15 ω = 5.0
0.1
ω =2.0
0.05
Imaginary Axis
0 ω=1 ω=0
-0.05
ω = 0.2
-0.1
-0.15
ω = 0.5
-0.2
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Real Axis
Fig. 7. Regions of uncertainty against a background of Nyquist plot

The ranges of values of Re{G( jω)} and Im{G( jω)} are computed with use of Taylor and
Bernstein representations.
2x
Re{G( jω)}∈ 1 + , for x = [0.95ω, 1.05ω]
(1 + x ) 2 2
Im{G( jω)}∈
(1 − x )x ,
2
for x = [0.95ω, 1.05ω]
(1 + x )
2 2
In Fig. 7 are presented the Nyquist plot for nominal value RC = 1 and the regions ABCD
(tetragon) and EFGH (rectangle) for two frequencies ω =0.2 and ω = 2.0. It gives us the
possibility to evaluate the envelope of Nyquist plot for these frequencies. It’s seen that
Kharitonov polynomials approach (rectangle) gives some overestimation compared with
linear interval equations method.
EXAMPLE 4. Consider a second-order low-pass Sallen - Key section of Fig.1
Let denote R1 = 1/G1 and R2 = 1/G2.
We have now a transmittance of the form
U2 1
K( s ) = = .
U 1 R 1 C 1 R 2 C 2 s 2 + (R 1 + R 2 )C 2 s + 1
Assuming R1C1 = R2C2/2= RC = [1-ε, 1+ε], ε = 0.1, we have
U2 1
K( s ) = =
U 1 [1.62 , 2.42 ]s 2 + [2.7 , 3.3]s + 1
Re{G( jω)}∈ 1 − x , for x = [1.62 ω 2 , 2.42 ω 2 ]
Im{G( jω)} = x , for x = [ 2.7 ω, 3.3ω]
In Fig. 8a and 8b are drawn fragments of Nyquist plot for nominal value RC = 1.0 and
appropriate regions for ω = 0.2 and ω = 1.0.
Although uncertainties in the Example 4 are greater then in previous one both methods
produce smaller regions. There are two reasons of such results: Firstly, the different
coefficients of the transfer function are sometimes dependent; secondly, improved division
defined by (49) is not optimal whereas relation (50) leads to the optimal enclosure.
4. Conclusions
An efficient and well motivated approach to the problem linear analog circuit tolerance was
described. One-parameter tolerance problem was solved for bilinear and biquadratic
transfer function. This unified method was based on the range evaluation of a quotient of
two polynomials of second or fourth order. It was done by computing coefficients of
Bernstein polynomials generated for some Taylor expansion (form) of a rational function.
The Taylor forms together with Bernstein expansions constitute a significant enhancement
of the toolkit of interval analysis, see also (Neumaier, 2002).
a)
-0.3
-0.4
ω = 0.2
Imaginary Axis
-0.5
-0.6
-0.7
0.5 0.6 0.7 0.8 0.9
Real Axis
b)
-0.1
-0.2
Imaginary Axis
ω = 1.0
-0.3
-0.4
-0.5
-0.2 -0.1 0 0.1 0.2
Real Axis
Fig. 8. Regions of uncertainty and Nyquist plot

The results presented in this chapter make it possible, by simple algorithms, to obtain the
Nyquist envelope (consequently the amplitude envelope and the phase envelope) of an
interval rational transfer function of a continuous-time system. It gives possibility to readily
check whether system with such uncertainty comply with frequency response specifications.
The results of the numerical calculations are quite satisfactory. It indicates that the interval
analysis seems to be a promising tool for robust analysis of linear systems. Numerical
studies show that it’s necessary next step to “more” optimal complex interval division
(Lohner & Wolff von Gudenberg, 1985; Moore et al., 2009).
5. References
Antreich, K.J.; Graeb, H.E., & Wieser, C.U. (1994). Circuit analysis and optimization driven
by worst-case distances, IEEE Trans. Computer-Aided Design, Vol. 13, No. 1,
pp. 57-71, ISSN: 0278-0070.
Bartlett, A.C.; Tesi., A. & Vicino, A. (1993). Frequency response of uncertain systems with
interval plants”, IEEE Trans. Automat. Contr., Vol. 38, No. 6, pp. 929-933,
ISSN: 1063-6536.
Chen, J.-J. & Hwang, C. (1998a). Computing frequency responses of uncertain systems”,
IEEE Trans. Circuits Syst. I, Vol. 45, No. 3, pp. 304-307, ISSN: 1057-7122.
Chen, J.-J. & Hwang, C. (1998b). Value sets of polynomial families with coefficients
depending nonlinearly on perturbed parameters, IEE Proc. – Control Theory and
Applications, vol. 145, No. 1, pp. 73-82, ISSN: 1751-8644.
Chen, W.-K. (2009). The Circuits and Filters Handbook, CRC Press, ISBN: 9781420055276, 3rd
ed., Boca Raton, FL.
Dasgupta, S. (1988). Kharitonov’s theorem revisited, Syst. Contr. Lett., Vol. 11, No. 5, 381-384,
ISSN:0167-691.
Elden, L. & Wittmeyer-Koch, L. (1990). Numerical Analysis, Academic Press,
ISBN: 0-12-236430-9, Boston.
Femia, N. & Spagnuolo,G. (1999). Genetic optimization of interval arithmetic-based worst
case circuit tolerance analysis, IEEE Trans. Circuits Syst. I, Vol. 46, No. 12,
pp.1441-56, ISSN: 1057-7122.
Garczarczyk, Z. (1993). An interval approach to finding all equilibrium points of some
nonlinear resistive circuits, In: Circuit Theory and Design’93, Dedieu, H. (Ed.),
pp.1281-86, Elsevier, ISBN: 0-444-81664-X, Amsterdam.
Garczarczyk, Z. (1995) An efficient method for computing the range values of a rational
function with application, Proceedings of the European Conference on Circuit Theory
and Design (ECCTD'95), pp. 459-462, ISBN: 975-561-061-8, Istanbul, 27-31 August,
1995, Istanbul Technical University, Istanbul, Turkey.
Garczarczyk, Z. (1999). Frequency responses of linear systems with interval parameters,
Proceedings of the ECCTD’99, pp.615-18, Stresa, Italy, 29 August - 2 September 1999,
Politecnico di Torino, Torino, Italy.
Garczarczyk, Z. (2002). Parallel schemes of computation for Bernstein coefficients and their
applications, Proceedigs of the International Conference on Parallel Computing in
Electrical Engineering (PARELEC 2002), pp. 334-337, ISBN: 0-7695-1730-7, Warsaw,
22-25 September, 2002, IEEE Computer Society, Los Alamitos, CA.
Geher, K. (1971). Theory of Network Tolerances, Akademiai Kiado, Budapest.

Hwang, C. & Yang, S.-F. (2002). Generation of frequency-response templates for linear
systems with an uncertain time delay and multilinearly-correlated parameter
perturbation responses of uncertain systems, IEEE Trans. Circuits Syst. I, Vol. 49,
No. 3, , pp. 378-383, ISSN: 1057-7122.
Kharitonov,V.L. (1979) Asymptotic stability of an equilibrium position of a family of
systems of linear differential equations, Diff. Equations, Vol.14, No. 11, 1483-1485.
Kolev ,L.V.; V. Mladenov, V. & Vladov, S. (1988). Interval mathematics algorithms for
tolerance analysis, IEEE Trans. Circuits Syst., Vol. 35, No. 8, pp.967-975,
ISSN: 0098-4094.
Lohner, L. & Wolff von Gudenberg, J. (1985). Complex interval division with maximal
accuracy, In: Proceedings of 7th Symposium on Computer Arithmetic, Hwang, K. (Ed.),
pp. 332-336, 1985, IEEE Computer Society, Urbana, IL.
Moore, R.E.; Kearfott, R.B. & Cloud, M.J. (2009). Introduction to Interval Analysis, SIAM Press,
ISBN: 978-0-898716-69-6, Philadephia, PA.
Nataraj, P.S.V. & Barve, J.J. (2003). Reliable computation of frequency response plots for
nonrational transfer functions to prescribed accuracy, Reliable Computing, Vol. 9,
No.5, 373-389, ISSN: 1385-3139.
Neumaier, A. (1990). Interval Methods for Systems of Equations, Cambridge University Press,
ISBN: 0-521-33196-X, Cambridge.
Neumaier, A. (2002). Taylor forms – use and limits, HTML document,
http://www.mat.univie.ac.at/~neum/papers.html#taylor.
Oettli W., Prager W., Compatibility of approximate solution of linear equations with given
error bounds for coefficients and right-hand sides, Numer. Math., Vol. 6, No. 1, 1964,
pp.405-409, ISSN: 0029-599X .
Petkovic, M.S. & Petkovic, L.D. (1998). Complex Interval Arithmetic ans Its Applications,
Wiley-VCH, ISBN: 3-527-40134-2, Berlin.
Ratschek, H. & Rokne, J. (1984). Computer Methods for the Range of Functions, Ellis Horwood,
ISBN: 085312703, Chichester.
Rokne, J. & Lancaster, P. (1971). Complex interval arithmetic, Comm. ACM., vol. 14,
pp. 111-112, ISSN:0001-0782.
Shi, C.-J.R. & Tian, M.W. (1999). Simulation and sensitivity of linear analog circuits under
parameter variations by robust interval analysis, ACM Trans. Design Automat.
Electron. Syst.,Vol. 4, No.3, pp.280-312, ISSN: 1084-4309.
Spence, R. & Soin, R.S. (1997). Tolerance Design of Electronic Circuits, Imperial College Press,
ISBN: 1-86094-040-4, London.
Tan, N. & Atherton, D.P. (2000). Frequency response of uncertain systems: A 2q-convex
parpolygonal approach, IEE Proc.-Control Theory Appl., Vol. 147, No. 9, pp. 547-555,
ISSN: 1350-2379.
Tan, N. (2002). Computing of the frequency response of multilinear affine systems, IEEE
Trans. Automat. Contr., Vol. 47, No.10, pp. 1691-1696, ISSN: 1063-6536.
Tian, M.W. & Shi, C.-J.R. (2000). Worst case tolerance analysis of linear analog circuits using
sensitivity bands, IEEE Trans. Circuits Syst. –I, Vol. 47, No. 8, pp.1138-1145,
ISSN: 1057-7122.
Tian, W.; Ling, X.-T. & Liu, R.-W. (1996). Novel methods for circuit worst-case tolerance
analysis, IEEE Trans. Circuits Syst. - I, Vol.43, No. 4, pp.272-278, ISSN: 1057-7122.
Trench, W.F. (1985). Explicit inversion formulas for Toeplitz band matrices, SIAM J. Alg.
Disc. Meth. Vol.6, No. 4, pp. 546-554, ISSN: 0895-4798.
8
Analog Design Issues for Mixed-Signal CMOS

Integrated Circuits
Gabriella Trucco1 and Valentino Liberali2
1 Department of Information Technologies, Università degli Studi di Milano
2 Department of Physics, Università degli Studi di Milano
Italy
1. Introduction
Today, due to the continuous miniaturization of electronic components, a single integrated
circuit (IC) contains many transistors and interconnections very close each other, and this
causes an increased number of unwanted interactions. Crosstalk is one of the main difficulties
to face. In a mixed-signal System-on-Chip (SoC), i.e., when analog and digital circuits are
integrated on the same silicon chip, performance limitations come mainly from the analog
section which interfaces the digital processing core with the external world. In such ICs, the
digital switching activity may affect the analog section.
VDDD,ext VDDA,ext
Interconnection
coupling
v DDD Digital v DDA
i DDD Analog
i SSD
v SSD Substrate coupling v SSA
Interconnection
coupling
VSSD,ext VSSA,ext
Fig. 1. Schematic diagram of a mixed-signal IC; in the digital section only the switching
currents iDDD and iSSD are modeled.
Fig. 1 illustrates a simplified scheme of digital/analog interactions: the switching currents
drawn from the voltage supplies (iDDD and iSSD ) cause a voltage drop across the
interconnection impedances, and the on-chip supply voltages (vDDD and vSSD ) differ from
the external voltages. Voltage fluctuations may propagate to the analog part of the chip, either
trough interconnection cross-capacitances and mutual inductances, or through the common
substrate of the silicon chip. This interaction, acting as a “digital noise” superimposed to
analog signals, is often the limiting factor affecting the overall system performance.
For this reason, the optimum “mixed-signal” design can be very different from the optimum
stand-alone design. The analog designer must choose the optimum circuit architecture
considering robustness and crosstalk immunity.
The objective of this chapter is to provide some guidelines for the design of analog blocks
suitable for mixed analog-digital integrated circuits. Three different design levels will be
considered.
• Modeling: the model must be as simple as possible; the designer has to consider everything
is important and to neglect the details that do not contribute to a remarkable improvement,
in order to obtain valuable results at a reasonable complexity level.
• Architectural design: the switching noise generated by digital circuits should be as low as
possible; analog structures should be insensitive to digital noise.
• Physical design: layout design must be optimized for the fabrication technology, to ensure
a proper isolation between digital and analog sections, and to achieve a correct biasing of
substrate and well areas.
2. Modeling
The choice of the optimum circuit architecture with respect to robustness and crosstalk
immunity requires the analysis of noise generation, noise propagation, and effects on sensitive
parts of the system. Hence, a correct design methodology should account for digital switching
noise from early stages of the design process, in order to evaluate different architectural
choices. To this end, analysis tools are required to evaluate current consumption during logic
transition, in order to understand the propagation path towards analog blocks, and to design
suitable protection structures.
Switching noise effects depend on total currents drawn from the positive and the negative
supplies of the digital circuit. Therefore, the calculation of the current consumption of each
single logic gate is a too much detailed information, with would require a huge computational
effort for simulation at circuit level. For this reason, a viable method should provide only
aggregate information.
Although logic transitions are a completely deterministic phenomenon, their effects are
complex. Noise effects depend on the values of currents and of their time derivatives,
and on propagation mechanisms, which in turn are related to both on-chip and off-chip
interconnections and on substrate parasitics (Donnay & Gielen, 2003). Then, for a large
integrated system, logic transitions can be considered as a cognitively stochastic process, due
to the huge number of logic blocks. For these reasons, a statistical distribution can model
the overall switching current of a large digital circuit, using only few global parameters. The
amplitude distribution and the power spectral density of the digital noise can be obtained
from a theoretical analysis.
For simplicity, let us consider a combinational network, made up with identical logic cells,
each of them driving equal capacitive loads. A simplified model of digital switching current
can be obtained under the following hypotheses.
1. Independence of logic transitions: the transition activity of a logic gate is independent of
transitions of other gates. Although this statement is not true, as the output of a logic
Analog Design Issues for Mixed-Signal CMOS Integrated Circuits 167
X u(t)
0 t
X d(t)
Fig. 2. Switching instants of logic gates modeled as two trains of Dirac impulses.
gate drives other cells, in a large system the huge number of logic gates makes each of
them dependent only on a very small number of neighboring cells. Therefore, each logic
transition is independent of almost all other transitions.
2. Input switching instants uniformly distributed in time: the transition activity of logic cells
occur at random instants with uniform distribution over time.
3. Logic gates with equal delay: all logic transitions require the same time, therefore all current
pulses have the same finite time duration tp .
4. Logic gates with equal current consumption: the current consumption due to switching
activity is equal for all logic cells.
Under the above assumptions, the digital switching noise is described by a shot noise process.
The instants when logic gates start switching can be considered as Poisson points.
Given a time interval of duration t, we define the random variable n (t) as the number of
transitions of signals within the considered time interval. The probability to have exactly
n = k events is given by:
(λt) k
Pr[ n (t) = k] = e−λt for k = 0, 1, 2, . . . (1)
k!
The number of Poisson points in an interval of length t is a Poisson distributed random
variable, and the parameter λ is the density of the points (Papoulis & Pillai, 2002).
Each logic transition can be described as a Dirac impulse, as shown in Fig. 2. Therefore, two
trains of impulses taken at random instants are the stochastic processes Xu (t) and Xd (t) which
represent the transitions of logic gates from 0 to 1 and from 1 to 0, respectively. Each of the
processes can be written as:
X ( t ) = ∑ δ ( t − t i ). (2)
i
Under the assumptions mentioned above, the convolution between the train of impulses and
the current drawn by the single logic gate gives the total current drawn by the whole digital
circuit:
I ( t ) = h ( t ) ∗ X ( t ) = ∑ h ( t − t i ), (3)
i
where h(t) is the impulse response, representing the current of a single gate in one logic
transition.
This process, known as shot noise, is based on the statistical independence of the events
(Papoulis & Pillai, 2002), which are, in our case, the transitions of logic gates. If the impulse
density λ is uniform over time, the process is stationary. Fig. 3 illustrates an example of a
stationary shot noise process.
X(t)
I(t)
h(t)
0 t1 t2 t3 t 0 t1 t2 t3 t
Fig. 3. Switching current as a shot noise process.
2.1 Amplitude and frequency distribution of switching noise

The amplitude distribution of the total current drawn by the digital circuit is represented by
the probability density function (p.d.f.) of the stochastic process I (t), which can be calculated
from the p.d.f. of the single current pulse f H (i ).
At an arbitrary time instant t1 , the total current I (t1 ) is a random variable, whose p.d.f.
depends on both the number of Poisson impulses falling in the interval [ t1 − tp , t1 ] (i.e., the
number of logic gates which have not yet completed the logic transition), and the p.d.f. of the
single current pulse f H (i ) (Boselli et al., 2010):
f (i ) = δ(i ) Pr[ n = 0] + f H (i ) Pr[ n = 1] + f H (i ) ∗ f H (i ) Pr[ n = 2] + . . . +

+ f H (i ) ∗ f H (i ) ∗ . . . ∗ f H (i ) Pr[ n = k] + . . . =

k factors (4)
∞
= ∑ f k (i ) Pr[ n = k],
k =0
where
f 0 (i ) = δ (i ),
f 1 (i ) = f H (i ),
f 2 (i ) = f H (i ) ∗ f H (i ),
f k (i ) = f H (i ) ∗ f H (i ) ∗ . . . ∗ f H (i ) .

k factors
By using the Poisson probability (1) in (4), we obtain:

∞ (λtp )k
f (i ) = ∑ f k (i )e−λtp . (5)
k =0
k!
If λtp < 1, i.e. the duration of current pulses is small compared to the average interval between
Poisson impulses, then we have a low-density shot noise, and the p.d.f. of the total current can
be obtained by adding just a few terms of the series (5), since the general term vanishes quickly
as k increases. If λtp > 1, then we have a high-density shot noise, and the p.d.f. of the total
current tends to be gaussian.
The frequency distribution of the switching current I (t) is given by its power spectral density
(p.s.d.) S I ( f ), which can be calculated as (Papoulis & Pillai, 2002):
S I ( f ) = S X ( f ) · | H ( f )|2 = λ2 δ( f ) · | H ( f )|2 + λ · | H ( f )|2 , (6)
where S X ( f ) = λ2 δ( f ) + λ is the power spectral density of the process X (t) and H ( f ) is the
Fourier transform of the impulse response h(t). As the Dirac’s impulse δ( f ) is zero for all
f = 0, the term δ( f )| H ( f )|2 in (6) can be replaced with δ( f ) H 2 (0) = δ( f ) Q2 , where Q is the
charge transferred during the complete switching of a single logic gate:
+∞
Q = H (0) = h(t)dt. (7)
−∞
Therefore, the power spectral density S I ( f ) of the stochastic process I (t) is:
S I ( f ) = λ2 Q2 δ( f ) + λ · | H ( f )|2 , (8)
and the normalized power PI of the switching current I (t) is:

+∞ +∞
PI = S I ( f ) d f = λ2 Q2 + λ | H ( f )|2 d f . (9)
−∞ −∞
In (9), the term λ2 Q2 is the dc component of the digital switching power (λQ is the average
+∞
value of the current drawn from the supply voltage), while the term λ −∞ | H ( f )|2 d f is the
ac component of the switching power. The rightmost term in (9) can be simplified by using
Parseval’s theorem, thus obtaining:
+∞
PI = λ2 Q2 + λ h2 (t)dt. (10)
−∞
For any impulse response h(t), the normalized power PI can be written as:
λ 2
PI = λ2 Q2 + α Q , (11)
tp
where α is a “pulse shape” factor, which depends on the single current pulse waveform in
time domain, and tp is the switching time of logic gates (Boselli et al., 2010).
2.2 Current pulses with different duration, amplitude, and time density
Although equations (4) to (11) were derived starting from restrictive assumptions, the theory
can be extended to digital systems made of logic cells with different switching time, different
switching currents, and switching activity variable over time.
Let us start considering different switching times. For simplicity, let us assume that the
combinational circuit is made of two types of logic cells, labeled “A” and “B”. In more
detail, gates of type “A” are characterized by the digital switching current i A (t), which can
be described as a shot noise with time density λ A and impulse response h A (t), and gates of
type “B” are characterized by the digital switching current i B (t), with time density λ B and
impulse response h B (t). The total current drawn by the whole circuit is:
i ( t ) = i A ( t ) + i B ( t ), (12)
which is the sum of two shot noise processes. The amplitude distribution f (i ) of the total
current i (t) is:
f (i ) = f A (i ) ∗ f B (i ), (13)
where f A (i ) and f B (i ) can be calculated separately using (5).
The power spectral density S I I ( f ) is given by the sum of the p.s.d. of the single processes and
their cross-spectra:
S I I ( f ) = S AA ( f ) + S BB ( f ) + S AB ( f ) + S BA ( f ). (14)
The cross-spectra S AB ( f ) and S BA ( f ) can be obtained by taking the Fourier transforms of the
cross-correlations R AB (τ ) and R BA (τ ), which are constant:
R AB (τ ) = R BA (τ ) = λ A λ B Q A Q B . (15)
Therefore, the cross-spectra S AB ( f ) and S BA ( f ) have a single component at f = 0:
S AB ( f ) = S BA ( f ) = λ A λ B Q A Q B δ( f ). (16)
By using (8) and (16) in (14), we obtain:
S I I ( f ) = (λ A Q A + λ B Q B )2 δ( f ) + λ A · | H A ( f )|2 + λ B · | HB ( f )|2 . (17)
Therefore, at f = 0 the power spectrum component is given by the square of the sum of dc
current; while at any frequency f = 0, the power spectral density is given by the sum of the
power spectral densities of all shot noise components.
Current pulses having different peak amplitudes can be described by considering Poisson
impulses with different intensities, proportional to the current drawn by logic gates. The
mathematical model is a generalized Poisson process (Papoulis & Pillai, 2002), given by:
X G (t) = ∑ c i δ ( t − t i ), (18)
i
where ci is a random variable representing the amplitude of Poisson impulses, with mean μ c
X ( τ ) is (Papoulis & Pillai, 2002):
and standard deviation σc . The autocorrelation R G
X ( τ ) = μ c λ + ( μ c + σc ) · λδ ( τ ),
RG 2 2 2 2
(19)
G ( f ) is given by the Fourier transform:
and the power spectral density S X
G
SX ( f ) = F (RG
X ( τ )) = μ c λ δ ( f ) + ( μ c + σc ) · λ.
2 2 2 2
(20)
The current consumption I G (t) due to switching activity of logic gates with different current
intensities can be calculated by filtering the process X G (t) through the linear, time-invariant
system h(t). The power spectral density S GI ( f ) is:
I ( f ) = S X ( f ) · | H ( f )| = λ Qavg δ ( f ) + λ (1 + σc ) · | H ( f )| ,
SG G 2 2 2 2 2
(21)
where Qavg represents the average charge transferred during the switching transitions
(assuming μ c = 1).
Finally, let us consider a non-uniform distribution of logic switching activity over time. In
this situation, the switching noise can be described by a non-stationary stochastic process.
In a sequential network driven by a master clock, we can assume that the time density of
logic transitions is periodic, and therefore we have a cyclostationary shot noise. Although
the p.s.d. cannot be defined for a non-stationary process, it is possible to define a “mean
energy spectrum” which has frequency components similar to (8), plus discrete frequency
components at the master clock frequency and its harmonics.
VDD R L
off−chip
on−chip v
Fig. 4. Equivalent circuit for bondwires.
2.3 Effects of parasitics on on-chip supply voltages

Digital switching noise propagates from the digital to the analog section through both
interconnections and substrate. Therefore, realistic models of interconnections (including
package, bonding and on-chip parasitics) and substrate must be adopted for simulations.
Such models are inherently technology dependent. The model of couplings through package
interconnections strongly depends on the package. Therefore, the designer should use the
correct model of the production package. For the same reason, the use of different package
types for prototyping is not recommended, as parasitic effects can be very different. Substrate
models can also be very different. We can distinguish two major categories of substrates:
heavily-doped bulk with epitaxial layer, and lightly-doped substrate. The heavily-doped bulk
has a very low resistance and can be considered as a single node. Therefore, any disturbance
injected into the bulk propagates into the whole chip, irrespective of the distance. On the
other hand, the lightly-doped substrate is resistive, and the substrate resistance attenuates the
injected disturbance. Some fabrication technologies allow to insert a buried n-well, that can
be used for shielding purposes. Such differences must be considered during the design of
the chip. Moreover, the same circuit integrated in different technologies can behave in a very
different way from the point of view of robustness to crosstalk. Indeed, effects of substrate
parasitics put a severe limit on design portability. The results obtained in previous subsection
can be used to calculate the on-chip noise voltage is due both to digital switching currents and
to parasitic elements.
Let us start considering the simplified circuit shown in Fig. 4, where the current generator
models the digital switching noise source, and bondwire parasitics are modeled as series
inductance L and resistance R. The bondwire impedance Z is:
Z = R + sL = R + j2π f L. (22)
The on-chip power supply v is affected by a noise voltage having the power spectral density:
SV ( f ) = S I ( f ) · | Z |2 = λ2 Q2 R2 δ( f ) + λR2 · | H ( f )|2 + λ(2π )2 L2 f 2 · | H ( f )|2 . (23)
The normalized power PV of the switching noise affecting the on-chip voltage supply v is:
+∞ +∞ +∞
h (t)dt,
2
PV = SV ( f )d f = λ2 Q2 R2 + λR2 h2 (t)dt + λL2 (24)
−∞ −∞ −∞
where we have used Parseval’s theorem for both h(t) and its time derivative h (t). The
+∞
first two terms in (24), λ2 Q2 R2 and λR2 −∞ h2 (t)dt, are the dc and ac components due to
+∞ 2
the voltage drop across the parasitic resistance R. The last term, λL2 −∞ h (t)dt, is the ac
VDD R L
off−chip
on−chip v
I Rw
Cw
Fig. 5. Equivalent circuit for calculation of bondwire and substrate parasitic effects.
component due to the parasitic inductance L. By comparing the voltage spectral density and
power in (23) and (24) with the current spectral density and power in (8) and (9), we can
observe that the noise voltage terms due to the parasitic resistance R are similar to the noise
current terms, since the resistance R gives a proportional relationship between current and
voltage. On the other hand, the last term in (23) and (24) accounts for the inductive voltage
drop Lh (t). Therefore, spectral characteristics of noise voltage are dependent on both the
impulse response h(t) and its time derivative h (t). The rms value of the on-chip noise voltage
is given by:

+∞ +∞
vrms = PV = λ2 Q2 R2 + λR2 h2 (t)dt + λL2 h 2 (t)dt. (25)
−∞ −∞
Now we suppose that, besides bondwire parasitic inductance L and resistance R, the n-well
and p-substrate are providing an additional ac path from on-chip supply towards ground,
modeled by the resistance Rw and the capacitance Cw , as shown in Fig. 5. The overall
impedance Z is:
R + s( L + RRw Cw ) + s2 LRw Cw
Z= . (26)
1 + s( R + Rw )Cw + s2 LCw
Since the impedance formula (26) has a second-order denominator, oscillations may arise in
the circuit in the underdamped case, i.e., when

L
R + Rw < 2 . (27)
Cw
If the values of parasitics satisfy (27), then the current pulses due to digital switching make the
on-chip voltage supply
to oscillate, giving rise to the well known “VDD bounce”. The lower
the ratio ( R + Rw )/ L
Cw , the longer the duration of the bouncing.
2.4 Interconnection parasitics

An accurate model of interactions between analog and digital parts of an integrated circuit
must account for off-chip parasitics. In particular, package and wire bonding parasitics may
give a remarkable contribution to propagation of switching noise. Indeed, in addition to
the parasitic elements of a single interconnection, an accurate model should consider also
capacitances and mutual inductances between adjacent wires, as shown in Fig. 6 (Boselli
iDD
VDD VDD
L R L R
(external) (on-chip)
K CGND K
C
v(t)
L R L R
vs CGND
Fig. 6. Equivalent circuit of bonding and package parasitics between two adjacent wires.
et al., 2007). In this model, each wire has series inductance and resistance, capacitance to
ground, and both capacitive and inductive couplings towards the other wires. The switching
current iDD affects both the on chip voltage supply and the signals coupled either through
cross-capacitances (C) or through mutual inductances (K). Coupling between neighboring
wires must be carefully considered, since it contributes to disturbance propagation from
digital supplies to analog supplies, even without galvanic connection.
The parameters R, L, C, and K in Fig. 6 strongly depends on the package. Therefore, the
designer should use the correct model of production package. Moreover, the use of different
package types for prototyping is not recommended, as parasitic effects can be very different
(Ferragina et al., 2010).
3. Architectural design
A careful evaluation of digital switching noise effects should allow the designer to select a
robust architecture for the analog blocks and to choose digital structures which generate less
switching noise as possible.
To reduce digital switching noise, transition activity of logic gates must be low, and load
capacitance must be minimized. To this end, a partitioning of logic circuitry into different
clock domains can reduce both the total capacitance and the switching activity, provided
that each part of the circuit is driven by the minimum clock frequency required for correct
operation.
The analog designer should use robust structures, insensitive to noise (Bonomi et al., 2006).
Fully-differential structures are useful to this end, since injected disturbances behave as
common-mode signals and are rejected. Moreover, on-chip decoupling capacitances help in
reducing digital switching noise, as they provide a low impedance path for high frequency
disturbance.
As an example, let us consider the voltage reference generator shown in Fig. 7. It is based on
a band-gap voltage reference and it provides the voltages used as references in a 3-bit flash
analog-to-digital converter (ADC). VBG is the band-gap voltage reference; V1 , V2 , . . . , V7 are
the voltage references of the flash ADC; Vbias is used to bias the operational amplifiers. The
band-gap reference voltage is not affected by switching noise. Indeed, the circuit exhibits a
low impedance to VSSA ; moreover, the reference output node is capacitively coupled by CBG
to VSSA . For these reasons, the output voltage is kept at a constant value VBG = 1.22 V (with
respect to the VSSA supply). On the other hand, the resistive string voltages V1 , V2 , . . . , V7 are
VDDA
M0
M2 M1
RBG Vref
Vbias R
− V7
VBG
+ R
V6
R4 R3 R2 R
V5
R
− V4
M3
+ R
CBG V3
R1
R
V2
CC R
V1
Q2 Q1
R
VSSA
Fig. 7. Schematic diagram of the analog voltage reference.
affected by the digital switching noise superimposed to VDDA , which is injected through the
MOS transistor M0 .
To understand the effect of the switching noise on the whole ADC, let us consider
the analog-to-digital conversion stage in Fig. 8, which is part of a pipeline converter
(Rodríguez-Vázquez et al., 2003). The input voltage Vin is stored into a sample-and-hold
circuit (S&H). A flash ADC converts the input voltage, by comparing it with each of the
reference voltages and by decoding comparator outputs to obtain a binary N-bit codeword,
which corresponds to the “segment” of the input range where Vin lies in. The 7 comparators
divide the range in 8 segments, which are coded with 3 bits. The binary code is converted
again into the corresponding (lower) reference voltage by a digital-to-analog converter (DAC),
and the difference between the input voltage and the voltage corresponding to the N-bit code
is amplified to obtain the output voltage Vout , which is passed to the next pipeline stage. By
cascading pipeline stages, it is possible to achieve a high resolution ADC.
However, it is worth pointing out that a pipeline ADC is a “mixed-signal” circuit, where
partial results from first stages must be digitally decoded and stored until the last pipeline
stage has completed its operation. To operate correctly, the pipeline converter must be driven
by a two-phase clock generator made up of digital gates. The clock generator acts as digital
noise source, which affects the voltage references of the ADC and DAC. If the clock frequency
is f ck = 100 MHz, with rise and fall times tr = tf = 100 ps, then, according with the model
presented in Sect. 2, the digital switching noise has a power spectral density with the following
characteristics: it depends on the shape of the single current pulse, it becomes negligible for
Vin
VDDA
S&H
Vref R
− V7 V7
+ R V6 V6 + Vout
R V5 V5
+ 2N
−
DECODER
R V4 V4
R +
V3 V3
−
R V2 V2
R V1 V1
R
V0
VSSA
SEL
N bits
Fig. 8. Schematic diagram of one stage of a pipeline ADC, with the resistor string for
reference voltage generation.
frequencies f > 2/tr = 20 GHz, and it exhibits peaks at multiples of f ck = 100 MHz (Boselli
et al., 2010). The switching noise propagation through substrate and interconnections leads
to fluctuations in the voltage references. Although both converters share the same voltage
reference levels, ADC and DAC operations occur at different time instants. Therefore, a
fluctuation of the voltages leads to an additional error, which is amplified and transferred
to the next stage, thus limiting the effective number of bits.
To improve the robustness of the ADC to the digital switching noise, it is necessary to improve
the power supply rejection ratio in the frequency range where digital switching noise is
generated. This can be achieved by modifying the voltage reference generator, as illustrated
in Fig. 9. A first improvement consists in the use of an NMOS transistor (M0 ), instead of the
PMOS transistor in Fig. 7. The NMOS transistor in common drain configuration increases the
impedance towards the positive supply, thus improving disturbance rejection. Moreover, the
addition of an on-chip decoupling capacitance (Cdec ) between analog supplies further reduces
voltage fluctuations, as noise peaks on reference voltages are inversely proportional to Cdec
(Boselli et al., 2007).
As a further example, we consider the effects of disturbances coming from the digital section
on a fully-differential voltage-controlled oscillator (VCO). The schematic diagram of the VCO
is illustrated in Fig. 10 (Liao et al., 2003). To reduce the effects of digital disturbance, the
VCO has a fully-differential structure and the output signal is differential: v1 − v2 . Since
VDDA
M0
Vref R
+ V7
− R V6
R V5
R V4
Cdec
R V3
R V2
R V1
VSSA
Fig. 9. Schematic diagram of the improved voltage reference generator.
VDD
Vc
v1 v2
VB
Fig. 10. Schematic diagram of the VCO.
the digital switching noise is a common mode signal, the differential output should not be
affected, provided that the differential structure is perfectly matched.
Fig. 11 shows a lumped model of on-chip parasitics affecting the control voltage of the
VCO (Trucco et al., 2004). The model accounts for capacitances between wires and substrate
charge pump
+ loop filter
Req
VSS Cc v1 v2
(external) bonding & package
parasitics Vc
VSS
(on chip) Cj,w p-well
buried n-well
Rsub
Cj,b
p-substrate
Fig. 11. Model for propagation of digital noise to the VCO through interconnections and
substrate.
1.5
without digital noise
with digital noise
1
0.5
sign(v1−v2)
−0.5
−1
−1.5
2 4 6 8 10 12
time (ns)
Fig. 12. Differential VCO output.
(Cc ), substrate resistance (Rsub ), well-to-well capacitance (Cj,w ) and well-to-bulk capacitance
(Cj,b ). Although the VCO structure is differential, the control voltage Vc is a single-ended
signal. Therefore, it is affected by switching noise, which propagates through interconnection
parasitics and through the substrate. Simulation result shown in Fig. 12 confirm this
conclusion. More details can be found in (Soens et al., 2006; Trucco et al., 2004).
4. Physical design
The IC layout must be designed to isolate the analog sensitive parts from the digital noise
injecting structures.
In principle, it is possible to shield both digital and analog structures, to reduce the amount
of injected noise. However, the designer must keep in mind that the best isolation strategy
depends on the fabrication technology and on the package. Moreover, it is worth pointing
out that in the frequency range of digital switching noise there is no integrated structure
analog transistor digital switching transistor
shield
Z j2
Z j1 Zb
Z s2 Z s1
p−substrate
Fig. 13. Simplified cross-section of a shielding layer inserted between analog and digital
parts, with equivalent impedances.
which operates either as an ideal short circuit, or as an ideal open circuit. In other words,
any integrated geometry has an electrical impedance, whose value is neither zero nor
infinity. Therefore, any shielding technique must be carefully evaluated, as it depends on the
frequency of both signals and disturbances and on the disturbance paths from digital to analog
devices. These paths can vary, due to both the fabrication technology and the frequency
range of signals. A shield is obtained inserting one or more layers with different impedance,
to collect noise current and to prevent disturbance from reaching sensitive devices (Jenkins,
2004). An example is triple-well shielding, where a buried n-well is used to separate the local
p-wells from the p-substrate. Fig. 13 shows a triple well shielding placed around an analog
MOS transistor. The shield exhibits a capacitive impedance Zj1 towards the p-substrate, and
has a non zero resistivity, modeled with lumped resistances Zs1 and Zs2 . For an NMOS device,
the impedance Zj2 is capacitive (due to the reverse biased junction between the p-well and the
buried n-well). For this reason, triple-well shielding can be an effective technique, provided
the frequency range is not too large. Fig. 14 shows a qualitative plot of the impedance of
the disturbance path as a function of the frequency. On the contrary, for PMOS transistors,
triple-well shielding can be harmful, as the impedance Zj2 is mainly resistive (Rossi et al.,
2003). Shielding is less effective in heavily doped substrates, as the low resistivity of the bulk
propagate disturbance across the whole chip (Liberali, 2002).
In lightly doped substrates, guard rings provide effective isolation, as disturbance paths are
near to the silicon surface. Guard rings around noise sources provide a low resistance path
to ground for the noise; therefore, they help minimizing the amount of noise injected into the
substrate. Again, efficiency of guard rings depends on the frequency range of injected noise
and on package inductance.
The relative position of analog and digital cells with respect to each other on the same die is an
important issue to consider. In lightly-doped substrates, physical separation helps in reducing
crosstalk.
On-chip interconnections can provide additional paths for injected disturbance. In a careful
design, the voltage supplies of the analog and of the digital sections must be completely
separated, and also pad rings and ESD protections should have their separate supplies.
Packaging affects performance and reliability in mixed-signal integrated circuits. One of the
most common used assembling technology is chip-in-package. When using this assembling
|Z p|
(log) shielded
unshielded
f (log)
Fig. 14. Qualitative plot of the impedance from the digital noise source to the sensitive
analog device.
technique, the designer should account for both bondwires and package parasitics. When
the digital part operates at high speed, inductive effects are a major source of performance
degradation. Multiple bonding helps in achieving a further reduction of parasitic equivalent
bondwire inductances (Ferragina et al., 2010). An assembling technology without bondwires
(flip-chip mounting) has even better noise immunity, due to reduced parasitic elements, and
must be considered for high-performance mixed-signal integrated systems. However, it is
worth noting that interconnection parasitics due to the circuit board remain unchanged.
Finally, special post-processing techniques for 3-D insulation of parts of the chip can be helpful
for critical applications, at the expense of additional wafer cost (Chong & Xie, 2008).
5. Conclusion
This chapter has presented some aspects of digital noise in mixed-signal CMOS ICs.
Digital switching noise can be modeled as a stochastic process. By considering switching
activity of logic gates as a random process, with transition instants randomly distributed
in time, digital switching currents can be modeled as shot noise processes, and small signal
analysis techniques can be applied to evaluate their impact on analog structures.
As a general rule, crosstalk between digital and analog sections increases with size
reduction and with clock frequency. Design techniques for crosstalk reduction are essential
for high-performance integrated systems. Differential structures and on-chip decoupling
capacitances can be helpful in reducing disturbance, thus improving crosstalk immunity. A
correct design approach should be based on a top-down methodology, including a crosstalk
analysis from early design stages, to improve the robustness and to reduce the risk of failure.
Physical design is also very important, since noise propagation depends on fabrication
and assembling technologies. Therefore, rules for the “best” mixed-signal design are
technology-dependent, and, in general, design portability is not guaranteed with respect to
crosstalk robustness.
6. References
Bonomi, D., Boselli, G., Trucco, G. & Liberali, V. (2006). Effects of digital switching noise on
analog voltage references in mixed-signal CMOS ICs, Proc. Brazilian Symposium on
Integrated Circuit Design (SBCCI), Ouro Preto (Minas Gerais), Brazil, pp. 226–231.
Boselli, G., Trucco, G. & Liberali, V. (2007). Effects of digital switching noise on analog circuits
performance, Proc. European Conf. on Circuit Theory and Design (ECCTD), Seville,
Spain, pp. 160–163.
Boselli, G., Trucco, G. & Liberali, V. (2010). Properties of digital switching currents in fully
CMOS combinational logic, IEEE Trans. VLSI Systems 18: 1625-1638.
Chong, K. & Xie, V.-H. (2008). Three-dimensional impedance engineering for mixed-signal
system-on-chip applications, Proc. Int. Conf. Solid-State and Integrated-Circuit
Technology (ICSICT), Beijing, China, pp. 1447–1451.
Donnay, S. & Gielen, G. (eds) (2003). Substrate Noise Coupling in Mixed-Signal ASICs, Kluwer
Academic Publishers, Boston, MA, USA.
Ferragina, V., Ghittori, N., Torelli, G., Boselli, G., Trucco, G. & Liberali, V. (2010). Analysis and
measurement of crosstalk effects on mixed-signal CMOS ICs with different mounting
technologies, IEEE Trans. Instr. and Meas. 59: 2015–2025.
Jenkins, K. A. (2004). Substrate coupling noise issues in silicon technology, Proc. IEEE Topical
Meeting on Silicon Monolithic Integrated Circuits in RF Systems, Atlanta, GA, USA,
pp. 91–94.
Liao, H., Rustagi, S. C., Shi, J. & Xiong, Y. Z. (2003). Characterization and modeling of the
substrate noise and its impact on the phase noise of VCO, Proc. Radio Frequency Integr.
Circ. Symp. (RFIC), Philadelphia, PA, USA, pp. 247–250.
Liberali, V. (2002). Evaluation of epi layer resistivity effects in mixed-signal submicron
CMOS integrated circuits, Proc. IEEE Int. Conf. on Microelectronics (MIEL), Niš, Serbia,
pp. 569–572.
Papoulis, A. & Pillai, S. U. (2002). Probability, Random Variables and Stochastic Processes, 4th ed.,
McGraw-Hill, New York, NY, USA.
Rodríguez-Vázquez, A., Medeiro, F. & Janssens, E. (eds) (2003). CMOS Telecom Data Converters,
Kluwer Academic Publishers, Boston, MA, USA.
Rossi, R., Torelli, G. & Liberali, V. (2003). Model and verification of triple-well shielding
on substrate noise in mixed-signal CMOS ICs, Proc. European Solid-State Circ. Conf.
(ESSCIRC), Estoril, Portugal, pp. 643–646.
Soens, C., Van der Plas, G., Badaroglu, M., Wambacq, P., Donnay, S., Rolain, Y. & Kuijk,
M. (2006). Modeling of substrate noise generation, isolation, and impact for an
LC-VCO and a digital modem on a lightly-doped substrate, IEEE J. Solid-State Circ.
41: 2040–2051.
Trucco, G., Boselli, G. & Liberali, V. (2004). An approach to computer simulation of
bonding and package crosstalk in mixed-signal CMOS ICs, Proc. Brazilian Symposium
on Integrated Circuit Design (SBCCI), Porto de Galinhas (Pernambuco), Brazil,
pp. 129–134.
9
Tunable Analog and Reconfigurable Digital

Circuits with Nanoscale DG-MOSFETs
Savas Kaya, Hesham F. A. Hamed & Soumyasanta Laha
Ohio University
USA
1. Introduction
1.1 CMOS downscaling to DG-MOSFETs
As device scaling aggressively continues down to sub-32nm scale, MOSFETs built on Silicon
on Insulator (SOI) substrates with ultra-thin channels and precisely engineered source/drain
contacts are required to replace conventional bulk devices (Celler & Cristoloveanu, 2009).
Such SOI MOSFETs are built on top of an insulation (SiO2 ) layer, reducing the coupling
capacitance between the channel and the substrate as compared to the bulk CMOS. The
other advantages of an SOI MOSFET include higher current drive and higher speed, since
doping-free channels lead to higher carrier mobility. Additionally, the thin body minimizes
the current leakage from the source to drain as well as to the substrate, which makes the SOI
MOSFET a highly desirable device applicable for high-speed and low-power applications.
However, even these redeeming features are not expected to provide extended lifetime
for the conventional MOSFET scaling below 22nm and more dramatic changes to device
geometry, gate electrostatics and channel material are required. Such extensive changes are
best introduced gradually, however, especially when it comes to new materials. It is the focus
on 3D transistor geometry and electrostatic design, rather than novel materials, that make the
multi-gate MOSFETs as one of the most suitable candidates for the next phase of evolution in
Si MOSFET technology (Skotnicki et al., 2005; Amara & Olivier, 2009).
The multi-gate MOSFET architectures can efficiently control the channel from multiple sides
of the channel instead of the top-side in planar bulk MOSFETs. The ability to alter channel
potential by multiple gates (i.e double, triple, surround) provides a relatively easier and
robust way to control the channel electrostatics, reducing the short channel effects and leakage
concerns considerably. Thus, the last decade has witnessed a frenzy of design activity
to evaluate, compare and optimize various multi-gate geometries, mostly from the digital
CMOS viewpoint (Skotnicki et al., 2005). While this effort is still ongoing, the purpose of
the present chapter is to underline and exemplify the massive increase in the headroom for
CMOS nanocircuit engineering, especially at the mixed-signal systems, when the conventional
MOSFET architecture is augmented with one extra gate. Being the simpler and relatively
easier to fabricate among the multigate MOSFET structures (FinFET, MIGFet, Π-MOSFET and
so on) the double gate (DG) MOSFET is chosen here to explore these new circuit possibilities.
The great potential of DG-MOSFETs for new directions in circuit engineering has been
explored also by others. For instance the Purdue group, led by Roy (Roy et al., 2009) has
explored the impact of DG-MOSFETs (specifically in FinFET device architecture) for power
reduction in digital systems and for new SRAM designs. Kursun (Wisconsin & Hong Kong)
has illustrated similar power/area gains in sequential and domino-logic circuits (Tawfik &
Kursun, 2008). Several French groups have recently provided a very comprehensive review
of their DG-MOSFET device and circuit works in a single book (Amara & Olivier, 2009). Their
works contain both simulation and practical implementation examples, similar to the work
carried out by the AIST XMOS initiative in Japan (AIST, 2006) as well as a unique DG-MOSFET
implementation named FlexFET by the ASI Inc.(ASI, 2009).
1.2 Context: Mixed-Signal & Adaptive Systems

In addition to features essential for digital CMOS scaling (Skotnicki et al., 2005; Mathew et al.,
2002) such as the higher ION /IOFF ratio and better short channel performance, DG-MOSFETs
possess architectural features also helpful for the design of massively integrated mixed-signal
and adaptive systems with minimal overhead to the fabrication sequence. Given the fact that
they are designed for sub-22nm technology nodes, the DG MOSFETs can effectively handle
GHz modulation, making them relevant for the mixed-signal system-on-chip applications
with wireless/RF connectivity and giga-scale integration. Also, they have reduced cross-talk
and better isolation provided naturally by the SOI substrate, multi-finger gates, low parasitics
and scalability. However, the DG-MOSFET’s potential for facilitating mixed-signal and
adaptive system design is highest when the two gates are driven with independent signals
(Pei & Kan, 2004; Raskin et al., 2006). It is the independently-driven mode of operation that
furnishes DG MOSFET with a unique capability to alter the front gate threshold via the back
gate bias. This in turn leads to:
• Increased operational capability out of a given set of devices and circuits.
• Reduction of parasitics and layout area in tunable or reconfigurable circuits
• Higher speed operation and/or lower power consumption with respect to the equivalent
conventional circuits.
On the digital end, gate-level tunability of DG-MOSFETs allow us to explore reconfigurable
logic architectures that can increase functionality and flexibility of logic blocks such as ALU
and programable arrays without significant overheads in terms of size, power or design
complexity. As a result, the DG-CMOS circuitry has gained steady and growing attention
for mixed-signal community in the last 5 years. Many works that utilizes DG-MOSFETs
in RF amplification and mixing applications (Reddy et al., 2005; Mathew et al., 2004), in
tunable analog circuit blocks, Schmitt triggers, filters have been already published (Kaya et al.,
2007). This chapter reviews some of these efficient and compact mixed-signal system blocks,
exploring their feasibility and capabilities. At a time when performance gains resulting from
circuit engineering is desperately needed to mitigate the impasse of aggressive device scaling,
this is believed to be timely and very useful.
1.3 DG-MOSFET structure

DG-MOSFETs considered in this work are chosen to comply with the mixed-signal circuit
design constraints that integrate analog circuits on the same substrate as digital building
Tunable Analog and Reconfigurable Digital Circuits with Nanoscale DG-MOSFETs 183
tox=2nm
Lgate=100nm 0.4
Top Gate tsi=10nm
800 0
Vth [V]
Symmetric
Drain Current [μA]

Source Drain -0.4
600 -0.8
Bottom Gate +0.75V
-0.4 0 0.4 0.8 +0.5V
Back Gate Bias [V]
400
+0.25V
D D
Vfg Vbg Vfg Vbg +0.0V
200
VBG=-0.5V
S S
SDDG (Vfg=Vbg) IDDG (Vfg≠ Vbg) 0
0 0.5 1
Front Gate Bias [V]
a) b)
Fig. 1. a) The DG-MOSFET device structure used in this work and its circuit symbols for SDDG and
IDDG modes, b) simulated characteristics of an n-type DG-MOSFET at different back-gate bias
conditions. For comparison, symmetric (V f g =Vbg ) drive case is also included. Inset shows the resulting
shift in the front gate threshold
blocks with minimal overhead to the fabrication sequence (Raskin et al., 2006; Kranti et
al., 2004). This implies using DG-MOSFETs with a minimal body thickness (tSi 20nm),
oxide insulator thickness (tox 2nm) and gate length (L 20nm), and maximum ION /IOFF
ratio optimized normally for minimum switching delay power product. It is assumed that
both gates have been optimized for symmetrical threshold VT = ± 0.25V using a dual-metal
process.
Fig.1a above illustrates the generic DG-MOSFET structure used in 2D simulations of all
devices and circuits. The device simulations in this work are accomplished using either TCAD
(DESSIS (Synopsys, 2008)) or UFDG-SPICE3 (Fossum, 2004) simulators in drift-diffusion
approximation for carrier transport, which is sufficient for low-power circuit-configurations
explored here. The transfer (ID -VG ) characteristics of a generic n-type DG-MOSFET simulated
using DESSIS is also available in Fig.1b. It is obvious that the top-gate threshold can be tuned
via the applied back-gate voltage. This ’dynamic’ threshold control is crucial to appreciate
the tunable properties of the circuit structures presented here. However, such independently
driven double gate (IDDG) devices have lower transconductance, and higher sub-threshold
slope than the symmetrically driven double gate (SDDG) counterparts under equal geometry
and bias conditions (Pei & Kan, 2004). Thus bottom-gate tunability comes with a reduction
in intrinsic DG-MOSFET performance, a price well justified by the wide variety of circuit
possibilities as explored below.
2. DG CMOS modeling & simulation

The last ten years have witnessed a sizable effort in migrating conventional compact
models to more sophisticated but numerically demanding novel approaches based on the
surface-potential. Such a move was inevitable given the aggressively scaled dimensions and
new physics such as tunneling and quantization effects that must be accounted for accurately.
Yet, there is no public-domain surface-potenial based DG-CMOS SPICE models that can be
accessible to the circuit and system engineers in terms of availability and usability. As a
result, we adapted using two commercial modeling approaches successfully to simulate the
DG-CMOS circuits, which are detailed below.
2.1 UFDG SPICE

The UFDG model is a process/physics and charge based compact model for generic DG
MOSFETs (Fossum, 2004). The key parameters are related directly to the device physics.
This model is a compact parameterized Poisson-Schrodinger solver for DG MOSFETs that
physically accounts for the charge coupling between the front and the back gates. The
UFDG allows operation in the independent gate mode and is applicable to fully-depleted SOI
MOSFETs. The quantum mechanical (QM) modeling of the carrier confinement, dependent
on the ultra-thin body (tSi ) as well as transverse electric field, is incorporated via Newton
Raphson iterations that link it to the classical formalism. The dependence of carrier mobility
on tSi on transverse electric field is also accounted for. In addition, the carrier velocity
overshoot and dependence on carrier temperature is characterized in the UFDG transport
modeling to account for the ballistic and quasiballistic transport in scaled DG MOSFETS
(Ge et al., 2001). The channel current is limited by the thermal injection velocity at the
source, which is modeled based on the QM simulation. The UFDG model also accounts
for the parasitic (coupled) BJT (current and charge) which can be driven by transient
body charging current (due to capacitive coupling) and/or thermal generation (Kim, 2001).
Lumped source and drain contact resistances, gate-induced barrier lowering and impact
ionization currents are also considered, the latter of which is characterized by a non-local
carrier temperature-dependent model for the ionization rate integrated across the channel
and the drain. The charge modeling which is patterned after that is physically linked to
the channel-current modeling. All terminal charges and their derivatives are continuous for
all bias conditions, as are all currents and their derivatives. Temperature dependence for
the intrinsic device characteristics and associated model parameters are also implemented
without the need for any additional parameters. This temperature dependence modeling is
the basis for the self-heating option, which iteratively solves for local device temperature in
DC and transient simulations in accord with a user defined thermal impedance. Hence UFDG
model has sufficient rigor to accurately model sub-100 nm devices commonly used for in the
proposed circuits.
2.2 TCAD
A secondary approach adapted in our simulations is the use of technology CAD (TCAD)
package by Synopsys (Synopsys, 2008), which can solve the appropriately coupled set of
electron/hole transport equations and electrostatic (Poission) equation over realistic 2D/3D
meshes. In TCAD no mathematical models are assumed for the terminal characteristics and
a precise device geometry can be accounted for to estimate the outcome of semiconductor
processing technologies and device characteristics. The TCAD device simulation tools are
applicable to a broad range of applications including Analog/RF devices and can be used as
an aid to gain insight to device performance and operation.
In the two-tiered TCAD packages, the process simulator deals with geometrical modeling of
the fabrication steps of semiconductor devices such as transistors and diodes. On the other
hand, the device simulator simulates the electrical characteristics of the devices, in response
to the external electrical, thermal or optical boundary conditions imposed on the structure.
Figs.1 & 2 shows the Id -V f g characteristics at different back-gate bias conditions for an
n-channel MOSFET an a DG-CMOS pair, respectively, as obtained from so-called mixed-mode
TCAD simulations that include multiple instances of devices in an outer SPICE-like network
solver. Due to the multiple transistors each containing upwards of 2000 mesh points and the
VDD
0.4
0V Sym
Vbgp
Vbgp 0.2
Output [V]
VOUT 0 0.1V -0.1V
VIN VIN
VOUT 0.2V -0.2V
-0.2 0.3V -0.3V
CL
p
Vbgp Vbg
-0.4
VSS -0.4 -0.2 0 0.2 0.4

Input Bias [V]
Fig. 2. a) The simple inverter implemented using the DG MOSFETs with additional inputs for tuning
transfer characteristics b) TCAD simulated DC transfer characteristics when the two back gates are
p
biased jointly (Vnbg = -Vbg ).
bipolar charge transport in each device these simulations are CPU intensive and require rather
large memory space. This situation is further compounded when the quantum mechanical
corrections and sophisticated dependence of mobility on parallel and perpendicular fields.
Therefore the TCAD approach must be carefully considered in large circuits and may be only
needed where accuracy is the prime concern.
3. Analog circuits blocks

In the following we provide examples for compact & low-power RF-CMOS system blocks
designed using independent gate DG-MOSFETs. In all cases, the bottom gate is used to tune
the circuit performance while also reducing overall system size (number of transistor and total
area). Many integrated signal processing platforms can use these system blocks to process
the signals from receivers and nanosensors. Using simulations, we explore how compact
low-power circuits including tunable single-ended and differential amplifiers, integrators,
filters and current and voltage controlled-oscillators may be built and tuned. Depending
on the nature of nanosensing devices and S/N ratio, more custom solutions may always be
possible.
3.1 CMOS voltage amplifier

The DG CMOS inverter pair (see Fig.2) can serve as a high-gain push-pull amplifier when
biased in the transition region. Depending on the selection of the sign and magnitude of
the bottom-gate bias, the simple amplifier’s characteristics can be altered in a number of
ways, which greatly enhances the variety of applications for this otherwise simple circuit.
p
For instance, Fig.2b shows that co-setting of the bottom gates at the same voltage (Vnbg = Vbg )
results in proportional shifts in the voltage window for amplification. This "window-shifting"
can be conveniently utilized in a number of ways such as in analog wave-shaping circuits
sensitive to DC bias levels or in Schmitt triggers (Kaya et al., 2007; Cakici et al., 2003).
An alternative scheme for programming the CMOS pair is conjugation, whereby the two
complementary bottom-gates are driven by separate signals of equal magnitude but opposite
p
polarity, i.e Vnbg = - Vbg . In a mixed-mode design using bipolar supply voltages, this biasing
scheme is indeed possible and provides a method of varying the amplifier gain. As shown
40 n p
Vbg =-Vbg
Sym
0.0V
0.1V
AC Gain [dB]
20 0.3V
0.5V
40
iddg
Band Width [MHz]

30
sddg
Gain [dB]
30 Gain
20
0
20 10
BW
0 0.2 0.4
n p
Vbg =-Vbg [V]
-20 5 6 7 8 9
10 10 10 10 10
Frequency [Hz]
Fig. 3. a) The simulated DC response of the tunable DG-CMOS pair for various joint back gate biases
p
(Vnbg = -Vbg ). The amplifier gain changes with the back gate bias and b) AC gain analysis
in Fig.3a, the slope (gain) of the transition region is a function of conjugate bias levels set on
the bottom gates and the change in the output impedance (inset, Rout =1/gd ) dominates the
simulated intrinsic gain (gm /gd ) response. For comparison, the output of SDDG CMOS pair
is also provided in the both plots above. While the gain of SDDG inverter is higher, without
any bias control, it offers neither design latitude nor alternative configurations. On the other
hand, the self-feedback arrangement also included in Fig.3a, where the output of the IDDG
p
CMOS pair drives their bottom-gates (Vnbg = Vbg = VOUT ), results in a inverting buffer with
a gain of one. This may be especially suitable in applications where a linear signal buffer is
required. The gain-bandwidth tradeoff of the IDDG-CMOS amplifier is illustrated in Fig.3b,
which shows the outcome of AC analysis with a load capacitor of C L =1 pF. Thus, it should be
possible to fine tune simple CMOS amplifier’s frequency response using the conjugate biasing
scheme in a very linear fashion.
3.2 Current mirrors

Another essential block used in the design of analog circuitry is the simple current mirror.
Normally the current copying characteristics of the simple current mirror (CM) (Fig.4a), is
fixed once the circuit is built and depends on the ratio of transistor width between the input
(reference) and output branch. In the case of DG-CMOS, however, a similar gain factor can be
easily obtained, and dynamically altered, by appropriate back biases of DG-MOSFETs used in
the mirror block, as shown in Fig.4b. The back bias can modulate overall conductivity of the
output transistor, thus effecting the copying ratio. Such tunability not only greatly enhances
the variety of applications for this otherwise simple circuit, but could also lead to area and/or
power savings over similar circuits built using bulk MOSFETs, as also discussed by others
(Kumar et al., 2004)
Even for the modest back-bias conditions at the output transistor (Voset 1 V), it is possible
to achieve mirror ratios around 100. Note that poor output impedance of the simple CM
is due to short gate length ( 100nm) devices employed here. Such compromise in the
output conductance can be easily dealt with by adapting a cascade CM, as shown in Fig.4c.
The cascade CM design retains all aspects of tuning in the simple CM, while increasing the
output impedance of the CM (Fig.4d). Once again, the above simulations not only show the
great potential in Independently Driven Double Gate (IDDG) tunable current mirrors but also
50 VDD
VDD Vseti=Vin
Vseto=0.4 V
Iin=3.1μA
Output Current [μA]

40
Vref
Vref
30
Vseto=0.35 V IIN IOUT
IIN IOUT VIN VOUT
20
VIN VOUT Vseto=0.3 V
Vseti Vseti Vseto

Vseto 10
Vseto=0.2 V
A1 A2
0 A1 A2
0 0.5 1 1.5 2
Vout [V]
a) b) c)
400
Mirror Ratio, IOUT/IIN

Vseti=Vin 100
Iin=133μA 10
Simple CM
Vseto = 1.0V
Output Current [μA]
0.8
1
300
Vseti = Vin 0.1
0.9V
0.6 0 0.2 0.4 0.6 0.8 1
Vseto [V]
200 0.8V
VIN [V]
0.4
0.7V IDDG CM
SDDG CM Vseti=0.8V
100
0.6V
0.2
0.5V IDDG CM
Vseti=1.2V
0.4V
0
0 0.5 1 1.5 2 0
Vout [V] 0 200 400 600
d) e) IIN [μA]
Fig. 4. a) simple DG current mirror and b) the simulated output I-V response as a function of tuning
voltage Voset . The output impedance is low due to short channel effects c) The improved DG cascade
current mirror d) The dependence of the I-V response of the cascade current mirror on Voset . e)
Comparison of the required voltage across the input of the simple CM in three configurations: SDDG (no
back gate control) and IDDG with two different back gate voltages
provide valuable insights for the more complicated current-mode circuits blocks investigated
in the following sections, which uses a number of such CM in a differential topology to form
amplifiers, filters and alike. Moreover, comparison at the same current levels shows that
the input voltage across DG current mirror can be significantly lower than that required for
conventional version (Fig.4d). Therefore, in addition to the tunability without the use of an
extra transistor (less area and parasitics), another major advantage of DG CM circuits is the
potential to lower voltage supply and power dissipation (lower V I N ).
3.3 Current amplifier

The dynamic alteration of mirror ratios is the principle of amplification behind the simple but
tunable current amplifier in Fig.5a, which can also be built using the cascade CM for higher
performance. The proposed current amplifier is built using a two-stage design consisting of
an amplification (A1:A2) block and an DC offset cancellation blocs (A2:A2). Without the error
cancellation stage this differential block would still operate but can result in DC offset errors
in driving similar differential blocks. Both of these blocks are built using DG CMs: the back
gates of lower transistor pairs (Vseto ) are used for scaling the output current, while the back
gates of input transistors (Vseti ) are used for scaling the input current. PMOS transistors bias
the amplifier to a DC operating point, which can be controlled also using the back-gate Vb .
It is possible to achieve appreciable gain and bandwidth programming in various biasing
schemes for the bottom-gate control voltages on the input and output sides (Vseti , Vseto ), as
shown in Fig.5a. Our simulations indicate that the bandwidth can be easily tuned by two
orders of magnitude and the gain by 15 dB using this amplifier. Moreover, by combining
VDD
P1 P2 P3 Vb P6 P5 P4
IIN(+) IIN(-)
a) IOUT (+)
IOUT(-)
N1 N2 N3 N6 N5 N4
Vseti Vseto Vseti
A1 A2 A2 A2 A2 A1
30 30 0.0V, 0.0V
0.5V, 0.5V
1.0V, 1.0V
25 25 -0.5V, -0.5V
-1.0V, -1.0V
Current Gain [dB]

Current Gain [dB]
Vseti, Vseto
20 20
30
100
25 15
BW f3dB [GHz]
15
20
10
Vseti, Vseto 10
10
15 0.0V, 0.0V
0.0V, 0.2V
5 0.0V, 0.4V 5
-0.8 -0.4 0 0.4 0.8 1
Bias Difference (Vseti-Vseto) [V] 0.2V, 0.0V -1 -0.5 0 0.5 1
0.4V, 0.0V Bias Vseti=Vseto [V]
0
1M 10 M 100 M 1G 10 G 100 10 M 100 M 1G 10 G 100 G 1T
Frequency [Hz] Frequency [Hz]
b) c)
Fig. 5. a) Current amplifier circuit implemented using simple DG CM components, b) the gain control
and c) bandwidth control in current amplifier via asymmetric and symmetric biasing schemes,
respectively.
these biasing schemes, it should be possible to concurrently tune the gain and bandwidth in
the same amplifier. Once again, this is achieved without the use of extra transistors found
in conventional tunable CMOS circuits, thus, in principle, reducing the area and power
requirements considerably. Moreover, this current amplifier may be realized also in the
single-ended fashion, i.e. a single CM stage, which can be used as a sense amplifier with
a tunable frequency response that can be very useful in nanosensor environments with a
cluttered spectrum.
3.4 Operational Transconductance Amplifiers - OTA

Operational transconductance amplifiers (OTA) produce differential output currents in
response to differential voltage inputs. They have become increasingly popular in the last two
decades due to ease of design and reduction in circuit complexity compared to operational
voltage amplifiers in certain applications (Sanchez-Sinencio & Silva-Martinez, 2000). They
often drive a capacitive load in a compact OTA-C block that can act as very efficient integrators
and appear also in other filter elements. Since the back-gate biasing in DG-CMOS architecture
offers real advantages to current mode circuit design to alter circuit operation with minimal
intrusion, the OTAs with current outputs are set best for taking advantage of the tunability in
amplifier designs. Accordingly, we focus below in two different OTA circuits.
3.4.1 Simple OTA

The first OTA topology explored is the simplest of all, as illustrated in Fig.6a, which is adapted
from bulk MOSFET implementation normally requiring 6 transistors (Szczepanski et al., 2004),
as opposed to 4 DG-MOSFETs in the new topology. The availability of the individual bottom
gates allows the elimination of the two extra transistors for transconductance (gm ) tuning
6
Transconductance, gm [mS/μm]
Vsetn = -Vsetp = 0.25V
OTA-C Filter Gain [dB]

4 20
Vsetn=0.30V, Vsetp=-0.25V CL=10fF

VDD VDD
2 Vsetp
Sym
0.0V 10 Vsetn=0.35V, Vsetp=-0.25V
-VOUT +VOUT 0.1V CL=0.1pF
0 0.3V Vsetn = -Vsetp
+VIN CL -VIN CL=1pF
0.5V 0.1V
0.25V
Vsetn
0 0.4V
-2 VSS VSS 0.5V
5 6 7 8 9 10 11 12 6 7 8 9 10 11
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]
a) b)
Fig. 6. a) Transconductance (gm ) of the unloaded (CL =0) OTA circuit (inset) versus frequency as a
function of the conjugate tuning bias. gm has a linear dependence on the bias setting and does not
trade-off with the bandwidth b) AC gain of OTA-C filter at various bias settings and for three
capacitance values. For a typical C = 10 fF, GHz operation is within reach. Although gain can be tuned
using conjugate bias pairs, a wider tuning range is possible via asymmetric bias (Vsetn = Vsetp )
across the two branches of the OTA, which should save both power and area while also
minimizing the parasitics.
Similar to the CMOS amplifier case, there are two tuning schemes available to this simple
p
OTA circuit: an asymmetric bias (Vset = Vnset ) to shift frequency response or a conjugate bias
p
(Vset =−Vset ) to alter the transconductance (gm ) of OTA. Fig.6a summarizes this latter case,
n
where the frequency dependence of gm on the conjugate programming voltage is plotted

against frequency. The most important figure of merit, gm , of OTA varies linearly with the
programming voltage and the bandwidth (BW) of the OTA is constant despite varying gm ,
which is one of the main hallmarks of OTAs (Sanchez-Sinencio & Silva-Martinez, 2000). The
gm is constant up-to ∼ 100 GHz range limited by small parasitic capacitances on SOI substrate.
When an asymmetric bias is used to tune the OTA, we can conveniently shift the frequency
p
response. For a fixed realistic load of C L = 10 fF and Vset =−Vnset =0.25V, the resulting OTA-C
circuit serves as a low-pass filter with a corner frequency ∼5 GHz, as shown in Fig.6b. Even
for a relatively large load of C L = 1 pF, the filter pass-band extends up to 200 MHz. The same
corner frequency can be tuned almost a decade depending on the asymmetric bias on the back
gates. This simple but powerful example aptly illustrates the potential of DG-MOSFET analog
circuits.
3.4.2 VHF OTA

Practical implementation of high-performance tunable OTAs requires more sophisticated
architectural elements that optimize the gain as well as the input and output impedance. Such
elements modify the transfer function by canceling poles and shifting zeros in the complex
plane to improve frequency performance and/or stability. However, a detailed account of
DG-CMOS OTA optimization is beyond the scope of this chapter. Instead, we shall attempt
to illustrate that improvements to the simple OTA structure above is indeed possible. For
instance, a more advanced version of the simple OTA circuit with cross feed-forward elements
intended to improve the output conductance is presented in Fig.7a. There are two sets of
tuning nodes in this circuit: the input side with nodes VCpI , VCnI and the load side with
b)
VDD
VCpI
IOUT(-)
VIN(+)
VCpL VCpL
VCnI
VSS
VDD
VCpI
VCnL VCnL
VIN(-)
IOUT(+)
VCnI
VSS
a) c)
Fig. 7. a) A tunable operational transconductor amplifier (OTA) based on simple DG-MOSFET inverters
with feedforward compensators. b) The simulated response of the differential OTA as a function of
conjugate bias VCpL =−VCnL at feedforward structure, and c) the AC characteristics of a simple gm − C
integrator with CL = 1 pF as a function various values of control bias VCpL =−VCnL for two cases of
VCpI =VCnI 0 and 0.5 V.
VCpL , VCnL . The former mostly impacts the transconductance term, while the later determines
the output conductance (Nauat, 1992). Normally, all control nodes are held at 0.0V, unless
otherwise noted, and the conjugate bias pairs may be varied. The resulting architecture
operates linearly up to large values ( 500mV or higher) of the input signal amplitude and
the gm (i.e. the slope) can be tuned using voltages VCpL , VCnL , as evident in Fig.7b.
The ability to tune the transconductance can be readily utilized in a variety of applications
such as the C-gm integrator shown in Fig.7c. A fairly large capacitor value of C=1 pF was
used in this circuit. The BW of the integrator can be tuned by the control nodes VCpI , VCnI
as well as the capacitor value, while the gain can be determined by the nodes VCpL , VCnL . In
comparison with the simple OTA (Fig.6a), the unloaded (C L =0) bandwidth of the VHF OTA
structure is found to improve by an order of magnitude, which compares well with the bulk
CMOS implementation (Nauat, 1992) as well as the loaded data (C=1 pF) in Fig.7c. A SDDG
version could operate at much higher frequencies, although it would require more power and
area, as discussed in the previous section. We also observe that the tuning range of DG-CMOS
OTA circuit is more limited than the current mode integrator, a point to be discussed in more
detail in the next section.
3.5 Current-Mode Integrator and High-Order Filters

To illustrate the power of the simple DG circuit blocks and address another important building
block used in almost all analog RF systems, this section is dedicated to examples of first and
second order filters. Hierarchically as well as pedagogically, it is appropriate to start the
discussion with first-order tunable integrators, which can then be used to build higher-order
VDD
Vsetp
a)
IIN(+) IIN (-)
IOUT(-) IOUT(+)
Vsetn
C C
300 300 -20

Vsetp=+0.5V Vsetn=+0.5V Vsetp=+1.0V
Vsetp=+1.0V Vsetn=0.0V
Vsetn=-0.5V -40
200 Vsetp=+1.5V 200
IDDG (Vsetp=+1.0)
IDDG (Vsetn=0) -60
100 100
Iout [μA]
-80 IDDG (Vsetn=-0.5V)

SDDG (Vsetn=0) IDDG (Vsetn=0)
HD3 [dB]
SDDG (Vsetn=+0.5V) IDDG (Vsetn=+0.5V)
0 Iin [μA] Iin [μA] 0 -100
-20
100 200 100 200 SDDG (Vsetp=+1.0V)
20
SDDG (Vsetp=+0.5V)
-100 IDDG IDDG 20
-100 -40
15
Error [μA]
Error [μA]
SDDG SDDG
Vsetp=1.0V
10 -60
-200 10 -200
5
-80 IDDG (Vsetp=+1.5V)
Vsetn=+0.5V IDDG (Vsetp=+1.0V)
-300 -300 Vsetn=0.0V IDDG (Vsetp=+0.5V)
-300 -150 0 150 300 -300 -150 0 150 300
-100
0 100 200 300 400
b) Iin [μA] c) Input Current [μA]
Fig. 8. a) A differential current-mode integrator implemented using only eight IDDG MOSFETs and two
capacitors C. b) Simulated DC transfer characteristics of the integrator for various Vsetp (Vsetn=0V), and
Vsetn (Vsetp=1.0V) values. The tuning is achieved by either the top (Vsetp ) or the bottom (Vsetn ) half of
the circuit, without causing any DC offsets. Its impact on the linearity (inset) is only slightly below the
SDDG performance at identical conditions. c) The third-order harmonic distortion (HD3) is a strong
function of the tuning voltage in IDDG integrator. Even though it is below in down-tuning conditions,
for up-tuning configurations (Vsetn>0 or Vsetp<1) the HD3 figures of IDDG design are quite comparable
to that of SDDG.
examples. Although there are many options and transfer function choices, again, we focus on
current mode integrators that can fully take advantage of DG-CMOS architecture.
As the first example, a current-mode integrator proposed in (Karsilayan & Tan, 1995) is
implemented using IDDG MOSFETs, as shown in Fig.8a. This design eliminates the additional
output blocks used in tunable bulk CMOS equivalent, reducing the transistor count from 16
to 8. Halving the number of transistors not only reduces the silicon layout area, but it can also
translate to reduction in power consumption and transistor parasitics, all of which are crucial
considerations in integrated RF systems (Kaya et al., 2009). In the present circuit, each parallel
pMOSFET pair have been realized with a single p-type DG-MOSFET with twice the width
of the n-type devices, i.e. (W/L)n =10 and (W/L) p =20. In the conventional circuits used for
comparison, every IDDG-MOSFET is replaced with twin SDDG or bulk CMOS transistors in
parallel. The conventional CMOS transistors used for this purpose have identical gate stack as
the DG-MOSFETs but 3 times deeper (30nm) junctions typically found in bulk Si technology.
The proposed integrator circuit is essentially composed of two balanced current-mirror blocks,
clamped together at the center nodes, and an input capacitor. The input current offsets the
balance between the n-type and p-type branches by (dis)charging the center node higher
(lower), resulting in a net deficit (excess) current at the output node. To facilitate tunability,
the back-gates all of n-type (p-type) DG-MOSFETs are tied together to a voltage Vsetn (Vsetp ).
Vsetp=1.0V Vsetn 20 VsetpO=0.5V

0 Vsetn=0V
- 1.0V 0.8V VsetpI=1.0V
- 0.5V
-10 0.0V 1.0V
0.5V 0
Gain [dB]
Gain [dB]
-20 1.0V 1.2V

Vsetn [V] VsetpO [V]
-30 -20
-1 0 1 2 0.4 0.6 0.8 1 1.2
1000 20
BW [MHz]
BW [MHz]
Gain [dB]
-40 Gain
100 60 10
-40
-50 10 40 0
IDDG BW
SDDG
1 20 -10
-60 -60
1M 10 M 100 M 1G 10 G 100 G 1M 10 M 100 M 1G 10 G
a) Frequency [Hz] b) Frequency [Hz]
Fig. 9. a) Simulated BW of the balanced integrator for C=1pF. The inset shows the extracted tuning
range for the same figures in the SDDG and IDDG cases b) Simulated gain tuning of the integrator for
C=1.0pF. The inset shows there is no trade-off between the BW and the gain in this current-mode circuit.
The tuning of the integrator can be accomplished either by adjusting voltage Vsetn for a fixed
Vsetp =VDD =1.0V or by setting Vsetp while Vsetn is grounded. The integrator can also be tuned
by concurrently setting the Vsetn and Vsetp .
Overall, the integrator circuit is found to have very good linearity and an impressive tuning
performance, indicated by the DC transfer data in Fig.8b. The unique feature of this circuit is
the common node between the upper and lower CM blocks, which prevents the development
of DC offsets by the concurrent modulation of these blocks by the input capacitance C. The
lack of DC offset at the output which often plague such tunable circuits (Sedighi & Bakhtiar,
2007; Zeki et al., 2001) is a distinguishing characteristic of this circuit.
Using the integral function method developed by Cardeira and co workers (Cerdeira et
al., 2004), it is possible to analyze the same DC transfer curves to calculate total harmonic
distortion as well as the 3rd harmonic distortion (HD3) as shown in Fig.8c. Even with
very large input currents we find that HD3 remains below −20dB. The linear relationship
between Iout and Iin is especially impressive for |Iin |<150μA. For |Iin |>150μA, down-tuning
(Vsetn <0.0 and Vsetp >1.0V) results in a less-linear circuit. However, at up-tuning (Vsetn >0.0
and Vsetp <1.0V) settings the errors in the output of IDDG circuit approaches that of the SDDG
counterpart for |Iin |<250μA and HD3 drops to −80 dB level. Such a wide variation in
linearity performance indicates that even though IDDG-MOSFETs are intrinsically capable
of matching SDDG performance for distortion, this is only possible at up-tuning that fully
activate the back gates.
The AC response of the integrator (Fig.9a&b) indicates that the BW and gain can be tuned
by using different but non-exclusive biasing schemes requiring only ±1V. The tuning of BW
by more than two decades can be obtained via a single control node (Vsetn or Vsetp ), whereas
the gain tuning by 30dB requires the asymmetric bias of Vsetn between the input (VsetnI ) and
output (VsetnO ) nodes. To illustrate the superiority of this IDDG integrator over conventional
counterpart, in terms of tunability, we also include in the inset of Fig.9a&b the simulated
response of the SDDG integrator with twice as many transistors. Since the SDDG devices have
intrinsically higher gm and employs additional transistors for tuning it has almost twice larger
BW, although with a limited tuning range. This limitation arises because the conventional
tuning is limited when the parallel MOSFET shuts off below its threshold. In the case of IDDG
tuning, the back gate can modulate the current in the front gate even when its own conductive
IFB+
VDD S1 S2
IoutBP-
∫ ∫
IoutLP- Vsetp
Iin+
1:10
a)
Iin- IoutLP+
IoutBP+
IIN (-)
N P
IFB- IFB(-)
IOUTBP(+)
C C
Vsetn
0
Current Gain [dB]
Vsetp=1.0V Vsetp=0.6
C C
-20
Vsetp [V] IOUTBP(-) IFB(+)
0.6 0.7 0.8 0.9 1 P N
400 15 IIN(+)
fo [MHz]
-40 300
10
Q
200
5 1:10
100 Vsetp
0 0
-60 VDD
10 100 1000
c) Frequency [MHz] b)
Fig. 10. a) The block diagram for the tunable current-mode 2nd order LP/BP filter using the integrator
above. b) The full circuit diagram for the 2nd-order BP filter using two integrator stages. The second
stage (S2) is simplified by using a simple C-gm integrator since an LP output is not used in this case. A
full LP filter would require the full integrator block in S2 but not the intermediate block S1 as BP output
is redundant. A size ratio 1:10 in the 1st stage is used to generate gain. c) Simulated frequency response
of the 2nd-order BP filter (C=1pF) as a function of control node Vsetp . The filter can be tuned only using
0.5V and without impacting Q.
channel ceases. It must be pointed out that the inset in Figure 10c also shows vividly the lack
of gain tradeoff in this current-mode circuit.
Two of the tunable integrators above can be employed to build a dual-response
low-pass/band-pass filter. The circuit topology for this low/band-pass filter is shown in
Fig.10a&b). To create a more compact design the second stage (S2) integrator is simplified
by using a basic current-mode C-gm integrator. It is sufficient to replace this stage with the
full design to provide also a low-pass output. Conversely, the secondary output of the first
stage (S1) may be eliminated if the band-pass output is not required. Either way, it possesses
very impressive tunable characteristics, as shown in Fig.10c, BW moving over a decade just by
tuning one of the control nodes, in this case Vsetp , by half a Volt. By combining the control node
for the n channel MOSFET block (Vsetn ) and extending the voltage range, it should be possible
to move the center frequency further or tune the quality factor, which is weakly dependent on
any one of the control signals, as shown in the inset of Fig.10c.
3.6 Oscillators
So far the oscillator circuit design has not extensively benefited from the DG-CMOS
architectures as the limited number of published works concentrate on the DG
implementations of known circuits. Yet the use of IDDG MOSFETs make these circuits tunable
oscillators, which have a very wide and significant application potential illustrated in the
examples below.
3.6.1 Voltage-Controlled Ring Oscillator (VCRO)

Normally a conventional ring-oscillator circuit has oscillation frequency fixed by the
architecture and the number of inverters used. Fig.11a shows that the basic IDDG-inverter
a) VIN 1 2 2n-1 Two Possible Designs:

1) All back gates tied together => Vbgn = Vbgp
2) n and p-type back gates biased oppositely => Vbgn = -Vbgp
initialization
20 1.0V (VBGp=VBGn)
Vbgn=Vbgp=0.55V
9-Stage RO 1.5V (VBGp=VBGn)
19-Stage RO
Out [V]
1 1.0V (VBGp+VBGn =VDD)

1.5V (VBGp+VBGn =VDD)
Frequency [GHz]
15
0
Vbgn=Vbgp=0.8V
Out [V]
1 10
0
Vbgn=Vbgp=1.1V
5
Out [V]
0
0
0 0.2 0.4 0.6 0.8 0 0.5 1 0 0.5 1 1.5
Time [ns]
b) c) Back Gate Bias [V]
Fig. 11. a) Simple ring oscillator becomes a versatile VCO in the IDDG implementation based on
back-gate biasing of inverters (L=50nm and W p /Wn =2) b) The transient response of the VCO to various
p
control biases when Vnbg = Vbg c) The proposed control characteristics of the VCO in single and dual
control schemes for two different sizes of the rings
can be used to build a simple yet efficient tunable ring oscillator with two different operation
modes depending on the back-gate biasing scheme used. These two modes correspond to
single and differential gate control, where the back gates of n- and p-type DG-MOSFETs are
p p
either tied together (Vnbg =Vbg ) or biased oppositely (Vnbg =−Vbg ). As the two biasing schemes
change the threshold or the delay of the IDDG inverters, respectively, the oscillation response
becomes sensitive to the bias voltages. The transient response of a nine-stage IDDG ring
p
VCRO is shown in Fig.11b for three cases of the control signal Vnbg =Vbg . The typical control
characteristics of the same circuit are shown in Fig.11c for two different designs (9 or 19
stages). The oscillation frequency of this VCRO is a strong, almost linear function of the
p
applied control bias, especially in the case of differential control bias (Vnbg =-Vbg ). The two
p
branches of tuning curve in the single bias control case (Vnbg = Vbg ) correspond to oscillations
with different duty ratios (tON /tOFF > 0.5 vs. tON /tOFF < 0.5) for the output signal.
Note that a wider tuning range is possible for larger Vdd values as well as longer chains of
inverters, which has been kept here fairly small to minimize the simulation times (hence the
GHz frequencies). A preliminary study of phase noise and jitter performance on the VCRO
structure has revealed that, the proposed VCRO circuit has mediocre characteristics in terms
of stability and may not be used for timing or system clock circuits. However, they will still
be very attractive options for simple sensing and counting circuits as well as local oscillators
in the communication circuits (Kaya & Kulkarni, 2008)
3.6.2 Current Controlled Oscillators (ICO)

For sensors producing current signals, such as found in most optoelectronic systems, it is more
convenient to alter the oscillator output via a current input. The inset of Fig.12a shows the
IDDG implementation of such an ICO, based on the ultra-low power RC relaxation oscillator
100 70
0.4V (SDDG)
0.5V (SDDG) 50
0.5V VBG=0.25V
0.6V (SDDG) VDD ; IIN
Frequency [MHz]
Frequency [MHz]
0.6V VBG=0.3V 30
10 0.7V (SDDG) 0.4V; 0.5nA
0.7 V, VBG=0.35V 0.5V; 0.5nA
IIN
0.5V; 5nA
Vbgn Vbgn
Mp1 Mp2
0.6V; 5nA
Vbgn
10 0.7V; 5nA
1 Vbgp
Q Q 7
Mn3 Vbgn Mn4
Vbgn Vbgn
5
Vbgn Vbgn
Mn1 Mn2
VSS VSS VSS VSS
0.1 3
0.01 0.1 1 0 0 0.1 0.2 0.3 0.4 0.5
Input Current [nA] Back Gate Bias [V], VBGp=VBGn
a) b)
Fig. 12. a) (inset) the switched-current ICO circuit built using IDDG MOSFETs (L=50nm and W p /Wn =2)
and a latch,and its ICO response. b) The same circuit can also operate as a VCO circuit using back-gate
biasing, albeit with a lower sensitivity. Note that the ICO response is extremely linear and spans for
orders of magnitude even on a log-log scale
circuit proposed by (Chunyan et al., 2003). The most interesting, and also attractive, feature
of this circuit is the lack of bias supply for the input block comprising two p-type MOSFETs
(M p1 ,M p2 ) and four n-type MOSFETs (Mn1 -Mn4 ). As a whole the circuit is a switched current
circuit driving a CMOS latch block at the center. The output (Q,Qb) of the latch is also the
oscillator output driven by the middle nodes (drains) of the inverters Mn1 /M p1 or Mn2 /M p2 .
For weak signals it takes much longer to (dis)charge these intermediate nodes so the latch
output does not alter frequently. When it does, the same input signal is directed to the opposite
branch to continue the (dis)charge operation over again and so on. Therefore the input stage
does not need a steady DC current source or large voltage drops beyond a single threshold.
The result is an ICO that have a very impressive sensitivity to the input current down to pA
range as shown in Fig.12a, only limited in the present simulations by the numerical accuracy
of the transistor models and convergence criteria. In principle sub-pA range of currents can
be detected (Chunyan et al., 2003), while at the higher frequency end the circuit is only limited
by the delay of the latch. Moreover the oscillations are possible at low Vdd values down to 0.3
V or so, depending on the DG-CMOS thresholds.
Strictly speaking, the ICO circuit above does not need a DG-CMOS for operation. However,
the DG-CMOS implementation has two advantages: i) it can be used also as a VCO by virtue
of the back gate bias and ii) operates more efficiently with a higher upper limit as a result
of higher transconductance of DG-MOSFETs. The former can be achieved in a variety of
fashions. For instance the back gates of the transistors in the input block can be biased using
either the single or the differential fashion as found in the precious circuits. Or the center latch
circuit can be back-biased. The outcomes of these two approaches is presented in (Chunyan
et al., 2003). Although the accessible frequency range in the VCO mode is dwarfed in contrast
to massive ICO response given in logarithmic scale, the VCO performance can be improved
by increasing Vdd above 0.5 V or by employing more number of inverter stages for the latch.
Thus the same circuit can be used as a universal ICO/VCO circuit capable of operation with
pW level of signals. More importantly, based on the above example, it should be possible to
convert any ICO circuit to a VCO using the back gates in IDDG-MOSFET equivalents. This
opens up exciting possibilities for ICO/VCO design and analog signal processing.
1 5
Vdd RF
FFT Spectrum (arb.)

Term
R LO (60MHz) LO Amplitude
0.9 Term
0
Vout (10MHz)
(RF±LO)
Conversion Gain [dB]

IF Terms Higher
(50±10MHz) Harmonics
0.8 LO RF
Vout [V]
-5
0.7 0 50
Frequency [MHz]
100 150
-10
0.6
-15
0.5
-20
LO DC Offset
0.4
-25
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.2 0.4 0.6 0.8 1
Bias [V]
a) Time [μs] b)
Fig. 13. a) A one transistor simple DG mixer circuit and its output in time and frequency domain (inset).
b) The dependence of the mixer down conversion performance as a function of local oscillator (LO)
amplitude and DC offset for an RF signal of 80 mV pp
3.7 RF Mixers
RF mixers are commonly used in transceiver and analog signal processing systems for up or
down conversion of input signals with respect to a reference signal, the local oscillator (LO).
They mix (multiply) the two input signals (RF and LO), to produce an output that contain
sum (w RF + w LO ) and difference (w RF − w LO ) term in the spectral content also known as
Intermediate frequency (IF) terms. These new terms at the output is a direct consequence
of the non linear transconductance of the transistors. For a good mixer, higher the IF terms
the better with respect to the incoming signal amplitude, whose ratio decided the conversion
gain. The smallest bulk CMOS circuit accomplishing this task requires 3 transistors, where as
the balanced Gilbert Cell is built with as many as 6 transistors. Therefore, the simplicity and
performance of a single transistor mixer realizable with a IDDG MOSFET is a truly efficient
and interesting one, which we investigate below.
The DG MOSFETs possess natural features suitable for signal cross modulation in efficient
and compact RF mixers. The availability of closely coupled and well matched pair of of
gates along with fully depleted body in a DG-MOSFET allows us to build a mixer circuit
using only one DG MOSFET as shown in Fig.13a (inset). The resistor is intended for setting
a reasonable DC bias as well as serving as an AC load for the mixer. Both the TCAD and
UFDG simulations have been found to produce almost equivalent temporal oscillations (main
panel) and the FFT spectra given in Fig.13a (also inset). Using an RF (50MHz) and LO signals
(10MHz) with equal and small amplitudes (100mV), we can clearly observe the IF terms in
the FFT spectrum. In order to avoid the bias-point related considerations DC level of both the
RF and LO signals are set at 0.5V. Since the DG-MOSFET used in this case has a threshold
around 0.25V a depletion mode device may be more suitable in physical implementation
when DC level is zero. The appearance of higher harmonics (w RF − 2w LO =40MHz and
w RF + 2w LO =80MHz) in the spectrum is indicative of the higher order non-linear terms
in DG-MOSFET transconductance as well as the non-balanced architecture of this single
transistor mixer. This can be easily remedied with the inclusion of well known balanced mixer
topologies canceling odd terms by current addition. The optimum operating conditions and
peak performance of this simple mixer architecture can be explored with the aid of Fig.13b.
The left panel shows that conversion efficiency already saturates when the RF amplitude
reaches 100mV, while the right panel indicates that a gate overdrive (VGS −VT of around)
0.4
VDD VDset 0.4
Output, out2 [V]

Vsetp
0.2 0.2
Output [V]
VIN
0 0 {VDset,VSset}
Vsetn
VSSet
{0.2V,-0.2V}
VSS -0.2
-0.2 VOUT1 VOUT2 {0.3V,-0.3V}
0.4V {0.4V,-0.4V}
0.7V
-0.4
-0.4 1.0V Vsetn=Vsetp=Vout1
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
a) Input [V] b) Input [V]
Fig. 14. a) Simulated DC response of the tunable inverting Schmitt Trigger with a large hystereses
p
obtained with relatively small control voltages (Vset = -Vnset ), thanks to large gain of CMOS pair used in
the second stage for feedback and the inversion is obtained at the output node out1 b) An alternative
design for a tunable non-inverting Schmitt Trigger, where tunable rail voltages (VSset and VDset ) and a
p
higher gain second stage with symmetric gates (Vnset =Vset =Vout1 ) are used. In this case the output is at
node out2 , thus non-inverting, and the hysteresis can be scaled both vertically and horizontally using
only 4 DG-MOSFETs
250mV is sufficient as the sweet spot in device operation where the non-linear terms are
maximized.
It is possible to build other interesting but more complex mixer architectures with DG-CMOS
devices, at the expense of circuit area. However, the simple mixer above may find a special
welcome for area-tight wireless communication circuits where space and price is a premium.
Moreover, the performance of the simple DG-mixer can be further enhanced by material and
electrostatic optimization of contacts, channel strain and even doping, known to play a big
role in linear device design.
4. Mixed-Signal & Digital Circuit Blocks

Designed as solutions to extend Si CMOS scaling, the DG MOSFETs are known to offer a
better switching performance in logic circuits, especially in SDDG configuration. However,
this is not the objective of this section and would not be suitable for the general theme of
this book. Yet, the inclosure of a section on logic blocks have relevance in two aspects: they
are important just as much in mixed-signal circuitry and they can be beneficial pedagogically
in exploring more creative designs in building adaptive systems. In some cases, notably in
non-linear analog circuits such as the Schmitt trigger circuits or Threshold Logic circuits to be
presented next, the boundary between the analog and digital world is even more blurred and
parts of the circuit effectively works as an analog computing element.
4.1 Schmitt Triggers

The ability to laterally shift the CMOS amplifiers transfer response paves the way for the
construction of a simple Schmitt Trigger circuit, a non linear analog circuit block very useful
in reducing noise in analog wave shaping and control circuits as well as in digital systems.
In our design, we use only four DG-MOSFETs as opposed to six MOSFETs needed in bulk
CMOS design (Cakici et al., 2003). Previous attempts with DG-MOSFETs were either not
tunable or needed six transistors for tunability (Kumar et al., 2004). To built a circuit that
A B VDD VDD
VDD
VDD
VG1 VG2 FOUT

F F
VSS F
A B A A
VG1 VG2 FOUT A B
0V 2V A
2V 0V B VSS
0V 0V A•B
B B
-2V -2V 1 VSS VSS
2V 2V 0
a) b) c) d)
Fig. 15. a) a fine-grain reconfigurable static CMOS digital logic gate realizing five different functions via
back gate control by (Beckett, 2008) b) Bulk CMOS NAND implementation, c) hybrid NAND circuit with
p-type DG-MOSFET and d) ultra-compact NAND gate realized using a high-VT n-type DG-MOSFET
(filled black) (Chiang et al., 2005)
overcomes these inadequacies, we consider in Fig.14a (inset) a two-stage CMOS circuit where
p
the conjugate programming of the second stage (Vset =−Vnset ) shifts the first stage’s response
to two extremes. The simulated output of the Schmitt Trigger circuit is shown in Fig.14a for
three different bias settings. The conjugate bias required to set the two extremes, i.e. the width
of the hysteresis, can be decided from Fig.2b. The relatively large gain of the second stage is a
key here in producing a very large hysteresis width. To design a small hysteresis, application
of a relatively large conjugate bias may be needed, limiting the output swing of the second
stage or the amount of shift for the first stage. An upper limit for the resulting power savings
in this Schmitt Trigger circuit with four transistors is expected to be around 11% to 14% as
shown by earlier works (Cakici et al., 2003).
It is also possible to scale the whole hysteresis by adopting a different topology in the second
stage. In this case the rail voltages are the programmable nodes (VDset and VSset ), and the
p
back gates are tied to front gates (Vset =Vnset =Vout1 ), i.e, the SDDG inverter configuration.
The simulated characteristics of such a circuit are given in Fig.14b for three rail voltage
combinations. The hysteresis is scaled both vertically and horizontally as the feedback
voltage from the output of the second stage changes. Also, the gain of the second stage is
higher, resulting in a noninverting Schmitt trigger with almost ideal shapes and more spacing
between them.
Yet another way of optimizing the Schmitt Trigger circuits would be to reduce bottom-gate
coupling by a thicker gate oxide, which would result in smaller shifts in Figure 9a between bias
settings. This requires process changes and may be a less desirable path than voltage tuning,
which can be realized in a number of alternative fashions besides the above approaches. In
any case, tuning via rail voltages may have its own limitations if the tuning circuitry cannot
tolerate low-impedance nodes in the circuits above.
4.2 Reconfigurable Static DG CMOS Logic

The DG-CMOS inverter, previously utilized as a high gain amplifier, is the first and foremost
reconfigurable logic block to consider in exploring the boundaries of reconfigurable logic
circuits designed with the DG-CMOS technology. It is a simple yet very important circuit.
Also known as the logic NOT gate in digital logic circuits, it has a very wide range of usage
in all digital systems at all levels of complexity, and determines power×delay product. The
switching threshold is usually a trade-off for power and speed and is likely to remain fixed
once the device is fabricated. The fabrication tolerances can result in unwanted switching
thresholds that are difficult to compensate, which can lead to logic errors or poor performance.
The DG-CMOS inverter, on the other hand, can modify the DC transfer curves in order to
compensate for the process, voltage, and temperature variations. Such a flexibility will only be
becoming more important as the device dimensions go below 20nm, beyond which parameter
fluctuations are much larger and more varied (Hwang et al., 2009) At the same time, even a
single IDDG-MOSFET can offer a lot as a programmable elements used for turning off power
to a complete logic block in an effort to cut down leakage in power-off modes (Tawfik &
Kursun, 2004). Therefore, the variable threshold in IDDG devices has many more avenues to
impact mixed-signal design than discussed in the following sections.
An interesting and powerful example for reconfigurable static CMOS logic may be found
in Fig.15a that uses the back-gate mediated extreme threshold swings to alter the output
functionality obtained from only 4 transistors. Obviously, what is interesting is not the actual
functions implemented, which are trivial, but the concept which can be extended to include a
more complex array of functions using only a fraction of transistors that would be needed in
conventional designs.
Another impressive approach to building compact reconfigurable circuits were proposed by
IBM group, who indicated that IDDG n-MOSFETs threshold can be selected high enough
so that it would only conduct when both inputs are high. This is of course the logic AND
functionality from a single transistor, which can be employed in CMOS NAND gates as
shown in Fig.15b. It provides impressive gains in Si area usage (∼50% reduction), switching
speed (11% improvement for a four-input NAND) and power dissipation (10% reduction),
which are experimentally confirmed (Chiang et al., 2006). While these result are impressive
in themselves, the elegancy of the concept and flexibility it can provide in reconfigurable and
programmable circuits are probably so far under-appreciated.
4.3 Compact Dynamic Digital Circuits

A dynamic CMOS digital circuit performs its functions in successive pre-charge Φ = 0
and evaluation pulses (Φ = 1) of a periodic clock signal. Dynamic digital circuits feature
a high-speed operation because the parasitic capacitance is minimized by abandoning the
pull-up network in favor of clocking a single p-channel MOSFET that always charges the
output node to logic ’1’ state before the output evaluation phase. Transistor sizing is a key
aspect for performance, as optimum transistor size in the pull-down network would lead
to a a faster discharging rate. In contrast to a static digital circuit, which would always
have twice the capacitive loading (pull-up and pull-down networks), this results in faster
operation and lower power dissipation. Two dynamic logic circuits (NAND and NOR) built
using IDDG-MOSFETs are studied in this section to illustrate the capabilities of DG-MOSFETs
for reconfigurable logic systems. The circuits Fig.16a&b also employ the high-VT transistors
at the logic kernel (see previous section), which leads to halving of the number of input
transistors as compared to the conventional CMOS design. It also shortens the long chains of
n-channel MOSFET in the path of discharge current by 50%, which is important for its speed
performance. Also, the clock inputs are designed using SDDG transistors in an effort to boost
VDD 1
VDD 1000 1100 1110 1111 1111
Clock [V]
(ABCD)
0.5
Φ
Φ 0
F F F
F
AND Output [V]

1
A B 0.5
A B C D
0
C D
Φ
1
OR Output [V]
Φ 0.5
VSS 0
VSS 0 5 10 15 20
Time [ns]
a) b) c)
Fig. 16. High-VT threshold DG-MOSFETs (filled symbols) is used in the logic kernels of the
ultra-compact a) 4-input domino F=AND logic gate and b) 4-input domino F=OR logic gate. c) The
corresponding timing diagrams obtained from SPICE simulations verifying correct operation as
recorded at the non-inverting output (F).
pre-charge and evaluation performance. Note that each pair of inputs driving the independent
gates of a single nMOSFET actually carries out an AND functionality as implied by the high-Vt
(Chiang et al., 2006). It is therefore important to choose and control DG-MOSFET threshold
accurately for this scheme to work.
The simulated timing diagrams obtained from transient SPICE simulations of these two
circuits are jointly plotted in Fig.16c, which verifies the correct operation for each input vector
indicated in the clock-panel. It is helpful to remember that the output evaluation is done at
the rising-edge of a clock signal. Although these circuit examples are simple, the implications
for an array of logic systems including memories have been well documented (Datta et al.,
2009). For instance, it has been reported that IDDG dynamic logic circuits with improve the
read stability of SRAMs by 62%, while reducing its idle mode leakage power, the write power,
and the cell area by up to 62%, 16.5%, and 25.53%, respectively (Tawfik & Kursun, 2004)
4.4 Power Efficient DG-XOR Circuit

A practical example of how the DG-CMOS devices can improve the static CMOS circuit
performance may be found in Fig.17a, which shows a compact XOR (⊕) circuit block based on
high-VT IDDG transistors. XOR circuits are crucially important for implementing a number
of common logic blocks such as the parity coders or adders. Thus improvements in this
circuit has large implications for a given technology. The number of transistors required to
implement this four-input circuit in conventional CMOS technology is eight. However, we
only use four transistors and shorter pull-up network thanks to AND functionality hidden
with the high VT IDDG transistors. An evaluation of the SPICE transient output given in
Fig.17b confirms that the circuit works accurately. The power dissipated in this DG-XOR
implementation VDD =1V is found to be 54% less than that of the conventional circuits with
eight single gate transistors. This is accompanied by a 20% speed improvement as well, which
resulted from the reduced parasitics.
A [V]
0
A B A B
1
B [V]
F=A B
A B 0
F [V]
A B
A B
0
0 5 10 15
a) b) Time [ns]
Fig. 17. a) DG XOR circuit with 4 IDDG-MOSFETs, two of which are high VT (filled black) and b) the
simulated output of this circuit
4.5 DG Threshold Logic Gates

In order to build reconfigurable logic systems, one can also use a threshold logic gates (TLG),
which is not as widely known as, but can be more powerful than the elementary Boolean
gates studied so far (Kaya et. al, 2007). TLGs are composed of two blocks: an input circuit
calculating weighted sums of the logic inputs (Σωi xi ) and an output block comparing this
weighted sum against a pre-set gate Threshold (T). If Σωi xi ≥T then the function output F=1,
otherwise F=0. Using a multiple input circuits with tunable T, it is possible to produce many
different logic functions with a single TLG.
To fully exploit the nature of reconfiguration in IDDG MOSFETs, an ultra-compact threshold
logic gate is presented in Fig.18 This circuit is designed with IDDG transistors in the input
block, resulting in fewer transistors, as compared to the original bulk CMOS circuit. The
back-gate of the front half-sized transistors are tied to power rails, ensuring that transistors
are constantly turned on to contribute the half weights as indicated in Fig.18. The half-sized
transistors serve to prevent undefined states when all input transistors are turned off or to
avoid a Vsum =0.5 condition. Both channels of double-gate transistors are used for input
signals in this design, so the number of input transistors is halved. The input signals applied
to p-channel and n-channel double-gate transistors contribute positive or negative magnitude
weights, respectively.
The correct operation of AND, MAJ and OR logic functions are verified using SPICE
simulations as shown in Fig.19. Although this 8-input circuit functions correctly, there is a
concern with the odd-number of inputs being active. When the number of active transistors
is not equal between the n- and p-input blocks, it has been found that noise margins may
deteriorate. This is because the IDDG transistors current increases typically ×2.5 as opposed
to simple doubling when both gates are turned on as in the SDDG case. This additional
current can upset circuit operation. However, it is possible to remedy the noise margin
problem problem using the tunable IDDG threshold at the inverter. Lowering the T slightly to
∼0.45V (VDD =1V) provides compensation for the asymmetry in the noise margin, such that
correct switching is restored. This demonstrates that T adjustment via back-gate biasing may
be used for erroneous output transitions or badly designed TLG circuits. Since the weight
Inputs [V]
00000001
00000011
00000111
00001111
00011111
00111111
01111111
11111111
T
VDD 0
VSS 1 Vsum
Fout [V]
xp2 xp4 xpi
xp0 xp1 xp3 xp(i-1) AND

0
1
Vsum
Fout [V]
VSUM OR
xn0 xn1 xn3 xn(i-1) F F
0
xn2 xn4 xni
1
Vsum
Fout [V]
VSS VDD
MAJ
Σ(xpi- xni)+0.5(xpo- xno) Φ 0
0 5 10 15 20 25
a) Time [ns]
b)
Fig. 18. a) A static-weight threshold logic gate designed using DG-CMOS devices with a minimalist
input block and a tunable gate threshold, and b) its simulated logic functionality
transistors can be eliminated and back-gates used as additional inputs, this implementation
offers remarkable gains in silicon area while also capable able to correct any design errors.
4.6 DG-TLG with Dynamic Weights

Expanding on the static weight DG-TLG design introduced above, an innovative circuit
with dynamic weight programming capability is possible when the back gates are used
weight programming nodes, as shown in Fig.19a. Although it has the more number of
input transistors as compared to the previous circuit, it takes advantage of the back-gates
to dynamically program the weights for all inputs. The back-gate biasing changes weights of
each transistor associated with the input at the front gate. The typical range of the back-biasing
voltages are needed for practical weights and can be found from the plot in Fig.19b. These
weights have been calculated by normalizing simulated currents with the IDDG-MOSFET
current as both gates held at 1.0V. The calculated weights have limited Vds biasing dependency
for weights less than 4. It must be noted that to have zero current at wi =0 or xi =0 case, the
input transistors must have high-VT (>1.0V) in Fig.19a. Therefore, only when both inputs
are high simultaneously (wi =xi =1) will the IDDG transistor be able to conduct current. The
identical half-sized double-gate transistors located in the front of the circuit are biased for
contributing half weights in the analog computation block so race conditions are less likely.
To verify the circuit performance and functionality, a SPICE simulation is conducted in
Fig.20a, which illustrates examples of weight programming for this highly adaptive digital
system. Using the same block with different weights and gate threshold, one can realize
different logic functions easily. Especially for large weights, however, a dedicated D/A
converter may be needed, which is the main drawback of this implementation. The TLG
functions work correctly in all cases, and designed to produce identical outputs, as would be
expected from the choice of weights and the gate threshold (T). Clearly, this circuit has an
expandable functionality, which is useful for fine-grain reconfigurability.
There is one complication in Fig.20a, however, which is associated with the slow speed of
the second function F2 =2x1 +2x3 +2x5 +2x7 . The speed of this circuit is slow mainly because the
headroom of the noise margin is inferior, as can be seen from the internal node voltage, Vsum ≈
T = 0.5V at the time of transition. This implies that the transistor sizes chosen in the design are
not optimum in this particular implementation. The delay in output transition is significantly
influenced by the noise margin as much as the size of input transistors and implemented
functions. Unlike the static-weight circuit, this variable-weight circuit has smooth transitions
T 5
VDD |VDS|=0.25V
VSS
4 |VDS|=0.5V
Mathematical Weights, ωi
wp1 wp2 wpi |VDS|=0.75V
xp0 xp1 xp2 xpi 3
VSUM
xn0 xn1 xn2 xni F F 2
wn1 wn2 wni

p-type n-type
1
VDD DG-MOSFET DG-MOSFET
VSS
Σ(wpi xpi- wni xni)+0.5(xpo- xno) Φ 0

-2 -1.5 -1 -0.5 0 0.5
Back Gate Bias [V]
1 1.5 2
a) b)
Fig. 19. a) DG-TLG circuit re-designed for dynamic weights and b) typical values of mathematical
weights accessible via back-gate biasing
and outstanding noise margins in terms of "stair-case" response shown in Fig.20b. As input
transistors are activated one at a time, no errors appear up to eight active inputs. Therefore,
no complications are expected in weight programming, except providing additional circuitry
to set appropriate back-bias voltages and routing such signals on the chip layout.
5. Future Directions & Summary

With the imminent arrival of public-domain surface-potential based SPICE models for
multiple gate SOI MOSFETs in general and DG-MOSFETs in particular, circuit engineering
is well poised to take advantage of the remarkable design latitude and functional flexibility
these transistors have in store for extending Si roadmap to the next decade. With these new
simulation engines and rapidly expanding system-level efforts led by several national and
international programs in Japan and Europe, along with the several companies and academic
centers now providing practical means to prototype DG circuits, we should expect a wide
range of tunable analog RF circuits, reconfigurable logic blocks, on-chip power management
blocks and mixed-signal system-on-chip applications to come into existence in the next few
years. It would not be surprising therefore to find in five years actual products containing
SDDG and IDDG MOSFETs in ’hybrid’ implementations, whereby a limited number of such
circuits and devices are employed to improve nanocircuits fault tolerance, and adaptability.
Although this timeline is probably rather speculative, once the Si scaling reaches sub-20nm,
it is conceivable to expect that all ’bets’ are open. Then all technologies that can provide
maximum amount of performance leverage (technology nodes) with minimum amount of
investment and departure from the established fabrication lines are in the race to extend
Moore’s Law. We believe DG-MOSFETs may offer what is just needed.
This chapter has provided multiple examples for many of the fundamental analog CMOS
building blocks (including amplifiers, oscillators, filters, mixers and logic gates) used in
today’s wireless communication, mobile computing, and signal sensing and mixed-signal
processing platforms. These building blocks have tunable performance and offer fine-grain
reconfigurable functionalities thanks to the DG-CMOS devices expected to make a big impact
in the final stretch of Si scaling. Especially in the independently driven configuration, the
DG devices are capable of providing the design latitude and flexibility that will be especially
valuable when conventional circuits can not be further pursued due to matching problems,
power dissipation or both. However, they will also bring their own challenges in terms of
layout, control signal routing and additional steps in fabrication.
Math. Weights, wi
1 8 N-weights
00000001
00000011
00000111
00001111
00011111
00111111
01111111
11111111
Inputs [V]
6 P-weights
4
0 2
1
Vsum 0
F1 [V]
Vsum [V]
F1 = 1.5x1+ 0.5x2+ 2x5+ 0.5x6+ 2x7+ 0.5x8, T=7
0
1
Vsum
F2 [V]
0
F2 = 2x1+ 2x3+ 2x5+ 2x7, T=8 1
0
Fout [V]
1
Vsum
F3 [V]
F3 = 0.5x1+ 0.5x3+ 0.5x5+ 0.5x7, T=2

0 0
0 5 10 15 20 25 0 10 20 30 40 50
Time [ns] Time [ns]
a) b)
Fig. 20. a) verification of correct operation of the dynamic weight DG-TLG circuit. b) stair case
simulation exploring the worst case scenarios for the noise margin in NAND/AND functions of
increasing size
6. References
Celler, G.K & Cristoloveanu, S J. Appl. Phys. 93, 4955 (2003).
Skotnicki, T. et al. The End of CMOS Scaling, IEEE Circuits & Devices Magazine, Jan/Feb 2005.
Amara, A. & Rozeau (Eds.), Olivier Planar Double-Gate Transistor, Springer, ISBN:
978-1-4020-9327-2, (2009)
Tawfik, S.A. & Kursun, V. Low-Power and Compact Sequential Circuits With Independent-Gate
FinFETs. IEEE Transactions Electron Devices, 55, pp.60-70 January (2008)
Roy, K.; Kulkarni, J.P. & Gupta S.K. Device/circuit interactions at 22nm technology node,
Proceedings of the 46th Annual Design Automation Conference, July 2009 (pp.
97-102) San Francisco, USA.
Liu, Yongxun et al., Advanced FinFET CMOS Technology: TiN-Gate, Fin-Height Control and
Asymmetric Gate Insulator Thickness 4T-FinFETs, Int. Electron Devices Meeting-IEDM,
December 2006 (pp.1-4), San Francisco, USA.
Meek, B. & Wilson, D.G. Flexfet Independently-Double-Gated CMOS for Dynamic Circuit Control,
IEEE Workshop on Microelectronics and Electron Devices - WMED , April 2009
(pp.1-4), Boise, USA.
Mathew, L. et al. Vertical CMOS double gate MOSFET with notched poly gates, IEEE Si
Nanoelectronics Workshop, June 2002 (pp. 5-6) Honolulu, USA.
Pei, G & Kan, E.C.C Independently driven DG MOSFETsxed signal circuits: Part I - quasi static and
nonquasi statiic channel coupling, IEEE Transactions on Electron Devices, 51, 2086-2093
(2004).
Raskin, J.P ; Chung, T.M. ; Kilchytska, V. ; Lederer, D. & Flandre, D. Analog/RF performance of
multiple gate SOI devices: wideband simulation and characterization, IEEE transactions on
Electronic devices, 53(5), 1088-1095 (2006), pp. 1088-1095 (2006)
Reddy, M.V.R.; Sharma, D.K. ;Patil, M.B. & Rao, V.R. Power-area evaluation of various double-gate
RF Mixer topologies, IEEE Elec Dev. Lett., 26(9), pp.664-666, (2005)
Mathew, L. et al. CMOS Multiple Independent Gate Field Effect Transistor (MIGFET), Poceeedings
of IEEE SOI Coference, 187-188 (2004)
Kaya, S.; Hamed, H.F.A & Starzyk, J. Low power Tunable Analog Circuit Blocks Based on
nanoscale Double-Gate MOSFETs, IEEE Transactions on Circuits and Systems II, 54(7),
pp 571-575 (2007)
Kranti, A.; Chung, T.M.; Flandre, D. & Raskin, J.P. Laterally assymmetric channel engineering in
fully depleted double gate SOI MOSFETs for high performance analog applications, Solid
State Electronics, 48(6), 947-959 (2004)
Fossum, J. G. UFDG User’s Guide (Ver. 3.0), University of Florida, Gainesville, Aug. 2004.
Ge, L.; Fossum, J.G. & Liu, B. Physical Compact Modelling and Analysis of velocity overshoot in
Extremely scaled CMOS Devices and circuits, IEEE Transactions on Electronic Devices,
vol. 48, pp. 2074-2080, Sep 2001.
Kim, K. Design and analysis of Double gate CMOS for Low Voltage Integrated Circuit Applications,
including physical modeling of Silicon-on-Insulator MOSFETs, Ph.D Dissertation,
University of Florida, Gainesville, 2001.
Online resource at http://www.synopsys.com/tools/tcad/pages/default.aspx
Cakici, T.; Bansal, A. & Roy, K. A low power four transistor Schmitt Trigger for asymmetric double
gate fully depleted SOI devices, Proceedings of IEEE SOI Conference, 21-23 (2003)
Kumar, A.; Minch, B.A. & S. Tiwari, Low Voltage and Performance Tunable CMOS Circuit Design
Using Independently Driven Double Gate MOSFETs, Proc. IEEE SOI Conference, p.119,
2004.
Sanchez-Sinencio, E. & Silva-Martinez, J. CMOS transconductance amplifiers, architectures and
active filters: a tutorial, IEE Proceedings - Circuits, Devices and Systems, 147, pp.3-12,
Feb 2000
Nauat, B. A CMOS Transconductance-C Filter Technique For Very High-Frequency, IEEE Journal
of Solid-State Circuits, Vol. 27, no.2, pp.142-153, Feb 1992.
Szczepanski, S.; Koziel, S. & Sanchez-Sinencio, E. Linearized CMOS OTA using active-error
feedforward technique, Proceedings of the International Symposium on Circuits and
Systems, p.549, May 2004.
Karsilayan, A. & Tan, M. Current Mode Tunable Integrators for low voltage applications, Electronic
Letters, 31, pp 1525-1526, 1995
Sedighi, B. & Bakhtiar, M.S. Variable-gain current mirror for high-speed applications. IEICE
Electronics Express, 4, pp.277âĂŞ281, 2007
Zeki, A.K.; Toker, A. & Ozoguz, S. Linearly tunable transconductor using modiïňAed ˛ CDBA,
Analog Integr Circ Sig Process, 26, pp.179âĂŞ183, 2001
Cerdeira, A.; AlemÃan, ˛ M.A.; Estrada, M. & Flandre D. Integral function method
for determination of nonlinear harmonic distortion. Solid-State Electronics, 48,
pp.2225âĂŞ2234, Dec 2004
Kaya, S.; Hamed, H.F.A. & Kulkarni, A. Widely tunable low power high-linearity current-mode
integrator built using DG-MOSFETs, Analog Integr Circ Sig Process, 30, pp.215-222,
June 2009
Hassoune, I.; O’Connor, I. & Navarro, D (2007). On the performance of double-gate MOSFET
circuit applications, IEEE Northeast Workshop on Circuits and Systems (NEWCAS)
pp.558-561.
Kaya, S. & Kulkarni, A (2008). A novel voltage-controlled ring oscillator based on nanoscale
DG-MOSFETs, Int. Conference on Microelectronics (ICM) pp.417-420.
Chunyan, W.; Ahmed, M.O. & Swamy, M.S (2003). A CMOS current controlled oscillator and
its applications, Proc. of ISCAS, Vol.1, pp. 793-796
Tawfik, S.A. & Kursun, V. Robust FinFET Memory Circuits with P-Type Data Access Transistors
for Higher Integration Density and Reduced Leakage Power. J of Low Power Electronics,
5, pp.1-12, 2009
Hwang, C.-H.; Li, T.-Y.; Han, M.-H; Lee, K.-L.; Cheng, H.-W. & Li, Y (2009) Statistical Analysis
of Metal Gate Workfunction Variability, Process Variation, and Random Dopant
Fluctuation in Nano-CMOS Circuits, Int Conf. Simulation of Semiconductor Processes
and Devices SISPAD, pp. 1-4.
Beckett, P. A low power reconfigurable logic array based on double gate transistors, IEEE trans. on
VLSI Systems, vol 16, no. 2, pp. 115-123, Feb 2008
Chiang, M.H; Kim, K.; Tretz, C. & Chuang, C.T. Novel High-Density Low-Power Logic Circuit
Techniques Using DG Devices, IEEE Trans. Electron Devices, vol. 52, no.10, pp.
2339-2342, Oct. 2005.
Chiang, M.H.; Kim, K.; Chuang, C.T. & Tretz, C. High-Density Reduced-Stack Logic Circuit
Techniques Using Independent-Gate Controlled Double-gate Devices, IEEE Trans. Electron
Devices, vol. 53 no.9, pp. 2370-2377, Sep. 2006.
Datta, A.; Goel, A.; Cakici, R.T.; Mahmoodi, H.; Lekshmanan, D. & Roy, K.; , Modeling and
Circuit Synthesis for Independently Controlled Double Gate FinFET Devices IEEE Trans.
CAD of Integrated Circuits and Systems, vol.26, no.11, pp.1957-1966, Nov. 2007
Kaya, S. et. al, Reconfigurable Threshold Logic Gates with Nanoscale DG-MOSFETs, Solid-State
Electronics, vol. 51, no.7,pp. 1301-1307, Oct. 2007.
Celinski, P. et al., State-of-the-Art in CMOS Threshold-Logic VLSI Gate Implementations and
Applications, Proc. SPIE, VLSI Circuits & Systems, vol. 5117, pp. 53-64, Apr. 2003.
10
Statistical Analog Circuit Simulation:

Motivation and Implementation
David C. Potts
Fairchild Semiconductor Corporation
USA
1. Introduction
New technologies are continually being developed that enable designers to create faster,
more complex circuits, packed within a shrinking die. However, along with the promise of
speed and density comes the challenge of variability, as intra-die device mismatch looms
proportionately greater. Analog designs typically employ multiple core building block
circuits, including current mirrors, band gap references, differential pairs and op amps, that
are especially sensitive to device mismatch. Understanding the impact and potential
interactions of variations between these matched devices can be critical in producing a
commercially viable product.
The first part of this chapter will provide a background on the statistical nature of the
semiconductor manufacturing process, with a particular focus on their implications on
device performance. Due to the complexity of interactions coupled with circuit-specific
design sensitivities, traditional corner models do not provide the designer with sufficient
accuracy and visibility to thoroughly assess and improve the quality of their designs.
Corner models also do not account for mismatch, which is a major concern for analog
designs. A statistical simulation system that realistically replicates process variability will
provide the designer with insights to optimize the design.
The second part of the chapter will delve into the extraction and use of statistical models
within a statistical simulation system. A properly implemented statistical design tool can
become one of the greatest assets available to the designer. Following a discussion of
various published statistical model formulations and extraction methodologies from
literature, we will consider how they might be incorporated and used within commercially
available simulators.
We conclude the chapter with a demonstration that systematically evaluates the
components of a band gap circuit to isolate matching sensitivities and refine the design for
optimized results. With the assistance of statistical design analysis, a designer can make
informed choices that will produce better circuit performance and manufacturability.
2. Semiconductor process variation

Semiconductor device and circuit performance will fluctuate due to the inherent underlying
statistical variation in the process itself. This variation can include both random and
systematic components. As illustrated in Figure 1, the overall total variance can be
partitioned into components reflecting the physical separation of the material during
processing.
Fig. 1. Classifications of Statistical Variation

Lot-to-lot variance is generally the largest of the components as it reflects significant sources
of variation not seen in the other groups, including variation across different tools that may
be used at a given process step, variation between batches of raw materials, along with time-
based trends and cycles relating to tool aging, preventive maintenance, upgrades and
adjustments. Wafer to wafer variance can result from the slight differences experienced
between wafers at single wafer processing steps as well as from gradients across batch
processed wafers, such as induced by temperature and flow gradients within a furnace tube.
Die-to-die variance can be an artifact of differences in exposures in stepper based
lithography or gradients or localized disturbances of wafer uniformity. Lot-to-lot, wafer-to-
wafer and die-to-die variance combined are often referred to as Global Variation, because all
devices found on any particular die will be simultaneously and equally affected by them in
the same way. In other words, in the world of that particular die, this is a global effect.
Within-die (device-to-device) variation may include a more localized contribution of some
of the wafer uniformity effects driving die-to-die variance, as well as individual device
definition effects resulting in slight non-uniformities in film thicknesses and edge
definitions, dopant distributions, junction depths, surface roughness, and so on. Within-die
variance is generally referred to as Local Variation, because the performance of each
individual device on a given die will be affected slightly differently by it.
This variation can include both random and systematic components. The designer may
have some limited control over certain systematic components relating to device layout, but
needs to be aware of and have some means to estimate the effects of variation on circuit
performance. Traditionally, this was done using so-called ‘corner’ models, intended to
represent the worst case corners of the process variation.
3. Issues with traditional corner models

In traditional corner methodologies, ‘worst case’ models were typically created by
evaluating the sensitivities of critical model parameters individually and then setting each of
them to their worst case values simultaneously. The accuracy of this approach, however,
would be highly dependent on the actual physical correlation between the parameters as
Statistical Analog Circuit Simulation: Motivation and Implementation 209
well as the cumulative probability that all would be worst case at the same time (Nardi et
al., 1999). The corner method also assumes a ‘one-size-fits-all’ solution, when in reality
different designs and circuit architectures will exhibit different worst case sensitivities.
Finally, fixed corner models do not account for the intra-device variations that can have a
major impact on analog circuit performance.
3.1 The issue of correlation

To demonstrate the impact of correlation, consider two standard normal variables, X and Y,
which are summed and scaled to create Z. Figure 2 depicts the results for 3 cases
representing negative, zero and positive correlation between X and Y:
Fig. 2. The Impact of Correlation

In this simple example, it is intuitively obvious that when X and Y are negatively correlated,
they would tend to cancel each other out, thus minimizing the resulting variability of Z.
Conversely, when they are positively correlated, they would tend to reinforce each other,
creating greater variability. Semiconductor processes, of course, are much more complex
with a great number of interacting variables. The fact that there are a large number of
variables brings in the next problem: how to determine which combinations of these
variables best define the corners?
3.2 The issue of corner selection

Assume we have a normally distributed process and we want to define a set of worst case
corners that encompass an interval of ± 3 standard deviations about its mean (μ ± 3σ). In
other words, the probability the process would fall outside of our μ ± 3σ corners would be
about 0.0027. The probability that two different uncorrelated normally distributed variables
would both simultaneously fall outside their respective μ ± 3σ is only (0.0027)2 = 0.00000729.
As the number of independent variables increases, the probability that they would all
simultaneously fall outside their respective μ ± 3σ windows drops off rapidly, as shown in
Figure 3a.
Instead of putting all variables at ± 3σ, we might prefer to find a ± kσ window such that the
probability of falling outside remains constant at 0.0027 (for n variables, this corresponds to
the standard normal z score for area of (0.00271/n)/2). As the number of independent
variables increases, the k value drops, as shown in Figure 3b.
Of course, there is nothing that forces us to select a corner that puts each variable at the
same k value. Figure 3c show the line that plots possible solutions of k values when there
are only 2 variables to consider (for 3 variables, the solution would be a surface and for n
variables, it would be an n dimensional space).
Fig. 3.
(a) Probability of Multiple Variables Falling Outside Their Respective μ ± 3σ Windows
(b) k Values vs. # Variables for Cumulative Probability Outside μ ± kσ = 0.0027
(c) Possible Solutions for k1 and k2 for Constant Probability Outside μn ± knσn = 0.0027
The more variables there are in a given process, the less likely that the uncorrelated
components within them will all be worst case at the same time. Ideally, a worst case corner
would place those parameters that have greatest impact on circuit performance at more
extreme values, while letting other less important parameters remain at more nominal levels.
In the context of semiconductor device and circuit performance, the relative importance of a
given process parameter often depends on the device architecture and operating conditions.
Figure 4 depicts the sensitivities of several simulated MOS IDS conditions to SPICE model
parameters lint (channel length offset fitting parameter), wint (channel width offset fitting
parameter), vth0 (threshold voltage @ Vbs=0), tox (gate oxide thickness) and rdsw (parasitic
resistance per unit width).
The underlying independent process variables that would contribute to that variation
include poly gate lithography, gate oxide deposition and source drain implant and anneal
(Mutlu & Rahman, 2005). Being independent, the probability of all of them being worst case
at the same time is quite low. Figure 5 further demonstrates this effect, showing the results
of a 10000 trial Monte Carlo simulation of the propagation delay of a simple inverter cell.
Although the Monte Carlo completely covers the range of values defined by the worst case
corner models for the individual model parameters, the resulting propagation delay
distribution falls well inside the values predicted by the corners, simply because the
occurrence of those simultaneous worst case conditions is so improbable:
Fig. 4. Some Underlying MOS IDS Sensitivities vs. Device Size and Bias Conditions
Fig. 5. All Parameters Simultaneously at Worst Case Yields Unrealistic Corners

Complicating the issue of corner selection is the fact that the worst case conditions may be
completely different for circuit performance criteria that are sensitive to different process
perturbations, such as the propagation delay of a CMOS digital logic circuit versus the gain
of an operational amplifier. Even between related circuit performance parameters within the
same circuit cell there can be notable differences. Consider the enable and disable
propagation delays of a sample CMOS digital logic circuit as present in Table 1. When set to
the worst case corners for disable (HZ/LZ) delay, TpZH encompasses less than 25% of the
delay window obtained when using worst case enable corners (0.4nS vs. 1.8nS). The
difference between the two corners is the placement of Tox. Ordinarily, Tox would be
reduced for a Fast corner as it provides higher drive. However, thinner Tox also means
higher oxide capacitance. The benefit of higher drive more than compensates for the penalty
of higher capacitance in active delays, but the impact of the higher capacitance dominates
for disable delays.
Statistical models are not tied to a particular fixed choice of conditions as corner models are.
They are generally formulated to reflect underlying process interactions by re-expressing
the correlated model parameters as functions of an appropriate set of uncorrelated
TpHZ TpLZ TpZH TpZL

Worst Case Corner Setting
(nS) (nS) (nS) (nS)
Corner 1: Worst Case Disable Times 3.2/4.6 3.3/4.1 2.3/2.7 2.0/2.5
Corner 1: Δ Slow - Fast 1.6 0.8 0.4 0.5
Corner 2: Worst Case Enable Times 3.6/4.2 3.5/3.9 1.6/3.4 1.5/2.9
Corner 2: Δ Slow - Fast 0.6 0.4 1.8 1.4
Table 1. Different Circuit Parameters may have Opposing Corner Conditions
parameters. When exercising a statistical model, the uncorrelated parameters are perturbed,
rather than the model parameters directly. These changes are then propagated through to
the model parameters to generate properly correlated model decks. While statistical models
do not inherently resolve the issues of circuit dependencies in and of themselves, they do
enable the use of exploratory statistical simulation strategies including design of
experiments and response surface model (DOE/RSM) techniques that can efficiently
evaluate the response of a given circuit over the entire process/design space to determine
the particular worst case conditions for a given circuit (Rappitsch et al., 2004; Sengupta et al.,
2004; Zhang et al., 2009).
3.3 The issue of localized matching variation

It is imperative for analog/mixed-signal designs, and is becoming increasingly important
for digital designs as well, that today’s simulation methodologies have the means to
evaluate the effects of localized device mismatch on circuit performance. Fixed corner
models applied uniformly across all device instances in a circuit do not provide any
allowance for mismatch. As seen in Figure 6, the impact of mismatch on analog circuit
blocks can easily exceed the variation that would otherwise be expected due to global
variation over the entire process range. Simulating under the effects of global process
variation only, the current mirror output current, IO, exhibited a standard deviation of
~50nA, traced predominantly to VT, with some residual sensitivity to LEFF/WEFF and
Fig. 6. Statistical Simulation of Basic Current Mirror

mobility. Adding in additional slight perturbations to the values of these parameters as

applied each individual device in the circuit, the standard deviation of IO increased about
17x to 0.85uA, almost entirely attributed to the slight difference in VT applied between the
critically matched MOS devices:
Local mismatch variation is observed by comparing two or more identical devices on a die.
In the absence of systematic variation, a normally distributed random mismatch variation
would induce a normal distribution upon a given parameter, P, such that P would be
expected to have a mean of μP, the average value of P across that die, and a standard
deviation of σP:
P ~ η( μP, σP2) (1)

The observed difference in P between any two identical devices would be expected to be
distributed with a mean of 0 and standard deviation of √2σP (variance of 2σP2):
ΔP ~ η(0, 2σP2) (2)

(Lakshmikumar et al., 1986) derived a 1/√(LW) scaling dependence for threshold voltage
and conductance mismatch. Using Fourier techniques, (Pelgrom et al., 1989) postulated a
generalized expression for the variance of ΔP between two rectangular devices as:
AP2
σ 2 (ΔP ) = + SP2 Dx2 (3)
WL
where: W and L are the width and length of each rectangle
Dx is the separation distance between the rectangles
AP , also known as A factor, is the area coefficient and
SP is the spacing coefficient
As indicated that model, the variance of ΔP would be expected to increase as the device sizes
decrease and as the devices are spaced farther apart from one another. The magnitude of
the A factor is typically a reflection of the process design itself as opposed to specifically
controllable manufacturing components (Tuinhout, 2002). For MOS devices, VT, gm and ID
matching is affected by multiple process architectural components, including S/D and
channel doping (Tuinhout et al., 2000 & Dubois et al., 2002) and gate poly/oxide definition
(Difrenza et al. 2003; Brown et al., 2007; Cathignol et al., 2008).
For analog designs in MOS technologies, threshold voltage mismatch is of particular
concern. (Pelgrom et al., 1998) presents a physical representation of AVT, the A factor for
MOS threshold voltage mismatch, as:
q ⋅ tox 2 Ntdepl
AVT = (4)
ε 0ε ox
where: N represents the total number of doping in the depletion region (Na+Nd)
tdepl represents the width of the depletion region
tox represents the gate oxide thickness
A direct relationship between tox and AVT is clearly evident. A former rule of thumb for
technology nodes over 0.1μm gate length suggested AVT, in saturation regions, would run at
about 1 mVμm per nm of gate oxide thickness (Pineda de Gyvez & Rodríguez-Montañés,
2003). Within equation (4), the reduction of tox is somewhat offset by the required increases
in doping levels at reduced geometries. Deep sub-100nm processes bring increasing effects
from lithography and other gate region uniformity challenges (Brown et al., 2007; Cathignol
et al., 2008 & Lewyn et al., 2009). Layout effects and neighbouring topology can all induce
additional mismatch deviations beyond those accounted for in AVT (Drennan et al., 2006 &
Wils et al., 2010).
From a design perspective, it is important to take in account the relationship of circuit bias
selections on resulting mismatch performance (Kinget, 2004). For instance, as VGS
approaches VT, the relative mismatch variation in ID increases, peaking in subthreshold
region as shown in Figure 7:
Fig. 7. MOS ID Relative Mismatch Variation Increases in Subthreshold Region
Fig. 8. Comparison of Current Mirror Data

The influence of biasing impacts can be seen in sample current mirror data. Figure 8 shows
results, measured over multiple mirror configurations and sizes, for the total observed range
of Io (expressed as +/- %) relative to the median operating Io value under various test
conditions. Mirrors intended to run at very low currents will be exhibit proportionately
greater mismatch sensitivities. Reducing this variation requires larger devices and/or more
complex mirror configurations, either of which can adversely impact manufacturing costs
due to a larger die area.
Statistical models can offer the designer the opportunity to evaluate and compare the effects
of mismatch on circuit performance under different design scenarios. Relative to corner
models, statistical models offer improved accuracy, by properly retaining key parameter
correlations, improved coverage, by not being tied to some arbitrary set of corners, and
improved capability, by incorporating localized mismatch as well as global process
variation effects.
4. Implementing statistical design

Implementing statistical design requires the development or procurement and integration of
3 key components: a simulation tool capable of exercising statistical models, the statistical
models themselves and finally the appropriate methodologies to use them efficiently and
cost effectively to validate and improve a circuit’s design (Duvall 2000). The goal of
statistical circuit modeling is to be able to replicate the observed pattern of global and local
variances such that their effects on a particular circuit design can be simulated and, if
necessary, design enhancements introduced prior to committing the design to silicon.
4.1 Extracting statistical models

Statistical models are formulated to retain correlation by re-expressing the correlated model
parameters as functions of an appropriate set of uncorrelated parameters. When exercising
a statistical model, the uncorrelated parameters are perturbed, rather than the model
parameters directly. These changes are then propagated through to the model parameters
to generate properly correlated model decks.
In its most generic representation, a statistical model would define the value of some
parameter P within the jth device on the ith die as:
Pij = μPROCESS + GOFFi + LOFFij (5)

where: μPROCESS = overall process mean for that parameter.
GOFFi = global offset associated with the ith die:
(μ=0, σ2=σ2GLOBAL)
LOFFij = local offset for the jth device on ith die:
(μ=0, σ2=σ2LOCAL)
As indicated in Figure 9, variations in the independent fabrication process variables (eg:
implant dose and energy, furnace temperature, ramp time, flow rate, etc.), interact to create
statistical distributions of the process characteristics (eg: junction depths, doping profiles,
etc.). Different characteristics may exhibit some degree of correlation to one another due to
common influences. For example, the annealing temperature/time of a poly implant will
have some effect on the ultimate doping profiles of earlier source/drain and well
implants/diffusions. The process architecture design and implementation will influence the
nature and strength of these correlations. The statistical variations and inter-correlations of
process characteristics will drive the statistical variations and inter-correlations of the device
characteristics, as influenced by the device architecture design, and so on.
Fig. 9. Progression of Increasingly Complex Parameter Interactions

It is effectively impossible to precisely track the propagation of the variation and their
impacts throughout the levels. We can get a general assessment of process variation from
inline process data, device variation from wafer electrical test (ET) and circuit performance
variation via wafer sort (WS) and final test (FT) data, but we have no way of knowing what
specific process conditions any particular die experienced. TCAD simulators can be coupled
together to cover the entire process (Hanson et al., 1996), but that requires very well
calibrated models as the effects of any errors/omissions would be compounded throughout
the system.
The inputs to the circuit simulator (referred to hereafter as the model parameters) are a
mixture of inter-correlated pseudo-physical as well as non-physical (fitting) parameters.
Since they are inter-correlated, it is not statistically (or physically) appropriate to perturb
their values independently of each other. Proper correlation between the model parameters
can be maintained by expressing the model parameters as functions of other independent
parameters which are more suitable for applying direct statistical perturbations. These
parameterized model expressions can be thought of as behavioral models, developed to
provide suitable proxy for device characteristic/model parameter distributions as inputs to
the circuit simulator such that reasonably realistic circuit performance projections can be
expected.
Establishing appropriate distributions and intercorrelations of the model parameters can be
a significant challenge. Wafer electrical test (ET) data is used to characterize a process and
extract the circuit model parameters. The cumulative effect of the underlying variations in
the process is manifest in the observed distribution of ET parametric data. That is:
E=f(p) (6)
where: E = an ET parameter
p = a vector of process parameters
ET data is used to extract the circuit model parameters. This is generally done by creating a
large database of ET results obtained over a wide array of device geometries, architecture
and operating conditions and using a specialized extraction tool, such as ICCAP, to optimize
the model parameters via curve fitting. Hence, we have:
M=f(E) (7)
where: M = a device model parameter

E = a vector of ET parameters
For statistical modeling, the challenge is to define how to alter the model parameters in a
statistically realistic manner. As stated earlier, it is not appropriate to vary the model
parameters directly since they are correlated with one another. It is also not feasible to
estimate the correlation between the model parameters from the model files themselves as
they are usually only directly extracted for a very limited number of ET sites (and even if a
suitably large set of model files were generated, there would be concerns over whether the
model extraction methodology itself might have influenced the results). TCAD simulation
can be used to develop models tied back to independent physical components, but this
introduces additional, compounding sensitivities to the inherent accuracy of each modeled
stage. Circuit designers and modelers often have less access to and familiarity with those
TCAD tools. They are generally quite familiar with ET data, however, and large samples are
often readily available from which the necessary statistical information can be determined
and utilized for statistical modeling (Chen et al., 1996; Potts & Luk, 1998; Singhal &
Visvanathan, 1999). The variation of several model parameters can be directly mapped to
the variation in measured or extracted ET characteristics, including vth0 (to measured
threshold voltage), xl/lint and xw/wint (to extracted LEFF and WEFF calculations,
respectively), tox (to inverse of gate oxide capacitance) and the sheet resistances of various
layers. Others can be proportionally mapped to functions of measured data, including
mobility (u0 ~ Gm/Cox*[L/W]) and saturation current (is ~ ln(vbe)).
The first step of the extraction process is to validate the ET data, removing any invalid
outliers, and transforming each parameter to a standardized normal distribution (keeping
track of the transformations so that we know how to reverse transform it back later). Next,
we perform principal component analysis (PCA) on the transformed data. PCA is a
technique that can be used to re-express a correlated set of variables in terms of uncorrelated
components [16]. An orthogonal transformation matrix, B, is found such that:
Y=B(E-Ē) (8)
Z=Λ-½ B(E- Ē)=AX (9)
S=B’ΛB (10)
where: E = matrix of correlated ET data, with means Ē
Y = matrix of principal components
Z = standardized PCA components
S = covariance matrix of X1,X2, ...,Xn
B’ΛB = spectral decomposition of S
Λ = diagonal matrix, diag(λ1,..λn), with λ1>λ2>...>λn the eigenvalues of S
Α = Λ-½ B and X = (E-Ē)
Each of the principal components in Y and Z has a mean of 0 and is uncorrelated with all
other principal components (that is, each Yi is uncorrelated with all other Yi and each Zi is
uncorrelated with all other Zi). The variance of each Yi is the value of the corresponding ith
eigenvalue, while the standardized PCA components, Zi, each have a variance of 1. If all Xi
are normal, then each of the Zi is standard normal, which is convenient for formulating the
statistical models. For example, to run a monte carlo, the statistical simulation tool would
generate vectors of Z, with each Zi being a random normal value. These random vectors of
Z would then be reverse transformed back into corresponding vectors of E, from which we
can map random, but properly correlated, perturbations of M!
Figure 10 demonstrates this technique. The black data points represent an actual sample of
data collected over a 4 month period. The original 6 correlated ET parameters are
decomposed into 4 uncorrelated PCA components. The matrix between them on the lower
right graphically depicts that transformation relation. LEFFN and LEFFP are strongly related to
PCA parameter A, TOXN and TOXP are strongly related to B, VTP is strongly related to D and
VTN is related to C with dependance on A and D as well. While the PCA solution is entirely
a mathematical construct, it may offer insights into the underlying physical relationships.
Physically, LEFFN and LEFFP would be highly dependent on the gate poly CD, TOXN and TOXP
on the gate oxide thickness, VTN would be dependent on multiple parameters, including NA,
TOX, xj and, for short/narrow devices, L/W, while VTP would have a strong dependence on
VT adjust implant. A PCA solution that does not appear to bear any resemblance to a logical
underlying physical relationship should merit greater scrutiny of the data for a possible
invalid readings or a need for normality transformation.
Fig. 10. Example of PCA Transformations: ET > PCA & PCA > ET
For parameters that cannot be directly mapped to physical data, it will be necessary to
indirectly estimate appropriate values that will yield appropriate results when used in
simulation. This includes all mismatch parameters. The backward propagation of variance
(BPV) technique is quite helpful in this process (McAndrew et al., 1997; Telang & Higman,
2001, Drennan & McAndrew 2003; McAndrew et al., 2010). Measured ET data is collected
over a wide spectrum of device geometries and bias conditions. Simulations are then set up
covering the same set of parameters. For the first pass of simulations, a small arbitrary value
of variation is assigned to each of the independent mismatch model parameter (such as 1%
of its corresponding global variance). These initial simulations are used to determine the
covariance matrix (or squared correlations) between the mismatch models parameters and
the resulting simulated mismatch variance. Regression analysis is then performed to fit an
appropriate vector of mismatch model parameter variance such that the simulated ET
mismatch variances would approximate the actual measured ET mismatch variances:
σ2ET=S*σ2Model (11)
where: σ2ET = vector of observed ET variance in measured data
S = covariance matrix of simulated ET results vs model parameters
σ2Model = vector of (fitted) variance to assign to model parameters
Fig. 11. Example of BPV to Fit Observed ET Data
4.2 Implementing statistical models

Over the past decade or so, Monte Carlo and other statistical simulation capabilities have
been added to commercial SPICE simulators. They enable the use of specially parameterized
and formulated expressions to implement the desired statistical model behavior (Lu et al.,
2009). Recent compact models are also incorporating new parameters that, when combined
with extracted layout information, can better predict important mismatch sensitivities, such
as stress and well proximity effects (Watts et al., 2006; Yang et al., 2008).
We have implemented our parameterized statistical models within the Cadence Analog
Design Environment, utilizing the monte carlo features available within their Statistical
Analysis Tool (Potts & Luk, 2005). This tool offers the ability to designate random variables
into two groups, process and mismatch, as declared within a statistics block within the
model library, prior to the models themselves:
statistics {
process {
vary G1 dist=gauss std=1
….
vary Gn dist=gauss std=1
}
mismatch {
vary L1 dist=gauss std=1
….
vary Lm dist=gauss std=1
}
}
Within the models, we then encode the ith model parameter, Pi, as a functions of these
independent variables by applying the statistical models we have derived for global
variation , e.g.: fGi(G1,...,Gn), and mismatch, e.g.: fLi(L1,...,Lm), such that:
Pi = PTYPICAL + fGi(G1,...,Gn) + fLi(L1,...,Lm) (5)

Since we have formulated our statistical models as functions of independent normal
variables, each of our global variables (G1 - Gn) has been declared as Gaussian distributions
with a mean of 0 and a standard deviation of 1. The local variables (L1 – Ln) are declared as
Gaussian distributions with a mean of 0 and a standard deviation of δi (0< δi <1), where δi
are fitted through a backwards propagation of variance technique.
With statistical SPICE models in hand, the simplest and most generic analysis methodology,
equally applicable to dc, transient or any other simulation set-up, utilizes Monte Carlo
simulations to detect and isolate potential trouble spots in the circuit. With the Cadence 6.x
ADE-XL/GXL platform, traceability can be enabled to monitor Monte Carlo values applied to
each instance during each trail, providing a means to quickly locate any design weaknesses.
The major drawback to Monte Carlo analysis is simulation time. A large number of trails
are needed, especially if one needs to accurately evaluate the tails of the distribution. This is
less of an issue for small circuits or individual circuit blocks which can be simulated on the
order of seconds or less per trial. As such, one strategy for larger circuits would be to break
it down into blocks, and fitting behavioural macromodels to express the variation of the
output of one block, which could then be applied as the input to the next block. Ignoring
correlation, this could simply be done by redefining a fixed voltage or current as a design
variable, say V1, set by an additional random variable of desired location and spread, e.g.:
parameters V1 = {desired mean value}

statistics {
process { ....
vary V1 dist=gauss std={desired standard deviation}
}
A more proper solution, however, would retain correlation by expressing the V1 voltage as
a function of the same Monte Carlo variables used in defining the SPICE statistical models
themselves. This would be done by running Monte Carlo simulations on the circuit block
that generates the V1 signal, applying regression techniques to fit the resulting V1 over the
values for the Monte Carlo parameters from each trial, and then using that regression
equation to define the V1 input to apply to the next block, e.g.:
parameters V1 = fGi(G1,...,Gn) + fLi(L1,...,Lm)

There are alternative methods that do not require Monte Carlo, including sensitivity
analysis, design of experiments (DOE) and response surface modelling (RSM) techniques.
Typically, a sensitivity analysis is performed to isolate the critical model inputs and then a
DOE is run over those variables (which generally requires far fewer trials than a Monte
Carlo), and then RSM is employed to analyze/optimize the results. These methodologies are
not as readily implemented within standard commercial SPICE simulators, requiring
significant additional pre-/post-processors for set-up and analysis. Commercial solutions
are available from 3rd party vendors, however, including Circuit Surfer® (PDF Solutions),
Variation Designer (Solido Design Automation) and WiCkeD™ (MunEDA GmbH).
5. Demonstrational analysis of a band gap circuit

In this section, we will demonstrate the use of our statistical CAD tools and methodologies
to characterize and optimize a Bi-CMOS band gap circuit consisting of a MOS bias
generator, PNP band gap reference and MOS op amp, as shown in Figure 12. The circuit
was initially designed and simulated to produce a stable reference voltage, VBGOUT, of about
1.18 +/- 20 mV over corner models.
Fig. 12. Band Gap Circuit used in this Example

The baseline process Monte Carlo projected a VBGOUT σ of 9.5mV – virtually all traced to
PNP Is variation.
Fig. 13. Process-Only Monte Carlo Results

The combined process and mismatch Monte Carlo generated a much larger variation along
with a prominent asymmetric low tail:
Fig. 14. Combined Process & Mismatch Monte Carlo

Partitioned mismatch Monte Carlos quickly pinpointed the source of the tail to MOS
mismatch sensitivities within the start-up & biasing block:
Fig. 15. Partitioned Mismatch Monte Carlo Results

Probing in the biasing block revealed “lurking cliff” ΔVt sensitivities between devices P1 &
P2 and N3 & N4 (where P1,P2,... refer to devices as labelled in Figure 12):
Fig. 16. Tail Traced to ΔVT in Bias Circuit

After removing the outlying values in the tail, the remaining mismatch sensitivities are
traced to the differential pair (P5/P6) and mirror (N5/N6) in the op amp and the PNP pair
(Q0/Q1) in the band gap:
Fig. 17. Non-Tail Sensitivities: Op Amp & Band Gap

Increasing the sizes of these identified critical devices by about 2x to 3x from their original
values reduces the Vbgout standard deviation under combined Process & Mismatch Monte
Carlo from ~ 35mV to ~ 10mV. At that point, the PNP Is process sensitivity becomes the
dominant factor in overall VBGOUT variability and any additional mismatch reduction yields
minimal benefit.
Fig. 18. Overall Variation Optimized @ 2x-3x
6. Conclusion
Statistical design offers considerable improvements over traditional worst case design
methodologies. New tools and methodologies are being developed and offered in the EDA
market that will enable the designer to use statistical models efficiently. A statistical design
simulation framework enables the opportunity to make more intelligent design choices up
front that will result in a more robust and manufacturable circuit design.
7. References
Brown, A.; Roy, G. & Asenov, A. (2007). Poly-Si-Gate-Related Variability in Decananometer
MOSFETs with Conventional Architecture. IEEE Trans. on Electron Devices, Vol. 54,
No. 11, (Nov 2007) pp. 3056-3063, ISSN 0018-9383.
Cathignol, A.; Cheng, B.; Chanemougame, D; Brown, A; Rochereau, K; Ghibaudo, G. &
Asenov, A. (2008). Quantitative Evaluation of Staistical Variability Sources in a 45-
nm Technological Node LP N-MOSFET. IEEE Electron Device Letters, Vol. 29, No. 6,
(Jun 2008) pp. 609-611, ISSN 0741-3106.
Difrenza, R.; Vildeuil, J.; Llinares, P. & Ghibaudo, G. (2003). Impact of Grain Number
Fluctuations in the MOS Transistor Gate on Matching Performance. International
Conference on Microelectronic Test Structures (ICMTS’03), pp. 244-249, ISBN 0-7803-
7653-6, Monterey, CA, USA, Mar 2003.
Drennan, P. & McAndrew, C. (2003). Understanding MOSFET Mismatch for Analog
Design. IEEE J. of Solid-State Circuits, Vol. 38, No. 3, (Mar 2003) pp. 450-456, ISSN
0018-9200.
Drennan, P.; Kniffen, M. & Locascio, D. (2006). Implications of Proximity Effects for Analog
Design. 2006 Custom Integrated Circuits Conference (CICC’03), pp. 169-176, ISBN 1-
4244-0076-7, San Jose, CA, USA, Sep 2006.
Dubois, J.; Knol, J.; Bolt, M.; Tuinhout, H.; Schmidtz, J.; & Stolk, P. (2002). Impact of
Source/Drain Implants in Threshold Voltage Matching in Deep Sub-micron CMOS
Technologies. 32nd European Solid-State Device Research Conference (ESSDERC 2002),
pp. 115-118, ISBN 88-900847-8-2, Bologna, Italy, Sep 2002.
Duvall, S. (2000). Statistical Circuit Modeling and Optimization. 5th International Workshop
on Statistical Metrology, pp. 56-63, ISBN 0-7803-5896-1, Honolulu, HI, USA,
Jun 2000.
Hanson, D.; Goosens, R.; Redford, M.; McGinty, J.; Kibarian, J. & Michaels, K. (1996).
Analysis of Mixed-Signal Manufacturability with Statistical Technology CAD
(TCAD). IEEE Trans. on Semiconductor Manufacturing, Vol. 9, No. 4, (Nov 1996) pp.
478-488, ISSN 0894-6507.
Lakshmikumar, K.; Hadaway, R. & Copeland, M. (1986). Characterization and Modeling of
Mismatch in MOS Transistors for Precision Analog Design. IEEE Journal of Solid-
State Circuits, Vol. 21, No. 6, (Dec 1986) pp. 1057-1066, ISSN 0018-9200.
Lewyn, L.; Ytterdal, T.; Wulff, C. & Martin, K. (2010). Analog Circuit Design in Nanoscale
CMOS Technologies. Proceedings of the IEEE, Vol. 97, No. 10, (Oct 2010) pp. 1687-
1714, ISSN 0018-9219.
Lu, N, Watts, J. & Springer, S. (2009). Elements of Statistical SPICE Models.
NSTI-Nanotech 2009, Vol 3, pp. 616-619, ISBN 978-1-4398-1784-1, Houston, TX, USA,
May 2009.
Kinget, P. (2005). Device Mismatch and Tradeoffs in the Design of Analog Circuits. IEEE J. of
Solid-State Circuits, Vol. 40, No. 6, (Jun 2005) pp. 1212-1224, ISSN 0018-9200.
McAndrew, C.; Bates, J.; Ida, R. & Drennan, P. (1997). Efficient Statistical Modeling, Why β is
More Than Ic/Ib. Bipolar/BiCMOS Circuits and Technology Meeting (BCTM’97), pp.
28-31, ISBN 0-7803-3916-9, Minneapolis, MN, USA, Sep 1997.
McAndrew, C.; Stevanović, I.; Li, X.; & Gildenblat, G. (2010). Extensions to Backward
Propagation of Variance for Statistical Modeling. IEEE Design & Test of Computers,
Vol. 27, No. 2, (Mar/Apr 2010) pp. 36-43, ISSN 0740-7475.
Mutlu, A. & Rahman, M. (2005). Statistical Methods for the Estimation of Process Variation
Effects on Circuit Operation. IEEE Trans. on Electronics Packaging Manufacturing,
Vol. 28, No. 4, (Oct 2005) pp. 364-375, ISSN 1521-334X.
Nardi, A.; Neviani, A.; Zanoni, E.; Quarantelli, M. & Guardiani, C. (1999). Impact of
Unrealistic Worst Case Modeling on the Performance of VLSI Circuits in Deep
Submicron CMOS Technologies. IEEE Transactions on Semiconductor Manufacturing,
Vol. 12, No. 4, (Nov 1999) pp. 396-402, ISSN 0894-6507.
Pelgrom, M.; Duimaijer, A. & Welbers, A. (1989). Matching Properties of MOS Transistors.
IEEE Journal of Solid-State Circuits, Vol. 24, No. 5, (Oct 1989) pp. 1433-1439, ISSN
0018-9200.
Pelgrom, M.; Tuinhout, H. & Vertregt, M. (1998). Transistor Matching in Analog CMOS
Applications. International Electron Devices Meeting (IEDM’98), pp. 915-918, ISBN 0-
7803-4774-9, San Francisco, CA, USA, Dec 1998.
Pineda de Gyvez, J. & Rodríguez-Montañés, R. (2003). Threshold Voltage Mismatch (ΔVT)
Fault Modeling. Proceedings of the 21st VLSI Test Symposium (VTS’03), pp. 145-150,
ISBN 0-7695-1924-5, Napa Valley, CA, USA, May 2003.
Potts, D. & Luk, T. (2005). Extraction and Implementation of Effective Mismatch Models.
DesignCon East 2005, Worcester, MA, USA, Sep 2005.
Rappitsch, G.; Seebacher, E.; Kocher, M. & Stadlober, E. (2004). SPICE Modeling of Process
Variation Using Location Depth Corner Models. IEEE Transactions on Semiconductor
Manufacturing, Vol. 17, No. 2, (May 2004) pp. 201-213, ISSN 0894-6507.
Sengupta, M.; Saxena, S.; Daldoss, L.; Kramer, G.; Minehane, S. & Cheng, J. (2005).
Application Specific Worst Case Corners using Response Surfaces and Statistical
Models. IEEE Transactions on Computer-Aided Design of Integrated Circuits and
Systems, Vol. 24, No. 9, (Sep 2005) pp. 1373-1380, ISSN 0278-0070.
Telang, N & Higman, J. (2001). Statistical Modeling Techniques: FPV vs. BPV. Proceedings of
the 2001 International Conference on Microelectronic Test Structures (ICMTS 2001), pp.
71-75, ISBN 0-7803-6511-9, Kobe, Japan, Mar 2001.
Tuinhout, H. (2002). Impact of Parametric Mismatch and Fluctuations on Performance
and Yield of Deep-submicron CMOS Technologies. European Solid-State Device
Research Conference (ESSDERC’02), pp. 95-102, ISBN 88-900847-8-2, Bologna, Italy,
Sep 2002.
Tuinhout, H.; Wils, N. & Andricciola, P. (2010). Parametric Mismatch Characterization for
Mixed-Signal Technologies. IEEE Journal of Solid-State Circuits, Vol. 45, No. 9, (Sep
2010) pp. 1687-1696, ISSN 0018-9200.
Watts, J.; Su, K. & Basel, M. (2006). Netlisting and Modeling Well-Proximity Effects.
IEEE Trans. on Electron Devices, Vol. 53, No. 9, (Sep 2006) pp. 2179-2186, ISSN 0018-
9383.
Wils, N. ; Tuinhout, H. & Meijer, M. (2010). Influence of Metal Coverage on Transistor
Mismatch and Variability in Copper Damascene Based CMOS Technologies, 2010
IEEE International Conference on Microelectronic Test Structures (ICMTS’10), pp. 182-
187, ISBN 978-1-4244-6915-4, Hiroshima, Japan, Mar 2010.
Yang, L. ; Cui, M. ; Ma, J. ; He, J. ; Wang, W. & Wong, W. (2008). Advanced Spice Modeling
for 65nm CMOS Technology, 9th International Conference on Solid-State and Integrated-
Circuit Technology (ICSICT’08), pp. 436-439, ISBN 978-1-4244-2185-5, Beijing, China,
Oct 2008.
Zhang, H. ; Chen, T. ; Ting, M. & Li, X. (2009). Efficient Design-Specific Worst-Case

Corner Extraction for Integrated Circuits, 46th ACM/IEEE Design Automation
Conference (DAC ’09), pp. 386-389, ISBN 978-1-6055-8497-3, San Francisco, CA, USA,
Jul 2009.
11
Advanced Statistical Methodologies for

Tolerance Analysis in Analog Circuit Design
Bruno Apolloni1 , Simone Bassis2 , Angelo Ciccazzo3 , Angelo Marotta4 ,
Salvatore Rinaudo5 and Orazio Muscato6
1,2 Department of Computer Science, University of Milan,
Via Comelico 39/41, 20135 Milano
3,4,5 STMicroelectronics, Stradale Primo Sole 50, 95121 Catania
6 Department of Mathematics and Informatics, University of Catania,
Viale Andrea Doria 6, 95125 Catania

Italy
1. Introduction
The influence of process variations is becoming extremely critical for nano technology nodes
(90nm and below), due to geometric tolerances and manufacturing non-idealities (such
as edge or surface roughness, or the fluctuation in the number of doping atoms). The
most worrying of all is the statistical variability introduced by discreteness of charge and
granularity matter in the transistors approaching molecular and atomic scale dimensions.
The main sources of statistical variability are the random distributions of discrete dopants
and charged defects, the line edge roughness of the photo resist and the granularity of the
materials (Bernstein et al., 2006; Boning & Nassif, 1999). As a result, production yields and
circuit figures of merit (such as performance, power, and reliability) have became extremely
sensitive to incontrollable statistical process variations (PV). The main sources of variations
are: environmental factors, whose transient arises during the operation of a circuit (e.g. power
supply or temperature variations), and physical factors due to the manufacturing process,
which result in a (permanent or aging) variation of the device structure and interconnections.
The latter reflect into random (possibly spatial) drifts of the design parameter.
Although already considered in the past, the increasing impact of these drawbacks constitutes
a completely new challenge. While process engineers have traditionally coped with die-to-die
fluctuations, the today within-die variations are more subtle since they imply that different
areas of the same die exhibit different values of the various parameters. With a further
shrinking of process technology, the on-chip variation is getting worse for each technology
node, thus having a direct impact on the design flows. By contrast, the latter conventionally
rely on deterministic models.
At a front end, parameter variability has a significant impact both on the power dissipation
and performance of a circuit, with a consequent yield decrease and remarkable cost
implications. Indeed, to maintain production efficiency we must raise up control costs and
cycle time, a drawback which dramatically increases with the process complexity. To contrast
it, the following two joint tasks become essential:
• to characterize statistically integrated circuits (IC) manufacturing process fluctuations;

• to predict reliably circuit performance spreads at the design stage.
Failure in the former can result in a low parametric yield, since ICs do not meet design
specifications. On the one hand, a successful statistical characterization promotes a robust
manufacturability reflecting in a high fabrication yield (i.e. a high proportion of produced
circuits which function properly). On the other hand, it requires managing complex design
flows in the design-verification-production life-cycle of ICs.
Summing up, random and systematic defects as well as parametric process variations have
a big influence on the design/production cycle, causing frequent re-spinning of the whole
development and manufacturing chain. This leads to high costs of multiple manufacturing
runs and entails extremely high risks of missing a given market window. One way to
overcome these drawbacks is to implement the DFM/DFY paradigm (Bühler et al., 2006)
where Design for Manufacturability (DFM) mates Design for Yield (DFY) to form a synergistic
manufacturing chain to be dealt with in terms of: i) relationships between the statistical circuit
parameters matching the production constraints, and ii) performance indicators ensuring
correctly functioning dies. This chapter introduces a pair of procedures aimed at identifying
these parameters exactly with the goal of maximizing performance indicators defined as a
function of the parameters’ confidence region.
The material is organized as follows. In Section 2 we discuss the statistical aspect of IC design
and introduce the lead formalism. In Section 3 we focus on the statistical modeling task
with special regard to two advanced solution methods. Hence we introduce benchmarks in
Section 4 to both provide a comparison between the performances of the above methods and
show their behaviors w.r.t. state-of-the-art procedures introduced by researchers in the last
years. Concluding remarks are drawn in the final section.
2. Statistics in IC design
Electronic devices are replicated multiple times on a wafer and different wafers are produced,
but each device cannot be produced in the same way in terms of electrical performance. Main
factors that make the fabrication result uncertain are: the imperfections characterizing the
masks and tolerances in their positionings, various changing effects of ion plant temperature
during production, tolerances in size, etc. Generally fluctuations’ processes produce fluctuations
in electrical performance. Consequently, an essential tool for electronic circuit designing is
represented by the statistical model which formally relates the former to the latter.
A circuit is classified as acceptable in performances if all specifications on its electrical behavior
are met. In the context of the microelectronics industry, the term yield phrases the ratio
between the number of acceptable chips and total number of produced chips:
# accetable chips
yield = (1)
# manufactured chips
The acceptability of each chip is decreed by checking that the questioned electrical parameters
individually fall into tolerance intervals. In addition, each wafer contains several sites with
special test structures that enable further performance measurements in order to verify the
manufacturing process. All the measurements are collected in a database which statistically
characterizes the electrical behavior of the devices.
As for the final product we may classify the integrated circuits into:
• acceptable chips, which satisfy all performance requests,
Advanced Statistical Methodologies for Tolerance Analysis in Analog Circuit Design 229
• functional failures, when malfunctions affect chips,

• parametric failures, when chips fail to reach performances.
Coming to their manufacturing, we are used to distinguish three categories of failures that we
synthesize through:
2.1. random yield (sometimes called statistical yield), concerning the random effects occurring
during the manufacturing process, such as catastrophic faults in the form of open or short
circuits. These faults may be a consequence of small particles in the atmosphere landing
on the chip surface, no matter how clean is the wafer manufacturing environment. An
example of a random component is that of threshold voltage variability due to random
dopant fluctuations (Stolk et al., 1988);
2.2. systematic yield (including printability issues), related to systematic manufacturability issues
deriving from combinations and interactions of events that can be identified and addressed
in a systematic way. An example of these events is the variation in wire thickness
with layout density due to Chemical Mechanical Polishing/Planarization (CMP) (Chang
et al., 1995). The distinction from the previous yield is important because the impact of
systematic variability can be removed by adapting the design appropriately, while random
variability will inevitably impact design margins in a negative manner;
2.3. parametric yield (including variability issues), dealing with the performance drifts induced
by changes in the parameter setting – for instance, lower drive capabilities, increased
leakage current and greater power consumption, increased resistance and capacitance (RC)
time constants, and slower chips deriving from corruptions of the transistor channels.
From a complementary perspective, the unacceptable performance causes for a circuit may be
split into two categories of disturbances:
• local, caused by disruption of the crystalline structure of silicon, which typically determines
the malfunctioning of a single chip in a silicon wafer;
• global, caused by inaccuracies during the production processes such as misalignment of
masks, changes in temperature, changes in doses of implant. Unlike the local disturbance,
the global one involves all chips in a wafer at different degrees and in different regions.
The effect of this disturbance is usually the failure in the achievement of requested
performances, in terms of working frequency decrease, increased power consumption, etc.
Both induce troubles on physical phenomena, such as electromagnetic coupling between
elements, dissipation, dispersion, and the like.
The obvious goal of the microelectronics factory is to maximize the yield as defined in (1). This
translates, from an operational perspective, into a design target of properly sizing the circuit
parameters, and a production target of controlling their realization. Actually both targets are
very demanding since the involved parameters π are of two kinds:
• controllable, when they allow changes in the manufacturing phase, such as the oxidation
times,
• non controllable, in case they depend on physical parameters which cannot be changed
during the design procedure, like the oxide growth coefficient.
Moreover, in any case the relationships between π and the parameters φ characterizing
the circuit performances are very complex and difficult to invert. This induces researchers
to model both classes of parameters as vectors of random variables, respectively Π and
Φ 1 . The corresponding problem of yield maximization reverts into a functional dependency

among the problem variables. Namely, let Φ = (Φ1 , Φ2 , . . . , Φ t ) be the vector of the
performances determined by the parameter vector Π = (Π1 , Π2 , . . . , Πn ), and denote with
DΦ the acceptability region of a given chip. For instance, in the common case where each
performance is checked singularly in a given range, i.e.:
φkl ≤ Φ k ≤ φku k = 1, . . . , t (2)
DΦ reads:
DΦ = Φ| φkl ≤ Φ k ≤ φku k = 1, . . . , t (3)
The yield goal is the maximization of the probability P that a manufactured circuit has an
acceptable performance, i.e.

P = P [ Φ ∈ DΦ ] = f Φ (φ)dφ (4)
DΦ
where f Φ is the joint probability density of the performance Φ.

To solve this problem we need to know f Φ and manage its dependence on Π. Namely,
methodologies for maximizing the yield must incorporate tools that determine the region
of acceptability, manipulate joint probabilities, evaluate multidimensional integrals, solve
optimization problems. Those instruments that use explicit information about the joint
probability and calculate the yield multidimensional integral (4) during the maximization
process are called direct methods. The term indirect is therefore reserved for those methods
that do not use this information directly. In the next section we will introduce two of these
methods which look to be very promising when applied to real world benchmarks.
3. Statistical modeling
As mentioned in the introduction, a main way for maximizing yield passes through mating
Design for Manufacturability with Design for Yield (DFM/DFY paradigm) along the entire
manufacturing chain. Here we focus on model parameters at an intermediate location
in this chain, representing a target of the production process and the root of the circuit
performance. Their identification in correspondence to a performances’ sample measured
on produced circuits allows the designer to get a clear picture of how the latter react to the
model parameters in the actual production process and, consequently, to grasp a guess on
their variation impact. Typical model and performance parameters are described in Table 1 in
Section 4.
In a greater detail, the first requirement for planning circuits is the availability of a model
relating input/output vectors of the function implemented by the circuit. As aforementioned,
its achievement is usually split into two phases directed towards the search of a couple of
analytic relations: the former between model parameters and circuit performances, and the
latter, tied to the process engigneers’ experience, linking both design and phisical circuit
parameters as they could be obtained during production. Given a wafer, different repeated
measurements are effected on dies in a same circuit family. As usual, the final aim is the model
1 By default, capital letters (such as X, Y) will denote random variables and small letters (x, y) their
corresponding realizations; bold versions (X, Y , x, y) of the above symbols apply to vectors of the
objects represented by them. The sets the realizations belong to will be denoted by capital gothic
symbols (X, Y).
identification, in terms of designating the input (respectively output) parameter values of the
aforementioned analytical relation. In some way, their identification hints at synthesizing
the overall aspects of the manufacturing process not only to use them satisfactory during
development yet to improve oncoming planning and design phases, rather than directly
weigh on the production.
For this purpose there are three different perspectives: synthesize simulated data, optimize
a simulator, and statistically identify its optimal parameters. All three perspectives share the
following common goals: ensure adequate manufacturing yield, reduce the production cost,
predict design fails and product defects, and meet zero defects specification. We formalize
the modeling problem in terms of a mapping g from a random vector X = ( X1 , . . . , Xn ),
describing what is commonly denoted as model parameters 2 , to a random vector Y =
(Y1 , . . . , Yt ), representing a meaningful subset of the performances Φ. The statistical features
of X, such as mean, variance, correlation, etc., constitute its parameter vector θX , henceforth
considered to be the statistical parameter of the input variable X. Namely, Y = g (X ) =
( g1 (X ), . . . , gt (X )), and we look for a vector θY that characterizes a performance population
where P(Y ∈ D Y ) = α, having denoted with D Y the α-tolerance region, i.e. the domain
spanned by the measured performances, and with α a satisfactory probability value. In turn,
D Y is the statistic we draw from a sample sy of the performances we actually measured
on correctly working dies. Its simplest computation leads to a rectangular shape, as in (3),
where we independently fix ranges on the singular performances. A more sophisticated
instance is represented by the convex hull of the jointly observed performances in the overall
Y space (Liu et al., 1999). At a preliminary stage, we often appreciate the suitability of θY by
comparing first and second order moments of a performances’ population generated through
the currently identified parameters with those computed on sy .
As a first requisite, we need a comfortable function relating the Y distribution to θX .
The most common tool for modeling an analog circuit is represented by the Spice
simulator (Kundert, 1998). It consists of a program which, having in input a textual
description of the circuit elements (transistors, resistors, capacitors, etc.) and their
connections, translates this description into nonlinear differential equations to be solved
using implicit integration methods, Newton’s method and sparse matrix techniques. A
general drawback of Spice – and circuit simulators in general – is the complexity of the
transfer function it implements to relate physical parameters to performances which hampers
intensive exploration of the performance landscape in search of optimal parameters. The
methods we propose in this section are mainly aimed at overtaking the difficulty of inverting
this kind of functions, hence achieving a feasible solution to the problem: find a θX
corresponding to the wanted θY .
3.1 Monte Carlo based statistical modeling

The lead idea of the former method we present is that the model parameters are the
output of an optimization process aimed at satisfying some performance requirements. The
optimization is carried out by wisely exploring the research space through a Monte Carlo
(MC) method (Rubinstein & Kroese, 2007). As stated before, the proposed method uses the
experimental statistics both as a target to be satisfied and, above all, as a selectivity factor
for device model. In particular, a device model will be accepted only if it is characterized by
parameters’ values that allow to obtain, through electrical simulations, some performances
which are included in the tolerance region.
2 We speak of X as controllable model parameters to be defined as a suitable subset of Π.
Performance Space Model Parameter Space

Y = (Y1 , . . . , Yt ) X = ( X1 , . . . , Xn )
Statistical Modeling
ẙ2 x̊2
central value
ẙ1 x̊1
Fig. 1. Proposed flow: from the experimental statistics we determine a statistical Spice model
for the device.
The aim of the proposed flow is the following: on the basis of the information which
constitutes the experimental statistics, we want to map the space Y of the performances (such
as gain and bandwidth) to the space X of circuit parameters (such as Spice parameters or
circuit components values), as outlined in Fig. 1. Variations in the fabrication process cause
random fluctuations in Y space, which in turn cause X to fluctuate (Koskinen & Cheung,
1993). In other words, we want to extract a Spice model whose parameters are random
variables, each one characterized by a given probability distribution function. For instance,
in agreement with the Central Limit Theorem (Rohatgi, 1976), we may work under usual
Gaussianity assumptions. In this case, for the model parameters which have to be statistically
described, it is necessary and sufficient to identify the mean values, standard deviations and
correlation coefficients. In general, the flow of statistical modeling is based on several MC
simulation steps (strictly related to bootstrap analysis (Efron & Tibshirani, 1993)), in order to
estimate unknown features for each statistical model parameter. The method will proceed by
executing iteratively the following steps, in the same way as in a multiobjective optimization
algorithm, where the targets to be identified are the optimal parameters θX of the model.
In the following procedure, general steps (described in roman font) will be specialized to the
specific scenario (in italics) used to perform simulations in Section 4.
Step 1. Assume a typical (nominal) device model m0 is available, whose model parameters’
means are described by the vector ν̊X (central values). Let D Y be the corresponding
typical tolerance region estimated on Y observations sy . Choose an initial guess of X
joint distribution function on the basis of moments estimated on given X observations sx .
Let M denote the companion device statistical model, and set k = 0.
In the specific case of hyper-rectangle tolerance regions defined as in (3), let ν̊Yj ± 3σ̊Yj , j = 1, . . . , t
denote the two extremes delimiting each admissable performance interval. Moreover, since model
parameters X of M follows a multivariate Gaussian distribution, assume (in the first iteration)
a null cross-correlation between { X1 , . . . , Xn }, hence θ Xi = {νXi , σXi }, i = 1, . . . , n, where by
default νXi = ν̊Xi , i.e. the same mean as the nominal model is chosen as initial value, and σXi is
assigned a relatively high value, for instance set equal to the double of the mean value.
Step 2. At the generic iteration k, an m-sized 3 sample sMk = {xr }, r = 1 . . . , m will be
generated according to the actual X distribution.
3
A generally accepted rule to assign m is: for an expected probability level 10−ξ , the sample size m
should be set in the range [10ξ +2 , 10ξ +3 ] (Johnson, 1994).
In particular, when Xi are nomore independent, the discrete Karhunen-Loeve expansion (Johnson,
1994) is adopted for sampling, starting from the actual covariance matrix.
Step 3. For each model parameter xr in sMk , the target performances yr will be calculated
through Spice circuit simulations.
Step 4. Only those model parameters in sMk reproducing performances lying within the
chosen tolerance region D Y will be accepted. On the basis of this criterion a subsample
sM of s having size m ≤ m will be selected.
k
Mk
In particular, by keeping a fraction 1 − δ, say 0.99, of those models having all performance values
included in D Y , we are guaranteeing a confidence region of level δ under i.i.d. Gaussianity
assumptions.

Step 5. On the basis of the subsample sM k , a new model Mk will be computed through
standard statistical techniques.
For each model parameter Xi , i = 1, . . . , n, the n standard deviations could be computed on
the sample sM through Maximum Likelihood Estimators (MLE) (Mood et al., 1974), Spearman
Rank-Order correlation coefficient (Lehmann, 2006; Press et al., 1993) may be used to estimate
cross-correlation, while, according to circuit designers’ report, the n means will be kept equal to the
nominal ν̊Xi , i = 1, . . . , n.
Step 6. If the number m of selected model parameters which have generated M is sufficiently
high (for instance they constitute a fraction 1 − δ, let’s say 0.99, of the m instances, then the
algorithm stops returning the statistical model M . Otherwise, set k = k + 1 and goto Step
2.
The iterative procedure described above is based on Attractive Fixed Point method (Allgower
& Georg, 1990), where the optimal value of those features to be estimated represents the
fixed point of the algorithm. When the number of the components significantly increases, the
convergence of the algorithm may become weak. To manage this issue, a two-step procedure
is introduced where the former phase is aimed at computing moments involving single
features Xi while maintaining constant their cross-correlation; the latter is directed toward the
estimation of the cross-correlation between them. The overall procedure is analogous to the
previous one, with the exception that cross-correlation terms will be kept fixed until Step 5 has
been executed. Subsequently, a further optimization process will be performed to determine
the cross-correlation coefficients, for instance using the Direct method as described in Jones
et al. (1993). The stop criterion in Step 6 is further strengthen, prolonging the running of the
procedure until the difference between cross-correlation vectors obtained at two subsequent
iterations will drop below a given threshold.
3.2 Reverse spice based statistical modeling

A second way we propose to bypass the complexity handicap of Spice functions passes
through a principled philosophy of considering the region DX where we expect to set the
model parameters as an aggregate of fuzzy sets in various respects (Apolloni et al., 2008).
First of all we locally interpolate the Spice function g through a polynomial, hence a mixture
of monomials that we associate to the single fuzzy sets. Many studies show this interpolation
to be feasible, even in the restricted form of using posynomials, i.e. linear combination of
monomials through only positive coefficients (Eeckelaert et al., 2004). The granular construct
we formalize is the following.
Given a Spice function g mapping from x to y (the generic component of the

performance vector y), we assume the domain DX ⊆ R n into which x ranges to be
the support of c fuzzy sets { A1 , . . . , Ac }, each pivoting around a monomial mk . We
consider this monomial to be a local interpolator that fits g well in a surrounding of
the Ak centroid. In synthesis, we have g(x) ∑ck=1 μ k (x)mk (x), where μ k (x) is the
membership degree of x to Ak , whose value is in turn computed as a function of the
quadratic shift ( g(x) − mk (x))2 .
On the one hand we have one fuzzy partition of DX for each component of y. On the other
hand, we implement the construct with many simplifications, in order to meet specific goals.
Namely:
• since we look for a polynomial interpolation of g, we move from point membership
functions to sets, to a monomial membership function to g, so that g(x) ∑ck=1 μ k mk (x).
In turn, μ k is a sui generis membership degree, since it may assume also negative values;
• since for interpolation purposes we do not need μ k (x), we identify the centroids directly
with a hard clustering method based on the same quadratic shift.
α kj
Denoting mk (x) = β k ∏nj=1 x j , if we work in logarithmic scales, the shifts we consider for
the single (say the i-th) component of y are the distances between zr = (log xr , log yr ) and the
hyperplane hk (z ) = wk · z + bk = 0, with wk = {αk1 , . . . , αkn } and bk = log β k , constituting
the centroid of Ak in an adaptive metric. Indeed, both wk and bk are learnt by the clustering
algorithm aimed at minimizing the sum of the distances of the zr s from the hyperplanes
associated to the clusters they are assigned to.
With the clustering procedure we essentially learn the exponents αkj through which the
x components intervene in the various monomials, whereas the β k s remain ancillary
parameters. Indeed, to get the polynomial approximation of g(x) we compute the mentioned
sui generis memberships through a simple quadratic fitting, i.e. by solving w.r.t. the vector
2
μ = {μ1 , . . . , μ c } the quadratic optimization problem: μ = arg minμ ∑r =1 ( g(xr ) − yr )) ,
m
where xrj denotes the j-th component of the r-th element of the training set sx , yrj its
approximation, with
c c n
α jki
yj = ∑ m jk (x) = ∑ μ jk ∏ xi (5)
k =1 k =1 i =1
where the index r has been hidden for notational simplicity, and μ k s override β k s.
3.2.1 A suited interpretation of the moment method

An early solution of the inverse problem:
Which statistical features of X ensure a good coverage (in terms of α-tolerance regions) of
the Y domain spanned by the performances measured on a sample of produced dies?
relies on the first and second moments of the target distribution, which are estimated on
the basis of a sample sy of sole Y collected from the production lines as representatives of
properly functioning circuits. Our goal is to identify the statistical parameters θX of X that
produce through (5) a Y population best approximating the above first and second order
moments. X is assumed to be a multidimensional Gaussian variable, so that we identify
it completely through the mean vector νX and the covariance matrix ΣX which we do not
constrain in principle to be diagonal (Eshbaugh, 1992). The analogous νY and ΣY are a
function of the former through (5). Although they could not identify the Y distribution in full,
we are conventionally satisfied when these functions get numerically close to the estimates
of the parameters they compute (directly obtained from the observed performance sample).
Denoting with νXj , σXj , σXj,k and ρ Xj,k , respectively, the mean and standard deviation of X j and
the covariance/correlation between X j and Xk , the master equations of our method are the
following:
1.
c
νYi = ∑ αikj νM ik
(6)
k =1
where Mik on the right is a short notation of mik (X ), and νMik denotes its mean.
2. Thanks to the approximations
νΞ log νX , σΞ σX /νX , ρΞi,j ρ Xi,j (7)
with Ξ = log X, coming from the Taylor expansion of respectively Ξ, (Ξ − νΞ )2 and (Ξi −
νΞi )(Ξ j − νΞ j ) around (νXi , νXj ) disregarding others than the second terms, the rewriting
of ΣY reads
c c
σY2i = ∑ σM
2
ik
+2 ∑ σMik,ir (8)
k =1 k,r =1
k <r
c
σYi,j = ∑ σMik,jr (9)
k,r =1
with
⎛ ⎞
⎜ n σX
2 n σXj σXr ⎟
2
σM ν2Mik ⎜ ∑ 2
+2 ∑ ρ Xj,r aikj aikr ⎟
j
⎝ a (10)
ik
j =1
ikj 2
νXj j,r =1
νXj μ Xr ⎠
j <r
n σXj σXw
σMik,ir νMik νMir ∑ aikj airw ρ Xj,w
νXj νXw
(11)
j,w =1
We numerically solve (6) and (8-9) in νX and ΣX when the left members coincide with the
target values of νY and ΣY , respectively, and νMik is approximated with its sample estimate
computed on samples artificially generated with the current values of the parameters. Solving
equations means minimizing the differences between left and right members, so that the
crucial point is the optimization method employed.The building blocks are the following.
The steepest descent strategy. Using the Taylor series expansion limited to second
order (Mood et al., 1974), we obtain an approximate expression of the gradient components of
νY w.r.t. νX through
2
∂νYi c
1 σX
∑ αikj + 3 νMik
j
(12)
∂νXj k =1
νX j νXj
Thus we may easily look for the incremental descent on the quadratic error surface accounting
for the difference between computed and observed means. Expression (12) confirms the scarce
sensitivity of the unbiased mean νX , and its gradient as well, to the second moments, so
that we may expect to obtain an early approximation of the mean vector to be subsequently
refined. While analogous to the previous task, the identification of X variances and
correlations owns one additional benefit and one additional drawback. The former derives
from the fact that we may start with a, possibly well accurate, estimate of the means. The
latter descends from the high interrelations among the target parameters which render the
exploration of the quadratic error landscape troublesome and very lengthy.
Identification of second order moments. An alternative strategy for X second moment
identification is represented by the evolutionary computation. Given the mentioned
computational length of the gradient descent procedures, algorithms of this family become
competitive on our target. Namely, we used Differential Evolution (Price et al., 2005), with
specific bounds on the correlation values to avoid degenerate solutions.
A brute force numerical variant. We may move to a still more rudimentary strategy
to get rid of the loose approximations introduced in (6) to (12). Thus we: i) avoid
computing approximate analytical derivatives, by substituting them with direct numerical
computations (Duch & Kordos, 2003), and ii) adopt the strategy of exploring one component
at a time of the questioned parameter vector, rather than a combination of them all, until
the error descent stops. Spanning numerically one direction at a time allows us to ask the
software to directly identify the minimum along this direction. The further benefit of this task
is that the function we want to minimize is analytic, so that the search for the minimum along
one single direction is a very easy task for typical optimizers, such as the naive Nelder-Mead
simplex method (Nelder & Mean, 1965) implemented in Mathematica (Wolfram Research Inc.,
2008). We structured the method in a cyclic way, plus stopping criterion based on the amount
of parameter variation. Each cycle is composed of: i) an iterative algorithm which circularly
visits each component direction minimizing the error in the means’ identification, until no
improvement may be achieved over a given threshold, and ii) a fitting polynomial refresh on
the basis of a Spice sample in the neighborhood of the current mean vector. We conclude the
routine with a last assessment of the parameters that we pursue by running jointly on all them
a local descent method such as Quasi-Newton procedure in one of its many variants (Nocedal
& Wright, 1999).
3.2.2 Fine tuning via reverse mapping

Once a good fitting has been realized in the questioned part of the Spice mapping, we
may solve the identification problem in a more direct way by first inverting the polynomial
mapping to obtain the X sample at the root of the observed Y sample, and then estimating
θX directly from the sample defined in the DX domain. The inversion is almost immediate
if it is univocal, i.e., apart from controllable pathologies, when X and Y have the same
number of components. Otherwise the problem is either overconstrained (number n of X
components less than t, dimensionality of Y components) or underconstrained (opposite
relation between component numbers). The first case is avoided by simply discarding
exceeding Y components, possibly retaining the ones that improve the final accuracy and
avoid numeric instability. The latter calls for a reduction in the number of questioned X
components. Since X follows a multivariate Gaussian distribution law, by assumption, we
may substitute some components with their conditional values, given the others.
4. Numerical experiments
The procedures we propose derive from a wise implementation of the Monte Carlo methods,
as for the former, and a skillful implementation of granular computing ideas (Apolloni et al.,
model parameter performance parameter

device
label description label description
U0 Mobility at nominal temperature
A0 Bulk charge effect coefficient GM conductance
VTH0 Threshold voltage at VBS = 0 for large L IDSAT source drain current
pMOS
K1 First order body effect coefficient VTH25−25 saturation voltage
B01 Bulk charge effect coefficient for channel lenght VTH25−08 saturation voltage
B11 Bulk charge effect coefficient for channel width
U0 Mobility at nominal temperature GM conductance
VSAT Saturation voltage IDSAT source drain current
nMOS
VTH0 Threshold voltage at VBS = 0 for large L VTH25−25 saturation voltage
K1 First order body effect coefficient VTH25−08 saturation voltage
Bf Ideal maximum foward Beta
HFE Current Gain
Re Emitter Resistance
NPN-DIB12 VA Early Voltage
Is Transport Saturation Current
Ic Collector Current
Vaf Forward Early Voltage
Table 1. Model parameters and performances of the identification problems.
2008), as for the latter, however without theoretical proof of efficiency. While no worse from
this perspective than the general literature in the field per se (McConaghy & Gielen, 2005),
it needs numerical proof of suitability. To this aim we basically work with three real world
benchmarks collected by manufacturers to stress the peculiarities of the methods. Namely,
the benchmarks refer to:
1. A unipolar pMOS device realized in Hcmos4TZ technology.
2. A unipolar nMOS device differentiating from the former for the sign (negative here,
positive there) of the charge of the majority mobile charge carriers. Spice model and
technology are the same, and performance parameters as well. However, the domain
spanned by the model parameters is quite different, as will be discussed shortly.
3. A bipolar NPN circuit realized in DIB12 technology. DIB technology achieves the full
dielectric isolation of devices using SOI substrates by the integration of the dielectric trench
that comes into contact with the buried oxide layer.
The related model parameter took into consideration and measured performances are
reported in Table 1.
We have different kinds of samples for the various benchmarks as for both the sample
size which ranges from 14, 000 (pMOS and nMOS) to 300 (NPN-DIB12) and the measures
they report: joint measures of 4 performance parameters in the former two cases, partially
independent measures of 3 performance parameters in the latter, where only HFE and VA are
jointly measured. Taking into account the model parameters, and recalling the meaning of t
and n in terms of number of performance and model parameters, respectively, the sensitivity
of the former parameters to the latter and the different difficulties of the identification tasks
lead us to face in principle one balanced problem with n = t = 4 (nMOS), and two unbalanced
ones with n = 6 and t = 4 (pMOS) and n = 4 and t = 3 (NPN-DIB12). In addition, only 4 of
the 6 second order moments are observed with the third benchmark.
4.1 Reverting the Spice model on the three benchmarks

With reference to Table 2, in column θX we report the parameters of the input multivariate
Gaussian distribution we identify in the aim of reproducing the θY of the Y population
observed through sy . Of the latter parameter, in the subsequent column θY /θ̂Y we compare
Advances in Analog Circuitsi
benchmark solution
dataset ( n, t ) m
θ /ˆ`
θ
1−δ/
X Y Y
benchmark μX σX ρX μY σY ρY 1 −δ
⎛
− 0.16582 ⎞
⎜ − 0.46312 ⎟ ⎛ ⎞
⎜ ⎟ 0.933746
⎜ − 0.41451 ⎟
⎜ ⎟ ⎜ 0.451486 ⎟
⎜ − 0.49665 ⎟ ⎛ ⎞ ⎛ ⎞ ⎜ ⎟
⎛ ⎞ ⎛ ⎞ ⎜ ⎟ − 0.835824 0.0118109 ⎜ − 0.287658 ⎟
⎜ − 0.35008 ⎟ ⎜ ⎟
233.424 3.63673 ⎜ ⎟ ⎜ − 0.838496 ⎟ ⎜ 0.0187507 ⎟ ⎜ − 0.282512 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ − 0.12573 ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜
0.28798
⎟ ⎜
0.01806
⎟ ⎜ ⎟ ⎜ − 0.971835 ⎟ ⎜ 0.0121665 ⎟ ⎜ − 0.389979 ⎟ 0.946713
⎜ ⎟ ⎜ ⎟ ⎜ − 0.47067 ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜
0.99185
⎟ ⎜
0.01083
⎟ ⎜ ⎟ ⎜ − 0.969196 ⎟ ⎜ 0.0164674 ⎟ ⎜ − 0.387441 ⎟ 0.9
pMOS (6, 4) 14, 000 ⎜ ⎟ ⎜ ⎟ ⎜ − 0.07056 ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜
0.45255
⎟ ⎜
0.03275
⎟ ⎜ ⎟ ⎜ 0.000973318 ⎟ ⎜ 0.000029378 ⎟ ⎜ − 0.254446 ⎟ 0.900398
⎝ 4.06626 × 10− 5 ⎠ ⎝ 4.48106 × 10− 6 ⎠ ⎜ − 0.39330 ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ 0.00097472 ⎟ ⎜ 0.000029348 ⎟ ⎜ − 0.0727698 ⎟ 0.8
⎜ 0.09484 ⎟ ⎝ ⎠ ⎝ ⎠ ⎜ ⎟
4.67824 × 10− 5 9.90006 × 10− 6 ⎜
⎜
⎟
− 0.16367 ⎟
0.00448103 0.000146626 ⎜
⎜
− 0.367477 ⎟
⎟
⎜ ⎟ 0.00447346 0.000130486 ⎜ − 0.174543 ⎟
⎜ 0.21068 ⎟ ⎝
⎜ ⎟ 0.900391 ⎠
⎜ 0.49711 ⎟
⎝ ⎠ 0.983658
0.22781
0.48312
⎛ ⎞
0.445093
⎜ 0.395429 ⎟
⎛ ⎞ ⎛ ⎞ ⎜ ⎟
0.552391 0.028568 ⎜ − 0.499279 ⎟
⎛ ⎜ ⎟
⎛ ⎞ ⎛ ⎞ − 0.765278 ⎞ ⎜ 0.550715 ⎟ ⎜ 0.0276768 ⎟ ⎜ − 0.432434 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
752.395 134.099 ⎜ − 0.467972 ⎟ ⎜ 0.66383 ⎟ ⎜ 0.0176982 ⎟ ⎜ − 0.637969 ⎟ 0.9008
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 152858.0 ⎟ ⎜ 9667.22 ⎟ ⎜ 0.756786 ⎟ ⎜ 0.664162 ⎟ ⎜ 0.0173677 ⎟ ⎜ − 0.640323 ⎟ 0.9
nMOS (4, 4) 14, 000 ⎝ 0.68184 ⎠ ⎝ 0.0186854 ⎠ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 0.306389 ⎟ ⎜ 0.00221691 ⎟ ⎜ 0.0000830626 ⎟ ⎜ − 0.298401 ⎟ 0.8304
⎝ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
0.521661 0.131933 − 0.786377 ⎠ ⎜ 0.00222077 ⎟ ⎜ 0.0000619134 ⎟ ⎜ − 0.271952 ⎟ 0.8
⎝ ⎠ ⎝ ⎠ ⎜ ⎟
− 0.468842 0.0100527 0.000355129 ⎜ − 0.375841 ⎟
⎜ ⎟
0.0100711 0.000280373 ⎜ − 0.354887 ⎟
⎝ ⎠
0.92015
0.950419
⎛ − 0.192107 ⎞ ⎛ 113.244 ⎞ ⎛ 6.82099
⎞
⎛ ⎞ ⎛ ⎞
138.302 8.3859 ⎜ 0.00139749 ⎟ ⎜ 113.242 ⎟ ⎜ ⎜ 6.95918 ⎟ 0.9054
⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ 4.96031 × 10− 6 ⎟
NPN-DIB12 (4, 3 ) ⎜ 0.67258 ⎟ ⎜ 0.263238 ⎟ ⎜ − 0.477207 ⎟⎟ ⎜ 0.0000654246 ⎟ ⎜ ⎟ − 0.490798 0.9
322 ⎝ 5.28102 × 10 − 18 ⎠ ⎝ 4.14306 × 10 − 19 ⎠ ⎜ − 0.980327 ⎟
⎜
⎜ ⎟ ⎜
⎜ 0.0000653275 ⎟ ⎜ × 10 − 6 ⎟ − 0.566678 0.8136
⎝ ⎠ ⎝ ⎠ ⎝
4.81021 ⎟
136.319 13.6538 0.167527 110.164 11.1459 ⎠ 0.8
− 0.0444712 110.238 11.2166
Table 2. Benchmarks used for testing the proposed procedure and analysis of the identification solution. Rows: benchmarks.
Columns: inferred model distribution parameters (indexed by X) and reconstructed performance parameters (indexed by Y ), plus
comparative levels of the tolerance regions (as a function of δ).
238
Y

0.03
Y

20.04

Y215

20.02

10

5

0.01

0.02

0.05

0.05

Y

1
Y

0.05

Y
1 0.1030

20

10

20 1

0.10

0.05

10 30

0.01

5

0.02

0.02

10

0.03

0.04

15
(a) pMOS (b) nMOS (c) npn-DIB
Fig. 2. Comparison between output data and reconstruction provided by Reverse Spice based
procedure for the devices listed in Table 2 when projected on the two principal components
of the target. Points: reconstructed population lying within (dark gray) and outside (light
gray) 0.90 tolerance region (black curves) identified by black points. Gray crosses: original
target output; black crosses: target output uniformly spread with noise terms.
the values computed on the basis of θX (referring to a reconstructed distribution – in

italics) with those computed through the maximum likelihood estimate from sy (referring
to the original distribution – in bold). As a further accuracy indicator, we will consider
tolerance regions obtained through convex hull peeling depth (Barnett, 1976) containing a
given percentage 1 − δ of the performance population. In the last column of Table 2, headed
by (1 − δ)/(1 − δ ), we appreciate the difference between planned tolerance rate (in bold),
as a function of the identified Y distribution, and ratio of sampled measures found in
these regions (in italics). We consider single values in the table cells since the results are
substantially insensitive to the random components affecting the procedure, such as algorithm
initialization. Rather, especially with difficult benchmarks, they may depend on the user
options during the run of the algorithm. Thus, what we report are the best results we obtain,
reckoning the overall trial time in the computational complexity consideration we will do later
on in this section.
For a graphical counterpart, in Fig. 2 we report the scatterplot of the original Y sample and an
analogous one generated through the reconstructed distribution, both projected on the plane
identified by the two principal components (Jolliffe, 1986) of the original distribution. We also
draw the intercept of this plane with a tolerance region containing 90% of the reconstructed
points (hence δ = 0.1).
An overview of these data looks very satisfactory, registering a relative shift between sample
and identified parameters that is always less than 0.17% as for the mean values, 45% for the
standard deviations and 25% for the correlation. The analogous shift between planned and
actual percentages of points inside the tolerance region is always less than 2%. We distinguish
between difficult and easy benchmarks, where the pMOS sample falls in the first category.
Indeed the same percentages referring to the remaining benchmarks decreases to 0.13%, 10%
and 9%.
Given the high computational costs of the Spice models, their approximation through cheaper
functions is the first step in many numerical procedures on microelectronic circuits. Within the
vast set of methods proposed by researchers on the matter (Ampazis & Perantonis, 2002a;b;
Daems et al., 2003; Friedman, 1991; Hatami et al., 2004; Hershenson et al., 2001; McConaghy
et al., 2009; Taher et al., 2005; Vancorenland et al., 2001) in Table 3 we report a numerical
comparison between two well reputed fitting methods and our proposed Reverse Spice
based algorithm (for short RS). The methods are Multivariate Adaptive Regression Splines
(MARS) (Friedman, 1991), i.e. piecewise polynomials, and Polynomial Neural Networks
θX θX

train test train test
⎛ 0.0000125623 ⎞ ⎛ 0.0000242739 ⎞ ⎛ 0.000228931 ⎞ ⎛ 0.000369871 ⎞

0.0000350975 0.0000759397 0.000751481 0.00131925
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
RS ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 0.0000151476 ⎟ ⎜ 0.0000211444 ⎟ ⎜ 0.000164105 ⎟ ⎜ 0.000159924 ⎟
⎜ 3.06034 × 10−10 ⎟ ⎜ 6.62265 × 10−10 ⎟ ⎜ 1.54286 × 10−8 ⎟ ⎜ 2.33858 × 10−8 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠
3.59774 × 10−9 1.10138 × 10−8 1.24052 × 10−7 2.92353 × 10−7
−
⎛ 8.68173 ∗ 10 6 ⎞ ⎛ 0.0000168024 ⎞ ⎛ 0.000124012 ⎞ ⎛ 0.0002805 ⎞
0.0000246876 0.0000528055 0.000401349 0.00100927
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
MARS ⎜ 0.0000100344 ⎟ ⎜ 0.0000143915 ⎟ ⎜ 0.0000946271 ⎟ ⎜ 0.000112503 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 2.80773 × 10−10 ⎟ ⎜ 5.92204 × 10−10 ⎟ ⎜ 5.3722 × 10−9 ⎟ ⎜ 6.07291 × 10−9 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠
4.66935 × 10−9 1.19291 × 10−8 6.47147 × 10−8 2.22601 × 10−7
0.0000602061 0.0000769737 0.000125976 0.000280898
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0.000230822 0.000293665 0.000409046 0.00101197
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
PNN ⎜ 0.0000100003 ⎟ ⎜ 0.0000142199 ⎟ ⎜ 0.0000948249 ⎟ ⎜ 0.000111354 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 2.7761 × 10−10 ⎟ ⎜ 5.70282 × 10−10 ⎟ ⎜ 4.14671 × 10−9 ⎟ ⎜ 7.14833 × 10−9 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠
2.38434 × 10−9 9.12621 × 10−9 2.84136 × 10−8 2.62591 × 10−7
Table 3. Performance comparison between fitting algorithms. Rows: algorithms; main

columns: benchmark parameterization; subcolumns: experimental environments (training
set, test set).
(PNN) (Elder IV & Brown, 2000). Namely, we consider the θX reported in Table 2 as the
result of the nMOS circuit identification. On the basis of these parameters and through Spice
functions, we draw a sample of 250 pairs (xr , yr ) that we used to feed both competitor
algorithms and our own. In detail we used VariReg software (Jekabsons, 2010a;b) to
implement both MARS and PNN. To ensure a fair comparison among the differente methods,
we: i) set equal to 6 the number of monomials in our algorithm and the maximum number
of basis functions in MARS, where we used a cubic interpolation, and ii) employ the default
configuration in PNN by setting the degree of single neurons polynomial equal to 2. Moreover,
in order to understand how the various algorithms scale with the fitting domain, we repeat
the procedure with a second set θX of parameters, where the original standard deviations
have been uniformly doubled. In the table we report the mean squared errors measured on a
test set of size 1000, whose values are both split on the four components of the performance
vector and resumed by their average. The comparison denotes similar accuracies with the
most concentrated sample – the actual operational domain of our polynomials – and a small
deterioration of our accuracy in the most dispersed sample, as a necessary price we have to
pay for the simplicity of our fitting function.
As for the whole procedure, we reckon overall running times of around half an hour. Though
not easily contrastable with computational costs of analogous tasks, this order of magnitude
results adequate for an intensive use of the procedure in a circuit design framework.
4.2 Stochastically optimizing the third benchmark model

The same NPN-DIB12 benchmark discussed in Section 4.1 was also used to run the two-step
MC procedure depicted in Section 3.1. In particular, estimation of the sole standard deviations
σXi s in the former phase alternates with cross-correlation coefficients’ in the latter, while the
means remain fixed to their nominal values νXi = ν̊Xi Namely, at each iteration a sample
sM = {xr }, r = 1 . . . , m = 5000 was generated, and the whole procedure was repeated 7
times, until over 99% of sample instances were included in the tolerance region. Fig. 3 shows
the number m of selected instances for each iteration of the algorithm.
100m/m
100
98
96
94
92
90
1 2 3 4 5 6 7
iter.
Fig. 3. Percentage of selected instances at each iteration of the two-step MC algorithm.
4.3 Comparing the proposed methods

In order to grasp insights on the comparative performances of the proposed methods, we
list their main features on the common NPN-DIB12 benchmark. Namely, in the first row of
Table 4 we report the reference value of the means and standard deviations of both X and Y
distributions. As for the first variable, we rely on the nominal values of the parameters for the
θX θY
μX σX μY σY
⎛ ⎞
135 ⎛ ⎞ ⎛ ⎞
⎜ ⎟ 113.242 6.9592
⎜ 0.8 ⎟ ⎜ ⎟ ⎜ ⎟
Reference ⎜ ⎟
⎜ 5.12 × 10−18 ⎟
− 5 −
⎝ 6.5328 × 10 ⎠ ⎝ 4.8102 × 10 ⎠6
⎝ ⎠
110.238 11.2166
138
⎛ ⎞ ⎛ ⎞
135 8.2375 ⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟ 110.5854 6.6418
⎜ ⎟ ⎜ 7.9064 × 10−2 ⎟ ⎜
MC ⎜ 0.8 ⎟ ⎜ ⎟ ⎝ 6.346 × 10−5 ⎟ ⎠
⎜ −
⎝ 4.691 × 10 ⎠6 ⎟
⎜ 5.12 × 10−18 ⎟ ⎜ 3.9744 × 10−19 ⎟
⎝ ⎠ ⎝ ⎠
110.039 7.507
138 9.4
⎛ ⎞ ⎛ ⎞
138.302 8.3859 ⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟ 113.244 6.821
⎜ ⎟ ⎜ ⎟
⎜ 0.6726 ⎟ ⎜ 0.2632 ⎟ ⎜ − ⎟ ⎜ − ⎟
⎟ ⎝ 6.5425 × 10 ⎠ ⎝ 4.9603 × 10 ⎠
RS 5 6
⎜ 5.281 × 10−18 ⎟ ⎜ 4.1431 × 10−19
⎝ ⎠ ⎝ ⎠
110.164 11.1459
136.319 13.6538
Table 4. Comparison between both model and performance moments re reference and
reconstructed frameworks.
means, leaving empty the cell concerning the standard deviations. As for the performances,
we just use the moment MLE estimate computed on the sample sy . In the remaining rows we
report the analogous values computed from a huge sample of the above variables artificially
generated through the statistical models we identify.
Both tables denote a slight comparative benefit of using the reverse modeling (row RS),
in terms of both a greater variance of the model parameters and a better similarity of
the reconstructed performance parameters with the estimated ones w.r.t. the analogous
parameters obtained with Monte Carlo method (row MC). The former feature reflects into
less severe constraints in the production process. The latter denotes some improvement in the
reconstruction of the performances’ distribution law, possibly deriving from both freeing the
νX from their nominal values and a massive use of the Spice function analytical forms.
5. Conclusions
A major challenge posed by new deep-submicron technologies is to design and verify
integrated circuits to obtain a high fabrication yield, i.e. a high proportion of produced
circuits that function properly. The classical approach implemented in commercial tools
for parameter extraction (IC-Cap by Agilent Technology (2010), and UTMOST by Silvaco
Engineered (2010)) requires a dedicated electrical characterization for a large number of
devices, in turn demanding for a very long time in terms both of experimental characterization
and parameter extraction.
Thus, a relevant goal with these procedures is to reduce the computational time to have
a statistical description of the device model. We fill it by using two non conventional
methods so as to get a speed-up factor greater than 10 w.r.t. standard procedures in literature.
The first method we propose is based on a Monte Carlo technique to estimate the (second
order) moments for several statistical model parameters, on the basis of characterizated data,
collected during the manufacturing process.
The second method exploits a granular construct. In spite of the methodology broadness the
attribute granular may evoke, we obtain a very accurate solution taking advantage from strict
exploitation of state-of-the-art theoretical results. Starting from the basic idea of considering
the Spice function as a mixture of fuzzy sets, we enriched its implementation with a series of
sophisticated methodologies for: i) identifying clusters based on proper metrics on functional
spaces, ii) descending, direction by direction, along the ravines of the cost functions of the
related optimization problems, iii) inverting the (X, Y ) mapping in case of unbalanced
problems through the bootstrapping of conditional Gaussian distributions, and iv) computing
tolerance regions through convex hull based peeling techniques. In this way we supply a very
accurate and fast algorithm to identify statistically the circuit model.
Of course, both procedures are susceptible of further improvements deriving from a more and
more deep statistics’ exploitation. In addition, nobody may guarantee that they will resist to
a further reduction of the technology scales. However the underlying methods we propose
could remain at the root of new solution algorithms of the yield maximization problem.
6. References
Agilent Technology (2010). IC-CAP Device Modeling Software – Measurement Control and
Parameter Extraction, Santa Clara, CA.
URL: http://www.home.agilent.com/agilent/home.jspx
Allgower, E. L. & Georg, K. (1990). Computational solution of nonlinear systems of equations,
American Mathematical Society, Providence, RI.
Ampazis, N. & Perantonis, S. J. (2002a). OLMAM Neural Network toolbox for Matlab.
URL: http://iit.demokritos.gr/ abazis/toolbox/
Ampazis, N. & Perantonis, S. J. (2002b). Two highly efficient second order
algorithms for training feedforward networks, IEEE Transactions on Neural Networks
13(5): 1064–1074.
Apolloni, B., Bassis, S., Malchiodi, D. & Witold, P. (2008). The Puzzle of Granular Computing,
Vol. 138 of Studies in Computational Intelligence, Springer Verlag.
Barnett, V. (1976). The ordering of multivariate data, Journal of Royal Statistical Society Series A
139: 319–354.
Bernstein, K., Frank, D. J., Gattiker, A. E., Haensch, W., Ji, B. L., Nassif, S. R., Nowak, E. J.,
Pearson, D. J. & Rohrer, N. J. (2006). High-performance CMOS variability in the
65-nm regime and beyond, IBM Journal of Research Development 50(4/5): 433–449.
Boning, D. S. & Nassif, S. (1999). Models of process variations in device and interconnect, in
A. Chandrakasan (ed.), Design of High Performance Microprocessor Circuits, chapter 6,
IEEE Press.
Bühler, M., Koehl, J., Bickford, J., Hibbeler, J., Schlichtmann, U., Sommer, R., Pronath, M.
& Ripp, A. (2006). DFM/DFY design for manufacturability and yield - influence
of process variations in digital, analog and mixed-signal circuit design, DATE’06,
pp. 387–392.
Chang, E., Stine, B., Maung, T., Divecha, R., Boning, D., Chung, J., Chang, K., Ray,
G., Bradbury, D., Nakagawa, O. S., Oh, S. & Bartelink, D. (1995). Using a
statistical metrology framework to identify systematic and random sources of die-
and wafer-level ILD thickness variation in CMP processes, in CMP processes, IEDM
Technology Digest, pp. 499–502.
Daems, S., Gielen, G. & Sansen, W. (2003). Simulation-based generation of posynomial
performance models for the sizing of analog integrated circuits, IEEE Transactions
on Computer-Aided Design of Integrated Circuits and Systems 22(5): 517–534.
Duch, W. & Kordos, M. (2003). Multilayer perceptron trained with numerical gradient,
Proceedings of the International Conference on Artificial Neural Networks (ICANN)
and International Conference on Neural Information Processing (ICONIP), Istanbul,
pp. 106–109.
Eeckelaert, T., Daems, W., Gielen, G. & Sansen, W. (2004). Generalized simulation-based
posynomial model generation for analog integrated circuits, Analog Integrated Circuits
Signal Processing 40(3): 193–203.
Efron, B. & Tibshirani, R. J. (1993). An Introduction to the Bootstrap, Chapman & Hall, New
York.
Elder IV, J. F. & Brown, D. E. (2000). Induction and polynomial networks. network models for
control and processing, in M. Fraser (ed.), Intellect, Portland, OR, pp. 143–198.
Eshbaugh, K. S. (1992). Generation of correlated parameters for statistical circuit simulation,
IEEE Transactions on CAD of Integrated Circuits and Systems 11(10): 1198–1206.
Friedman, J. H. (1991). Multivariate Adaptive Regression Splines, Annals of Statistics 19: 1–141.
Hatami, S., Azizi, M. Y., Bahrami, H. R., Motavalizadeh, D. & Afzali-Kusha, A. (2004).
Accurate and efficient modeling of SOI MOSFET with technology independent
neural networks, IEEE Transactions on Computer-Aided Design of Integrated Circuits and
Systems 23(11): 1580–1587.
Hershenson, M., Boyd, S. & Lee, T. (2001). Optimal design of a CMOS OP-AMP via geometric
programming, IEEE Trans. on Computer-Aided Design of Integrated Circuits and Systems
20(1): 1–21.
Jekabsons, G. (2010a). Adaptive basis function construction: an approach for adaptive
building of sparse polynomial regression models, Machine Learning, In-Tech p. 28. In
Press.
Jekabsons, G. (2010b). VariReg software.
URL: http://www.cs.rtu.lv/jekabsons/
Johnson, G. E. (1994). Constructions of particular random processes, Proceedings of the IEEE
82(2): 270–285.
Jolliffe, I. T. (1986). Principal Component Analysis, Springer Verlag.
Jones, D. R., Perttunen, C. D. & Stuckman, B. E. (1993). Lipschitzian optimization without the
Lipschitz constant, Journal of Optimization Theory and Applications 79(1): 157–181.
Koskinen, T. & Cheung, P. (1993). Statistical and behavioural modelling of analogue integrated
circuits, Circuits, Devices and Systems, IEE Proceedings G 140(3): 171–176.
Kundert, K. S. (1998). The DesignerâĂŹs Guide to SPICE and SPECTRE, Kluwer Academic
Publishers, Boston.
Lehmann, E. (2006). Nonparametrics, Statistical Methods Based on Ranks, Vol. XVI, Prentice-Hall.
1st edition in 1975, revised edition in 2006.
Liu, R. Y., Parelius, J. M. & Singh, K. (1999). Multivariate analysis by data depth: Descriptive
statistics, graphics and inference, The Annals of Statistics 27: 783–858.
McConaghy, T. & Gielen, G. (2005). Analysis of simulation-driven numerical performance
modeling techniques for application to analog circuit optimization, Proceedings of
IEEE International Symposium on Circuits and Systems.
McConaghy, T., Palmers, P., Gao, P., Steyaert, M. & Gielen, G. G. E. (2009). Variation-Aware
Analog Structural Synthesis: A Computational Intelligence Approach, Springer.
Mood, A. M., Graybill, F. A. & Boes, D. C. (1974). Introduction to the Theory of Statistics,
McGraw-Hill, New York.
Nelder, J. A. & Mean, R. (1965). A simplex method for function minimization, Computer Journal
7: 308–313.
Nocedal, J. & Wright, S. J. (1999). Numerical Optimization, Series: Springer series in operations
research, Springer, New York.
Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. (1993). Numerical Recipes in
Fortran; the Art of Scientific Computing, Cambridge University Press, New York, NY,
USA.
Price, K. V., Storn, R. M. & Lampinen, J. A. (2005). Differential Evolution, A Practical Approach to
Global Optimization, Vol. 538 of Natural Computing Series, Springer.
Rohatgi, V. K. (1976). An Introduction to Probablity Theory and Mathematical Statistics, Wiley
Series in Probability and Mathematical Statistics, John Wiley & Sons, New York.
Rubinstein, R. Y. & Kroese, D. P. (2007). Simulation and the Monte Carlo Methods, Probability
and Statistics, 2nd edn, John Wiley and Sons Inc.
Silvaco Engineered (2010). UTMOST III – SPICE Modeling Software, Santa Clara, CA.
Stolk, P. A., Widdershoven, F. P. & Klaassen, D. B. M. (1988). Modeling statistical dopant
fluctuations in MOS transistors, IEEE Transactions on Electron Devices 45(9): 1960 –
1971.
Taher, H., Schreurs, D. & Nauwelaers, B. (2005). Extraction of small signal equivalent circuit
model parameters for statistical modeling of HBT using artificial neural networks,
Gallium Arsenide Applications Symposium (GAAS 2005) 3-7 ottobre 2005.
Vancorenland, P., Van der Plas, G., Steyaert, M., Gielen, G. & Sansen, W. (2001). A
layout-aware synthesis methodology for RF circuits, ICCAD’01: Proceedings of
the 2001 IEEE/ACM International Conference on Computer-Aided Design, IEEE Press,
Piscataway, NJ, USA, pp. 358–362.
Wolfram Research Inc. (2008). Mathematica 7.
URL: http://www.wolfram.com/products/ mathematica/index.html
Part 3
Applications
12
Analog-aware Schematic Synthesis

Yuping Wu
Institute of Microelectronics, Chinese Academy of Sciences,
China
1. Introduction
An analog circuit has great requirements of constraints on circuit and layout optimization
for the purpose of functionality. Various constraint generation methods were provided, but
there are too many limitations even the circuit topology has a bit variance due to no
knowledge of the circuit functionality. To get the requirements exactly, you must know the
circuit functionality exactly before, so analog circuit functionality analysis is very important
for analog circuit design, especially for automatic analog/mixed signal design, but until
now there is few method research report for automatic analog circuit functionality analysis
except for the digital system design. The conventional way is that most of the work is done
from an analog structural feature highlighted circuit schematic by the engineer manually,
that is to say a good circuit schematic is the precondition for manual analysis on circuit
functionality, which brings another issue about analog circuit schematic generation for
analog / mixed signal design automation.
It should be appreciated that the circuit schematic generation has been in use for years
with digital designs, functional clustering based analog circuit schematic generation was
reported in [37, 39-43], which is rule-based and only feasible for some simple functional
blocks due to the limitation of the description of rules. In the commercial tools from
Cadence, Synopsys, and Magma, they use the methods from digital [8] for analog as
instead, user cannot get the analog structural features insight, so it is hard to get the
constraints for circuit and layout optimization from the schematic, although some
previous works have been done [9][44].
In the long term, analog schematic generation is also necessary for future analog synthesis
and analog design migration. The complete analog design automation flow is a far-away
perfect expectation, as the part of such synthesis flow, analog behavioral synthesis will
transform the behavioral description into circuit netlist, and the circuit netlist will be
transformed into analog schematic, also such analog-aware schematic synthesis is the
technical base to schematic optimization / retuning for analog design technology migration.
To overcome such issues, we studied a structural feature-based analog circuit analysis and
partition technique, generated the constraints for schematic generation, circuit optimization
and layout optimization after circuit analysis; based on that, we proposed an algorithm to
generate analog aware circuit schematic [12] from the partitioning results with analog
functionality and structural features highlighted, the constraints for circuit and layout
optimization are identified on that schematic, and also analog functionality and structural
feature can be got insight intuitively, which is helpful to circuit designers and layout
engineers for circuit optimization and layout optimization.
This chapter describes the implementation of such analog-aware circuit schematic synthesis,
and is organized as: section 1 gives the technical background necessitates for analog-aware
circuit schematic synthesis; section 2 will present the analog-aware schematic synthesis
flow; section 3 will detail structure features of analog functional circuits and descriptions,
which includes low level analog structure features, high level analog structure features,
structure feature library composition, structure feature associated attributes, and structure
feature recognition; section 4 will describe analog circuit functionality analysis and
partitioning, which includes input information, pre-processing, tracing direct current paths,
tracing signal paths, encoding for blocks, checking isomorphism and quasi-isomorphism,
and partitioning into hierarchy; section 5 will describe the constraint generation, which
includes constraints for schematic generation and optimization, constraints for circuit design
and optimization, and constraints for layout design and optimization; section 6 will describe
analog schematic generation, which includes the symbol generation based on functionality,
symbol placement, wiring, and constraint identification; section 7 will describe analog-
aware schematic synthesis with companion circuits, which includes common feature
extraction, functionality analysis and partitioning, constraint extraction with companion
circuits, and analog schematic generation with companion circuits; and finally we will show
some experimental results of such analog-aware circuit schematic synthesis technology.
2. Analog circuit schematic synthesis flow

As shown in Fig. 1(a), the traditional analog circuit schematic synthesis consists of 1) netlist-
in; 2) data-in for mapping between devices and symbols; 3) cell symbol generation; 4)
symbol placement for devices, cell instances, and ports; 5) wire routing; and 6) schematic-
out. In comparison, the new analog circuit schematic synthesis flow consists of 1) netlist-in;
2) data-in for mapping between devices and symbols; 3) template-in for functionality
analysis; 4) functionality analysis and partitioning for new hierarchy; 5) port analysis; 6)
constraint generation; 7) analog-aware symbol generation; 8) analog-aware symbol
placement; 9) analog-aware wire routing; 10) analog-aware constraint identification; and 11)
schematic-out as shown in Fig. 1(b).
In the two schematic synthesis flows, as the common parts, circuit netlist-in can be spice-
compatible netlist or netlist-in-database consisting of devices and connections; data-in for
mapping between devices and symbols will set up one-to-one relation between devices and
symbols for correct device symbol use; and schematic-out pushes the schematic data into the
EDA platform database, such as DFII or OA, so that the schematic viewer/editor can
display the schematic directly.
The differences between the traditional flow and the new flow are in red color. The first
difference between them is the introducing of the templates-in. The templates-in includes
circuit templates, symbol templates, and constraint templates. A circuit template has a
couple of associated symbol templates and constraint templates.
Circuit templates are used for functionality analysis and partitioning with bottom unit
circuit description and complex high level block composition description. The template for
unit circuit must describe the device composition and connections of the unit circuit with
transistor level in detail and stamp the functionality correctly; while the template for
complex high level circuit must describe composition of sub-functionalities and connections
among functional blocks, and also the functionality of the complex circuit must be stamped
Analog-aware Schematic Synthesis 249
with functionality name correctly. All the functionality names are used for functionality
analysis of complex high level circuit based on the specified name conventions.
Circuit netlist-in Circuit netlist-in
Data-in for mapping between devices Data-in for mapping between devices & symbols
& symbols
Template-in
Circuit functionality analysis and partitioning for new

hierarchy
Port analysis
Constraint generation
Symbol generation Analog-aware symbol generation
Symbol placement Analog-aware symbol placement
Wire routing Analog-aware wire routing
Analog-aware constraint identification
Schematic-out Schematic-out
(a) (b)
Fig. 1. Comparison of traditional analog circuit schematic synthesis flow (a) and novel
analog circuit schematic synthesis flow (b)
Symbol templates are for symbol generation based on the functionality, designers can get
functionality from the shapes of symbols, due to the symbol shape reflecting the
functionality intuitively.
Constraint templates are for generating sizing, floorplanning, and layout constraints, which
will speed up analog schematic synthesis, circuit sizing, floor-planning, and layout synthesis
by reducing the possible exploration space and making the solution candidates more
reasonable and acceptable [10]. The template for constraint generation can be built by
designers manually or from good designs by automatic extraction tools.
The second difference between the flows is the introducing of analog circuit functionality
analysis and partitioning for new hierarchy, which is the most solid base of the new flow
and will be a bit detailed in next section.
The third difference between the flows is the introducing of port analysis. In traditional
schematic synthesis flow, due to lacking of port analysis, all of the ports for each cell are
treated as inputs/outputs no matter what they are in purpose exactly, so the synthesized
schematic looks confused from the ports. Correct identification of port attribute is very
important in schematic, so the port attribute should be captured before, but it is impossible
to specify the port attributes manually for all the cells in a design especially when the design
is in large scale, designers can only input some for several of them. Hence, it is necessary to
use an automatic program to solve such issue. We introduce the port analysis for it, it
determines the port types for each sub-cell automatically based on the combination of
functionality partitioning, circuit template, signal flow analysis, dummy connection, ESD
connection, substrate connection, name convention, and so on. The port analysis result will
be used for pin placement on cell symbol generation and port terminal symbol selection and
placement on analog-aware symbol placement step.
The fourth difference between the flows is the introducing of constraint generation for
schematic synthesis, circuit sizing, floor-planning, and layout optimization, which is based
on the combination of functionality partitioning, constraint templates, signal flow analysis,
port analysis, dummy connection, ESD connection, MOSCAP connection, and so on. The
constraints include symmetry requirements in a DC path, device matching requirements
among DC paths, symmetry requirements between DC paths, dummy devices, protection
devices and the associated protected devices, MOSCAP devices, critical signal nets, net
current, and net wiring width, etc.
After analog-aware symbol placement and wire routing steps, as the fifth difference, analog
constraint identification on the schematic is necessary to make circuit designers and layout
engineers have a good insight on the design for circuit optimization, physical floor-
planning, and layout optimization. The identifications include symmetry requirements in a
DC path, device matching requirements among DC paths, symmetry requirements between
DC paths, dummy devices, protection devices and the associated protected devices,
MOSCAP devices, critical signal nets, net current and net wiring width, and so on. All the
identification contents are results from the steps of functionality analysis and partitioning,
port analysis, and constraint generation.
In summary, the great differences between traditional flow and novel flow are the
introducing of template-in for functionality analysis, functionality analysis and partitioning
for new hierarchy, port analysis, and constraint generation by the novel flow, which makes
it possible for analog-aware symbol generation for cells, symbol placement, wire routing,
and constraint identification on schematic based on the functionality, port types, and other
constraints, so the innovation of the flow is the functionality analysis and partitioning
technique, port analysis, automatic constraint generation, and constraint-driven analog-
aware schematic generation.
3. Structure features of analog functional circuits and descriptions

Structure features of analog functional circuits are the intuitive bases for setting up the circuit
templates directly and setting other associated constraint templates. The structure feature of
analog functional circuits includes low level analog structures and high level analog
structures; the first focuses on the composition of devices and their connections, and the later
focuses on the composition of basic or complex function blocks and their connections.
3.1 Low level analog structure features[1-3]

3.1.1 Structure features for basic amplifier circuits
Fig. 2. Structure features for CE amplifier
Fig. 3. Structure features for CC amplifier
Fig. 4. Structure features for CS amplifier

Fig. 5. Structure features for CS amplifier

Fig. 6. Structure features for differential amplifiers

3.1.2 Structure features for amplifier output circuits
Fig. 7. Structure features for OTL circuit
Fig. 8. Structure features for OCL circuit
Fig. 9. Structure features for BTL circuits

3.1.3 Structure features for current source circuits
Fig. 10. Structure features for current mirror / current source circuits
Fig. 11. Structure features for stack cascade current source circuits
Fig. 12. Structure features for cascode current source with wide output swing circuits
3.1.4 Structure features for oscillators
Fig. 13. Structure features for ring oscillators

Fig. 14. Structure features for cascade oscillators

3.1.5 Structure features for charge pump
Fig. 15. Structure features for charge pump circuits
3.1.6 Structure features for band-gap circuits

Fig. 16. Structure features for band gap circuits
[1-3]
3.2 High level analog structure features
3.2.1 Structure features for OPA and OPA-based circuits
Amplifier
Amplifier
shifter
Ouput
Level
Input
Fig. 17. Structure features for OPA circuits

Fig. 18. Structure features for INV-Ratio circuit
Fig. 19. Structure features for PASS-Ratio circuit
Fig. 20. Structure features for sum circuit

Fig. 21. Structure features for differentiator circuit
Fig. 22. Structure features for integrator circuit
Fig. 23. Structure features for logarithm circuit

Fig. 24. Structure features for exponential circuit
3.2.2 Structure features for active filtering circuits
Fig. 25. Structure features for Low-pass (1st-order) filter circuit
Fig. 26. Structure features for Low-pass (2nd order) filter circuit
Fig. 27. Structure features for high-pass filter circuit
Fig. 28. Structure features for band-pass filter circuit
Fig. 29. Structure features for Band-resistive filter circuit

3.2.3 Structure features for signal transformation circuits
Fig. 30. Structure features for voltage / current transformation circuit
Fig. 31. Structure features for AC/DC transformation circuit
Fig. 32. Structure features for Voltage / frequency transformation circuit

3.2.4 Structure features for PLL
Fig. 33. Structure features for PLL circuits

Fig. 34. Structure features for D-FF as PD
Fig. 35. Structure features for PFD circuit
3.2.5 Structure features for A/D Converters
Fig. 36. Structure features for integrating ADC

Fig. 37. Structure features for successive approximation ADC
Fig. 38. Structure features for charge-redistribution SA-approximation ADC
Fig. 39. Structure features for flash ADC

Fig. 40. Structure features for Σ-Δ ADC
3.2.6 Structure features for DAC
Fig. 41. Structure features for R/2nR DAC
Fig. 42. Structure features for R/2R DAC

Fig. 43. Structure features for voltage scaling DAC
Fig. 44. Structure features for voltage and charge scaling DAC
Fig. 45. Structure features for charge scaling DAC

3.3 Structure feature library composition

The structure feature library mainly contains structure feature description enclosed with a
cell in SPICE netlist format, the cell name consists of keyword as prefix, the separator char
“-“, and a normal string for making cell name be unique, where the keyword represents the
functionality of the analog structure.
For the bottom level analog structure feature description, device level netlist is used to
describe the devices and their interconnections; and for the high level analog structure
feature description, the block level netlist is used to describe the member block
instantiations and their interconnections, the member block instantiation comes from a low
level block of specific functionality, i.e., the template cell name quoted in the member block
instantiation must be a keyword representing functionality rather than a specific cell name,
which means that the instantiation represents the instantiation of functionality rather than
the instantiation of a specific structure, which makes high level structure feature description
independent from the specific detail low level or bottom level analog structure.
3.4 Structure feature associated attributes

Structure feature associated attributes include the constraints for schematic synthesis, sizing,
floorplanning, layout, symbol shape, pin-out attributes, and others.
3.4.1 Schematic constraint knowledge

Constraints for schematic generation and optimization should include the constraints within
a direct current path, the constraints between direct current paths, the constraints between
blocks, and terminal placement constraints.
The constraints within a direct current path include the device list of direct current path, the
top to down device sequence from power to ground based on power reaching level, and the
device symmetry between direct current path branches.
The constraints between direct current paths include the device symmetry among the direct
current paths, the parallel direct current paths of same signal reaching level, and the left to
right direct current path sequence from input to output based on signal reaching level for
direct current paths.
The constraints between blocks include the symmetry between the blocks, the left to right
sequence from input to output based on signal reaching level for blocks, the ring sequence of
the blocks based on signal path ring, and the parallel blocks based on signal reaching level.
The terminal placement constraints include the side constraint, the top to down sequence for
left side and right side terminals, and the left to right sequence for top side and bottom side
terminals.
3.4.2 Sizing constraint knowledge

Constraints for circuit design and optimization [11][13][22] can merge the optimization
parameters, reduce the exploration space, and speed up the optimization for sizing
procedure, so it is very important to generate such constraints no matter how the sizing step
is implemented in hand or in automation.
Structure constraints for transistor pairs can be set up for differential pairs, level shifter,
complementary pairs, current mirrors, matched direct current path, and matched blocks in
future, so the first step for structural constraint generation is to execute the low level
structure feature base matching exploration and high level structure feature based matching
exploration, which is described before, then set up such structure constraints for those
device pairs with the following considerations.
For good mismatch properties and an area efficient layout, the channel lengths and the
finger channel widths of the two transistors must be the same respectively. The ratio of the
two transistor finger numbers must be equal to the ratio of the currents, although the ratio is
1 for differential pairs and current mirrors, and 1 or other integer values for others.
LM1 = LM2, FWM1 = FWM2, and I1 / I2 = FMM1 / FMM2

The smaller the area of a transistor, the higher is its mismatch sensitivity. Therefore the
transistor channel width and length must not fall below a minimum value Wmin and Lmin for
differential pairs, level shifter, complementary pairs, current mirrors, and current sources:
FWi * FMi ≥ Wmin and Li ≥ Lmin, i ∈{M1, M2, …}

Both transistors operate as voltage-controlled current sources (vccs) and thus they must be
in saturation for current mirrors and current sources:
0 < VDSi < VGi = VGSi - VT, i ∈{M1, M2, …}

For a low VT-mismatch sensitivity, the effective gate voltage must not fall below a minimum
value VGmin for current mirrors and current sources:
0 < VGmin < VGSi – VT, i ∈{M1, M2, …}

For a low λ sensitivity the difference of the drain source voltages must not exceed a
maximum value VDSmax for current mirrors and current sources:
|VDSM1 - VDSM2| < VDSmax
3.4.3 Layout constraint knowledge

Constraints for layout design and optimization [4-7][16-21][23-36] include the symmetry
constraints for devices, direct current path branches, direct current paths, blocks and upper
level circuits, the matching constraints for group of devices, the neighboring constraints, the
protection constraints, the signal path and sequence constraints for direct current paths, and
the direct current path and power reaching sequence constraints for group of devices.
The symmetry constraints can be used for minimizing the mismatch by mirroring placement
of devices, direct current path branches, direct current paths, blocks, or upper level circuits,
and mirroring the wiring of interconnections to reduce the mismatch on devices and the
mismatch on wires, in further to reduce mismatch on direct current path branches, direct
current paths, blocks and upper level circuits during layout design and optimization, and
such constraints can be gotten with encoding based symmetry direction.
The matching constraints can be used for minimizing the mismatch on devices, direct
current path branches, direct current paths, and upper level circuits by optimal placement of
matching mode and dummy insertion to reduce the mismatch due to parasitic and process
variations, such constraints can be gotten from structural feature based recognition for
devices, encoding based match recognition for direct path braches, direct current paths,
blocks, and upper level circuits.
The neighboring constraints can be used for minimizing the interconnection parasitic,
interconnection interference, and interference among neighboring devices, which includes
closing-necessary, neighboring-forbidden, and less than / far away from a specified distance.
The protection constraints can be used for preventing the critical devices or critical device
groups interfered electrically by others, such constraints can be gotten from the previous
signal path tracing and matching device exploration method.
The signal path and sequence constraints for direct current paths can be used for
minimizing the interconnection parasitic on signal path to ensure the circuit frequency
performance while layout design and optimization, and such constraints can be gotten from
the signal path tracing method.
The direct current path and power reaching sequence constraints for group of devices can be
used for minimizing the interconnection parasitic on direct current path so as to reduce the
dc operation point variation due to parasitic on such path and ensure the DC performance
while layout design and optimization, and such constraints can be gotten from the direct
current path tracing method.
3.4.4 Constraint knowledge extraction based on good example circuits

Structure feature associated constraints are obvious in part, such as matching between
differential pair devices and matching among current mirror / current source devices, but
most of them are not so clear, so they need to be setup by hand based on the designer’s
professional experiences, it is very effective, but low efficiency due to handwork. There also
exists another way to setup part of those constraints with the leverage of some good
example circuits, which have embedded more professional design experiences.
Constraint knowledge extraction based on good example circuits mainly includes 1) analog
structure feature analysis, 2) locating for analog structure feature devices / blocks, and 3)
constraint capture for analog structure features from good schematic and layout data using
geometry calculation, such as one level symmetry and multi-level symmetry, matching and
matching mode, neighboring, protection, and so on.
3.5 Structure feature recognition

Recognition of low level analog structure feature is mainly graph-isomorphism of devices
and connections, and recognition of high level analog structure is mainly graph-
isomorphism of function blocks and interconnections with the ignorance of detail bottom
devices and interconnections among them, it is to say that two high level blocks may have
same functions if they have same composition of basic or high level functional blocks and
interconnections although their corresponding low level functional blocks of the identical
functionality may have different composition of devices and interconnection.
3.5.1 Recognition for low level analog structure features

Recognition for low level analog structure features is a direct searching procedure for
complete matching on detail devices and connections among them between the source
analog structure and the analog structure feature template with a bit tricky for speeding up.
As shown in Fig. 46, the main steps include graph setting-up, encoding for source analog
structure, finding matched low level analog structure templates from template map using
source structure coding value, and getting the functionality coding value for up level
structure feature recognition and the associated attributes. The template map is setup from
the analog structure feature template library.
start
graph setting-up
encoding for source analog structure
finding matched low level analog

structure template
getting the functionality coding

value and associated attributes
End
Fig. 46. Procedure for low level analog structure feature recognition
start
recognition for low level

analog structure features
abstracting
encoding for the abstract circuits
finding upper level matching

analog structure features
N upper level matching analog

structure features found?
Y
End
Fig. 47. Procedure for high level analog structure feature recognition
3.5.2 Recognition for high level analog structure features

Recognition for high level analog structure features is an iterative abstracting and searching
procedure for complete matching on functional blocks and connections among them but
with the ignorance of their bottom detail devices and connections between the source analog
structure and analog structure feature template with a bit tricky for speeding up.
As shown in Fig. 47, the main steps include 1) recognition for low level analog structure
features, 2) abstracting, i.e., replacing low level analog structure with virtual functional
block with ignorance of detail composition, 3) encoding for the abstract circuits, and 4)
finding the upper level matching templates with encoding value comparison, repeat step 2)
to step 4) until no any upper level matching templates are found.
4. Analog circuit functionality analysis and partitioning

The proposed analog circuit functionality analysis and partitioning flow is shown as in Fig.
47. The input information includes the necessary information, such as circuit netlist and
structural feature template libraries, and optional information: model type information and
port information. The analysis and partitioning flow includes pre-processing netlist, tracing
DC paths, tracing signal paths, encoding for DC paths and above block, checking
isomorphism, and partitioning & res-constructing design in new hierarchy.
Circuit netlist-in and templates-in
Pre-processing the input netlist
Tracing the direct current (DC) paths
Tracing the signal paths
Encoding for DC paths and above
Checking isomorphism
Partitioning & re-constructing for new hierarchy
Fig. 48. Functionality analysis and partitioning flow

Analog functionality analysis is one of the bases for analog-aware circuit schematic
synthesis; it is very different with traditional symbol analysis, it analyzes circuit
functionality based on the functionality-known detail bottom level unit circuit templates,
and the functionality-known complex high level circuit template with functionality
abstraction but without detail circuit descriptions for bottom unit circuits, which means that
analog functionality analysis is an accurate pattern matching for low level unit circuits, and
fuzzy pattern matching for high level circuits because the bottom devices and connections
are ignored as possible and the bottom level unit circuits are represented by functionality
and port connection only. The pattern matching is supported by encoding of graphic of
devices, functional blocks, and connections among them and encoding value matching.
After functionality analysis, the analog design needs to be reconstructed with a new
hierarchy based on functionality so as to use symbol templates to generate symbols and use
the constraint templates to produce the accurate sizing, floor-planning, and layout
constraints of the current analog circuit for future use. Also performance spec can be
allocated into new hierarchy for future parallel on circuit optimization.
4.1 Input information

The input information for analog schematic synthesis includes the circuit netlist in spice
netlist format, the data-in for mapping between devices & symbols, and the templates for
analog structure features and associated templates as necessary inputs, and the partial port
attributions or port name conventions as optional inputs.
4.2 Pre-processing
To make analog schematic synthesis more effectively, the pre-processing is necessary before
core analog schematic synthesis procedure. The pre-processing includes identifying the
aided devices, such as dummy devices and electronic static discharge (ESD) devices [45],
removing them for analog structure feature analysis, port attribution passing, and internal
power supply recognition.
The port attribution passing includes the top to down passing and the bottom up to top
passing, which should be executed iteratively until all the port attributions are set for each
cell especially when internal voltage regulation circuits are used for whole or part of the
circuit, because the port attribution may be passed from one cell A to another cell B of same
hierarchy level, for an example, cell A is a voltage regulator providing power supply to cell B.
Port attribution passing can set up the port attribution of each terminal for each cell, which can
reduce the complexity of analog functionality analysis and other derived analysis, because the
port attribution, such as power terminals, ground terminals, signal input terminals, and signal
output terminals, can be used to limit the start points and the end points for current flow
spreading and signal flow spreading, and the port attribution, such as power terminals and
ground terminals can be used reduce the complexity of circuit-based graph especially.
To make port attribution passed smoothly, the internal power supply recognition is a
necessary to make the internal power supply be regarded as power terminals of other
internal circuits when the internal voltage regulation circuits are used so as to ease the
analysis of other internal circuits. The internal power supply recognition should include
band-gap structure feature recognition, band gap reference circuit identification by finding
the OPA associated with the band-gap feature, and determination of output terminal(s) of
the band gap reference circuits.
4.3 Tracing direct current paths

In the method operation of tracing the direct current paths, tracing can be spread along the
direct currently flow direction, as shown in Fig. 49, or along the inverse of direction, which
is determined according to the presented terminal types, such as the positive power supply
terminals, the ground terminals, the negative power supply terminals, the current mode
input terminals, and the current mode output terminals. The detail tracing can be done as
the following descriptions.
D S C E
G I G I B I
B I
S D E C
P M P P1
I I
I I
N P2
M P
Fig. 49. Direction of current flow through devices

As the first operation method, the direct current path tracing can start from the positive power
supply terminals or current mode input terminals; spread along the drain to source or source
to drain for MOSFET and JFET, the collector to emitter for NPN BJT devices, the emitter to
collector for PNP BJT, the positive terminal to negative terminal for diode, and one terminal to
another terminal for some resistors and inductors, as shown in Fig. 50 and stop while reaching
the ground terminals, negative power supply terminals, current output mode input terminals,
or current mode output terminals. From such traversing, it gets the list of devices of a direct
current path, calculates the minimum distance to the positive power supply terminals or
current mode input terminals for each device, then sorts the device based on the distance
values from min to max to get the device sequence of the current path.
As the second operation method, the direct current path tracing can start from the ground
terminals, spread as above description, as shown in Fig. 51, and stop while reaching the
negative power supply terminals. From such traversing, it gets the list of devices of a direct
current path, calculate the minimum distance to the ground terminal for each device, then
sort the device based on the distance values from min to max to get the device sequence of
the current path.
As the third operation method, the direct current path tracing can start from the ground
terminals, spread as the inverse of current flow direction, as shown in Fig. 52, and stop
while reaching the current mode input terminals or the current mode output terminals.
From such traversing, it gets the list of devices of a direct current path, calculate the
minimum distance to the ground terminal for each device, then sort the device based on the
distance values from max to min to get the device sequence of the current path.
Positive Positive Positive Power Positive Power

Power Supply Power Supply Supply Supply
Terminal Terminal Terminal Terminal
Negative Power Ground Current Mode Current Mode

Supply Terminal Input Terminal Output Terminal
Terminal
Current Mode Current Mode

Input Input
Terminal Terminal
Negative Power Ground

Supply Terminal
Terminal
Fig. 50. Find the direct current path from the positive power supply terminal to the negative
power supply terminal, the ground terminal, the current mode input terminal, and or the
current mode output terminal, and from the current mode input terminal to the negative
power supply terminal or the ground terminal with normal direct current direction
Ground
Terminal
Negative Power
Supply Terminal
Fig. 51. Find direct current path from the ground terminal to the negative power supply
terminal with the normal direct current direction
Current Mode Current Mode

Input Terminal Output
Terminal
Ground Ground
Terminal Terminal
Fig. 52. Find the direct current path from the ground terminal to the current mode input
terminal and from the ground terminal to current mode output terminal with reverse of
direct current direction
Ground Current Mode

Terminal Output Terminal
Negative Power Negative Power

Supply Terminal Supply Terminal
Fig. 53. Find the direct current path from the negative power supply terminal to the ground
terminal or the current mode output terminal with reverse of direct current direction.
As the fourth operation method, the direct current path tracing can start from the negative
power supply terminals, spread as the inverse of current direction, as shown in Fig. 53, and
stop while reaching the ground terminals or current mode output terminals. From such
traversing, it gets the list of devices of a direct current path, calculate the minimum distance
to the negative power supply terminal for each device, then sort the device based on the
distance values from max to min to get the device sequence of the current path.
For a typical circuit, any one of the above operation method cannot dig out all the direct
current paths, so in practice, the combination of them is used, although there are some
overlaps among the above four operation methods. To filter out the overlapping direct
current path result, a map for identifying the handled devices is used so as to avoid
unnecessary repeat operations.
As an addition, grouping devices of the current source are not in the same direct current
path, but they are searched out, such as the companion devices from different direct current
paths of current sources circuit; also the other devices from different current paths but with
same power reaching levels or same ground reaching levels are searched out, so that such
devices can be placed on one horizontal line for easy wiring in schematic view.
4.4 Tracing signal paths

In the method operation of tracing the signal paths [14], tracing starts from the input signal
terminals, and spreads along gate to drain/source or drain/source to source/drain for
MOSFET and JFET, base to collector/emitter or collector/emitter to emitter/collector for
BJT, the positive terminal to negative terminal for diode, and one terminal to another
terminal for some resistors/capacitors/inductors, as shown in Fig. 54 other than feedback or
bypassing filtering devices. The signal spreading is terminated while reaching power supply
terminals, ground terminals, or output terminals.
During signal path tracing, the signal input terminal node is put into the signal node list,
handle the devices connected to the signal node, spread the signal based on the above signal
flow direction rules so as to find next possible signal nodes to which these devices are
connected to, put the new signal nodes into the signal node list, and traverse the signal node
list until all the signal nodes are handled. To speed up tracing signal path, a device map and a
node map should be used for a circuit. A flag is marked for a device in the device map while a
signal spreading is handled on that device in case of repeating signal spreading on the same
device in the future. Also, a flag is marked for a node in the node map while a signal spreading
is handled on that node in case of repeating signal spreading on the same node in the future.
The distance between an input signal port and a device is defined as the signal reaching
level of that device under that signal; the signal reaching level of a device may consist of
signal reaching minimum level and signal reaching maximum level, which reflects different
signal flow paths to that devices.
D C E
G s B B s
s
S E C
P P1 P
P1
s s s
s
P2 N
M P2
Fig. 54. Direction of signal flow through devices
Also, signal reaching level for a direct current path consists of the signal reaching minimum
level and the signal reaching maximum level, they can be gotten from the minimum of
signal reaching minimum levels and maximum of signal reaching maximum levels of all
devices in such direct current path respectively.
In further, signal reaching level for a block consists of the signal reaching minimum level
and the signal reaching maximum level, they can be gotten from the minimum of signal
reaching minimum levels and the maximum of signal reaching maximum levels of all the
direct current paths in such block respectively.
4.5 Encoding for blocks

Encoding from bottom level to up level, the bottom level is for direct current (DC) path only,
and the up level is the combination of direct current paths and more.
To encode for a direct current path, try to find the matched DC path structural feature from
the template libraries with ignorance of some auxiliary devices including the dummy
devices, protection devices, MOSCAP devices, power-down devices, and biasing devices,
assign the functionality name and functionality identification number to that DC path so
that it is encoded with such identification number in a bit fuzzy logic.
To encode for a cell/block, each DC path is considered as a virtual block of a specific
functionality, each sub-cell/block in the current cell domain is also considered as a black
box of a specific functionality, so the encoding step is to try to find the matched template of
same functional blocks and same signal connectivity among blocks, which can be handled as
pattern matching issue on quasi one-dimension, the functionality name and functionality
identification number is assigned to the current cell.
4.6 Checking isomorphism and quasi-isomorphism

In contrast to traditional sub-graph isomorphism algorithm [46-48], the checking issue is a
quasi one-dimension graph due to the simplification from each DC path or clusters of DC
paths to a functionality vertex, and also some unimportant connectivity is ignored, so it is a
bit fuzzy logic. The computing complexity is closing to O(n) due to the one dimension
approximation and sequenced, so the encoding and code value comparison can be used
efficiently for isomorphism checking.
4.7 Partitioning into hierarchy

The source circuit is abstracted in several hierarchy levels after the recognition of low level
analog structure features and the recognition of high level analog structure features, and
each block of any level in the abstract tree represents has a specific functionality.
Reconstruct the circuit netlist based on such functionality recognition abstract tree, which
includes the following main steps: 1) Determine the out connections of a block to build the
port terminal information for that functional block 2) build the netlist for the functional
block based on direct sub-blocks and their interconnection with sub-block handled as an
instantiation of the corresponding sub-cell, and 3) build the netlist for the bottom level
block: based on the detail devices and their interconnections. After that, such circuit
partitioning can make the new hierarchical circuit more intuitive for designer to understand
it and get more advantages on later circuit sizing, floorplanning and layout automation.
5. Constraint generation
Constraint generation is a very important step in analog schematic synthesis procedure [10].
After analog structure feature recognition, the analog structure feature associated constraint
templates can be used to generate the constraints for schematic synthesis, circuit synthesis, and
layout synthesis if the associated constraint template exists. The key is to find the device-to-
device mapping relation and block-to-block mapping relation so as to replace the virtual
device name or virtual block name with practical device name or practical block name of
source circuits, it is very easy, herein we do not discuss about it. Here we focus on the case
without constraint templates, as a complementary, the constraints can be generated with
leverage of part of the analog structure feature recognition result and further analysis results.
5.1 Constraint generation for schematic generation and optimization

Constraints for schematic generation and optimization should include the constraints within
direct current path, the constraints between direct current paths, the constraints between
blocks, and the terminal placement constraints.
The constraints within a direct current path include the device list of direct current path, the
top to down device sequence from power to ground based on power reaching level, and the
device symmetry between direct current path branches. The first three constraints can be
gotten as the result of tracing the direct current paths, and the constraint of device
symmetry between direct current path branches can be checked out with the devices of the
same power reaching level as a symmetry pair.
The constraints between direct current paths include the device symmetry among the direct
current paths, the parallel direct current paths of same signal reaching level, and the left to
right direct current path sequence from input to output based on signal reaching level for
direct current paths. The first constraint can be checked out using sub-graph isomorphism
method, the head line of the method can be overviewed as: 1) setup graph for each direct
current path; 2) encode for each graph; 3) compare the encoding values; 4) if the encode values
are matching, put the two direct current pats as symmetry candidate; and 5) check the signal
reaching minimum level and signal reaching maximum level of the direct current paths of the
candidate; regard them as symmetry pair if matching occurs. The second constraint can be
checked out if any two direct current paths have identical the signal reaching minimum level
and signal reaching maximum level. The third constraint can be checked out using the sorting
based on the signal reaching minimum level and signal reaching maximum level.
The constraints between blocks include the symmetry between the blocks, the left to right
sequence from input to output based on signal reaching level for blocks, the ring sequence
of the blocks based on signal path ring, and the parallel blocks based on signal reaching
level. The first constraint can be checked out if the two blocks are matched completely and
have identical the signal reaching minimum level and signal reaching maximum level. The
second constraint can be checked out by sorting the blocks with their signal reaching
minimum levels and signal reaching maximum levels. The third constraint can be checked
out by signal flow spreading, if a signal flow circle is checked, i.e., signal flow spreading
meets a past checked signal points, all blocks on such signal flow circle construct the ring,
the ring sequence of blocks are gotten by sorting with the signal reaching minimum levels
and signal reaching maximum levels of those blocks. The fourth constraints can be checked
out if any two blocks of a circuit have the identical signal reaching minimum levels and
signal reaching maximum levels.
The terminal placement constraints include the side constraint, the top to down sequence for
left side and right side terminals, and the left to right sequence for top side and bottom side
terminals. For the side constraints, in principle, the input terminals are presented with left
side constraint, the output terminals are presented with right side constraints, the positive
power supply terminals are presented with the top side constraints, and the ground
terminals and the negative terminals are presented with the bottom side constraints.
5.2 Constraint generation for circuit design and optimization

Constraints for circuit design and optimization can merge optimization parameters, reduce
the exploration space, and speed up the optimization for sizing procedure, so it is very
important to generate such constraints no matter how the sizing step is implemented in
hand or in automation.
Structure constraints for transistor pairs can be set up for differential pairs, level shifter,
complementary pairs, current mirrors, matched direct current path, and matched blocks in
future, so the first step for structural constraint generation is to execute the low level
structure feature base matching exploration and high level structure feature based matching
exploration, which is described before, then set up such structure constraints for those
device pairs with the following considerations.
For good mismatch properties and an area efficient layout, the channel lengths and the
finger channel widths of the two transistors must be the same respectively. The ratio of the
two transistor finger numbers must be equal to the ratio of the currents, although the ratio is
1 for differential pairs and current mirrors, and 1 or other integer values for others.
LM1 = LM2, FWM1 = FWM2, and I1 / I2 = FMM1 / FMM2

The smaller the area of a transistor, the higher is its mismatch sensitivity. Therefore the
transistor channel width and length must not fall below a minimum value Wmin and Lmin for
differential pairs, level shifter, complementary pairs, current mirrors, and current sources:
FWi * FMi ≥ Wmin and Li ≥ Lmin, i ∈{M1, M2, …}

Both transistors operate as voltage-controlled current sources (vccs) and thus they must be
in saturation for current mirrors and current sources:
0 < VDSi < VGi = VGSi - VT, i ∈{M1, M2, …}

For a low VT-mismatch sensitivity, the effective gate voltage must not fall below a minimum
value VGmin for current mirrors and current sources:
0 < VGmin < VGSi – VT, i ∈{M1, M2, …}

For a low λ sensitivity the difference of the drain source voltages must not exceed a
maximum value VDSmax for current mirrors and current sources:
|VDSM1 - VDSM2| < VDSmax
5.3 Constraint generation for layout design and optimization

Constraints for layout design and optimization include the symmetry constraints for
devices, direct current path branches, direct current paths, blocks and upper level circuits,
the matching constraints for group of devices, the neighboring constraints, the protection
constraints, the signal path and sequence constraints for direct current paths, and the direct
current path and power reaching sequence constraints for group of devices.
The symmetry constraints can be used for minimizing the mismatch by mirroring placement
of devices, direct current path branches, direct current paths, blocks, or upper level circuits,
and mirroring the wiring of interconnections to reduce the mismatch on devices and the
mismatch on wires, in further to reduce mismatch on direct current path branches, direct
current paths, blocks and upper level circuits during layout design and optimization, and
such constraints can be gotten with encoding based symmetry direction.
The matching constraints can be used for minimizing the mismatch on devices, direct
current path branches, direct current paths, and upper level circuits by optimal placement of
matching mode and dummy insertion to reduce the mismatch due to parasitic and process
variations, such constraints can be gotten from structural feature based recognition for
devices, encoding based match recognition for direct path braches, direct current paths,
blocks, and upper level circuits.
The neighboring constraints can be used for minimizing the interconnection parasitic and
interconnection interference.
The protection constraints can be used for preventing the critical devices or critical device
groups interfered electrically by others, such constraints can be gotten from the previous
signal path tracing and matching device exploration method.
The signal path and sequence constraints for direct current paths can be used for
minimizing the interconnection parasitic on signal path to ensure the circuit frequency
performance while layout design and optimization, and such constraints can be gotten from
the signal path tracing method.
The direct current path and power reaching sequence constraints for group of devices can be
used for minimizing the interconnection parasitic on direct current path so as to reduce the
dc operation point variation due to parasitic on such path and ensure the DC performance
while layout design and optimization, and such constraints can be gotten from the direct
current path tracing method.
6. Analog schematic generation

6.1 Symbol generation based on functionality
Generating the cell/block symbol based on its functionality includes the following sub-
operations: determining the symbol pattern from a symbol shape template based on the
functionality of the cell/block; determining the port terminal pattern for each port terminal
symbol based on its port type; determining the side location for each port terminal symbol
based on its port type; determining the sequence of the ports on each side based on the port
terminal attribute; and determining the exact location for each port terminal pattern.
6.2 Symbol placement based on functionality

Determining the placement of the symbols of the devices, the ports, and the cells/blocks
includes the following sub-operations: determining the placement of device symbols for the
devices in the DC path; binding the placement of device symbols for the devices in the DC
path as virtual block; determining the placement of the virtual blocks for the DC paths;
tuning the placement for the device symbols; and placing the port terminal symbols.
In the operation of determining the placement of device symbols for the devices in the DC
path, the symbols in a direct current path must be placed from up to down associated with
the current flow direction (POWER to GROUND), which is identified with the direct current
path analysis, the associated dummy devices and protection devices are also placed closing
to the corresponded device symbols, and also symmetry requirement in a DC path is
followed in this operation.
In the operation of binding the placement of device symbols for the devices in the DC path
as virtual block, a DC path (including the associated dummy devices and protection
devices) is regarded as a virtual block, and a rectangle is used as its symbol.
In the operation of determining the placement of the virtual blocks for the DC paths, the
virtual block symbol placement is based on the signal reaching level which is determined by
signal reaching level analysis step, and the virtual block is placed with signal reaching level
incremental order from left to right. DC path symmetry requirements are also followed by
specifying the symmetry axis and put make the virtual blocks of the symmetry pair
mirrored with it to each other.
In the operation of tuning the placement for the device symbols, fine tuning includes: tuning
the powered devices on the same top horizontal grid line; tuning the grounded devices on
the same bottom horizontal grid line; tuning the MOSCAP devices direction for bridging
source net and POWER/GROUND net; tuning the matching device symbols from the
current mirror/source pair to make all the associated gate terminals on the same horizontal
grid line; mirroring the diode-connected device symbol of current mirror/source; and
tuning the rotation status and location of the symbol for the devices(no DC current)
bridging between DC paths for shortest wiring length.
The block symbols in a cell are placed with the signal path folding minimized and the total
wiring length minimized, and also parallel stages must be followed.
In the operation of placing the port terminal symbols, the port terminal placement is
executed as: determining the side location for each port terminals based on the port type
with IN on left side, OUT on right side, VCC on top side, and GND and VSS on down side;
determining the port order(top to down on left and right sides, left to right for top and
down sides for each side); binding the differential nets and bus nets in neighboring
sequence; and tuning the exact location for wiring length minimized.
6.3 Wiring based on functionality

In the operation of wiring for schematic, the wiring includes the special wiring, wiring in a
direct current path, wiring between neighboring direct current paths, wiring among
multiple direct current paths, and wiring among cell instances and blocks.
Special wirings include the wiring for the net among differential devices and tail current
devices, the wiring for differential net pair, the wiring for the bus/bundle nets, the wiring
among current mirror and current source devices, the wiring for dummy devices, the wiring
for MOSCAP devices, the wiring for cross link between two DC paths, the wiring for
dummy devices, the wiring for the protection devices, the wiring for the bridging devices,
and the wiring for POWER and GROUND nets.
Wiring in a DC path includes the major vertical wiring with high weight and the minimum
horizontal wiring with low weight as transition only.
Wiring between the neighboring DC paths includes the major horizontal wiring with high
weight and the minimum vertical wiring with low weight for transition only.
Wiring among DC paths is similar with the wiring between neighboring DC paths, the most
difference is that the wiring needs to take the wiring overlapping the device symbol and
other wiring cross-points into account. For this reason, a line exploration algorithm can be
used with device symbol and other wiring cross-points handled as the obstacles with safety
halos.
Wiring among cell / block instances is similar with the wiring among DC paths, the most
difference is that both horizontal and vertical wiring have the same possible occurrence, so
they have the same weights in the cost of wiring.
6.4 Constraint identification

After the placement and the wiring processes, the identification on the schematic is
necessary to make the circuit engineer and the layout engineer have a good insight on the
design for circuit optimization, floorplanning and layout optimization. The identification
includes the identification of the symmetry requirements in a DC path, the identification of
the device matching requirements among DC paths, the identification of the symmetry
requirements between DC paths, the identification of the dummy devices, the identification
of the protection devices and the associated protected devices, the identification of the
MOSCAP devices, the identification of the critical signal nets, and the identification of the
net current and net wiring width.
All the identification contents are generated by structural feature based circuit functionality
analysis and partitioning engine.
7. Analog-aware schematic synthesis with companion circuits

The professional designers have a very good thumb of rules on drawing analog circuit
schematic, and it is necessary to follow such rules to make circuit schematic more analog-
aware while drawing the new analog circuits, especially in the case of analog schematic
synthesis, such a very good thumb of rules can be dug out from the companion circuits,
which were drawn before by the professional designers in hand. Also, such analog-aware
schematic synthesis with companion circuits is very useful to analog migration between
different technologies, which is very common in analog design due to the integrated-circuit
technology progress.
Analog-aware schematic synthesis with companion circuits accepts the new circuit netlist,
and the companion circuit schematic, mainly goes through such three steps: rule extraction
from companion circuit schematic, rule extraction for new circuit, and rule application for
new circuit schematic synthesis, and output the new circuit schematic in last, as shown in (a)
of Fig. 55.
7.1 Rule extraction from companion circuit schematic

Rule extraction from companion circuits accepts the companion circuit schematic, mainly
goes through the five steps: pre-processing, tracing direct current paths, tracing signal flow
paths, exploring structural features, and exploring schematic rules from companion circuit
schematic, and outputs the schematic rules for companion circuits, as shown in (b) of Fig. 55.
To leverage the schematic rules for new circuit as possible, rule extraction from companion
circuits should cover group device level, direct current path level, block level, and more
high level.
New circuit Companion Companion

netlist circuit circuit
Pre-Processing
Rule extraction from companion circuits
Tracing direct current paths
Tracing signal flow paths

Rule extraction for new circuit
Exploring structural features
Rule application for schematic synthesis Exploring schematic rules

from companion circuit
New Schematic rules for

circuit companion circuits
(a) (b)
New Schematic New Schematic rules

circuit rules for circuit for new circuit
netlist companion
Pre-Processing Constraint generation
Tracing direct current paths Constraint merge
Tracing signal flow paths

Symbol generation
Exploring structural features

Symbol placement
Exploring schematic rules
from structural feature analogy
Interconnection wiring
Schematic rules Schematic for

for new circuit new circuit
(c) (d)
Fig. 55. Analog-aware schematic synthesis with companion circuits
7.2 Rule extraction for new circuit

Rule extraction from new circuits accepts the schematic rules from companion circuits and
new circuit netlist, mainly goes through the five steps: pre-processing, tracing direct current
paths, tracing signal flow paths, exploring structural features, and exploring schematic rules
from structural feature analogy, and outputs the schematic rules for new circuits, as shown
in (c) of Fig 55.
Most of the steps are same as previous descriptions except exploring schematic rules from
structural feature analogy. Exploring schematic rules from structural feature analogy can be
done on device level, direct current path branch level, direct current path level, block level
and more high level, and in procedure the exploration should be started from low level
structural feature comparison to high level structure feature comparison.
If a group of devices in new circuit has the same structural feature as a group of devices in
companion circuits, the schematic rules for the group of devices in companion circuits will
be copied for the group of devices in the new circuit.
If a direct current path in new circuit has the same structural feature as a direct current path
in companion circuits, the schematic rules for the direct current path in companion circuits
will be copied for direct current path in the new circuit.
If a block in new circuit has the same structural feature as a block in companion circuits, the
schematic rules for the block in companion circuits will be copied for block in the new circuit.
If a new circuit has the same structural feature as a companion circuit, the schematic rules
for the companion circuit will be copied for the new circuit.
7.3 Rule application for new circuit schematic synthesis

Rule application for new circuit schematic synthesis accepts the net circuit netlist and the
schematic rules for new circuit, mainly goes through the five steps: constraint generation, merge
constraints with schematic rules, symbol generation, symbol placement, and interconnection
wiring, and outputs the schematic for new circuits, as shown in (d) of Fig 55.
Symbol generation includes the shape of symbols and the side location and side sequence
for each terminal pin-out, which should refer that of companion circuits if the identical
structural feature is found from the companion circuits, so the program needs to make a
comparison for checking out the functional matching relations for circuits and the
corresponding relation for terminal-to-terminal between new circuit and companion circuit.
The symbol placement includes the relative position, mirroring, rotating, symmetry, and
alignment rules, which should refer that of companion circuits if the identical structural
feature is found from the companion circuits, so the program needs to make a comparison
for checking out the functional matching relations for circuits and the corresponding
relation for device-to-device and block-to-block between new circuit and companion circuit.
The interconnection wiring includes the net self-symmetry, the net pair symmetry, and
quasi-bus wiring, which should refer that of companion circuits if the identical structural
feature is found from the companion circuits, so the program needs to make a comparison
for checking out the functional matching relations for circuits and the corresponding
relation for net-to-net between new circuit and companion circuit.
8. Experiments
We test the analog circuit schematic synthesis method with a flattened DAC circuit. After
the functionality analysis and partitioning, new hierarchy is re-constructed; the constraints
for schematic generation, circuit and layout optimization are generated; and also the
schematics are generated from the new hierarchy design, port types, and constraints. Part of
the hierarchical design schematic is shown as in Fig. 56 – Fig. 59; the analog structural
features can be got from the schematics intuitively.
The top circuit schematic is shown in Fig. 56, the top circuit is a digit-to-analog converter
circuit, which consists of two op-amp circuits, one band-gap circuit, one bias circuit, and one
DAC-core circuit. In this schematic, good layout symbols are generated, especially for op-
amp, and the symbol placement follows the signal flow clearly, which gives an intuitive
requirement on future floor-planning.
The DC-core circuit schematic is shown in Fig. 56, where the devices in a DC path are placed
from top to down; all the DC paths are aligned; T-ladder circuit can be captured intuitively;
the power down circuit (two inverters) are shown clearly; and mos-cap devices can be got
from the power line directly. All those give a better feeling for the requirements of device
placement in layout stage.
The op-amp circuit schematic is shown as Fig. 58, where the symmetry for differential pair
devices, load devices, and tail current devices (self-symmetry) is reflected correctly; DC
paths are also shown clearly and DC paths are placed with signal flow followed. All those
give a better feeling for the requirements on symmetry, dc connection wiring minimization,
signal wiring minimization, and necessary protections of the op-amp circuit in layout stage.
The band-gap circuit schematic is shown in Fig. 59, where the devices in a DC path are
placed from top to down; the quasi-symmetry between two band-gap branches is followed;
the power-down control logic circuits (two inverters) can be got from the schematic clearly;
and the power-connected mos-cap devices and the ground-connected mos-cap devices can
be got from the power line and ground line directly.
For clearness on circuit schematic, part of the constraints is not displayed, and due to the
page number limitation, the non-analog-aware circuit schematic generation results from
NLview and Cadence for this test case is not presented here, no any analog functionality are
reflected there correctly.
Fig. 56. Schematic of DAC

Fig. 57. Schematic of OPAMP
Fig. 58. Schematic of DAC-core

Fig. 59. Schematic of BANDGAP
9. Summary
Functionality analysis and partitioning technique can determine the functionality of analog
design accurately and partition it into functionality-based hierarchy; further template based
constraint generation can produce the constraints for schematic synthesis, circuit sizing,
floor-planning, and layout optimization. With leverage of them, a novel analog schematic
synthesis flow can produce analog-aware circuit schematics with functionality and
structural features highlighted, also analog constraints are identified on schematic for circuit
sizing, floor-planning, and layout optimization, which can be work as one of the base of
analog synthesis to bridge topology synthesis and synthesis of circuit, floor-planning, and
layout.
10. Reference
[1] Paul R. Gray, et al, “Analysis and design of analog integrated circuits”, 4th edition, 2001.
[2] Behzad Razavi, “Design of analog CMOS integrated circuits”, 2001.
[3] Phillip Allen, “CMOS analog circuit design”, 2nd edition, 2002.
[4] Bemardinis, F.; Sangiovanni Vincentelli, A.; "Efficient analog platform characterization
through analog constraint graphs", IEEE ICCAD-2005, pp415-421, Nov. 2005
[5] Malavasi, E.; Charbon, E.; Felt, E.; Sangiovanni-Vincentelli, A.; "Automation of IC layout
with analog constraints", IEEE Trans. On CAD, vol. 15, no. 8, pp923 - 942, Aug. 1996
[6] Yiu-Cheong Tam; Young, E.F.Y.; Chu, C.; "Analog Placement with Symmetry and Other
Placement Constraints", IEEE ICCAD-2006, pp349 - 354, Nov. 2006
[7] Jiayi Liu; Sheqin Dong; Xianlong Hong; Yibo Wang; Ou He; Goto, S.,"Symmetry
constraint based on mismatch analysis for analog layout in SOl technology", ASP-
DAC 2008, pp772 - 775, Mar. 2008
[8] Concept Engineering, "Nlview' Widgets: Customizable Schematic Generation Engines for
EDA Tools"
[9] Wei-Ting Chen, Wen-Tsong Shiue, "Circuit schematic generation and optimization in
VLSI circuits", The Proceedings of IEEE Asia-Pacific Conference on Circuits and Systems
2004, vol. 1, pp553 - 556, Dec. 2004
[10] Yuping Wu, "Research Reports on Analog Synthesis", unpublished.
[11] Graeb, H.; Zizala, S.; Eckmueller, J.; Antreich, K. “The sizing rules method for analog
integrated circuit design”， IEEE/ACM International Conference on ICCAD
2001，pp 343 – 349, 2001.
[12] Yuping Wu, “Novel method of analog circuit schematic synthesis”, IEEE 8th
International Conference on ASIC, pp1209-1212, 2009.
[13] Massier, T.; Graeb, H.; Schlichtmann, U.. “The Sizing Rules Method for CMOS and
Bipolar Analog Integrated Circuit Synthesis”, IEEE Transactions on Computer-
Aided Design of Integrated Circuits and Systems, Volume: 27, Issue: 12, pp2209 –
2222, 2008.
[14] Pengfei Zhang, Xisheng Zhang, and Yuping Wu, “Signal flow driven circuit analysis
and partitioning technique”, United States Patent 7448003.
[15] Balasa, F.; Maruvada, S.C.; Krishnamoorthy, K.; “On the exploration of the solution
space in analog placement with symmetry constraints”, Computer-Aided Design of
Integrated Circuits and Systems, IEEE Transactions on Volume 23, Issue 2, Feb.
2004 Page(s):177 - 191
[16] Koda, S.; Kodama, C.; Fujiyoshi, K.; “Linear Programming-Based Cell Placement With
Symmetry Constraints for Analog IC Layout”, Computer-Aided Design of
Integrated Circuits and Systems, IEEE Transactions on Volume 26, Issue 4, April
2007 Page(s):659 - 668
[17] Changxu Du; Yici Cai; Xianlong Hong; Qiang Zhou; “A shortest-path-search algorithm
with symmetric constraints for analog circuit routing”, ASIC, 2005. ASICON 2005.
6th International Conference On Volume 2, 24-0 Oct. 2005 Page(s):844 - 847
[18] Qiang Ma,; Young, Evangeline F. Y.; Pun, K. P.; “Analog placement with common
centroid constraints”, Computer-Aided Design, 2007. ICCAD 2007. IEEE/ACM
International Conference on 4-8 Nov. 2007 Page(s):579 - 585
[19] Koca, O.; Karl, H.; Weigel, R.; “A Novel Method Based Upon Nonlinear Optimization
for Analog Filter Design with Mask Constraints”; Signals, Systems and Electronics,
2007. ISSSE '07. International Symposium on July 30 2007-Aug. 2 2007 Page(s):9 - 12
[20] Koca, O.; Karl, H.; Weigel, R.; “A New Approach for Analog Filter Design with Mask
Constraints Utilizing Linear Programming”; Signals, Systems and Electronics, 2007.
ISSSE '07. International Symposium on July 30 2007-Aug. 2 2007 Page(s):5 - 8
[21] Dhanwada, N.R.; Nunez-Aldana, A.; Vemuri, R.; “Component characterization and
constraint transformation based on directed intervals for analog synthesis”, VLSI
Design, 1999. Proceedings. Twelfth International Conference On 7-10 Jan. 1999
Page(s):589 - 596
[22] Schwencker, R.; Eckmueller, J.; Graeb, H.; Antreich, K.; “Automating the sizing of
analog CMOS circuits by consideration of structural constraints”, Design,
Automation and Test in Europe Conference and Exhibition 1999. Proceedings 9-12
March 1999 Page(s):323 - 327
[23] Yiu-Cheong Tam; Young, E.F.Y.; Chu, C.; “Analog Placement with Symmetry and Other
Placement Constraints”, Computer-Aided Design, 2006. ICCAD '06. IEEE/ACM
International Conference on 5-9 Nov. 2006 Page(s):349 - 354
[24] Naiknaware, R.; Fiez, T.; “CMOS analog circuit stack generation with matching
constraints”, Computer-Aided Design, 1998. ICCAD 98. Digest of Technical Papers.
1998 IEEE/ACM International Conference on 8-12 Nov 1998 Page(s):371 - 375
[25] Mogaki, M.; Kato, N.; Shimada, N.; Yamada, Y.; “A layout improvement method based
on constraint propagation for analog LSI's”, Design Automation Conference, 1991.
28th ACM/IEEE June 17-21, 1991 Page(s):510 – 513.
[26] Donzelle, L.-O.; Dubois, P.-F.; Hennion, B.; Parissis, J.; Senn, P.; “A constraint based
approach to automatic design of analog cells”, Design Automation Conference,
1991. 28th ACM/IEEE June 17-21, 1991 Page(s):506 - 509
[27] Felt, E.; Charbon, E.; Malavasi, E.; Sangiovanni-Vincentelli, A.; “An efficient
methodology for symbolic compaction of analog ICs with multiple symmetry
constraints”, Design Automation Conference, 1992. EURO-VHDL '92, EURO-DAC
'92. European 7-10 Sept. 1992 Page(s):148 - 153
[28] Malavasi, E.; Charbon, E.; Felt, E.; Sangiovanni-Vincentelli, A., “Automation of IC
layout with analog constraints”, Computer-Aided Design of Integrated Circuits
and Systems, IEEE Transactions on
Volume 15, Issue 8, Aug. 1996 Page(s):923 - 942
[29] De Bernardinis, F.; Sangiovanni Vincentelli, A.; “Efficient analog platform
characterization through analog constraint graphs”, Computer-Aided Design, 2005.
ICCAD-2005. IEEE/ACM International Conference on 6-10 Nov. 2005 Page(s):415 -
421
[30] Fernanda Gusmão de Lima, Marcelo de O. Johann, José Luís Güntzel, Luigi Carro,
Ricardo Reis, “A tool for analysis of universal logic gates functionality”, Integrated
Circuits and Systems Design, 1999. Proceedings. XII Symposium on 29 Sept.-2 Oct.
1999 Page(s):184 – 187
[31] Choudhury, U.; Sangiovanni-Vincentelli, A.; “Automatic generation of parasitic
constraints for performance-constrained physical design of analog circuits”,
Computer-Aided Design of Integrated Circuits and Systems, IEEE Transactions on
Volume 12, Issue 2, Feb. 1993 Page(s):208 – 224.
[32] Zhe Zhou; Sheqin Dong; Xianlong Hong; Qingsheng Hao; Song Chen; “Analog
constraints extraction based on the signal flow analysis”; ASIC, 2005. ASICON
2005. 6th International Conference On Volume 2, 24-0 Oct. 2005 Page(s):825 - 828
[33] Jiayi Liu; Sheqin Dong; Fei Chen; Xianlong Hong; Yuchun Ma; Di Long; “Symmetry
Constraint for Analog Layout with CBL Representation”, Solid-State and Integrated
Circuit Technology, 2006. ICSICT '06. 8th International Conference on 23-26 Oct.
2006 Page(s):1760 - 1762
[34] Dhanwada, N.R.; Nunez-Aldana, A.; Vemuri, R.; “Hierarchical constraint
transformation using directed interval search for analog system synthesis”, Design,
Automation and Test in Europe Conference and Exhibition 1999. Proceedings 9-12
March 1999 Page(s):328 - 335
[35] Choudhury, U.; Sangiovanni-Vincentelli, A.; “Constraint generation for routing analog
circuits”, Design Automation Conference, 1990. Proceedings., 27th ACM/IEEE 24-
28 June 1990 Page(s):561 - 566
[36] Charbon, E.; Malavasi, E.; Sangiovanni-Vincentelli, A.; “Generalized constraint
generation for analog circuit design”, Computer-Aided Design, 1993. ICCAD-93.
Digest of Technical Papers., 1993 IEEE/ACM International Conference on 7-11
Nov. 1993 Page(s):408 - 414
[37] Zhe Zhou; Sheqin Dong; Xianlong Hong; Qingsheng Hao; Song Chen, “ Analog
constraints extraction based on the signal flow analysis”; ASIC, 2005. ASICON
2005. 6th International Conference On
Volume 2, 24-0 Oct. 2005 Page(s):825 – 828
[38] Kumar Arya, Swaminathan Misra, “Automatic Generation of Digital System Schematic
Diagrams”, Design Automation 1985. 22nd Conference on 23-26 June 1985
Page(s):388 – 395.
[39] Kumar Arya, Swaminathan Misra, “Automatic Generation of Digital System Schematic
Diagrams”，Design & Test of Computers, IEEE Volume 3, Issue 1, Feb. 1986
Page(s):58 – 65.
[40] Swinkels, G.M.; Hafer, L，”Schematic generation with an expert system”, Computer-
Aided Design of Integrated Circuits and Systems, IEEE Transactions on Volume 9,
Issue 12, Dec. 1990 Page(s):1289 – 1306.
[41] Wei-Ting Chen, Wen-Tsong Shiue, “Circuit schematic generation and optimization in
VLSI circuits”, Circuits and Systems 2004 Proceedings. The 2004 IEEE Asia-Pacific
Conference on Volume 1, 6-9 Dec. 2004 Page(s):553 - 556 vol.1.
[42] Kim, C.B., “Multiple mixed-level HDL generation from schematics for ASIC design”,
ASIC Conference and Exhibit, 1991. Proceedings Fourth Annual IEEE International
23-27 Sept. 1991 Page(s):P8 - 2/1-4.
[43] Tzi-Cker Chiueh , “HERESY: a hybrid approach to automatic schematic generation [for
VLSI]”, Design Automation. EDAC. Proceedings of the European Conference on
25-28 Feb. 1991 Page(s):419 – 423.
[44] Lee T.D., McNamee, L.P, “Structure optimization in logic schematic generation”,
Computer-Aided Design, 1989. ICCAD-89. Digest of Technical Papers, 1989 IEEE
International Conference on 5-9 Nov. 1989 Page(s):330 – 333.
[45] Green Andersen, “Automated generation of analog schematic diagrams”, Circuits and
Systems, 1990, IEEE International Symposium on 1-3 May 1990 Page(s):3197 - 3200
vol.4.
[46] Zhan R., Feng H., Wu Q., Chen G., Guan X., Wang A.Z., “A new algorithm for ESD
protection device extraction based on subgraph isomorphism”, Circuits and
Systems, 2002. APCCAS '02, 2002 Asia-Pacific Conference on Volume 2, 28-31 Oct.
2002 Page(s):361 - 366 vol.2
[47] J.R. Ullmann, “An Algorithm for Subgraph Isomorphism,” J. Assoc. for Computing
Machinery, vol. 23, pp. 31-42, 1976.
[48] Luigi P. Cordella, Pasquale Foggia, Carlo Sansone, and Mario Vento, “A (Sub)Graph
Isomorphism Algorithm for Matching Large Graphs”, IEEE TRANSACTIONS ON
PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 26, NO. 10,
OCTOBER 2004, Page(s):1367-1372.
[49] Bilal Radi A’Ggel Al-Zabi, Andriy Kernytskyy, Mykhaylo Lobur, Serhiy Tkatchenko,
“On Graph Isomorphism Determining Problem”, MEMSTECH’2008, May 21-24,
2008, Polyana, Page(s):84.
13
An SQP and Branch-and-Bound Based

Approach for Discrete Sizing of Analog Circuits
Michael Pehl and Helmut Graeb
Technische Universitaet Muenchen
Germany
1. Introduction
Analog circuits form an important part in integrated circuits and in particular in ASICs
(Application Specific Integrated Circuits). However, due to the high complexity, design of
this part has become a bottle-neck in the design flow (Gielen, 2007; Rutenbar et al., 2007). To
overcome this problem and to guaranty that the analog part can be designed in reasonable
time even for future technologies, methods supporting automatic design of analog circuits
must be advanced.
This chapter focuses on sizing of analog circuits. It starts from the point where a topology is
given. The task now is to choose design parameters, e.g., lengths and widths of transistors,
such that certain properties of the circuit are fulfilled.
Current tools to solve the sizing task mostly treat it as a continuous optimization problem
and use, e.g., certain gradient-based approaches to solve the problem in the continuous
domain (Graeb, 2007). However, many design parameters are discrete in reality, e.g.,
transistor multipliers (i.e., the number of transistors connected in parallel), or must be
discretized for some practical purposes, e.g., transistor lengths and widths which should
match to a manufacturing grid. Furthermore, for some future technologies as, e.g., FinFETs
(Knoblinger et al., 2005), the transistor parameters must fulfill certain geometrical properties,
and accordingly have to be discrete.
Considering discrete parameters, it is not sufficient to treat the sizing task as a continuous
optimization problem and rounding the result. This can be followed from mathematical
theory, where it is shown that continuous optimization with sub-sequent rounding might not
solve the original discrete optimization task (i.e., a optimization task that considers discrete
and continuous parameters) and leads to a suboptimal result (Li & Sun, 2006; Nemhauser &
Wolsey, 1988). This can be confirmed by experiments.
To solve discrete optimization problems, statistical and evolutionary approaches have been
proposed (Alpaydin et al., 2003; Cao et al., 2000; Gielen et al., 1990; Ochotta et al., 1996; Phelps
et al., 2000; Somani et al., 2007). However, for practical approaches these tools are usually
more slowly in comparison to deterministic gradient-based tools if a good initial solution
can be given for the task (what is normally true for analog sizing). Even if statistical and
evolutionary approaches might be the first choice if a global search is necessary, for many
cases deterministic gradient-based approaches are more suitable. Deterministic approaches
for discrete sizing of analog circuits have barely been published till today (Pehl & Graeb, 2009;
Pehl et al., 2008). In this chapter a new deterministic gradient-based approach is presented. It
consists of Sequential Quadratic Programming and Branch-and-Bound.

For the approach in this chapter, the problem is sub-divided into a non-linear program (NLP)
and a discrete program (DP). Afterward, a discrete program is modeled by a discrete quadratic
program (DQP) to speed up the algorithm.
Before the algorithm is presented, it is shown in Section 2 how the task of analog sizing can be
formulated as a discrete minimization program. The task is said to be solved if any parameter
set is found, where sizing constraints as well as performance specifications are fulfilled.
Introducing a relaxation of the parameters (i.e., all parameters are considered to be
continuously scalable), a non-linear, but continuous sub-problem can be defined, called the
relaxed program. To solve this NLP, in Section 3.1 of the chapter a sequential quadratic
programming (SQP) algorithm is introduced.
Obviously, the result of the relaxed program is a point in the relaxed - i.e., continuous -
domain. So, in Section 3.2 of the chapter a Branch-and-Bound approach is introduced to find
a discrete solution to the sizing task.
The algorithm based on SQP and Branch-and-Bound can be used to solve the discrete sizing
problem. However, to improve the run time of the approach, in Section 3.3 of the chapter a
modification to speed up the algorithm is described. In the modification, the quadratic model
of the objective function - which is computed in the SQP algorithm - is used to get a discrete
quadratic model of the original sizing task. By solving the discrete quadratic program a
discrete point can be found which gives an approximation for the obtainable discrete solution.
This approximation can be used to cut non-promising parts of the Branch-and-Bound tree and
to speed up the algorithm.
Experimental results in Section 4 show that in contrast to continuous optimization with
subsequent rounding the presented approach is able to find a discrete feasible solution in each
test case. Furthermore, it can be seen that the modification described in Section 3.3 decreases
the run time of the algorithm significantly without reducing the result quality.
Section 5 concludes and gives an outlook to future research.
2. Problem formulation
2.1 Sizing task
In the analog sizing step appropriate values for the design parameters d of a given topology
must be computed such that certain properties of the circuit are fulfilled. Typical design
parameters are, e.g., lengths and widths of transistors, which were normally considered
as continuous scalable in previous gradient-based approaches. However, in reality most
parameters in the circuit sizing step are discrete, e.g., due to manufacturing grids, due to
modern transistor types as FinFETs, or due to properties from the layout step.
For the approach presented in this chapter, the sizing task is formulated as a discrete
optimization task, i.e., a sizing task considering scalable discrete and continuous parameters.
For this purpose the vector of design parameters d can be subdivided into three parts
corresponding to different parameter classes:
1. Continuous parameters d c are used to model design parameters which do not require the
consideration of any grid and which lie in an Nc -dimensional domain D Nc that is bounded
by any upper bound d c,U and any lower bound d c,L
d c ∈ D Nc = {d | d c,L ≤ d ≤ d c,U } (1)

An SQP and Branch-and-Bound Based Approach for Discrete Sizing of Analog Circuits 299
2. Scalable discrete parameters d d are used to model design parameters which can only lie
on a – not necessarily uniform – Nd -dimensional grid. These parameters d d are subset of a
domain D Nd :
Nd
d d ∈ D Nd = ×D i (2)
k=1
D i is a set corresponding to the i-th discrete parameter di . Furthermore, D i is ordered by a
relation < (Pehl & Graeb, 2009). Assuming n i discrete parameter values for parameter di ,
the ordered set can be formulated as:

di ∈ D i : = (D, <) = di,1 , ..., di,k , ..., di,ni
∀ di,k < di,k+1

(3)
1 ≤ k < ni
3. Non-scalable discrete parameters d x can be used to consider design options which can not
be expressed by a scalable parameter and must be enumerated instead, e.g., the exchange
of different technologies. This class of parameters is non-numerical in many cases. One
way to consider this class of parameters, which fits to the approach presented in this
chapter, is to define binary surrogate parameter for each design option. Assuming n i
discrete design options di,1 , ..., di,ni for a parameter di , i.e.,

di ∈ D i : = di,1 , ..., di,k , ..., di,ni (4)
the values are collected in a vector d x,i:

T
d x,i = di,1 , ..., di,k , ..., di,ni (5)
Additionally the vector d b,i of surrogate design parameters is defined as:

T
d b,i = db,1 , ..., db,k , ..., db,ni
(6)
1; when option di should be chosen
db,k =
0; otherwise
Thus, the vector of surrogate design parameters can be mapped to the value of the
corresponding non-scalable discrete parameter di by
di = d b,i
T
d x,i (7)
To avoid that different options are chosen for the same parameter, an additional constraint
must be added to the optimization task defined below for each non-scalable discrete
parameter:
d bT d b = 1 (8)
The set of all binary variables corresponding to options for non-scalable parameters is
assigned as D Nb .
In this chapter only continuous and scalable discrete parameters are used. However, using
the binary surrogate parameters defined above, the approach in Section 3 can be applied
accordingly.
For continuous parameters d c , and scalable discrete parameters d d the domain D N of the
design parameters d can be defined as:
d ∈ D N = D Nc × D Nd (9)
Thus, vector d is composed by a continuous part d c and a discrete part d d :

T
d = d dT d cT (10)
The task of choosing a parameter point, such that certain circuit properties are fulfilled, now
is formulated as:
min ϕ(d ) s.t. c(d ) ≥ 0 (11)
d ∈D N
wherein c(d ) are sizing constraints, which ensure a reasonable sizing of the circuit (Graeb
et al., 2001; Massier & Graeb, 2008). ϕ(d ) is the objective function, which maps a multi
objective optimization task to a scalar minimization problem.
The objective function for analog sizing should support improvement of any circuit property
when the specification for a certain performance is fulfilled as well as when the specification
is violated. To build up such a function, an error ε(d ) for each performance f i (d ) is defined,
which is the normalized distance from the current performance value to the specification
bound f B,i of the performance:
f i (d ) − f B,i
ε(d ) = (12)
f N,i
f N,i is a normalization factor which ensures that the values for all performances are
comparable. Without loss of generality it can be assumed that f B,i is a lower bound for the
performance such that
ε(d ) ≥ 0 when specifications are fulfilled

(13)
ε(d ) < 0 when specifications are not fulfilled
This is illustrated in Figure 1. To support improvement of the performances when the

specifications are violated as well as when the specifications are fulfilled, in this approach
an exponential sum of the normalized errors for all N f performances is used:
Nf
ϕ(d ) = ∑ e−ε (d)
i
(14)
i =1
Although the given formulation leads to a Pareto-optimal point, i.e. a solution where one
performance can not be further improved without deteriorating another performance, we
choose to stop the optimization problem – and consider the sizing task solved – as soon as
a point is found which fulfills all specifications. Thus, the minimization is stopped as soon as
a point is found with:
∀ ε(d ) ≥ 0 (15)
i = 1, ..., N f
dcont specification
fulfilled
ε1 ≥ 0
ε1 < 0
dcont specifications
d1 d2 d3 ddisc fulfilled
performance 1
ϕ(d )
dcont
specification
d1 d2 d3 ddisc
fulfilled
scalar task
ε2 ≥ 0
ε2 < 0
d1 d2 d3 ddisc
performance 2
Fig. 1. Sizing task with two performances, one discrete parameter ddisc ∈ {d1 , d2 , d3 }, and one
continuous parameter dcont is mapped to a scalar optimization task by ϕ(d ).
2.2 Relaxation
To set up the relaxation of a discrete optimization task, the domain for each discrete parameter
is replaced by a continuous domain. As the domain for the discrete parameters in (3) can be a
ordered, the lower bound di,L and upper bound di,U for a discrete parameter di can be defined
as the first and the last element of the ordered set D i
di,L : = di,1 and di,U : = di,ni (16)
For all discrete parameters, the lower bounds can be collected in a vector d d,L

T
d d,L = ...di,1... (17)
and, respectively, the upper bounds can be collected in a vector d d,U

T
d d,U = ...di,ni ... (18)
Thus, a vector of lower and upper bounds for discrete and continuous parameters can be built
up:
d TL = d d,L
T T
d c,L T
; dU = d d,U
T T
d c,U (19)
For the relaxed optimization task all parameter points must be in the domain
N
Drel = { d | d L ≤ d ≤ dU } (20)
and the relaxed program can now be defined as:
min ϕ(d ) s.t. c(d ) ≥ 0 (21)

d ∈D rel
N
The relaxation of a problem is illustrated in Figure 2.

Obviously, the discrete parameter set D N is a subset of its relaxation Drel
N and
dcont dcont
D2 D2rel
dU specifications specifications
fulfilled dU
fulfilled
relaxation
dL dL
d1 d2 d3 ddisc d1 d3 ddisc
discrete program relaxed program
Fig. 2. Relaxation of the parameter domain D2 (see Figure 1) into the domain D2rel
D N = D N ∩ Drel
N
(22)
must be true.
The evaluation of the circuit performances in this approach is done by simulations. Thus, in
the rest of this chapter it is assumed, that – even if parameters are discrete – simulation of the
circuit is possible for each continuous point. In the future work the algorithm will be extended
to use exclusively simulation results from discrete points.
3. Discrete sizing approach

3.1 Sequential Quadratic Programming
In the approach, presented in this paper, the relaxed optimization problem in (21) is solved
by a Sequential Quadratic Programming (SQP) approach (e.g., (Nocedal & Wright, 1999)).
SQP converts a constrained nonlinear optimization problem in the continuous domain into a
sequence of unconstrained quadratic programming problems.
Using a vector of Lagrange multipliers λ, the Lagrangian function of the problem can be given
as
L(d, λ) = ϕ(d ) − λ T c(d ) (23)
and the optimization problem (21) can be reformulated as the unconstrained optimization
task:
min L(d, λ) (24)
d,λ
From the first-order necessary optimality conditions for unconstrained optimization, it
follows that ∇d,λ L(d, λ) =0 must hold true in the optimum point. Thus, to find the optimum
of the Lagrangian function, Newtons method can be applied to the gradient ∇d,λ L(d, λ).
Using Δd and Δλ for the change in parameters and μ as a linearisation index, the equation
system which must be solved can be given as:
⎡ ⎤
∇2dd ϕ(d ( μ)λ( μ) ) − ∑ λi,0 ci d ( μ) λ( μ) −∇d c T Δd ( μ+1) −∇d L(d ( μ) λ( μ) )
⎣ i ⎦ · =
−∇ c 0 Δλ( μ+1) c(d ( μ) )
d
(25)
or with H = ∇2dd ϕ(d ( μ) ), J = ∇d c(d ( μ) ), Ci = ∇2dd ci (d ( μ) ), and g = ∇d ϕ(d ( μ) ).

H − ∑ λi,0 Ci −JT Δd −g
i · = (26)
J 0 Δλ c ( d0 )
Algorithm 1: Branch and Bound(d inc ,D N , D̂rel

N)
Input: incumbent d inc , parameter domain D N , relaxed domain D̂rel

N
//compute minimum for given relaxed domain

d ∗ = arg min ϕ(d ) s.t.: c(d ≥ 0) 1
d ∈D̂ rel
N
if No feasible solution in D̂rel

N then 2
// pruning by infeasibility
return dinc 3
else if ϕ(d ∗ ) ≥ ϕ(d inc ) then 4
// pruning by value dominance
return dinc 5
else if d ∗ ∈ D N and ϕ(d ∗ ) < ϕ(d inc ) then 6
// pruning by optimality
return d∗ 7
else 8
// branching
Choose parameter di for branching 9

N d ≤ d∗

D̂ down
N = d ∈ D̂rel 10

N = d ∈ D̂ N d ≥ d∗
i i
D̂ up rel i i 11
d inc =Branch and Bound(d inc , D N ,D̂ down

N ) 12
d inc =Branch and Bound(d inc , D , D̂ up

N N ) 13
return dinc 14
end 15
If the second derivative of the Lagrangian function is convex, the optimization problem in
(24) can be solved by iteratively solving the equation system in (26). The result describes
the direction from the current point to the minimum of the quadratic model of the objective
function subject to the constraints. In the SQP approach a model of the matrix is usually
built up iteratively. This can be realized by different approaches, e.g., BFGS (Broyden Fletcher
Goldfarb Shanno) update formula, which is used in this approach.
After computing the direction by solving (24), a step size is computed at the original relaxed
program using line search. In this approach a Wolfe Powell step size algorithm is used.
The solution which is computed by SQP on the relaxed program is obviously no discrete
feasible point in general. In the next section of this chapter a Branch-and-Bound method is
described, which can be used to find a discrete solution for the original sizing task based on
the solution of the relaxed problem.
3.2 Branch and Bound

Branch and Bound (e.g., (Nemhauser & Wolsey, 1988)) is one of the most popular approaches
in discrete optimization. In the form which is used in this work, it decomposes the discrete
optimization task in a sequence of relaxed optimization tasks which are nonlinear but can be
solved in the continuous domain. A description of the recursive method is given in Algorithm
1 and in the following.
The algorithm is primarily based on two principles:
1. As the domain of discrete points is a subset of its continuous relaxation (22), the optimum
of the relaxed program is better than or equal to the continuous solution. For the
minimization in (11) and with (21):

min ϕ(d ) s.t. c(d ) ≥ 0 ≤ min ϕ(d ) s.t. c(d ) ≥ 0 (27)
d ∈D rel
N d ∈D N
Consequently, the minimum of the relaxed program is a lower bound for the discrete
minimum.
2. Each discrete point with objective function value better than the best discrete point so far,
is an upper bound for the discrete solution of the optimization task.
Initially in each recursion the relaxed optimization task is solved in the current relaxed domain
D̂rel
N (Algorithm 1, line 1 and Figure 3(a)). In the approach presented in this chapter SQP is
used at this point (Section 3.1). Following the first principle, the objective function value at the
minimum of the relaxed task ϕ(d ∗ ) is smaller than or equal to the value at the best discrete
solution in the current sub-domain ϕ(d ∗disc ) , i.e.,

ϕ(d ∗ ) = min ϕ(d ) s.t. c(d ) ≥ 0 ≤ ϕ(d ∗disc ) = min ϕ(d ) s.t. c(d ) ≥ 0 (28)
d ∈D̂ rel
N d ∈D N ∩D̂ rel
N
Thus, even if d ∗disc is not explicitly known at this point, the objective function value at the
optimum of the relaxed sub-problem ϕ(d ∗ ) is a lower bound for the sub-domain.
The minimum d ∗ which is computed for the relaxed optimization task is not necessarily
discrete. Thus, one of the parameters di ∈ D i (3) which must be discretized is chosen, and a
constraint is set on it to be greater than the next higher or smaller than the next lower discrete

value (called
branching). In Algorithm 1 (lines 10, 11) this is assigned by di ≥ d∗i and
di ≤ d∗i , with
d∗i
= max s.t. d < di and d∗i = min s.t. d > di (29)
d ∈D i d ∈D i
Figure 3(b) shows that adding one pair of such constraints can be considered as building up
two new relaxed sub-problems with reduced parameter domain (D̂ up N , D̂ N
down in Algorithm
1 and D̂1 , D̂2 in the example in Figure 3(b)). Typically, this is represented by a branching
2 2
tree (Figure 4): If the parent node of this tree represents the current domain, branching is
equivalent to adding two child nodes which correspond to the subsets D̂ up N and D̂ N
down . The
edges of the branching tree correspond to the constraints, which are added to define the
sub-problems.
For each sub-problem which is set up in the branching step, Algorithm 1 is executed
recursively (Algorithm 1, lines 12, 13). Following the heuristic order in Algorithm 1,
always D̂ down
N is explored before considering D̂ up
N . Thus, in Figure 3 D̂ 2 is explored before
1
considering D̂2 . In the example, after computing the continuous solution of sub-domain D̂21
2
the sub-problem is further branched, as the continuous solution of the sub-problem is not
element of the original discrete domain (Figure 3(c)).
If for each sub-problem branching constraints are added, until the solution of the relaxed
sub-problem is a discrete point and thus a leaf of the search tree is reached, the discrete
solution with the best objective function value is the optimum of the discrete optimization
task. However, without further modification, the computational effort of the method is
d1 Constraint 1 d1
D̂22 d1 ≥ d1∗
Constraint 2
d ∗disc d∗ d∗
better d1 ≤ d1∗
D̂21
d2 d2
ϕ(d)
(a) The solution of the relaxed optimization (b) Branching leads to two sub-sets e.g.,
task d ∗ is a lower bound for the solution of D̂ down
N = D̂21 and D̂ uNp = D̂22 , and
the discrete minimization d ∗disc, which is not two corresponding sub-problems. The
explicitly known at this point of time. parameter which is chosen for branching, is
computed by equation (30).
d1 d1
d∗ d inc = d ∗
D̂23 D̂24 D̂23 D̂24
d2 d2
(c) Following Algorithm 1, the lower (d) Following Algorithm 1, D̂ down

N = D̂23 is
sub-domain D̂21 is considered next. The considered next. The solution of domain D̂23
relaxed solution of the optimization task is the discrete optimum d inc. The region can
in sub-domain D̂21 is non-discrete in be pruned by optimality.
parameter d2 . Thus branching constraints
are added for this parameter and the
sub-domains D̂ down
N = D̂23 , D̂ uNp = D̂24 are
generated.
d1 d1
D̂22
d∗
d inc d inc
D̂23 D̂24
d2 d2
(e) Domain D̂24 is considered next. It does (f) Finally sub-domain D̂22 is explored. The
not include any feasible point and can be continuous optimum in this region has a
pruned by infeasibility. higher value than d inc from Figure 3(d). It
can be pruned by value dominance
Fig. 3. Illustration of Branch and Bound for a two-dimensional optimization task
Continuous solution in relaxed

domain D2rel gives lower
bound for discrete
minimum (Figure ??).
d1 ≤ d1∗
d1 ≥ d1∗
Solution of sub-problem Continuous solution of sub-region

in domain D̂21 is not D̂22 is worse than discrete upper
discrete. Further branching bound d inc. Node can be cut from
is necessary. d2 ≤ d2∗
d2 ≥ d2∗ the search tree
(pruning by value dominance)
Solution in sub-region Sub-region D̂24 does not
D̂23 discrete. No further include any feasible point.
branching necessary Node can be cut from search
(pruning by optimality). tree (pruning by infeasibility).
Fig. 4. Illustration of Branch and Bound from Figure 3 as a branching tree.
extremely high, as many non promising sub-problems must be solved. Thus pruning rules
(e.g., (Nemhauser & Wolsey, 1988)) are introduced to reduce the run time of the algorithm.
The pruning step can be understood as cutting non promising nodes from the search tree.
Three pruning rules can be found:
1. If a relaxed sub-problem does not include any feasible solution, the corresponding node
can be cut from the search tree. For this rule (known as pruning by infeasibility) it is
considered that each discrete domain is subset of its relaxation ( Algorithm 1, line 2, 3, and
Figure 3(e)).
2. If a discrete solution d inc has been found which is smaller than the value of the current
relaxed sub-problem, the node corresponding to the relaxed sub-problem can be cut off.
This pruning by value dominance uses that – due to principle 1 from above – the relaxed
program can not include a discrete point which is better than d inc (Algorithm 1, line 4, 5,
and Figure 3(f)).
3. If the solution of the relaxed program is discrete and better than the best solution found
so far, it is a new best solution (d inc ) for the discrete sizing task. However, at the same
time it is a lower bound for the corresponding relaxed sub-problem which can not include
any better point. Thus, no further branching is necessary in the sub-region. This is called
pruning by optimality (Algorithm 1, line 6, 7, and Figure 3(d)).
The recursive approach described so far realizes a "Depth-First" search. For branching always
the most fractional parameter is used (most fractional or most infeasible branching, e.g.,
Achtenberg et al. (2005)), i.e., assuming an index i = 1, ..., Nd for the discrete parameters di
with value d∗i , the branching index i is chosen by

d∗i − d∗i

i = arg max 0.5 − ∗ (30)
i =1,...,Nd di − d∗i

However, some problem specific properties can be used to speed up the algorithm in case of
analog sizing. This is described in the following sub-section.
Algorithm 2: Modified Branch and Bound(d inc ,D N , D̂rel

N)
Input: incumbent d inc , parameter domain D N , relaxed domain D̂rel

N
if d inc solves the sizing task then 1

// stop due to pruning rule 2’
return dinc 2
end 3
Run SQP (Section 3.1) on optimization problem 4
min ϕ(d ) s.t.: c(d ≥ 0)

d ∈D̂ rel
N
until the sizing task is solved or an optimum is found.

Get solution d ∗ and quadratic model from SQP. 5
if d ∗ does not solve the sizing task (i.e., no solution in D̂rel
N ) then 6
// pruning rule 1’
return dinc 7
else if d ∗ ∈ D N (i.e., d ∗ solves the problem) then 8
// pruning rule 2’
return d∗ 9
else 10
// compute incumbent
Get d ∗model by solving the discrete quadratic optimization task (31) using Algorithm 1 11
if d ∗model solves the sizing task then 12
// stop
return d∗model 13
end 14
// branching
Choose parameter
di for
branching 15
Compute d∗i and d∗i according to (34) and (35) 16

N d ≤ d ∗
Set D̂ down
N = d ∈ D̂rel 17

N = d ∈ D̂ N d ≥ d∗
i i
Set D̂ up rel i i 18
d inc =Modified Branch and Bound(d inc , D N ,D̂ down N ) 19
d inc = Modified Branch and Bound(d inc , D N , D̂ up N ) 20
return dinc 21
end 22
3.3 Modification of Branch and Bound

To reduce the computational effort of Branch and Bound, the most promising way is to
improve the pruning and the branching heuristic. Certain properties of the underlying
optimization problem can be used to speed up the process. The modified approach is shown
in Algorithm 2 and explained in the following.
3.3.1 Consideration of the quadratic model

As the continuous solution of the relaxed program is computed by an SQP approach, beside
the continuous solution a quadratic model of the objective function is computed during
Linearisation of
constraint 1
d1 Constraint 1 d1 Linearisation of
constraint 2
Constraint 2
d∗ Quadratic model d∗
of the program
better d ∗model
d2 d2
ϕ(d) quadratic model
Fig. 5. The quadratic model (right) and the continuous solution of the optimization task d ∗
are computed for the original program (left) during SQP. Solving the quadratic model in the
discrete domain gives a discrete solution d ∗model which is an upper bound for the original
program. In this example, d ∗model is equal to the discrete optimum of the original task.
solving the relaxed program. Furthermore, a linear model of the constraints is computed
(Algorithm 2, lines 4, 5). As these models are good local approximations for the relaxed
program, they are also a good local approximation for the discrete approach. Thus a quadratic
optimization task with linear constraints can be formulated as a surrogate optimization task
for (11):
1
min · d T · H · d + g T · d s.t. J · d + c(d0 ) ≥ 0 (31)
d ∈D̂ 2
wherein H and g are the Hessian matrix and the gradient for the objective function at the
solution point of the relaxed program, J is the Jacobian matrix for the constraints and c0 are
the constraint values at this point. D̂ N is the set of discrete points D in a relaxed sub-problem,
i.e.,
D̂ N = D N ∩ D̂ upN
or D̂ N = D N ∩ D̂ down
N
(32)
By solving the program in (31) using the discrete domain of the original task, the discrete
optimum d ∗model for the approximation of the objective function can be found (Figure 5;
Algorithm 2, line 11). Due to the second principle from Section 3.2, the value of d ∗model in
the original objective function – i.e., ϕ(d ∗model ) – is an upper bound for the discrete optimum if
it is feasible for the original task. Consequently, sub-regions with a continuous solution worse
than the discrete optimum of the model can be cut from the search tree. Thus, early pruning
by pruning rule 2 is possible, as – in contrast to standard branch and bound – a discrete upper
bound exists in the first branching node and not after discretizing all parameters, i.e., in the
first leaf of the search tree. This fact is especially important if many discrete parameters exist.
Additionally, solving the quadratic surrogate problem is computational much less expensive,
as no circuit simulations are necessary, which cause the highest time consumption in solving
the sizing task. Thus, the Branch and Bound algorithm from Section 3.2 can be used to solve
the discrete quadratic program with linear constraints in (31).
3.3.2 Consideration of non-optimality

For analog sizing the SQP approach is stopped as soon as any point is found which fulfills
specifications and constraints. Thus, the solution which is found is in general non-optimal
in terms of the objective function. Taking into account that it is a binary decision if a certain
point solves the sizing task or not, the branching rules can be reformulated.
Linearisation of
Specifications Specifications constraint 1
d1 fulfilled Constraint 1 d1 fulfilled Linearisation of
constraint 2
Constraint 2 d∗
model
Quadratic model
−g
of the program
d∗ d∗
d2 d2
ϕ(d) quadratic model
Fig. 6. SQP can be stopped as soon as a continuous point d ∗ is found which fulfills
constraints and specifications (left). At this point the quadratic model (right) is set up. In the
example, the objective function value at the discrete optimum of the quadratic model d ∗model
is better than the value at d ∗ and specifications and constraints are fulfilled at d ∗model .
Assuming that there is at least one discrete solution for the sizing task, obviously only these
sub-domains must be considered during Branch and Bound which include such a point. As
– due to (22) – the discrete solutions must be also in the relaxed domain, all sub-domains
can be cut which do not include a solution of the sizing task in their relaxation. This can be
considered by reformulating pruning rule 1 as:
1’. If a relaxed sub-problem does not include any solution for the sizing task, the
corresponding node can be cut from the search tree (Algorithm 2, lines 6, 7).
If pruning rule 1 is replaced by 1’ the discrete point d inc – which represents a solution
candidate – is only set up, if a discrete solution is found. The Branch and Bound algorithm
can be stopped in this case. Thus, pruning rule 2 (pruning by value dominance) becomes
redundant and can be left out. Pruning rule 1 is reformulated as a stop criterion, to set up
the discrete solution correctly and to avoid insufficient computational effort when the discrete
solution has been found:
3’. If any discrete solution for the sizing task has been found, no further branching is required
(Algorithm 2, lines 1, 2 and 8, 9).
The modifications of the pruning rules have an even stronger influence if the quadratic model
from Section 3.3.1 is considered. In this case, the quadratic model is set up once again in
the point d ∗ which is computed by SQP and solves the sizing task in the relaxed domain.
The point d ∗ can be non-optimal in terms of the objective function and thus in may cases the
solution of the quadratic optimization problem in the relaxed domain is also a better solution
for the underlying sizing task. The continuous solution of the quadratic model is of course
not evaluated by simulation. However, as the quadratic model is set up once again at d ∗ , it
is a locally better approximation of the objective function than the quadratic model used for
the last SQP step. Thus, even the discrete optimum d ∗model of the quadratic problem in (31)
computed by use of the quadratic model at d ∗ is often a better solution for the sizing task than
d ∗ itself (Figure 6).
Hence, in many cases the discrete solution of the model solves the sizing task in the initial
node of Branch and Bound and Branch and Bound can be stopped after computing the discrete
solution of the quadratic model (Algorithm 2, lines 12, 13).
If the initial solution of the quadratic model does not fulfill the specifications, the
non-optimality of the SQP solution d ∗ can also be used to improve the branching heuristic
which has significant influence on the runtime of standard Branch and Bound. The gradient
at a non-optimal point d ∗ is not equal to zero. Thus, it can be assumed that discrete solution
candidates can be found in direction of degression of the objective function. The gradient g at
the solution of the SQP algorithm has been already computed to improve the quadratic model
and comes without additional cost. For the branching heuristic used in this approach, now the
parameter which should be discretized and which corresponds to the gradient component gi
with the strongest influence to the objective function is discretized first (Algorithm 2, line 15).
In the "Depth First" search, then the sub-region is chosen which lies in direction of greatest
improvement (Algorithm 2, line 16), i.e., assuming the next discrete values for the parameter
di in domain D i from (3) are d a and db , with
d a = max s.t. d < di and db = min s.t. d > di (33)

d ∈D i d ∈D i
the rounding operator • and •

in Algorithm 1 is modified such that

d a ; if gi > 0
di = (34)
db ; if gi ≤ 0
and, respectively,
da ; if gi < 0
di
= (35)
db ; if gi ≥ 0
Thus, discrete points in gradient direction are considered first during branch and bound.
4. Experimental results
To show the effectiveness and efficacy of the algorithm, the sizing process of three different
circuits will be presented in this section. For each example, the results and the runtime of SQP
with sub-sequent rounding, of SQP and modified Branch and Bound (BaB) without quadratic
model, and of SQP and modified Branch and Bound considering the quadratic model (Section
3.3) is presented. The modified Branch and Bound algorithm considering the quadratic model
is presented in Section 3.3 Algorithm 2. The modified Branch and Bound algorithm without
the quadratic model is implemented identically, but the consideration of the quadratic model
(lines 11 - 14 in Algorithm 2) is switched off. I.e., both Branch and Bound approaches stop as
soon as a discrete solution for the sizing task is found. Branching in both Branch and Bound
algorithms is realized according to (34) and (35).
The circuit in the first example is the Miller amplifier in Figure 7. For the sizing tasks the
lengths, widths, and multipliers of the transistors are used as discrete parameters. The lengths
of all transistors shall be equal. Furthermore some multipliers and transistor widths (e.g.,
multipliers and widths of the differential pair) are set equal to avoid mismatch effects. For
transistor lengths and widths a 5nm manufacturing grid is assumed. The Miller capacitance is
represented by a continuous parameter. A 0.5pF load capacitance and a 2V supply voltage are
given for the circuit and the 45nm low power predictive technology (PTM; (Balijepalli et al.,
2007; Cao et al., 2000; Zhao & Cao, 2006)) from (Nanoscale Integration and Modelling Group,
Arizona State University, 2008) is used.
The simulated performance values of the amplifier before and after sizing are shown in Table
1. It can be seen from the results, that – as proposed in Section 1 – the continuous optimization
VDD
w5 ,l1 w6 ,l1 w7 ,l1

m5 m6 m7
bias out
w1 ,l1 w1 ,l1 in +
in −
m1 m1
Cc w8 ,l1
m8
w2 ,l1 w2 ,l1
m2 m2 gnd
Fig. 7. Miller Amplifier
Fig. 8. Runtime for up to 8 times parallelized algorithm on a 16 core 2.67GHz computer for
sizing of the Miller amplifier
and subsequent rounding violates two specifications in this case. In contrast, the goal of the
discrete sizing task was achieved if Branch and Bound with or without quadratic model has
been used. The result quality of Branch and Bound with quadratic model is as good as the
result quality achieved without the modification. However, the runtime comparison in Figure
8 clearly shows that the additional runtime for Branch and Bound considering the quadratic
model presented in this paper, is significantly smaller, than without the modification and the
additional cost compared to the optimization with subsequent rounding is neglectable in this
case.
In the second example the sizing of the more complex amplifier in Figure 9, which is
proposed in (Martins, 1998), is shown. For this example the 45nm high performance
predictive technology model from (Nanoscale Integration and Modelling Group, Arizona
State University, 2008) is used and again a 5nm manufacturing grid is assumed. The lengths
of all transistors and the widths of transistors which are in the same current mirror or in the
same differential pair are set equal. Additionally, some multipliers are set equal considering
the symmetries of the circuit. Thus, 14 multipliers, 11 widths, and the length are considered
as discrete parameters. Additionally, the compensation capacitance Cc and the bias voltages
Vbias,1 and Vbias,2 are represented by continuous parameters. A 20pF load capacitance and a 2V
supply voltage are given for the circuit. Again the sizing rules from (Massier & Graeb, 2008)
are used which define 93 constraints in this case. Specifications and simulated performances
SQP + BaB SQP + BaB

SQP
Perfor- Specifi- Initial w/o with
+
mance cation values quadratic quadratic
Rounding
model model
PSRR
> 135 134 138 139 137
[dB ]
Gain
> 85 89 89 90 89
[dB ]
CMRR
> 135 172 167 167 166
[dB ]
f transit
> 15 19 24 22 23
[ MHz]
59
ϕ [◦ ] > 60 50 (violates 61 60
spec)
SR (rising)
>15 10 18 16 17
[ μs
V
]
|SR (falling)|
>15 14 37 31 32
[ μs
V
]
Area
< 10 7 9 9 9
[( μm )2 ]
52
Power
< 50 56 (violates 49 49
[μW ]
spec)
Table 1. Specification and performance values for Miller amplifier using 45nm PTM, 2V
supply voltage, 1uA bias current
VDD
l0 l0
w3 w3
m3 m3
l0 l0 Vbias,1
w4 w4 l0 l0 l0 l0
w8 w4 w4 w10
m4 m14
m8 m14 m4 m10
in + in −
Cc out
l0 l0 Vbias,2 l0 l0 l0 l0
w2 w1 w9 w1 w2 w11
m2 m1 m9 m1 m2 m11
l0 l0 l0 l0 l0 l0
w5 w5 w6 w6 w5 w5
m5 m13 m6 m12 m13 m5
l0 l0
w7 w7
gnd m7 m7
Fig. 9. Low-voltage low-power operational amplifier from (Martins, 1998)

sizing of amplifier in Figure 9
are presented in Tabular 2.

Also in this case specifications are violated when SQP and sub-sequent rounding is used and
also here Branch and Bound with and without quadratic model solves the sizing task. The
runtime comparison in this case shows that also here Branch and Bound using the quadratic
model is much faster than without consideration of the quadratic model. As the number of
discrete parameters is much higher in this case, also the runtime of Branch and Bound on the
quadratic model is relatively large. Thus potential for further improvement of the algorithm
can be seen: The runtime of the algorithm can be reduced if the Branch and Bound algorithm
presented in (3.2) is advanced, which is used to find the discrete optimum on the quadratic
model and needs approximately half of the computational time in this experiment.
The third example shows the sizing process for the sense amplifier from (Yeung & Mahmoodi,
2006) (see Figure 11). Considering the symmetry of the circuit, 5 multipliers, 5 transistor
widths, and the transistor length are used as parameters. For the sizing process a 16nm low
power PTM is used and a 2nm manufacturing grid is assumed. Specifications and results are
listed in Table 3. For the simulation of the delay it is assumed that the inputs (bit line BL and
negative bit line BLB) are preloaded to VDD = 1.5V and the input signal is a voltage reduction
by 10mV at one of them. “Delay +” in Table 3 is defined as the time between the change of the
input signal at the positive input BL and the point of time when the positive output reaches
0.95 · VDD . Accordingly, the value of “Delay −” is defined as the time between the change of
the signal at the negative input BLB and the point of time when the positive output reaches
0.05 · VDD .
The results for this experiment show that in this case continuous optimization with
subsequent rounding leads to a solution of the sizing task. This especially happens, if only
a few or week constraints and specifications are defined and if only a small number of
parameters is used. However, the additional runtime for the modified Branch and Bound
approach is only a few seconds. Further analysis of the results shows, that the additional
SQP + BaB SQP + BaB

SQP
Perfor- Speci- Initial w/o with
+
mance fication values quadratic quadratic
Rounding
model model
63
PSRR
> 70 59 (violates 77 79
[dB ]
spec)
Gain
> 60 62 65 69 65
[dB ]
69.7
CMRR
> 70 57 (violates 81 83
[dB ]
spec)
f transit
> 150 92 274 256 221
[ MHz]
ϕ [◦ ] > 60 85 65 62 60
SR (rising)
> 10 8 23 16 23
[ μs
V
]
|SR (falling)|
> 10 8 17 14 15
[ μs
V
]
Table 2. Specification and performance values for the amplifier in Figure 9 using 45nm PTM,
2V supply voltage, 20pF load capacity
VDD w3 w1 w1 w3
l1 l1 l1 l1
m3 m1 m1 m3
w4 ,l1 ,m4
w4 ,l1 ,m4
BLB
BL
w2
out1 w 2 out2
l1 l1
m2 m2
en
w5 ,l1 ,m5
gnd
Fig. 11. Sense amplifier
runtime (approximately 30 seconds) is used for computing the gradient and setting up the
quadratic model. In contrast to Branch and Bound with consideration of the quadratic model,
Branch and Bound without the quadratic model has a significant higher runtime.
5. Conclusion
Sizing of analog circuits is one important task in the analog design flow. In this chapter a new
deterministic and gradient-based method has been presented to solve this task. The method
solves the relaxed, i.e., continuous sizing task using SQP. Discretization of the result is done
by a subsequent Branch and Bound approach under consideration of the quadratic model
which is computed during SQP. Additionally certain properties of the underlying sizing task
are used to speed up the approach.
The experimental results show that SQP with subsequent rounding can not solve the sizing
task in general. In contrast, SQP combined with Branch and Bound is a reasonable approach
for sizing analog circuits with discrete parameters. Furthermore, the experimental results
show, that the efficacy and efficiency of SQP and Branch and Bound can be increased
sizing of the sense amplifier
SQP + BaB SQP + BaB
SQP
Perfor- Speci- Initial w/o with
+
mance fication values quadratic quadratic
Rounding
model model
Delay +
< 60 402 59 46 55
[ ps ]
Delay −
< 60 423 48 46 48
[ ps ]
static power
<5 27.5 1.4 1.4 2.0
[μW ]
Area
< 0.025 0.225 0.018 0.022 0.019
[( μm )2 ]
Table 3. Specification and performance values for sens amplifier using 16nm PTM, 1.5V
supply voltage, 1 f F Load capacity.
significantly, if the modifications in Section 3.3 are used.

The task presented so far is not able to solve the discrete sizing task, if the circuit performances
can only be evaluated for discrete points. Thus, in the future work this problem will be tackled.
Additionally, the experiments have shown, that the runtime of the algorithm can be reduced
by accelerating the Branch and Bound approach which is used to solve the quadratic model.
Also the consideration of non-scalable discrete parameters mentioned in Section 2 is an open
task for the future work.
6. References
Achtenberg, T., Koch, T. & Martin, A. (2005). Branching rules revisited, Operations Research
Letters 33(1): –42– –54.
Alpaydin, G., Balkir, S. & Dundar, G. (2003). An evolutionary approach to automatic synthesis
of high-performance analog integrated circuits, IEEE TEC 7(3).
Balijepalli, A., Sinha, S. & Cao, Y. (2007). Compact modeling of carbon nanotube transistor for
early stage process-design exploration, ISLPED.
Cao, Y., Sato, T., Sylvester, D., Orshansky, M. & Hu, C. (2000). New paradigm of predictive
mosfet and interconnect modeling for early circuit design, IEEE CICC.
Gielen, G. G. E. (2007). Design tool solutions for mixed-signal/RF circuit design in CMOS
nanometer technologies, ASP-DAC.
Gielen, G., Walscharts, H. & Sansen, W. (1990). Analog circuit design optimization based on
symbolic simulation and simulated annealing, IEEE JSSC 25(3).
Graeb, H. (2007). Analog Design Centering And Sizing, Springer.
Graeb, H., Zizala, S., Eckmueller, J. & Antreich, K. (2001). The sizing rules method for analog
integrated circuit design, ICCAD.
Knoblinger, G., Kutter, F., Marshall, A., Russ, C., Haibach, P., Patruno, P., Schulz, T., Xiong,
W., Gostkowski, M., Schruefer, K. & Cleavelin, C. R. (2005). Design and evaluation
of basic analog circuits in an emerging MuGFET technology, IEEE International SOI
Conference 2005.
Li, D. & Sun, X. (2006). Nonlinear Integer Programming, Springer.
Martins, R. (1998). On the Design of Very Low Power Integrated Circuits, PhD thesis, Vienna
University of Technology.
Massier, T. & Graeb, H. (2008). The sizing rules method for CMOS and bipolar analog
integrated circuit synthesis, IEEE TCAD 27(12).
Nanoscale Integration and Modelling Group, Arizona State University (2008). URL:
http://ptm.asu.edu/ [date: 08.06.2010].
Nemhauser, G. L. & Wolsey, L. A. (1988). Integer and Combinatorial Optimization, Jon Wiley &
Sons, Inc.
Nocedal, J. & Wright, S. (1999). Numerical Optimization, Springer.
Ochotta, E. S., Rutenbar, R. A. & Calrley, L. R. (1996). Synthesis of high-performance analog
circuits in ASTRX/OBLX, IEEE TCAD 15(3).
Pehl, M. & Graeb, H. (2009). RaGAzi: A random and gradient-based approach to analog
sizing for mixed discrete and continuous parameters, ISIC 2009.
Pehl, M., Massier, T., Graeb, H. & Schlichtmann, U. (2008). A random and pseudo-gradient
approach for analog circuit sizing with non-uniformly discretized parameters, ICCD
2008.
Phelps, R., Krasnicki, M., Rutenbar, R., Carley, L. & Hellums, J. (2000). Anaconda:
simulation-based synthesis of analog circuits via stochastic pattern search, IEEE
TCAD 19(6).
Rutenbar, R. A., Gielen, G. G. E. & Roychowdhury, J. (2007). Hierarchical modeling,
optimization, and synthesis for system-level analog and RF designs, Proceedings of
the IEEE, Vol. 95, IEEE.
Somani, A., Chakrabarti, P. & Patra, A. (2007). An evolutionary algorithm-based approach to
automated design of analog and rf circuits using adaptive normalized cost functions,
IEEE TEC 11(3).
Yeung, J. & Mahmoodi, H. (2006). Robust sense amplifier design under random dopant
fluctuations in nano-scale cmos technoloties, IEEE ISOCC.
Zhao, W. & Cao, Y. (2006). New generation of predictive technology model for sub-45nm early
design exploration, IEEE Transactions on Electron Devices 53(11).
14
Analog Circuit for Motion Detection Applied to

Target Tracking System
Kimihiro Nishio
Tsuyama National College of Technology
Japan
1. Introduction
It is necessary for the system such as the robotics vision and the monitoring camera to detect
the motion of the object and recognize the target in real time. However, this is difficult in
conventional image processing systems constructed with a charge coupled device (CCD)
camera and Neumann-type computer since information processing in this setup is
accomplished in a time-sequential way. On the other hand, real-time image processing is
easily performed in biological systems constructed with the retina and the brain since
information processing is achieved in massively parallel nerve networks which have a
hierarchical structure.
The biological vision system constructed with the retina and brain can detect the motion of
the object in real time and judge the target instantly. The complementary metal oxide
semiconductor (CMOS) circuits based on the biological vision system can be expected to
realize the high speed processing system since each unit circuit operates in parallel as well
as the signal processing of the biological vision system. Many researchers proposed the
CMOS circuits for edge detection and motion detection based on the biological vision
system (Mead, 1989.; Moini, 1999.; Asai et al., 1999b.; Liu., 2000.; Yamada et al. 2001.; Nishio
et al. 2003). These circuits are characterized by the high speed processing.
Particularly, there are neurons for tracking the target in the superior colliculus of the brain.
The simple target tracking model was proposed based on the signal processing of the brain.
The cells for generating the motion signal were introduced at the first stage of the model.
The motor for tracking the target was controlled by the motion signal.
Recently, analog CMOS circuits were proposed based on the model for tracking the target
(Asai et al., 1999a.; Liu et al., 2001.; Moini, 1999). At the first stage of the circuits, analog
motion detection CMOS circuits (Asai et al., 1999b.; Liu., 2000.) based on the biological
vision system were introduced for generating the motion signal.
Recently, we proposed simple analog CMOS circuits for generating the motion signal based
on the biological vision system (Nishio et al. 2004.; Nishio et al. 2007). The circuit consists of
the half of the number of transistors utilized to previous proposed motion detection circuit,
which is used at the first stage of the tracking system. The realization of the simple system
for tracking the target can be expected by using our circuits to the first stage of the tracking
system.
In this study, simple analog CMOS circuit for motion detection was proposed based on the
biological vision system. And, I tried to develop the test system for tracking the target based
on the biological vision system. The system was constructed with the analog CMOS circuit
for motion detection.
The analog motion detection circuit is characterized by high speed processing because the
unit circuits process in parallel as well as the information processing of the retina and brain.
The analog motion detection circuit is characterized by compact structure. The unit circuit is
constructed with about 17 MOS transistors by using analog technology.
In this chapter, the following topics (1)-(4) are described.
1. Motion detection model based on the biological vision system
2. Simple analog CMOS circuit for motion detection
3. Target tracking model based on the biological vision system
4. Test system for tracking the target using analog motion detection circuit
2. Motion detection model based on the biological vision system

Figure 1 shows the unit model for motion detection (Reichardt, 1961). We call the model the
correlation model. The motion direction and velocity of the target can be detected by the
output signal generated by the model. In this section, I describe the details of the model.
The model (elementary motion detector; EMD) is constructed with the large monopolar cell
L, the delay neuron D and the correlator C. The photoreceptor P is the input part.
The transient response of each cell when the target (object) moves toward the right side is
shown in Fig. 1(b). The P outputs the signal which is proportional to light intensity. The
signal of P1 is input to L1. When the target moves on P1, L1 outputs the pulsed signal. The
pulsed signal of L1 is input to D. Then, the signal of D shows the maximum value. After the
target moves away from P1, the signal of D decreases. When the target moves on P2, L2
generates the pulsed signal by inputting the signal of P2. The signal VE which is proportional
to the signal of D is output when the pulsed signal of L2 is input to C. The time between the
generation of the pulsed signal of L1 and that of L2 is equal to the time that the target moves
from P1 to P2. The time is inversely proportional to the velocity of the target. Thus, VE is
proportional to the velocity of the target.
When the target moves toward the left side, VE is 0. When the target moves toward the right
side, the model generate the signal VE. This model can detect the motion of the right
direction. Thus, it is able to detect the various motion direction by using the model.
3. Simple analog CMOS circuit for motion detection

Figure 2 shows the unit analog motion detection circuit. The circuit was proposed by
mimicking EMD in Fig. 1. The circuit can generate the signals for detecting the motion
direction and velocity. The operation principles of the circuit are described in this section.
The functions of D and C in Fig. 1 are added to our simple circuit (Nishio et al. 2004.; Nishio
et al. 2007). The proposed circuit is simple structure, which consists of 17 MOS transistors
and 3 capacitors. The photodiode PD is utilized to the input part. When the target (light)
moves on PD1, the voltage VL1 shows about the supply voltage VDD. After the time tL, the
voltage VLD becomes about VDD by the capacitor CL. Since the pMOS transistor MP1 and
nMOS transistor MN1 used as the switches turn on for tL, the current IL1 flows into MP1 and
MN1. Then, the voltage VD shows the maximum value by the integration circuit constructed
with the capacitor CD and the nMOS transistor MN2 where the voltage VG1 is set to constant
value. VD is converted to the current ID by the nMOS transistor MN3. After tL, MP1 turns off
Analog Circuit for Motion Detection Applied to Target Tracking System 319
Target
v
P1 P2
P : Photoreceptor
L1 L2
L : Large monopolar cell
D D : Delay neuron
C C : Correlator
EMD
VE
(a)
P1
Time t
P2
Time t
L1
Time t
L2
Time t
D
Time t
Motion
(VE)
C
signal
Time t
(b)
Fig. 1. Unit model for motion detection. (a) Model. (b) Transient response of each cell.
and VD and ID are decreased by MN2. The current IC is 0 since the nMOS transistor MN4
turns off when the target is not projected on PD2.
The target moves toward the right side, and the target projected on PD2. Then, the voltage
VL2 becomes about VDD and IC is equal to ID since MN4 turns on. IC is converted to the output
voltage VE by the integration circuit constructed with the capacitor CO and the nMOS
transistor MN5 where the voltage VG2 is set to the constant value. VE is proportional to the
velocity of the target.
In the case that the circuit is applied to the target tracking system, the voltage Vcenter
described in section 4 is generated by the PD located on the center of the array. When the
target locates on the center of the input part, VE shows about 0 by the nMOS transistor MN6.
Photoreceptor P1 and Photoreceptor P2

Correlator C
Large monopolar cell L1 and L2
VDD
IC
MP1
VLD VL2 MN 4
PD 1 PD 2 VE
MN1 IL1 Vth
CL
VL1
VD
Vth ID VG2 Vcenter
MN3
VG1
CD MN2 CO MN5 MN6
Delay neuron D
Fig. 2. Unit analog motion detection circuit.
4. Target tracking model based on the biological vision system

Figure 3 shows the model for tracking the target based on the biological vision system. The
unit model EMD in Fig. 1 are arrayed in one-dimensionally. By using this model, it is able to
track the target and capture the target in the center of the input parts. In this section, I will
describe the details of the model.
The input part of the model is the photoreceptor P array. P generates the signal which is
proportional to light intensity. The signal of P is input to each EMD. EMDR generates the
signal VER when the target moves toward the right side. EMDL generates the signal VEL
when the target moves toward the left side.
I describe about the model in Fig. 3 in the case that the target moves toward the right side.
When the target moves toward the right side, VEL1 and VEL2 are not generated, and VER1 and
VER2 are sequentially generated. The signal Vright is generated by summing VER1 and VER2. Vright
and Vleft are signals for controlling the motor M. Since Vleft is generated by summing VEL1 and
VEL2, Vleft is not generated in this case. Table 1 shows the method for controlling the motor. In
this table, VDD means that the signal is generated and 0 means that the signal is not generated.
When the target moves toward the right side, Vright is VDD and Vleft is 0. Then, the motor
normally rotates for tracking the target. The visual area (P array) turns to the target by the
rotation of the motor. When the target is captured on the center of the input array, PC located
on the center of the array generates the signal Vcenter. Vright and Vleft are decreased by Vcenter.
Then, Vright and Vleft become 0 and the motor stops. The model repeats the tracking toward the
right (rotation of the motor) and the capture of the target (stop of the motor). When the target
moves toward the right side, the model can track the target well.
When the target moves toward the left side, VER1 and VER2 are not generated, and VEL1 and
VEL2 are sequentially generated. Then, Vleft is VDD and Vright is 0, and the motor rotates
inversely for tracking the target. When the target is captured on the center of the input
array, VPC is generated. Vright and Vleft become 0 and the motor stops. The model repeats the
tracking toward the left (rotation of the motor) and the capture of the target (stop of the
motor). When the target moves toward the left side, the model can track the target well.
Target
v
PL4 PL3 PL2 PL1 PC PR1 PR2 PR3 PR4
EMD L2 EMD L1 EMD R1 EMD R2
VEL2 VEL1 Vcenter VER1 VER2
Vleft Vright
M
M : Motor
Fig. 3. Model for tracking the target based on the biological vision system.
Vleft Vright Motor
0 0 Stop
0 VDD Normal rotation (track toward the right side)
VDD 0 Reverse rotation (track toward the lef t side)
VDD VDD Stop
Table 1. Method for controlling the motor.
5. Test system for tracking the target using analog motion detection circuit
The test system for tracking the target was fabricated based on the model in Fig. 3. Figure 4
shows the photograph of the fabricated test system for tracking the target. It is able to track the
target by arranging the unit circuits in Fig. 2 in one-dimensionally. The PD array fabricated on
the printed board was placed on the rotating table which rotates with 360 degrees.
I describe the test system for tracking the target in this section. In the subsection 5.1, the
measured results of the test circuit for motion detection are described. The operation
principle of the circuit for controlling the motor is also described in the subsection 5.2. The
measured results of the test system are shown in subsection 5.3.
5.1 Motion detection circuit

The test circuits of Fig. 2 were fabricated on the printed board by using discrete MOS
transistors (nMOS:2SK1398, pMOS:2SJ184, NEC). I measured the test circuit based on EMD
applied to the tracking system. The supply voltage VDD was set to 5 V. Vth, VG1 and VG2 were
set to 1 V, 0.8 V and 2 V, respectively.
The relationship between PD and the target (light) is shown in Fig. 5(a). The light is provided
as the object. The light was moved toward the right side, i.e., the light moved on PD1 and PD2
sequentially. The output voltage VE was monitored by the oscilloscope. The measured result
of the output voltage of the motion detection circuit is shown in Fig. 5(b). When the light
moved on PD2, VE showed about 4.3 V. The test circuit could generate the motion signal. Thus,
it is clarified from the results that the proposed circuit can operate normally.
Analog CMOS circuit
based on EMD Power supply equipment
Motor driver Input part Motor Rotating table

(H bridge circuit) (PD array)
Fig. 4. Photograph of the fabricated test system for tracking the target.
Target PD1 PD2

(Light)
(a)
Motion signal
500 ms
4.3 V
(b)
Fig. 5. Measured result of the test circuit for motion detection. (a) Relationship between PD
and the target. (b) Result.
5.2 Motor driver

The motor driver (TA7257P, TOSHIBA) was used as the H bridge circuit, which was
connected with the DC motor, as shown in Fig. 4. The H bridge circuit is used to control the
motor by the voltages Vleft and Vright genenrated by the tracking system in Fig. 3. Figure 6
shows the H bridge circuit. This circuit can control the normal rotation, inverse rotation and
stop of the motor.
The motor rotates normally when the switches SW1 and SW4 turn on and SW2 and SW3 turn
off, as shown in Fig. 6(a). When the SW1 and SW4 turn off and SW2 and SW3 turn on, as
shown in Fig. 6(b), the motor rotates inversely. The motor stops when all switches turn off
or turn on, as shown in Figs. 6(c) and (d).
To realize the condition table 1, Vright controls SW1 and SW4. And Vleft controls SW2 and SW3.
When Vright is about VDD and Vleft is 0, SW1 and SW4 turn on and the motor rotates normally.
When Vleft is about VDD and Vright is 0, SW2 and SW3 turn on the motor rotates inversely.
VDD VDD VDD VDD

SW1 SW3 SW1 SW3
(ON) (OFF) (OFF) (ON)
M M
SW2 SW4 SW2 SW4
(OFF) (ON) (ON) (OFF)
(a) (b)
VDD VDD VDD VDD

SW1 SW3 SW1 SW3
(OFF) (OFF) (ON) (ON)
Stop Stop
M M
SW2 SW4 SW2 SW4
(OFF) (OFF) (ON) (ON)
(c) (d)
Fig. 6. H bridge circuit. (a) Normal rotation. (b) Inverse rotation. (c) Stop. (d) Stop.
5.3 Measured results of the test system

The fabricated test system for tracking the target in Fig. 4 was measured. Bias voltages set in
subsection 5.1 were provided to the circuits based on EMD. As the target, the light was
projected on PD array.
The measured results of the test system, when the target moves toward the left side, are shown
in Fig. 7. The light was moved toward the left side until t=5 s from t=0 s. At t=5 s, the light was
stopped. The system tracked the light, as shown in images at t=4 and 5 s. At t=6 s, the motor of
the system stopped, and the system could capture the target on the center of the PD array.
PD array Target (Light)
t=0s t=2s
t=3s t=4s
Target (Stop) Motor (Stop)
t=5s t=6s
Fig. 7. Measured results of the test system when the target moves toward the left side.
The measured results of the test system, when the target moves toward the right side, are
shown in Fig. 8. The light was moved toward the right side until about 3 s. The light was
stopped at about 3 s. The system tracked the light toward the right side, as shown in images
between t=0.5 s and t=3 s. As shown in the image at t=4 s, the motor stopped and the system
could capture the target. Thus, it was clarified from the results that the fabricated system
can track the target and capture the target on the center of the PD array.
t=0s t = 0.5 s
t=1s t=2s
Target (Stop) Motor (Stop)
t=3s t=4s
Fig. 8. Measured results of the test system when the target moves toward the right side.
6. Conclusion
In this study, the simple analog CMOS motion detection circuit was proposed based on the
biological vision system. The simple circuits for motion detection were applied to the first
stage of the target tracking system. The test circuit for motion detection was fabricated on
the printed board by using discrete MOS transistors. The test system for tracking the target
was fabricated by using the test circuit. The test circuit could generate the motion signal for
controlling the motor of the system. The test system could track the target and capture the
target on the center of the input part. By using proposed basic circuits and system for
tracking the target, we can expect to realize the novel visual sensor for robotics system,
monitoring system and others.
7. References
Asai, T.; Ohtani, M.; Yonezu, H. & Ohshima, N. (1999a). Analog MOS Circuit Systems
Performing the Visual Tracking with Bio-Inspired Simple Networks, Proc. of the 7th
International Conf. on Microelectronics for Neural Networks, Evolutionary & Fuzzy
Systems, pp. 240-246
Asai, T.; Ohtani, M. & Yonezu, H. (1999b). Analog MOS Circuits for Motion Detection Based
on Correlation Neural Networks, Jpn. J. Appl. Phys., Vol.38, pp.2256-2261
Liu, S. (2000). A Neuromorphic a VLSI Model of Global Motion Processing in the Fly, IEEE
Trans. Circuits and Systems II, Vol. 47, pp. 1458-146
Liu, S. & Viretta, A. (2001). Fly-Like Visuomotor Responses of a Robot Using a VLSI Motion-
Sensitive Chips, Biological Cybernetics, Vol. 85, pp. 449-457
Mead, C. (1989) Analog VLSI and neural systems, Addison Wesley, New York
Moini, A. (1999) Vision Chips, Kluwer Academic, Norwell, MA
Nishio, K.; Yonezu, H.; Ohtani, M.; Yamada, H.; & Furukawa, Y. (2003). Analog Metal-
Oxide-Semiconductor Integrated Circuits Implementation of Approach Detection
with Simple-Shape Recognition Based on Visual Systems of Lower Animals, Optical
Review, Vol. 10, pp. 96-105
Nishio, K.; Matsuzaka, K. & Irie, N. (2004). Analog CMOS Circuit Implementation of Motion
Detection with Wide Dynamic Range Based on Vertebrate Retina, Proc. of 2004 IEEE
Conf. on Cybernetics and Intelligent Systems, 2004
Nishio, K.; Matsuzaka, K. & Yonezu, H. (2007). Simple Analog Complementary Metal Oxide
Semiconductor Circuit for Generating Motion Signal, Optical Review, Vol. 14, pp.
282-289
Reichardt, W. (1961) Principles of Sensory Communication, Wiley, New York
Yamada, H.; Miyashita, T.; Ohtani, M.; Nishio, K.; Yonezu, H.; & Furukawa, Y. (2001). Signal
Formation of Image-Edge Motion Based on Biological Retinal Networks and
Implementation into an Analog Metal-Oxide-Silicon Circuit, Optical Review, Vol. 8,
pp. 336-342
15
Analog Circuits Implementing a Critical

Temperature Sensor Based on Excitable Neuron
Models
Gessyca M., Tovar Nunez
Hokkaido University
Japan
1. Introduction
Temperature is the most often-measured environmental quality. This might be expected since
temperature control is fundamental to the operation of electronic and other systems. In the
present, there are several passive and active sensors for measuring system temperatures,
including thermocouples, resistive-temperature detectors (RTDs), thermistors, and silicon
temperature sensors (Gopel et al., 1990) (Wang et al., 1998). Among present temperature
sensors, thermistors with a positive temperature coefficient (PTC) are widely used because
they exhibit a sharp increase of resistance at a specific temperature. Therefore, PTC
thermistors are suitable for implementation in temperature-control systems that make
decisions, like shutting down equipments above a certain threshold temperature or to turning
cooling fans on and off, general purpose temperature monitors.
Here I propose a sub-threshold CMOS circuit that changes its dynamical behavior; i.e.,
oscillatory or stationary behaviors, around a given threshold temperature, aiming to the
development of low-power and compact temperature switch on monolithic ICs. The
threshold temperature can be set to a desired value by adjusting an external bias voltage.
The circuit consists of two pMOS differential pairs, small capacitors, current reference
circuits, and off-chip resistors with low temperature dependence. The circuit operation was
fully investigated through theoretical analysis, extensive numerical simulations and circuit
simulations using the Simulation Program of Integrated Circuit Emphasis (SPICE). Moreover,
I experimentally demonstrate the operation of the proposed circuit using discrete MOS
devices.
2. The model
The temperature sensor operation model is shown in Fig. 1. The model consists of a
nonlinear neural oscillator that changes its state between oscillatory and stationary when it
receives an external perturbation (temperature). The key idea is the use of excitable circuits
that are strongly inspired by the operation of biological neurons. A temperature increase
causes a regular and reproducible increase in the frequency of the generation of pacemaker
potential in most Aplysia and Helix excitable neurons (Fletcher & Ram, 1990). Generation
of the activity pattern of the Br-type neuron located in the right parietal ganglion of Helix
pomatia is a temperature-dependent process. The Br neuron shows its characteristic bursting
Frequency
Tc = Critical Temperature
Tc Temperature
Oscillatory Stationary
Fig. 1. Critical temperature sensor operation model.
activity only between 12 and 30◦ C. Outside this range, the burst pattern disappears and the
action potentials become regular. This means that excitable neurons can be used as sensors to
determine temperature ranges in a natural environment.
There are many models of excitable neurons, but only a few of them have been implemented
on CMOS LSIs, e.g., silicon neurons that emulate cortical pyramidal neurons (Douglas et
al., 1995), FitzHugh-Nagumo neurons with negative resistive circuits (Barranco et al., 1991),
artificial neuron circuits based on by-products of conventional digital circuits (Ryckebusch et
al., 1989) - (Meador & Cole, 1989), and ultralow-power sub-threshold neuron circuits (Asai et
al., 2003). Our model is based on the Wilson-Cowan system (Wilson & Cowan, 1972) because
it is easy to both analyze theoretically and implement in sub-threshold CMOS circuits.
The dynamics of the temperature sensor can be expressed as:
exp (u/A)
τ u̇ = −u + , (1)
exp (u/A) + exp (v/A)
exp (u/A)
v̇ = −v + , (2)
exp (u/A) + exp (θ/A)
where τ represents the time constant, θ is an external input, and A is a constant proportional
to temperature. The second term of the r.h.s. of Eq.(1) represents the sigmoid function, a
mathematical function that produces an S-shaped (sigmoid) curve. The sigmoid function can
be implemented in VLSIs by using differential-pair circuits, making this model suitable for
implementation in analog VLSIs.
To analyze the system operation, it is necessary to calculate its nullclines. Nullclines are curves
in the phase space where the differentials u̇ and v̇ are equal to zero. The nullclines divide the
phase space into four regions. In each region the vector field follows a specific direction.
Along the curves the vector field is either completely horizontal or vertical; on the u nullcline
the direction of the vector is vertical; and on the v nullcline, it is horizontal. The u and v
nullclines indicating the direction of vector field in each region are shown in Fig. 2.
The trajectory of the system depends on the time constant τ, which modifies the velocity field
of u. In Eq. (1), if τ is large, the value of u decreases, and for small τ, u increases. Figures 3(a)
and (b) show trajectories when τ = 1 and τ << 1. In the case where τ << 1, the trajectory on
the u direction is much faster than that in the v, so only close to the u nullcline movements of
vectors in vertical direction are possible.
Temperature Sensor Based on Excitable Neuron Models 329
1
u& < 0 nullcline v u& < 0
v& < 0 v& > 0
0.8
nullcline u
Trajectory
0.6
v(V)
0.4
0.2
u& > 0 u& > 0
v& < 0 v& > 0
0
0 0.2 0.4 0.6 0.8 1
u(V)
Fig. 2. u and v nullclines with vector field direction.
1
nullcline v Trajectory
0.9
0.8 Trajectory
0.7 nullcline v
0.6
v (V)
0.5 nullcline u
0.4
0.3
0.2
0.1
nullcline u
0
0 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 1
u (V) u(V)
(a) (b)
Fig. 3. Trajectory when a) τ = 1 and b) τ << 1.
Let us suppose that θ is set at a certain value where the critical temperature (Tc ), which is
proportional to A is 27◦ C. The critical temperature represents the threshold temperature we
desire to measure. When θ changes, the v nullcline changes to a point where the system will be
stable as long as the external temperature is higher than Tc . This is true because the system is
unstable only when the fixed point exists in a negative resistive region of the u nullcline. The
fixed point, defined by u̇ = v̇ = 0 is represented in the phase space by the intersection of the u
nullcline with the v nullcline. At this point the trajectory stops because the vector field is zero,
and the system is thus stable. On the other hand, when the external temperature is below Tc ,
the nullclines move, and this will correspond to a periodic solution to the system. In the phase
space we can observe that the trajectory does not pass through the fixed point but describes a
closed orbit or limit cycle, indicating that the system is oscillatory. Figure 4 shows examples
when the system is stable (a) and oscillatory (b). In (a) the external temperature is greater
than the critical temperature, hence, the trajectory stops when it reaches the fixed point, and
the system is stable. In (b), where the temperature changes below the critical temperature, the
trajectory avoids the fixed point, and the system becomes oscillatory.
Deriving the nullclines equation (u̇ = 0) and equaling to zero, I calculated the local minimum
(u− , v− ) and local maximum (u+ , v+ ), representing the intersection point of the nullclines
1
T>Tc T<Tc
Trajectory
0.9 Trajectory
0.8
0.7
0.6 nullcline v nullcline v
v(V)
0.5
0.4 Fixed Point
nullcline u
0.3 nullcline u
0.2
0.1
0
0 0.2 0.4
u(V) 0.6 0.8 1 0 0.2 0.4
u(V) 0.6 0.8 1
(a) (b)
Fig. 4. Nullclines showing the fixed point and the trajectory when a) system is stable b)
system is oscillatory.
given by: √
1 − 4A 1±
, u± = (3)
2
1
v± = u± + A ln ( − 1), (4)
u±
The nullclines giving the local minimum and local maximum (u± , v± ) are shown in Fig. 5(a).
From the local minimum and maximum equations (Eq. (3) and Eq. (4)), the nullcline equation
(v̇ = 0) and remembering that A is proportional to temperature, I determined the relationship
between θ and the temperature, to be given by:
1
θ± = u± + A ln ( − 1). (5)
v±
When τ << 1 the trajectory jumps from one side to the other side of the u nullcline, so
only along the u nullcline movement in the v direction are possible as shown in Fig. 3(b).
It is necessary to emphasis this fact because this characteristic is necessary for the system
operation; thus, I assume τ << 1.
2.1 Stability of the Wilson-Cowan system

Wilson and Cowan (Wilson & Cowan, 1972) studied the properties of a nervous tissue
modeled by populations of oscillating cells composed of two types of interacting neurons:
excitatory and inhibitory ones. The Wilson-Cowan system has two types of temporal
behaviors, i.e. steady state and limit cycle. According with the stability analysis in (Wilson
& Cowan, 1972), the stability of the system can be controlled by the magnitude of the all the
parameters. Equations (1) and (2) are a simplified set representing the Wilson-Cowan system
a) b)
1
θ=x θ=y
0.9
u+, v+
0.8
0.7 u nullcline
nullcline v-
0.6
rea
v(V)
nullcline u
0.5 nullcline v+ ea
ycl
ti c
0.4 Lim
0.3
0.2
0.1 u-, v-
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
u(V) u (V)
Fig. 5. a) u and v local maximum and local minimum. b) Threshold values x and y showing
the area where the system is oscillatory.
1 1
trajectory θ = 0.1 trajectory θ = 0.09
0.8 0.8
u nullcline u nullcline
0.6 0.6
v (V)
v (V)
v nullcline v nullcline
0.4 0.4
0.2 0.2
Fixed point
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
u (V) u (V)
(a) (b)
Fig. 6. Nulclines and trajectories when a) θ = 0.1 and b) θ = 0.09.
equations with and excitatory node u and an inhibitory node v. The nullclines of this system,
which are pictured in Fig. 2, are given by:
1
v = u + A ln( − 1) (6)
u
for the u nullcline ((Eq. 1) = 0), and
eu/a
v= (7)
eu/a + eθ/A
for the v nullcline ((Eq. 2) = 0).
For an easy analysis, let us suppose that A is a constant. In this case, there are some important
observations for the stability of the system.
• There is a low threshold value of θ bellow which the limit cycle activity can not occurs.
• There is a high threshold value of θ above which the system saturates and the limit cycle
activity is extinguished.
• Between these two values (x for the lower threshold and y for the higher threshold), the
system exhibit limit cycle oscillation.
1
θ = 0.09
θ = 0.1
0.8
u nullcline
0.6
v (V) v nullcline
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
u (V)
θ = 0.09
θ = 0.1
Fig. 7. v nullcline when θ = 0.1 and θ = 0.09.

1 1
trajectory θ = 0.9 trajectory θ = 0.91
0.8 0.8
u nullcline u nullcline
0.6 0.6
v (V)
v (V)
0.4 0.4
v nullcline v nullcline
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
u (V) u (V)
(a) (b)
Fig. 8. Nulclines and trajectories when a) θ = 0.9 and b) θ = 0.91.
Let us suppose that the value of A is fixed to 0.03, in this cases, depending on the magnitude
of the parameter θ (that is the external input of the system) the Wilson-Cowan oscillator will
show different behaviors. Figure 5(b) shows the area inside which the system exhibits a limit
cycle. The threshold values x and y are shown in the figure.
The nullclines and trajectories for different values of θ are shown in Figs. 6 and 8. In Figure 6
(a), θ was set to 0.1, we can observe that the system is exhibiting limit cycle oscillations. Thus,
for this case the system is unstable. When the value of θ is reduce to 0.09, as show in Fig. 6
(b). It can be observed that the trajectory stops at the fixed point. The fixed point in this area is
Vdd Vdd Vdd
Ia Ia Ib
M1 M2 M3 M4
u v u θ
I1 I2
Sensor
C1 g C2 g
Fig. 9. Critical temperature sensor circuit.
100
90
80
Tc ( R C)
70
60
50
40
30
20
10
0
0.085 0.09 0.095 0.1 0.105 0.11 0.89 0.895 0.9 0.905 0.91 0.915
θ(V)
Oscillatory Stable Stable Oscillatory
T < Tc Tc T ≥ Tc T < Tc Tc T ≥ Tc
Fig. 10. Relation between θ± and Tc .
an attractor, i.e. a stable fixed point. Thus, the system is stable. Figure 7 show the position of
the v nullclines when θ = 0.09 and θ = 0.1. The other case (for a high threshold), is shown is
Fig. 8. In figure 8 (a) θ is set to 0.9, at this point the system is oscillatory. When θ is increased,
(θ = 0.91) the system is stable.
We could observed that depending on the parameter θ (external input) the stability of the
system can be controlled. It is important to note that the stability also depends on the
magnitude of A, and that A is proportional to the temperature. These observations are the
basis of the operation of the temperature sensor system.for example, by setting the value of
the input θ, when the external temperature changes the system behavior also changes i.e.
stable and oscillatory.
3. CMOS circuit
The critical temperature sensor circuit is shown in Fig. 9. The sensor section consists of two
pMOS differential pairs (M1 − M2 and M3 − M4 ) operating in their sub-threshold region.
External components are required for the operation of the circuit. These components consist
of two capacitors (C1 and C2 ) and two temperature-insensitive off-chip metal-film resistors (g).
In addition, for the experimental purpose, two current mirrors were used as the bias current
of differential pairs. Note that for the final implementation of our critical temperature sensor
a current reference circuit with low-temperature dependence (Hirose et al., 2005) should be
used.
Differential-pairs sub-threshold currents, I1 and I2 , are given by (Liu et al., 2002):
exp (κu/v T )
I1 = Ia , (8)
exp (κu/v T ) + exp (κv/v T )
exp (κu/v T )
I2 = Ia , (9)
exp (κu/v T ) + exp (κθ/v T )
where Ia represents the differential pairs bias current, v T is the thermal voltage (v T = kT/q),
k is the Boltzmann’s constant, T is the temperature, and q is the elementary charge.
The circuit dynamics can be determined by applying Kirchhoff’s current law to both
differential pairs, which is represented as follows:
Ia exp (κu/v T )
C1 u̇ = − gu + , (10)
Ia exp (κu/v T )
C2 v̇ = − gv + , (11)
where κ is the sub-threshold slope, C1 and C2 are the capacitances representing the time
constants, and θ is bias voltage.
Note that Eqs. (10) and (11) correspond to the system dynamics (Eqs. (1) and (2)) previously
explained. Therefore, applying the same analysis, I calculated the local minimum (u− , v− )
and local maximum (u+ , v+ ) for the circuit equations, expressed by:

Ia /g ± ( Ia /g)2 − 4v T Ia /(κg)
u± = , (12)
2
v Ia
v± = u± + T ln ( − 1), (13)
κ gu±
and the relationship between the external bias voltage (θ) and the external temperature (T):
vT Ia
θ± = u± + ln ( − 1). (14)
κ gv±
where the relation with the temperature is given by the thermal voltage defined by v T = kT/q.
At this point the system temperature is equal to the critical temperature which can be obtained
from:
qκ (θ± − u± )
Tc = . (15)
k ln ( gvIa± − 1)
The threshold temperature Tc can be set to a desired value by adjusting the external bias
voltage (θ). The circuit changes its dynamic behavior, i.e., oscillatory or stationary behaviors,
depending on its operation temperature and bias voltage conditions. At temperatures lower
than Tc the circuit oscillates, but the circuit is stable (does not oscillate) at temperatures higher
than Tc . Figure 10 shows the relation between the bias voltage θ± and the critical temperature
a) b)
1.2
nullcline v nullcline v
1
0.8
Trajectory Trajectory
v(V)
0.6
0.4
nullcline u
nullcline u
0.2
0
-0.2 0 0.2 0.4 0.6 0.8 1 1.2 -0.2 0 0.2 0.4 0.6 0.8 1 1. 2
u(V) u(V)
Fig. 11. Trajectory and nullclines obtained through simulation results when a) the system is
oscillatory. b) the system is stationary
Tc with κ = 0.75; θ− for u and v local minimums and θ+ for u and v local maximums. When
θ− is used to set Tc , the system is stable at external temperatures higher than Tc ; while when
θ+ is used, the system is stable when the external temperature is lower than Tc and oscillatory
when it is higher than Tc .
4. Simulations and experimental results

Circuit simulations were conducted by setting C1 and C2 to 0.1 pF and 10 pF, respectively, g
to 1 nS, and reference current (Ib ) to 1 nA. Note that for the numerical and circuit simulations,
two current sources were used instead of the current mirrors. The parameter sets I used for
the transistors were obtained from MOSIS AMIS 1.5-μm CMOS process. Transistor sizes were
fixed at L = 40 μm and W = 16 μm. The supply voltage was set at 5 V. Figure 11(a) shows
the nullclines and trajectory of the circuit with the bias voltage (θ) set at 200 mV and the
external temperature (T) set at 27◦ C; the system was in oscillatory state. Figure. 11(b) shows
the nullclines when the system is stationary with the bias voltage (θ) set at 90 mV.
The output waveform of u for different temperatures is shown in Fig. 12. The bias voltage
θ was set to 120 mV, when the external temperature was 20◦ C the circuit was oscillating,
but when the temperature increases up to 40◦ C the circuit becomes stable. Figure 13 shows
the simulated oscillation frequencies of the circuit as a function of the temperature, the bias
voltage set to 120 mV. The frequency was zero when the temperature was above the critical
temperature Tc = 36◦ C, and for temperatures lower than Tc the frequency increased, as shown
in the figure.
Through circuit simulations, by setting the values for the critical temperature (Tc ) and
changing the bias voltage (θ) until the system changed its state, I established a numerical
relation between Tc and θ. When comparing this relationship between θ and Tc obtained
through different methods, I found a mismatch between the numerical simulations and
the circuit simulations. This difference might be due to the parameters that are included
in the SPICE simulation but omitted in the numerical simulation and theoretical analysis.
Many of these parameters might be temperature dependent; thus, their value changes with
temperature, and as a result of this change, the Tc characteristic changes. The difference
between the two simulations is shown in Fig. 14
T=20 (ºC)
1
0.8
0.6
u (V)
0.4
0.2
-0.2
T=30 (ºC)
1
0.8
0.6
u (V)
0.4
0.2
-0.2
T=40 (ºC)
1
0.8
0.6
u (V)
0.4
0.2
-0.2
0 1 2 3 4 5
time (ms)
Fig. 12. Waveform of u at different temperatures (from T = 20◦ C to T = 40◦ C).
20 1
T c 36 ºC
17.5 0.9
0.8
15
frequency (kHz) 0.7
Amplitud (V)
12.5
0.6
10 0.5
7.5 0.4
0.3
5
0.2
2.5 0.1
0 0
-20 0 20 T c40 60 80 100
Temperature (ºC)
Fig. 13. Oscillation frequencies of the circuit. (Tc = 36◦ C).
100
80 SPICE
60
numerical
T
40
20
-20
0.085 0.09 0.095 0.1 0.105 0.11 0.115 0.12 0.125
θ(V)
Fig. 14. Relation between θ± and Tc obtained through numerical and circuit simulations.
I successfully demonstrated the critical temperature sensor’s operation using discrete MOS
circuits. Parasitic capacitances and a capacitance of 0.033 μF were used for C1 and C2
respectively, and the resistances (g) were set to 10 MΩ. The input current (Ib ) for the current
mirrors was set to 100 nA and I obtained an output current (Ia ) of 78 nA.
Measurements were performed at room temperature (T = 23◦ C). With the bias voltage (θ) set
to 500 mV the voltages of u and v were measured. Under these conditions, the circuit was
oscillating. The voltages of u and v for different values of θ were also measured. The results
showed that for values of θ lower than 170 mV, the circuit did not oscillate (was stable), but
that for values higher than 170 mV, the circuit became oscillatory. Figures 15 and 16 shows the
oscillatory and stable states of u and v with θ set to 170 and 150 mV, respectively.
In addition, I also measured the nullclines (steady state voltage of the differential pairs). The
v nullcline (steady state voltage v of differential pair M3 − M4 ) was measured by applying a
variable DC voltage (from 0 to 1 V) on u and measuring the voltage on v. For the measurement
0.8
0.7
0.6 u
0.5
u,v (V) 0.4
0.3 v
0.2
0.1
-0.1
-0.1 -0.05 0 0.05 0.1
t (s)
Fig. 15. Experimental results: θ =170 mV at T= 23◦ C (oscillatory state).
1
0.8
u,v(V)
0.6
0.4
v
0.2
u
0
-0.1 -0.05 0 0.05 0.1
t(s)
Fig. 16. Experimental results: θ =150 mV at T= 23◦ C (stationary state).
of the u nullcline (steady state voltage u of differential pair M1 − M2 ), a special configuration

of the first differential pair of the circuit was used. Figure 17 shows the circuit used for the u
nullcline measurement. I applied a variable DC voltage (from 0 to 1 V) on v. For each value of
v I changed the voltage on u1 (from 0 to 1) and then measured the voltage on uo and u1 . This
enabled us to obtain the u nullcline by plotting the points where uo and u1 had almost the
same value. In this way, I obtained a series of points showing the shape of the u nullcline. The
series of points was divided into three sections, and the average was calculated to show the u
nullcline. Figure 18 shows the u nullcline divided into the three sections used for the average
calculation. The trajectory and nullclines of the circuit with θ set to 500 mV are shown in Fig.
19.
Notice that in the experimental results there is a difference in the amplitude of the potentials
u and v with respect to results obtained from the numerical and circuit simulations. This is
due to the difference in the bias current of the differential pairs. From Eqs. (12) and (13), we
can see that by making g and Ib (used in numerical and circuit simulations) the same value,
they cancel each other out; however, the output currents of the current mirrors were in the
I
M1 M2
u
v
u R
C1 g
Fig. 17. Circuit used for calculation of the u nullcline.
1
Section 1 Section 2 Section 3
0.8
original
Average data
0.6
Average
v(V)
Average
0.4
original
data
original
0.2 data
0
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
u(V)
Fig. 18. Sections used for the calculation of the u nullcline.
order of 78 nA, and g was set to 100 nS. This difference caused the decrease in the potentials
amplitudes, as shown in Figs.11 and 19.
Measurements performed at different temperatures were made. The bias voltage (θ) was set
to a fixed value and the external temperature was changed to find the value of the critical
temperature (Tc ) where the circuit changes from one state to the other. With the bias voltage
θ set to 170 mV at room temperature (T = 23◦ C), the circuit oscillated. When the external
temperature was increased to (T = 26◦ C), the circuit changed its state to stationary (did not
oscillate). Once again, when the external temperature was decreased one degree (T = 25◦ C),
the circuit started to oscillate; therefore, the critical temperature was Tc = 26◦ C. Measures of
the critical temperature (Tc ) for different values of the bias voltage (θ) were made.
In order to compare experimental results with, SPICE results and theoretical ones, the actual κ
(subthreshold slope) of the HSPICE model was measured and found to be in the order of 0.61.
The critical temperature for each value of θ obtained experimentally compared with the critical
1
u nullcline
0.8
0.6 trajectory
v(V)
0.4 original
u data
0.2
0 v nullcline
-0.2
-0.2 0 0.2 0.4 0.6 0.8 1
u(V)
Fig. 19. Experimental nullclines and trajectory.

80
Theoretical
70 Results
60 HSPICE
Results Experimental
Tc ( °C)
50 Results
40
30
20
10
0
0.05 0.1 0.15 0.2 0.25 0. 3 0.35 0. 4
θ (V)
Fig. 20. Bias voltage vs temperature, experimental results.
temperature obtained with theoretical analysis using Eq. (14) (with κ = 0.61) is shown in Fig.
20. The curves have positive slopes in both cases. This is because the temperature difference
between one value of bias voltage and the other decreases as the bias voltage increases. For
θ= 140 and 150 mV the experimentally obtained critical temperatures (Tc ) are 0◦ C and 13◦ C,
respectively, a difference of 13◦ C. For θ= 240 and 250 mV the critical temperatures (Tc ) are
54◦ C and 56◦ C, respectively: a difference of only 2◦ C.
The difference between the experimental, HSPICE, theoretical results is due to the leak current
caused by parasitic diodes between the source (drain) and the well or substrate of the discrete
MOS devices, and the mismatch between the MOS devices. In addition, because of the leak
current, when temperature increases, the stable voltages of u and v also increase. Figures 21(a)
and 21(b) shows the stationary state with θ set to 140 mV and temperature set to 23 and 75◦ C,
respectively.
a) b)
1
0.8
0.6
u,v(V)
0.4
u
0.2
v
u
0
-0.1 -0.05 0 0.05 0.1 -0.1 -0.05 0 0.05 0.1
t(s) t(s)
Fig. 21. Stationary state with a) θ= 140 mV and T= 23◦ C. b) θ= 140 mV and T= 75◦ C
Vu Vdd
Vdd Vs I1
Id
Vg p+ n+ n+
I ds
I db
p-type substrate
Vs
I d = I d +I db
Fig. 22. nMOS transistor structure showing leak current
5. nMOS Transistor with temperature dependence

The structure of a nMOS transistor showing the temperature-sensitive drain to bulk leakage
current (Idb ) is shown in Fig. 22. The drain current of the transistor is thus given by the sum
of the drain-bulk current (Idb ) and the channel current (Ids ).
Id = Ids + Idb (16)
and remembering that the saturated drain to source current when the transistor is operating
in the subthreshold region is given by
Ids = I0 eκ (Vg −Vs )/VT (17)
the drain current becomes

Id = I0 eκ (Vg −Vs )/VT + Idb (18)
where I0 represents the fabrication parameter, and Vs the common source nd bulk voltage.
The drain-bulk current (Idb ) is given by:
Idb = Gdb (Vdd − Vb ) (19)

16
14
12
10
Idb(nA)
8
6
4
2
0
-20 0 20 40 60 80 100 120 140
Temp(°C)
Fig. 23. Drain-bulk current Idb vs Temperature.
where Vdd is the supply voltage, Vb the bulk potential, and Gdb the temperature-dependent
drain-bulk conductance expressed as:
Eg ( Tnom ) Eg ( T )
VTnom − VT
Gdb = GS e (20)
where GS represents the bulk junction saturation conductance (1 × 10−14 ), Eg ( X ) is the energy
gap, and Tnom the nominal temperature (300.15 K). The temperature dependence of the energy
gap is modeled by
αT 2
Eg ( T ) = Eg (0) − (21)
β+T
Si experimental results give Eg (0) = 1.16 eV, α = 7.02 × 10−4 , and β = 1108.
Numerical simulations where carried out. Figure 23 shows the drain-bulk current of a single
transistor as the temperature changes. We can observe that when the temperature is less than
80 ◦ C the drain-bulk (Idb ) current is in the order of pF (≈ 30 pF), but as temperature increases,
Idb also increases in an exponential manner reaching values in the order of nA (≈ 16 nA for
T = 140 ◦ C).
The same analysis can be applied to pMOS transistors, but in addition the leak current from
the p-substrate to the n-Well is added to the drain current.
6. Differential pair with temperature dependence

Figure 24 shows a differential pair circuit consisting of two nMOS transistors (m1 and m2 ),
and an ideal current source (Ib ). According with the analysis done in the previous section, the
drain currents (I1 and I2 ) are
I1 = I0 eκ (u−Vs )/VT + Idb (22)
I2 = I0 eκ (v−Vs )/VT + Idb (23)
I1 I2
u m1 m2 v
Vs
Ib
r
Fig. 24. Differential pair.
Since Ib = I1 + I2 , we obtain
Ib − 2Idb
e−κVs /VT = κu/V
(24)
I0 (e T + eκv/VT )
From Eqs. (22) and (23), the drain currents become

( Ib − 2Idb )eκu/VT
I1 = + Idb (25)
eκu/VT + eκv/VT
( Ib − 2Idb )eκv/VT
I2 = + Idb (26)
eκu/VT + eκv/VT
From Eq. (24) the common source voltage Vs is
V
Vs = T ln I0 + ln (eκu/VT + eκv/VT ) − ln ( Ib − 2Idb ) (27)
κ
Equations (25) and (26) were plotted and compared with the SPICE simulations results (see
figure 25). I used the MOSIS AMIS 1.5-μm CMOS parameters (LEVEL 3). Transistor sizes
were set to W/L = 4 μm/1.6 μm. Ib was set to 100 nA, and v was set to 0.5 V. From the SPICE
simulations, the measured κ 0.47, I0 was 18.8 pA when T = 300.15 ◦ K, and 62.6 pA when
T = 350.15 ◦ K. We can observe that the theoretical results agreed with the SPICE results.
7. Dynamics of the CTS circuit

The critical temperature sensor circuit is shown in Fig. 9. The circuit dynamics Eqs. (10) and
(11) with the temperature dependence analysis become
( Ia − 2Idb − 2Iws ) exp (κu/v T )
C1 u̇ = − gu + + Idb + Iws , (28)
( Ia − 2Idb − 2Iws ) exp (κu/v T )
C2 v̇ = − gv + + Idb + Iws , (29)
To confirm the effect of the leak currents in the temperature sensor system, I conducted
a comparative analysis between HSPICE and the theoretical results without and with leak
current. The comparison between HSPICE results and theoretical results without leak currents
effect with the bias voltage θ set to 0.5 V and the external temperature set to T = 127 ◦ C, is
shown in Fig. 26(a). It can be seen that in this case the results between the theory and the
SPICE are very different, but in the same conditions when the effect of the leak current is
include in the theory the results are very similar, Fig. 26(b).
100
Theoretical( , )
80
SPICE( - )
60
I1 (nA)
40
20 T=350.15 K T=300.15 K
0
0 0.2 0.4 0.6 0.8 1
u (V)
Fig. 25. Theoretical and SPICE results of differential pair’s current I1 when temperature is
300.15 K and 400.15 K.
a) b)
HSPICE Theoretical HSPICE Theoretical
1
u (V)
0
2.6 2.8 3 3.2 3.4 3.6 3.8 4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
time (ms) time (ms)
Fig. 26. Comparison of CTS oscillations, between HSPICE results and theoretical results with
T = 127 ◦ C. a) without leak currents. b) with leak currents.
8. Conclusion
This research focused on the studied and the implementation of artificial neural systems. As
a small contribution, to reach the final goal all researchers have in common, the building o f
an arti f icial brain. To accomplish this, I proposed the design of a critical temperature sensor
strongly inspired by the operation of biological neurons of sea slugs and snails.
The sensor consists of a sub-threshold CMOS circuit that changes its dynamic behavior, i.e.,
oscillatory or stationary behaviors, at a given threshold temperature.
I analyzed the circuit’s operation theoretically, giving a mathematical model of its operation.
Also, I conducted extensive numerical and circuit simulations. Furthermore, I demonstrated
the operation of the circuit, using discrete MOS devices through experimental results.
The threshold temperature, can be set to a desired value by adjusting the external bias voltage
(θ). I demonstrated that the circuit changed its state between oscillatory and stationary when
the external temperature was lower or higher than the threshold temperature. Moreover, I
experimentally calculated the circuit nullclines, indicating the trajectory of the circuit when it
is in oscillatory state.
Future work
This kind of system can be used as a sensory system for first stage of perception (a receptor).
In other words a temperature receptor circuit, which detects a tranduces physical stimuli
(temperature) into electrical impulses.
The combination of such kind of simple circuit will allow the design of hardware system
that are capable of detecting, transforming, transferring, processing and interpreting sensory
stimuli. The possibility to built complex neuromorphic systems which sense and interact
with the environment will hopefully contribute to advancements in both, basic research
and commercial applications. This technology is likely to become instrumental for research
on computational neuroscience, and for practical applications that involve sensory signal
processing, in uncontrolled environments
9. References
Asai, T., Kanazawa, Y., & Amemiya, Y. (2003). A subthreshold MOS neuron circuit based on
the Volterra system. IEEE Trans. Neural Networks, Vol. 14(5): 1308-1312.
Barranco, B. L., Sinencio, E. S., Vazquez, A. R., & Huertas, J. L. (1991) A CMOS implementation
of FitzHugh-Nagumo neuron model. IEEE J. Solid-State Circuits., Vol. 26: 956-965.
Douglas, R., Mahowald, M., & Mead, C., (1995). Neuromorphic analogue VLSI. Ann. Rev.
Neurosci. Vol. 18: 255-281.
Fletcher D. S. & Ram L. J. (1990). High temperature induces reversible silence in Aplysia R15
bursting pacemaker neuron. Comp. Biochem. Physiol.. Vol. 98A: 399-405.
Gopel, W., Hesse, J., & Zermel J. N. (1990). Sensors. A comprehensive survey, Thermal sensors
T. Ricolfi and J. Scholz, Eds. Vol. 4, VCH, pp. .
Hirose, T., Matsuoka, T., Taniguchi, K., Asai, T., & Amemiya, Y. (2005). Ultralow-power
current reference circuit with low-temperature dependence. IEICE Transactions on
Electronics, Vol. E88-C(6): 1142-1147.
Meador J. L. & Cole, C. S. (1989). A low-power CMOS circuit which emulates temporal
electrical properties of neurons. Advances in Neural Information Processing Systems 1.,
D. S. Touretzky, Ed., Los Altos, CA: Morgan Kaufmann, pp.678-685.
Murray, A. F., Hamilton, A., & Tarassenko, L. (1989). Programmable analog pulse-firing neural
networks. Advances in Neural Information Processing Systems 1., D. S. Touretzky, Ed.,
Los Altos, CA: Morgan Kaufmann, pp. 671-677. .
Liu, S., Kramer, J., Indiveri, G., Delbruck, T., & Douglas, R. (2002). Analog VLSI: circuit and
principles. Massachusetts Institute of Technology Cambridge, Massachusetts. The MIT
press, London, England.
Ryckebusch, S., Bower, J. M., & Mead, C. (1989). Modeling small oscillating biological
networks in analog VLSI, In Advances in Neural Information Processing Systems 1., D. S.
Touretzky, Ed., Los Altos CA: Morgan Kaufmann, pp. 384-393.
Wang, C. C., Akbar, S. A., & Madou M. J. (1998). Ceramic based resistive sensors. Journal of
Electroceramics, Vol. 2(4): 273-282.
Wilson, H. R. & Cowan, J. D. (1972). Excitatory and inhibitory interactions in localized
populations of model neurons. Biophys. J., Vol. 12: 1-24
16
16
Evolvable Metaheuristics on Circuit Design

Felipe Padilla1,2, Aurora Torres1, Julio Ponce1,
María Dolores Torres1, Sylvie Ratté2 and Eunice Ponce-de-León1
1Aguascalientes University,
2École de Technologie Supérieure
1México
2Canada
1. Introduction
Evolutionary computation algorithms are stochastic optimization methods; they are
conveniently presented using the metaphor of natural evolution: a randomly initialized
population of individuals evolves following a simulation of the Darwinian principle. New
individuals are generated using genetic operations such as mutation and crossover. The
probability of survival of the newly generated solutions depends on their fitness
(Michalewicz et al., 1995). Evolutionary algorithms (EAs) have been successfully used to
solve different types of optimization problems (Back, 1996). In the most general terms,
evolution can be described as a two-step iterative process, consisting of random variation
followed by selection.
The structure of any evolutionary computation algorithm is shown in the figure 1.
procedure evolutionary algorithm

t Å0
initialize P(t)
evaluate P(t)
while (not termination-condition) do
begin
tÅt + 1
select P(t) from P(t - 1)
alter P(t)
evaluate P(t)
end
Fig. 1. Structure of any evolutionary algorithm
The term evolutionary computation is used to describe techniques such as genetic
algorithms, evolution strategies, evolutionary programming and genetic programming. The
different approaches are distinguished by the genetic structures under adaption and the
genetic operators that generate new candidate solutions (Cordon et al., 2001).
Evolvable hardware (EHW) is an exquisite combination of evolutionary computation and
electronic hardware. While the most common techniques of evolutionary computation are
genetic algorithms and genetic programming, electronic hardware implies not only digital
but analog circuits also. This field has earned importance since the early 1990´s because of
the advent of reconfigurable hardware.
The ultimate objective of this field is to design and construct intelligent hardware, capable of
online adaptation (Yao and Higuchi, 1999).
The first classification of evolvable hardware can be found in (De Garis, 1993). In this work
De Garis established there are extrinsic and intrinsic EHW. While Extrinsic EHW simulates
evolution by software and downloads to hardware only the best configuration; intrinsic
EHW simulates evolution directly in hardware.
Nowadays the scope of this discipline has grown vastly. According to Zebulum (Zebulum,
1996), evolvable hardware can be classified by several criterion like hardware evaluation,
evolvable computation approach, application area and evolvable platform. In regard to its
application area EHW in divided in: Circuit design, robotics and control, pattern
recognition, fault tolerance and very large scale integration (VLSI). We are interested in
discuss about the first one.
Circuit design is the art of constructing a sized circuit from user specifications (Das and
Vemuri, 2009). This task is divided according to the kind of circuits that are handled in
digital and analog circuit design.
Nowadays there are different algorithms that can be used to solve problems of optimization
of circuits like: Genetic Programming, Genetic Algorithm, Estimation of the Distribution
Algorithms, Ant Colony Optimizations, Others.
The more amenable nature of digital circuits made researchers like Louis (Louis, 1993) and
Koza (Koza, 1992) to focus first on the production of functional logic circuits. Afterwards,
the goal was not only to obtain functional circuits, but optimum ones. The work of Louis
(Louis, 1993) was pioneer on the use of genetic algorithms on the design of combinational
circuits; Thompson et al (Thompson et al., 1996) were the first in coding logic gates and its
connections. Other outstanding researches on digital design are Higuchi et al. (Higuchi et
al., 1996) specially focused on intrinsic evolution based on neural networks; Hernández and
Coello (Hernández and Coello, 2003) first worked with genetic algorithms and later with
genetic programming and Information Theory. A very interesting case is the use of ACO on
the optimization of combinatorial circuits (Mendoza, 2001).
The analog synthesis world also has numerous successful implementations of different
metaheuristics like genetic algorithms (Lohn and Colombano, 1998), (Zebulum et al., 2000),
(Goh and Li, 2001), (Das and Vemuri, 2007), (Khalifa et al., 2008), (Torres et al., 2010); genetic
programming (Koza et al., 1997), (Hu et al., 2005)(Chang et al., 2006) and estimation of the
distribution algorithms (Torres et al., 2009). Analog circuit synthesis is a process composed
of two phases: the selection of a suitable topology and the sizing of all its components
(Torres et al., 2010). While topology consists on the determination of the type of components
and its connections; sizing refers to the selection of the components values. Further on this
document, will be discuss some of the mentioned approaches.
Others types of evolutionary algorithms are based in biological systems in which complex
collective behaviour emerges from the local interaction of simple components. Some
examples of these algorithms are Swarm Intelligence, Ant Colony, Bees Algorithm, etc. We
will speak of an ant colony, this algorithm is based in the foraging behaviour of some
species of ants. Ant colonies are capable of finding the shortest paths between their nest and
food sources, through a substance denominated pheromone.
Evolvable Metaheuristics on Circuit Design 349
2. Optimization algorithm
Actual trends in VLSI technology are towards integration of mixed analog-digital circuits as
a complete system-on-a-chip. Most of the knowledge intensive and challenging design effort
spent in such systems design is due to the analog building blocks (Balkir et al., 2004). Analog
design has been traditionally a difficult discipline of integrated circuits (IC) design. In circuit
design optimization, a circuit and its performance specifications are given and the goal is to
automatically determine the device sizes in order to meet the given performance
specifications while minimizing a cost function, such as a weighted sum of the active area or
power dissipation (Baghini et al., 2007). This is a difficult and critical step for several
reasons: 1) most analog circuits require a custom optimized design; 2) the design problem is
typically under constrained with many degrees of freedom; and 3) it is common that many
(often conflicting) performance requirements must to be taken into account, and tradeoffs
must be made that satisfy the designer (Rutenbar et al., 2007).
Fuzzy techniques have been successfully applied in a variety of fields such as automatic
control data classification, decision analysis, expert systems, computer vision, multi-criteria
evaluation, genetic algorithms, ant colony systems, optimization, etc.
Works showing the possibility of application of fuzzy logic in computer aided design (CAD)
of electronic circuits started to appear in late 1980s and early 1990s. An argument for fuzzy
logic application in CAD is derived from the nature of the algorithm used for solving design
problems. The majority of algorithms for synthesis use heuristics that are based on human
knowledge acquired through experience and understanding of problems. Another
important source of knowledge is numerical data. Fuzzy logic systems are appropriate in
such situations because they are able to deal simultaneously with both types of information:
linguistic and numerical.
Also, fuzzy systems being universal appoximators can model any nonlinear functions of
arbitrary complexity. This is very useful in modelling complex circuit functions of high
accuracy at low cost, necessary in performance evaluation.
Design optimization of an electronic circuit is a technique used to find the design parameter
values (length and width of MOS transistors, bias current, capacitor values, etc.) in such a
way that the final circuit performances (de gain, gain-bandwidth, slew rate, phase margin,
etc.) meet as close as possible the design requirements.
There is no general design procedure independent of the circuit; also, there is no formal
representation to connect the circuit functions on its structure in a consistent manner. The
major obstacle consists in the peculiarity of the analog signals: the continuous domain of the
signals` amplitude and their continuous time dependency. Hereby the analog circuit design
is known like an iterative, multi-phase task that necessitates a large spectrum of knowledge
and abilities of designers.
3. Genetic algorithms
Genetic algorithms originally were called "reproductive plans" by John Holland (Holland,
1975), and were the first emulators of the genetic evolution that produced practical results.
In 1989, when Goldberg (Goldberg, 1989) published his book, mentioned more than 70
successful applications of this paradigm that continues winning popularity nowadays.
According to Coello (Coello, 1996), a good definition of genetic algorithm was established
by Koza in his book of 1992 (Koza, 1992), he says the following: "The genetic algorithm is a
highly parallel mathematical algorithm that transforms a group (population) of individual

mathematical objects (that usually have the form of chains of characters of fixed longitude),
each one with an associate aptitude value, in new populations (for example the following
generation) using modelling of operations under the Darwinian principle of the
reproduction and survival of the “most capable”, naturally, after the occurrence of the
genetic operators (sexual recombination)".
Ponce de León (Ponce de León, 1997) summarizes the mechanism of operation of the simple
genetic algorithm in the following way; "it is generated a population of n structures
aleatorily (chains, chromosomes or individuals) and then, some operators act transforming
the population. The transformation is carried out by means of the application of three
operators; once this culminates, it is said that a generational cycle has finished". The three
operators Ponce references are: selection, crossover and mutation.
The genetic algorithm in the form like Holland illustrates it (Holland, 1975) has the
following characteristic elements:
1. Representation of binary chains.
2. Proportional selection.
3. Crossover like the main method to produce new individuals.
After the Holland's proposal, have been carried out different modifications; either by means
of the use of different representation outlines, or until certain modifications to the selection
operators, crossover, mutation and elitism.
The diagram shown in the following figure presents the simplest version in the genetic
algorithm, well-known as SGA (for the initials in English of "Simple Genetic Algorithm").
Simple Genetic Algorithm
1. Let t=0 generations counter

2. Initialization P(t)
3. Evaluate P(t)
4. While stop criterion do not be reached, do:
for i=1,…,N/2 do
i. Select 2 individuals from P(t)
ii. Apply crossover with probability pc
iii. Apply mutation to the offspring with probability pm
iv. Introduce the 2 new individuals to P(t+1)
end-for
Fig. 2. Pseudocode of SGA

Although the general mechanism of this algorithm is extremely simple, it can be
demonstrated by means of Markov’s chains that the evolutionary algorithms that use elitist
selection mechanisms, will converge to a good global solution of certain functions whose
domain can be an arbitrary space (Torres, 2010). Günter Rudolph in 1996, generalized the
previous developments in theory of the convergence for binary search spaces and Euclidian
ones to general search spaces (Rudolph, 1996).
3.1 Genetic algorithms in automated analog design

Due to the high level of complexity that implies the task of designing and also to the strong
dependence that this task has with the knowledge and experience experts; the automatic
design of analogical circuits is a challenge and a necessity. Some researchers of the area believe
that the automation of the design should be preceded by a change in the process of current
design, for example, governed by the execution of the restrictions (Jerke, 2009). The fact is that
nowadays, it has not still been possible to automate this process in a complete way.
One of the metaheuristics that have shown better benefits in the realization of this task are
the genetic algorithm and the genetic programming; this space belongs to the genetic
algorithm.
Lohn and Colombado (Lohn and Colombado, 1998) used the genetic algorithm to design
two analog filters, one of low complexity and one of medium complexity. The contribution
of these researchers resides in that they demonstrated that it was feasible to use a very
simple lineal representation. They proposed a code outline in which each element was
represented by a fixed number of bytes called bytecodes in which they included an
operation code that dictated the connection of each element and three bytes more they used
to code its value.
Koza on the other hand, continued making use of the genetic programming in the synthesis
of computational circuits (Koza, 1997b) and controllers, filters and other kind of circuits
(Koza, 1997).
According to Ricardo Zebulum and his collaborators (Zebulum et al., 1998), the
Evolutionary Electronics is an area that seeks to find new techniques of automatic design
based on Darwinian concepts. The authors of the mentioned work, made the comparison of
three different methodologies in the design of electronic filters. Their work was put on
approval with two cases of study: A low-pass filter discussed in (Koza, 1996) and a filter
pass-band with band in passing between 2000 and 3000 Hz and the bands of rejection above
4000 and below 1000Hz. The methodologies on approval were the following:
"Outline of representation of variable longitude in combination with an evolutionary
algorithm that restricts the topology of the filter (parallel meshes of two elements each one).
For the simulation, an own tool was used in C, based on Laplace´s analysis.
"Outline of representation of fixed longitude in combination with an evolutionary algorithm
that doesn't restrict the topology of the circuit. To analyse the circuits they used Smash and
SPICE, obtaining the same results.
"Outline of representation of variable longitude in combination with an evolutionary
algorithm that doesn't restrict the topology of the circuit. For the simulation of the circuits
they used as much Smash as SPICE, obtaining the same results.
In this work, Zebulum and his collaborators demonstrated that making use of an
evolutionary algorithm based on the "Genetic Algorithm of Adaptation of Species (SAGA)
of Harvey (Harvey, 1993), they could be obtained results comparable with those obtained
using genetic programming, as for the answer in frequency of the obtained circuits using
much smaller populations. This work concludes settling down that as for time, the first
methodology was better, however this can explain to you for the rigidity of the used
topology that allowed the use of a tool of quicker simulation. In spite of the success of this
work, all the methodologies had inducer circuits whose values were so big as a result (2.2H
for example) that are not very practical. On the other hand, investigators as Grimbleby and
their collaborators (Grimbleby et al., 1995) they were working with mechanisms of numeric
optimization in combination with genetic algorithms for the synthesis of analogical circuits
using a chromosome of fixed length and a type of null component to fight with the variable
size of the real circuits.
The XXI century has also been witness of numerous efforts made toward the automation of
the synthesis of the analogical circuits, for example, in the year 2000, Zebulum et al.
(Zebulum et al., 2000), established some advantages of variable length representation
systems. Among other things, they argued that when using a fixed size, it is not only
required expert knowledge of the problem, but the potential of the evolutionary algorithms
is also limiting. That same year, they also proved an outline of representation of variable
longitude that they understood passive elements, connected nodes and disconnected nodes.
The authors emphasize the use of resistances and capacitors with programmable values in
their architecture. These investigators intend to work the two phases of the evolution of an
electric circuit (topology and adjustment of the parameters) in a sequential way, instead of
making it simultaneously.
In the year 2001, the investigating Goh and Li (Goh and Li, 2001) they began to outline some
of the weaknesses that persisted in the process of design of analogical circuits that they were
commented later by investigators as Khalifa and their collaborators (Khalifa et al., 2008),
(Das, 2008) among others.
The weaknesses that these investigators declare that they should be assisted, the reduction of
the enormous computational effort that implies the evaluation of big generations of circuits
Year Author Application

Horrocks and Active low-pass
1993
Spittle filter
Horrocks and
1994 Low-pass filter
Khalifa
1995 Grimbleby High-pass filter
Horrocks and
Khalifa
Lohn and
Colombano
Low-pass filter
1998 Zebulum et al.
Band- pass filter
1999 Krasnicki et al. OP-AMP
2000 Ando and Iba Passive filters
2000 Zebulum et al. Passive filters
Low-pass filter
2001 Goh and Li
High-pass filter
2007 Das and Vemuri Low-pass filter
Low-pass filter
2008 Khalifa et al.
High-pass filter
2008 Das and Vemuri OP-AMP
2010 Torres et al. Low-pass filter
Table 1. Relevant research on analog circuit synthesis using Genetic Algorithms (Torres,
2010).
that they don't always produce results and the reduction of the breach between the evolved
circuits and those that finally are taken to the physical implementation, due to the restrictions
of commercial physical devices. Other equally important aspects are related with the
elaboration of tools that due to their complexity, they require expert personnel's manipulation
or with a considerable level of knowledge (Krasnicki, 2001); as well as the execution in teams
whose level of sophistication is outside of the reach of a great number of people.
4. Estimation of Distribution Algorithms

Estimation of distribution algorithms (EDA's) constitute a relatively new field of the
Evolutionary Computation (Larrañaga, 2002) that replaces genetic operators (crossover and
mutation) for the estimation of the distribution of the selected individuals and the sampling
from the distribution to obtain the new population.
The objective of this paradigm is to avoid the use of arbitrary operators as crossover and
mutation, to modeling explicitly the most promising solutions for sampling solutions from
its distribution.
Pseudocode of the algorithm EDA:
Step 1: Random generation of M individuals (initial Population)
Step 2: Repeat the steps 3-5 for the generation l=1, 2,… until an stop criterion is reached
Step 3: Select N <= M individuals from Dl-1 according to a selection method
Step 4: Estimate the distribution of probability pl(x) from the group of selected individuals
Step 5: Sample M individuals (new population) from pl(x)
EDAs can be classified according to two fundamental approaches. The first is the level of
interdependences of variables, and the second is the type of involved variables. With regard
to the level of interdependences EDAs are divided in 3, when the variables are independent,
when there are bivaluated dependences and when there are multiple dependences. With
regard to the type of involved variables, they can be discrete, continuous or mixed.
The easiest version of an EDA is the “Univariate Marginal Distribution Algorithm” (UMDA)
introduced by Mühlenbein (Mühlenbein and PaaE, 1996). This algorithm works on the
supposition of complete independence among variables. Pseudocode of this algorithm in
presented in figure 3.
UMDA_AC
1. Begin
2. D0 m Generate M individuals at random
3. RepeatSefor l= 1,2,… until the stopping criteria met
a) Dl 1 m Select N d M individuals from Dl-1 according to the selection
¦
N
method G j ( X i xi | D lSe 1 )
b) pl(x) = p ( x | D Se
l 1 ) 3 n
p ( x )
= i 1 l i = 3i 1
n j 1
m
N
Estimate
Se
the joint probability distribution
c) Dl 1 m Sample M individuals from pl(x)
Fig. 3. Pseudocode for UMDA (Larrañaga, 2002).

Another very common approach for the estimation of the distribution supposing
independence among the variables is the algorithm PBIL ("Population-based incremental
learning") (Baluja, 1994) that contrary to UMDA, doesn't estimate a new model in each
generation, but refines it.
The main problem of the distribution of the estimation algorithms, is to estimate the model;
because as it gets more complicated, the dependences among the variables are captured in a
better way, however, its estimation becomes more expensive (Larrañaga, 2002). Regarding
models that consider bivariated dependences (dependences among pairs of variables), the
most outstanding methods according its use in the literature are those that use chains like
the “MIMIC” algorithm (Mutual Information Maximizing Input Clustering Algorithm) (De
Bonet et al., 1996), those that use trees, as the case of the COMIT (Baluja and Davies, 1997)
that uses the method of Chow and Liu [Chow 1968] based on the concept of mutual
information and the BMDA (Pelikan, 1999), in which Pelikan and Mühlenbein propose a
factoring of the distribution of joint probability. This algorithm is based on the construction
of an acyclic directed graph of dependences that is not necessarily connected.
Finally, the most common n-varied models are those that allow estimating a model in a
Bayesian-net form. This approach has originated a great variety of algorithms according to
the learning method, according to the nature of the variables (discrete or continuous),
according to the imposed restrictions, etc. (Larrañaga, 2002).
The great success genetic algorithms (GAs) have shown on several synthesis problems, has
motivated some researches to explore the EDA´s world in analog circuit synthesis. Next
table show some examples.
Year Author Application Used metaheuristic

2002 Mühlenbein et al. Low-pass UMDA
2007 Zinchenko et al. Mixed circuit UMDA
2009 Torres et al. Filters UMDA
2010 Torres et al. Filters MITEDA
Table 2. Relevant works on analog circuit synthesis by means of Estimation of the
Distribution Algorithms
From table 2 it can be seen UMDA is the most common approach implemented on the
analog circuit synthesis, nevertheless, MITEDA represents an effort on exploring the
behavior of more complex EDAS. This algorithm was developed inspired by the COMIT
and it uses the concept of mutual information used by Baluja and Davies (Baluja, 1997) to
build the tree of dependences. Later this tree is sampling in order to create new generations.
This algorithm represents the first tool that considers bi-valuated dependencies used in the
design of analogical circuits we know until this moment.
5. Ant Colony Optimization

The Ant Colony Optimization Algorithm is a meta-heuristic bio-inspired in the behavior of
real ant colonies. The first algorithm which can be classified within this framework was
presented in 1991 by Marco Dorigo. In his PHD thesis with Title: “Optimization, learning,
and Natural Algorithms”, modeling the way real ants solve problems using pheromones.
Real ants are capable of finding the shortest path from a food source to their nest. The ants
deposit a concentration of pheromone in theirs paths, and they follows with more
probability the way with more concentration of pheromone that it was previously deposited
by other ants, the essential trait of ACO algorithms is the combination of a priori
information about the structure of a promising solution with a posteriori information about
the structure of previously obtained good solutions. In the Ant Colony Algorithms a
number of artificial ants (agents) build solutions for an optimization problem and exchange
information on their quality via a scheme of global communication that is reminiscent of the
one adopted by real ants.
When exist paths without any amount of pheromone, the ants explore the neighbourhood
area in a totally random way. In presence of an amount of pheromone, the ants follow a
path with a probability based in the pheromone concentration. The ants deposit additional
pheromone concentrations during his travels. Since the pheromone evaporates, the
pheromone concentration in non-used paths tends to disappear slowly.
To find the shortest path, a moving ants lay some pheromone on the ground, so an ant
encountering a previously trail can detect it and decide with high probability to follow it. As
a result, the collective behavior that emerges is a form of a positive feedback loop where the
probability with which each ant choose the next path increases with the number of ants that
previously chose the same path.
The Ant Colony System (ACS) models the behavior of ants, which are able to find the
shortest path from their nest to a food source. Although individual ants move in a quasi-
random form, performing relatively simple tasks, the entire colony of ants can collectively
accomplish sophisticated movement patterns. Ants accomplish this by depositing a
substance called a pheromone as they move. This chemical trail can be detected by other
ants, which are probabilistically more likely to follow a path rich in pheromone. This trail
information can be utilized to adapt to sudden unexpected changes to the terrain, such as
when an obstruction blocks a previously used part of the path.
5.1 Application of ant colony to the design of combinatory logic circuits

To apply Ant Colony Algorithm to the design of logic circuits, in (Mendoza, 2001) is shown
as the design of logic circuits with ACO. In the case of the logic circuits, the treatment of the
problem does not seem to be so immediate.
5.2 Circuit representation

The circuits are represent used a bidimensional matrix. Where each element of the matrix is
a triplet of the type [Entrance 1, Entrance 2, Type of floodgate] (see figure 5). Was used five
types of floodgates: AND, OR, NOT, XOR and WIRE, although this last one is not a
floodgate, but rather it is a connection (a wire) that unites an element of certain column with
another one of the previous column. Each element of the matrix receives its entrances solely
of the exits of the previous column.
Fig. 4. Basic floodgate Not, Or, And

The first column directly receives its entrances of the table really of the given circuit. The last
column provides the exits of the circuit. The first N rows corresponds to the N exits of the
circuit. This form to represent a circuit has been used successfully.
In the following figure are shown the basic floodgate.
Fig. 5. Matrix used to circuit representation
5.3 Implementation
The route of an ant or agent will be a complete circuit. While each ant crosses a route, it
constructs a circuit. In the TSP the ants find the route in terms of distance, do it here in terms
of the number of floodgates.
A state or city is a column, which is made up of several elements to which it is called
substates to them, being these each one of the floodgates of a column and the number of
combinations of possible entrances of each floodgate of this column. The first N substates (N
is the number of exits in the circuit) is chosen with a selection factor P, and the others are
chosen randomly.
The distance between cities or states is measured as the increase or diminution from the
successes to the exits of the circuit when changing from a level to another one.
Unlike the problem of the TSP, in a same route (circuit), they do not have to visit all the
states.
The pheromones keep in a matrix called Trails. The length of this matrix corresponds to the
number of exits of the circuit. Each element of Trails is a three-dimensional matrix as well.
Next it is explained what they represent each one of the dimensions of the element. The first
dimension of this matrix corresponds to the combination of possible entrances to the
floodgate and goes from 0 to 6. The possible combinations of entrances, independent of the
incoming number of the table really.
The second dimension corresponds to the number of floodgate, that is to say, goes of 0 to the
number of floodgates except one (NumGates-1). The third dimension corresponds to the
number of successes that take until the level (column) previous and really goes of 0 to the
number of lines in the table, because the number of successes that can be had in any level is
between 0 and the number of lines of the true table.
5.3.1 The construction of a solution (route)

As it was already mentioned before, a state is a column of the matrix, each element of the
column is a floodgate with its respective entrances and their exit. Because of that, the
election of a state is a process that becomes by parts (floodgate by floodgate), reason why we
will call to each floodgate (element of the column) a substate. A state a combination of three
elements (floodgate, IN1, IN2). In order to choose a substate of anyone of the first N rows, a
value is assigned to him to each one of the possible combinations, call selection factor P,
with which it will compete remaining in that position.
The distance is a heuristic value and is given by the number of successes that the portion of
the circuit constructed until the moment produces with respect to exit l of the True table.
This is analogous to the distance in the TSP.
Once it has assigned a factor of selection to all the combinations, is chosen what of them
remains in the position in game. This is repeated with all the substates that belong to one of
the rows that represent an exit of the circuit. The other substates, are chosen randomly. This
is repeated until arriving at the last state from the circuit or column of the matrix. When all
the ants finish their route, the pheromone signs are updated. This becomes in two steps:
1. First the amount is due to update pheromone in the ways, simulating the pheromone
evaporation of the ways by the artificial ants to the passage of time.
2. The ways are due to update or to increase according to the routes constructed by each
ant in the algorithm. This becomes of the following form: If the circuit result of the route
is not valid (that it does not produce all the exits).
6. Multiobjective optimization
A population based evolutionary multiobjective optimization approach (Coello, 2009) to
design combinatorial circuit was proposed for first time by Coello and Hernández in 2000
(Coello and Hernández, 2000). This approach reduced the computational effort required by
genetic algorithm to design circuit at gate level. The main motivation was the reduction of
fitness function evaluations while keeping the capabilities of the GA to generate novel
designs. The main ideas behind MGA algorithm are:
1. Circuit representation as a matrix (originally proposed by Louis in 1991 (Louis and
Rawlins, 1991)) and an n-cardinality alphabet.
2. Incremental method to resized of matrix used to fit a circuit.
3. Fitness function in two stages. At the beginning only validity of the circuit outputs is
taken into account, and at the ending the fitness function is modified such that any
valid designs produced are rewarded for each WIRE gate that they include. (WIRE gate
indicates a null operation, that is, the absence of gate)
4. Use a multi-objective optimization technique (Fonseca and Fleming, 1995) (Coello,
1999). In general, it redefines the single-objective optimization of as a multiobjective
optimization problem in which we will have ൅ ͳ objectives, where m is the number
of constraints. There is a new vector, ത ൌ ሺǡ ଵ ǡ ǥ ǡ ୬ ሻ), where is the objective
functionǤଵ ǡ ǥ ǡ ୬ are the original constraints of the problem. An ideal solution X
would thus have ୧ ሺሻ ൌ Ͳ for ൌ ͳǡ ǥ ǡ ǡ and ሺሻ ൑ ሺሻ for all feasible (assuming
minimization). For combinatorial logic circuit design this technique consists on using a
population based multiobjective optimization technique such as VEGA (Schaffer, 1984)
to handle each of the outputs of the circuit as an objective. At each generation, the
population is split in to ൅ ͳ sub-populations, ൌ ʹ୬ (outputs), n: inputs of the
circuit. The main mission of each sub-population is to match its corresponding output
with the value indicated by the user in the truth table. After one of these objectives is
satisfied, its corresponding sub-population is merged with the rest of the individuals in
what becomes a joint effort to minimize the total amount of mismatches produced
(between the encoded circuit and the truth table). Once a feasible individual is found,
all individuals cooperate to minimize its number of gates (Coello and Hernández, 2002).
The MGA algorithm outperformance the GA algorithm in quality of solution and decreased
the evaluation amount of fitness function. This approach made a path in solving
evolutionary design of combinational logic circuits.
6.1 Formulation of multiobjective optimization problem

The multiobjective optimization problem can be formulated as follows (Coello and
Hernández, 2000):
A General Multiobjective Optimization Problem ሺࣩࣧ࣪ሻ: Find the vector ሬԦ ‫ כ‬ൌ ሾଵ‫ כ‬ǡ ǥ ǡ ୬‫ כ‬ሿ୘
which will satisfy the m inequality constraints:
୧ ሺሬԦሻ ൒ Ͳǡ ൌ ͳǡ ǥ ǡ (1)
the ‫ ݌‬equality constraints
݄௜ ሺ‫ݔ‬Ԧሻ ൌ Ͳǡ ݅ ൌ ͳǡ ǥ ǡ ‫݌‬ (2)

and optimizes the vector function
݂Ԧሺ‫ݔ‬Ԧሻ ൌ ሾ݂ଵ ሺ‫ݔ‬Ԧሻǡ ǥ ǡ ݂௞ ሺ‫ݔ‬Ԧሻሿ் (3)

்
where ‫ݔ‬Ԧ ൌ ሾ‫ݔ‬ଵ ǡ ǥ ǡ ‫ݔ‬௡ ሿ is the vector of decision variables.
That is, we wish to determine from among all ‫ݔ‬Ԧ ൌ ሾ‫ݔ‬ଵ ǡ ǥ ǡ ‫ݔ‬௡ ሿ் , which satisfy the
inequality and equality constraints above, the particular ‫ݔ‬Ԧ ‫ כ‬ൌ ሾ‫ݔ‬ଵ‫ כ‬ǡ ǥ ǡ ‫ݔ‬௡‫ כ‬ሿ் which yields the
optimum values of all the ݇ objective functions of the problem. Let be π the set defined as all
vectors ‫ݔ‬Ԧ ൌ ሾ‫ݔ‬ଵ ǡ ǥ ǡ ‫ݔ‬௡ ሿ் , that do not violate the constraints.
Pareto Optimality Definition: We say that ‫ݔ‬Ԧ ‫ כ‬ൌ ሾ‫ݔ‬ଵ‫ כ‬ǡ ǥ ǡ ‫ݔ‬௡‫ כ‬ሿ் ‫ א‬πǡ π ‫ ك‬Թ௡ ǡ ݂௜ ǣ Թ௡ ՜ Թ, is Pareto
optimal if for every ‫ݔ‬Ԧ ൌ ሾ‫ݔ‬ଵ ǡ ǥ ǡ ‫ݔ‬௡ ሿ் , and ‫ ܫ‬ൌ ሼͳǡ ǥ ǡ ݇ሽ either,
‫ٿ‬
௜‫א‬ூ ሺ݂௜ ሺ‫ݔ‬
Ԧ ሻ ൌ ݂௜ ሺ‫ݔ‬Ԧ ‫ כ‬ሻሻ (4)
Or, there is at least one ݅ ‫ܫ א‬such that
`݂௜ ሺ‫ݔ‬Ԧ ሻ ൐ ݂௜ ሺ‫ݔ‬Ԧ ‫ כ‬ሻ (5)

‫ݔ‬Ԧ ‫ כ‬ൌ ሾ‫ݔ‬ଵ‫ כ‬ǡ ǥ ǡ ‫ݔ‬௡‫ כ‬ሿ் is Pareto optimal if there exists no feasible vector ‫ݔ‬Ԧ ൌ ሾ‫ݔ‬ଵ ǡ ǥ ǡ ‫ݔ‬௡ ሿ் which
would decrease some criterion without causing a simultaneous increase in at least one other
criterion.
Pareto Dominance Definition: A vector ‫ݑ‬ ሬԦ ൌ ሺ‫ݑ‬ଵ ǡ ǥ ǡ ‫ݑ‬௡ ሻ is said to dominate ‫ݒ‬Ԧ ൌ
ሺ‫ݒ‬ଵ ǡ ǥ ǡ ‫ݒ‬௡ ሻ (denoted by ‫ݑ‬ ሬԦ ‫ݒ ع‬Ԧ ) if and only if ‫ݑ‬
ሬԦ is partially less than ‫ݒ‬Ԧ , i.e., ‫א ݅׊‬
ሼͳǡ ǥ ǡ ݇ሽǡ ‫ݑ‬௜ ൑ ‫ݒ‬௜ ‫ א ݅׌ ٿ‬ሼͳǡ ǥ ǡ ݇ሽǣ ‫ݑ‬௜ ൏ ‫ݒ‬௜ .
Pareto Optimal Set Definition: For a given गङच, ݂Ԧሺ‫ݔ‬Ԧሻ ൌ ሾ݂ଵ ሺ‫ݔ‬Ԧሻǡ ǥ ǡ ݂௞ ሺ‫ݔ‬Ԧሻሿ் , the Pareto
optimal set ሺ࣪ ‫ כ‬ሻ is defined as:
࣪ ‫ כ‬ൌ ሼ‫ݔ‬Ԧ ‫ א‬π ോ ‫ݔ׍‬Ԧ ᇱ ൌ ሾ‫ݔ‬ଵᇱ ǡ ǥ ǡ ‫ݔ‬௡ᇱ ሿ் ‫ א‬πሺ݂Ԧሺ‫ݔ‬Ԧ ᇱ ሻ ‫݂ ع‬Ԧሺ‫ݔ‬Ԧሻሻሽ (6)

Pareto Front Definition: For a given गङच, ݂Ԧሺ‫ݔ‬Ԧሻ ൌ ሾ݂ଵ ሺ‫ݔ‬Ԧሻǡ ǥ ǡ ݂௞ ሺ‫ݔ‬Ԧሻሿ் ǡand Pareto optimal set
࣪ ‫ כ‬, the Pareto front ሺ࣪࣠ ‫ כ‬ሻ is defined as:
ሬԦ ൌ ݂Ԧሺ‫ݔ‬Ԧሻ ൌ ሾ݂ଵ ሺ‫ݔ‬Ԧሻǡ ǥ ǡ ݂௞ ሺ‫ݔ‬Ԧሻሿ் ോ ‫ݔ‬Ԧ ‫ כ ࣪ א‬ሽ

࣪࣠ ‫ כ‬ൌ ሼ‫ݑ‬ (7)
7. Application
Due to the enormous success genetic algorithms has proved on the field of circuit design,
this section has the purpose of show how this metaheuristic could be used for the synthesis
of analog circuits.
In order to implement a genetic algorithm for the artificial evolution of any kind of process,
is indispensable to find a way to represent a solution of the given problem, to find the way
to generate possible solutions, to be able to evaluate the quality of the solutions and to have
a group of operators that let transform one solution into another. Figure 6 shows the general
flow used to implement a genetic algorithm in the analog circuit design according to Azizi
(Azizi, 2001).
Population
Initialization
Fitness Decode Spice Fitness

Calculation Chromosome Simulation Measure
Selection ,
Crossover, Mutation
Fitness
Calculation
No Found
Yes
Finish
Solution?
Fig. 6. Genetic Algorithm Flow for Analog Circuit Synthesis.

Representation mechanism
In order to initialize the first population, the programmer has to establish how each solution
is going to be represented and how the population can be generated.
A genetic encoding for artificial evolution of analog networks must be capable of representing
both; the topology and the sizing of the network (Mattiussi and Floreano, 2007). While
topology refers to the way each element is going to be connected to each other; sizing refers to
the type and dimension of each element on a net. Other important aspects of the
representation mechanism are its ability to capture any kind of circuit and the chance to reduce
the process and time inverted in translate the circuit into a netlist (net description list). The
representation mechanism has also to be flexible enough to be used with a wide range of
components values but sufficiently short to be computational handling. (Torres et al., 2009).
Torres et al (2009), reported a representation mechanism for passive elements of two
terminals. This mechanism uses a gene of six parts to represent an analog element as figure
7 shows. Each circuit is a linked list of several genes.
Node 1 Node 2 Current N Type Decade Value
Topology Sizing
Fig. 7. Gene description.
While node 1 and node 2 refers to the terminals of an electrical device; current N is a pointer
that is going to be used to build the network. Type, decade and value are the parameters
that completely characterize a specific element (Torres, 2009). These parameters use integer
coding according to table 3.
Type Decade Value

C(0) 10-6 – 10-9 (0-3) E6 (0-5)
R(1) 10+3 – 10+6 (0-3) E12 (0-11)
L(2) 10-3 – 10-6 (0-3) E12 (0-11)
Table 3. Sizing encoding system
Next figure shows an element and its corresponding gene. We refers to initial node, that
represents the beginning of an analog circuit.
10 μF
Node I Node 2
I 2 2 0 0 0
Actual
node
Fig. 8. An element of circuit and its corresponding gene.
Generation mechanism
Once, a representation mechanism has been selected, the generation routine need to be
established. The generation mechanism proposed by Torres et al. (2009) is based on an
operation code randomly generated. The operation code establishes the connection that has
to be done in the construction process of an admitted topology. The process begins in
“Initial node” and ends when certain termination criterion is reached. This criterion could
be one of two possibilities: the connection is done with the “Final Node” or the circuit
reaches a preset amount of elements.
Next figure describes how the generation mechanism works (CNode refers to the current
node, and INode corresponds to Initial Node) (Torres et al., 2009).
Generation mechanism
1. begin
2. CNode <- INode
3. while(Not meet termination criterion)
• Node1 = Cnode
• Generate OP-Code
• Execute_connection (Update Node2 and Cnode)
• Generate Type, Decade and Value
4. end_while
5. end
Fig. 9. Algorithm for the generation of each solution.

The circuit creation process performed by the former algorithm is very flexible. Once the
operation code has been chosen and the connection has been done, type, decade and value
of each element are generated. All operation codes used and their meaning are depicted in
table 4.
Op code Instruction
0 Connect to grown
1 Connect to final node
2 Connect to x node
3 Connect to new node
Table 4. The operation code of the generation mechanism (Torres et al. 2010)
Evaluation function
Evolvable process depends on the ability to distinguish good and bad solutions, because it
consists in continuously improve solutions from one generation to another. Therefore, a
fitness function that describes how close a circuit is from the target is needed.
Within the scope of analog circuit design, filters and amplifiers are the most frequently
discussed. Fitness function used on the synthesis of low-pass filter will be presented below.
Filters are circuits that block certain frequencies or bands of frequencies (Curtis, 2003). A
low pass filter is the one that let pass low frequencies while blocks high frequencies. Next
figure, illustrates the frequency response of an ideal and a real filter.
pass band stopband pass band stopband

Amplitude
Amplitude
fp fc fs fp fc fs Frequency (Hz)
Ideal low-pass filter. Real low-pass filter.
Fig. 10. Frequency response of an ideal and a real low-pass filter.
The fitness function used by Torres et al, is based on the measurement of the distance
between the ideal and the real (evolved) filter. This function is an adaptation of the one used
by Koza (Koza et al., 1997) and Hilder and Tyrrell (Hilder and Tyrrell, 2007) among other
researchers. This function is the sum of errors between the ideal frequency response and the
actual candidate, along N sampling points. Equation 8 describes the fitness measure for
filters.
1
F = (8)
1[
Where :
N
[ = ¦ O (H i ) *H i (9)
i 1
Hi = M( f i )T arg et M( f i )Actual (10)
“[” represents the error over the N points of frequency. If the deviation from target
magnitude is inacceptable according to the frequency band, then a penalty factor “O” has to
be assigned to the error function.
A sample error function “H” give us the absolute deviation between the actual output
response and the target response over the “i” sampling point. M(fi)Target denotes target
magnitude at a fi frequency, M(fi)Actual is the magnitude of the actual evolved circuit at a fi
frequency and fi is the sampling frequency.
Transformation of a solution
Finally, when representation, generation and evaluation of candidate solutions have been
solved, the programmer needs to find a group of operators to transform one solution into
another. Starting from two parents chosen by any selection routine, an offspring is produce
through two possible operators: crossover and mutation.
There are several selection algorithms; one of the more popular is the roulette-wheel.
Roulette-wheel selection is an operator used for selecting potentially useful solutions for
recombination. The fitness level of each solution is used to associate a probability of
selection. If fi is the fitness of individual i in the population, its probability of being selected
ϐ୧
is ൌ σ౤ , where n is the number of individuals in the population.
ౠసభ ϐ୨
Crossover operation, introduces new solutions into the genetic algorithm starting from
previous circuits; this operator is the responsible of changing some parts of a circuit by parts
from another one. According to Dastidar et al., (Dastidar et al.,2005) and Das and Vemuri
(Das and Vemuri, 2007), the use of some suitable connectivity rules, can reduce the
unwanted search space not only for active, but for passive circuit synthesis. The crossover
operator proposed by Torres et al., generates topological modifications because it alters the
connection order of the offspring. This operator can be applied to one or two crossover
points.
Next figure shows how this operator can be executed on the condense chromosome of two
progenitors, using the representation mechanisms proposed by Torres et al. In the figure
“T” refers to ground connection and “F” represents the final node of the analog circuit. This
condense representation of each solution only has connection nodes and type of each
element.
Crossover point =2
Offspring
Progenitor 1 I20 2T2 211 2F2

Offspring 1 I20 2T2 2T1 230 341 4F2
Progenitor 2 I21 230 3T1 340 451 5F2

Offspring 2 I21 230 321 3F2
Fig. 11. Crossover operator (Torres et al., 2010)

Mutation is an operator that traditionally introduces new solutions modifying only one
chromosome. There are several ways to implement mutation, Goh and Li (Goh and Li, 2001)
show a nice group on operators. The mutation exhibit in this section was proposed by
Torres et al. This mutation operator is executed at gene level; and it works by altering a
randomly chosen gene with another randomly generated. A mutated gene corresponds to a
different type of element with different value, but connected to the same pair of nodes
(Torres et al., 2010). Next figure shows an example of the use of this operator.
Original gene 2T021
Mutated gene 2T120
Fig. 12. Mutation operator.

Using all elements discuss in this section, the interested reader can implement an effective
genetic algorithm for the automated synthesis of a passive filter.
8. Conclusion and the future research

Nowadays exist applications in real life problems, where is possible used evolvable
metaheuristics based on populations to the circuit design process, in this chapter was
present some algorithms and applications, like Genetic Algorithms, Estimation of the
distribution algorithms, Ant colony optimization.
As shown there are multiple metaheuristics that can be used to circuit design trough
different representations. We describe how is the representation with Genetic Algorithms.
Since avoiding non valid topologies and non simulable networks, implies a very high
reduction on time and computational resources in our problem; mainly three algorithms
were compared at designing a low pass filter; a genetic algorithm (GA-AC), Ant Colony
Systems (ACO-AC) and an estimation of distribution algorithm (UMDA-AC). Experimental
results demonstrated that the group of mechanisms used in theses algorithms, worked
better with GA-AC than with UMDA-AC and ACO-AC, according to the Pearson's Chi-
squared tests with respect to the generation of low rate of non spice-simulable circuits.
Although UMDA-AC and ACO-AC performed faster the execution, and found a better
individual on 200 generations’ execution; statistically it cannot be said, the time difference is
significant.
With respect to the number of fitness evaluations, it can be said with statistical base, that
UMDA-AC performs less evaluations than GA-AC per execution. In order to improve the
performance of this algorithms, next step is the creation of a tool that blends the strengths of
each metaheuristic. The work team is already working on the design of some new operators
to be inserted on the EDA-AC and ACO-AC.
GA-AC could be improved by enhancing the algorithm with some mechanisms of diversity
control, like other kind of operators and another type of selection, in order to improve its
exploration and delays its convergence.
As future work is to continue working with various tools and algorithms that allow us to
improve new circuit design.
A new Artificial Intelligence that can be in charge of these systems, continues being distant
into the horizon, in the same way that we still lack of methods to understand the original
and peculiar things of each form to represent circuits.
9. References
Back, T.: (1996). Evolutionary Algorithms in Theory and Practice. Oxford University Press,
New York.
Baghini, M.; Kanphade, R.; Wakade, P.; Gawande, M.; Changani, M.; Patil, M. (2007). GP-
based Design and Optimization of a Floating Voltage Source for Low-Power and
Highly Tunable OTA Applications, WSEAS Transactions on Circuits and Systems,
Issue 10, Volume 6, October 2007, pp. 588-582.
Balkir, S.; Dundar, G.; Alpaydin, G. (2004). Evolution Based Synthesis of Analog Integrated
Circuits and Systems, IEEE NASA/DoD Conference on Evolution Hardware,
EH`04. Pp 26-29.
Baluja, S. (1994). Population-Based Incremental Learning: A Method for Integrating Genetic
Search Based Function Optimization and Competitive Learning. Technical Report
TR CMU-CS 94-163, Carnegie Mellon University.
Baluja, S. and Davies, S. (1997) Combining Multiple Optimization Runs with Optimal
Dependency Trees. Technical Report TR CMU-CS-97-157, Carnegie Mellon
University.
Chang, S.; Hou, H. and Su, Y. (2006). Automated Passive Filter Synthesis Using an Novel
Tree Representation and Genetic Programming. IEEE ransactions on Evolutionary
Computation, Vol. 10. No.1, February 2006. Pp. 93-100.
Coello, C. (1996). An Empirical Study of Evolutionary Techniques for Multiobjective
Optimization in Engineering Design, PhD thesis, Department of Computer Science,
Tulane University, New Orleans, Louisiana, USA..
Coello, C. A. A Comprehensive Survey of Evolutionary-Based Multiobjective Optimization

Techniques. Knowledge and Information Systems. An International Journal,
1(3):269–308, August 1999.
Coello, C. A. and Hernández, A. (2002) Design of combinational logic circuits through an
evolutionary multi-objective optimization approach. Artificial Intelligence for
Engineering, Design, Analysis and Manufacture, 16(1): 39-53.
Coello, C. A. Hernández, A. and Buckles, B. P. (2000) Evolutionary Multiobjective Design of
Combinational Logic Circuits, eh, pp.161-172, The Second NASA/DoD Workshop
on Evolvable Hardware (EH'00).
Coello, C. A., Lamont and, G. B., and Van Veldhuizen, D. A. (2007) Evolutionary
Algorithms for Solving Multi-Objective Problems, Second Edition, Springer, New
York, ISBN 978-0-387-33254-3.
Cordon, O.; Herrera, F.; Homann, F. and Magdalena, L. (2001). Genetic Fuzzy Systems:
Evolutionary Tuning and Learning of Fuzzy Knowledge Bases, World Scientic.
Das A. and Vemuri R. (2007). An Automated Passive Analog Circuit Synthesis Framework
using Genetic Algorithms. IEEE Computer Society Annual Symposium on VLSI
ISVL ’07. pp. 145-152.
Das, A. (2008) Algorithms for Topology Synthesis of Analog Circuits. Doctoral thesis.
University of Cincinnati. November.
Das, A. and Vemuri, R. (2009). A Graph Grammar Based Approach to Automated Multi-
Objective Analog Circuit Design. Design, Automation and Test in Europe
Conference and Exhibition 2009. IEEE Conferences., pp. 700-705.
De Bonet, J.; Isbell, C. and Viola, C. (1996) MIMIC: Finding Optima by Estimating
Probability Densities. Proceeding of Neural Information Processing Systems. Pp.
424-430.
De Garis, H. (1993). Evolvable Hardware: Genetic Programming of a Darwin Machine, in
Artificial Neural Nets and Genetic Algorithms, Albretch, R.F., Reeves, C.R., and
Steele, N.C., Eds., Springer-Verlag, New York.
Dorigo, M. (1991). Positive Feedback as a Search Strategy. Technical Report. No. 91-016.
Politecnico Di Milano, Italy.
Fonseca, C. M. and Fleming, P. J. Genetic Algorithms for Multiobjective Optimization:
Formulation, Discussion and Generalization. In S. Forrest, editor, Proceedings of
the Fifth International Conference on Genetic Algorithms, pages 416–423, San
Mateo, California, 1993. University of Illinois at Urbana-Champaign, Morgan
Kauffman Publishers.
Goh, C. and . Li (2001). GA Automated Design and Synthesis of Analog Circuits with
Practical Constraints. Proc. IEEE Int. Conf. Evol. Computation. pp. 170-177.
Goldberg, D. (1989) Genetic Algorithms in Search Optimization & Machine Learning.
Addison-Wesley .
Grimbleby, J. (1995) Automatic Analogue Network Synthesis Using Genetic Algorithms,
Proceedings of the First IEE/IEEE International Conference on Genetic Algorithms
in Engineering Systems (GALESIAS-95), pp.53-58, UK.
Harvey, I. (1993). The Artificial Evolution of Adaptive Behaviour, PhD Thesis, University of
Sussex, School of Cognitive and Computing Sciences, September, 1993.
Hernandez A. and Coello C. (2003). Evolutionary Synthesis of Logic Circuits Using

Information Theory. Artificial Intelligence Review 20: 445–471, Kluwer Academic
Publishers. Printed in the Netherlands.
Higuchi, T.; Iwata, M.; Kajitani, I.; Murakawa, M.; Yoshizawa, S. and Furuya, T. (1996).
Hardware Evolution at Gate and Function Levels. In Proceedings of the
International Conference on Biologically Inspired Autonomous Systems:
Computation, Cognition and Action, Durham, North Carolina.
Holland, J. (1975). Adaptation in Natural and Artificial Systems. University of Michigan
Press. Ann Arbor, MI, 1975; MIT Press, Cambridge, MA 1992.
Hu, J.; Zhong, X. and Goodman, E. (2005). Open-ended Robust Design of Analog Filters
Using Genetic Programming Proceedings of the 2005 conference on Genetic and
evolutionary computation, June 25-29, Washington, DC, USA, pp. 1619-1626.
Jerke, G. and Lienig, J. (2009) “Constraint-driven Design — The Next Step Towards Analog
Design Automation”. Proceedings of the 2009 international symposium on Physical
design. San Diego, California, USA. Pp. 75-82. 2009.
Khalifa, Y.; Khan, B. and Taha, F. (2008). Multi-objective Optimization Tool for A Free
Structure Analog Circuits Design Using Genetic Algorithms and Incorporating
Parasitics. Hindawi Publishing Corporation. Journal of Artificial Evolution and
Applications. Volume 2008, PP. 0-9.
Koza, J. (1992) Genetic Programming. On the Programming of Computers by Means of
Natural Selection. MIT Press. Cambridge, Massachussetts, 1992.
Koza, J.; Bennett, F.; Andre, D. and Keane, M.. (1996) Toward Evolution of Electronic
Animals Using Genetic Programming. Artificial Life V: Proceedings of the Fifth
International Workshop on the Synthesis and Simulation of Living Systems.
Cambridge, MA: The MIT Press.
Koza, J.; Bennethh, F.; Lohn, J.; Dunlap, F.; Keane M. and Andre D. (1997b) Automated
synthesis of computacional circuits using genetic programming” in Proc. 1997 IEEE
Conf. Evolutionary Computation. Piscataway, NJ: IEEE Press, pp. 447–452, 1997.
Koza, J.; Bennethh, F.; Andre, D. and Keane, M. (1997). Automated Synthesis of Analog
Electrical Circuits by Means of Genetic Programming. IEEE Transactions on
Evolutionary Computation, Vol 1, No.2, pp. 109-128.
Krasnicki, M.; Phelps, R.; Hellums, J.; McClung, M.; Rutenbar, R. and Carley, L. (2001). ASF:
a practical simulation based methodology for the synthesis of custom analog
circuits, In Proceedings of ICCAD 2001, pp. 350–357.
Larrañaga P. and Lozano, J. (2002) Estimation of Distribution Algorithms: A New Tool for
Evolutionary Computation”. Kluwer Academic Publishers.
Lohn, J. and Colombano, S. (1998). Automated Analog Circuit Synthesis using a Linear
Representation. Proc. of the Second Int'l Conf on Evolvable Systems: From Biology
to Hardware, Springer-Verlag, Berlin, pp. 125-133.
Louis, S. (1993). Genetic Algorithms as a Computational Tool for Design. PhD Thesis,
Department of Computer Science, Indiana University.
Louis, S. J. and Rawlins, G. J. E. (1991) Using Genetic Algorithm to Design Structures.
Technical Report 326. Computer Science Department, Indiana University,
Bloomington, Indiana.
Mendoza, B. (2001). Uso de del Sistema de la Colonia de Hormigas para Optimizar Circuitos
Lógicos Combinatorios. Tesis de Maestría en Inteligencia Artificial de la
Universidad Veracruzana. México.
Michalewicz, Z.; Dasgupta, D.; Le Riche, R. and Schoenauer M. (1995). Evolutionary
Algorithms for Constrained Engineering Problems.
Muhlenbein, H. and PaaE G. (1996). From Recombination of Genes to the Estimation of
Distributions I. Binary Parameters, in H.M.Voigt, et al., eds., Lecture Notes in
Computer Science 1411: Parallel Problem Solvingfrom Nature - PPSN IV, pp. 178-
187.
Pelikan M. and Mühlenbein, H. (1999). The Bivariate Marginal Distribution Algorithm.
Advances in Soft Computing-Engineering Design and Manufacturing. Pp. 521-535.
Ponce de León, E., (1997) Algoritmos Genéticos y su Aplicación a Problemas de
Secuenciación”. PhD. Tesis. Centro de Inteligencia Artificial. Instituto de
Cibernética, Matemática y Física.
Rudolph, G. (1996) Convergence of Evolutionary Algorithms in General Search Spaces, In
Proceedings of the Third IEEE Conference on Evolutionary Computation .
Rutenbar, R.; Gielen, G.; Roychowdhury, J. (2007). Hierarchical Modeling, Optmization, and
Synthesis for System-level Anlog and RF Designs, Proc. of the IEEE, Vol. 95, Issue
3, March 2007, pp. 640-669.
Schaffer, J. D. (1984) Multiple Objective Optimization with Vector Evaluated Genetic
Algorithms. PhD thesis, Vanderbilt University.
Shragowitz, E.; Lee,J.; Kang, Q. (1998). Application of Fuzzy Logic in Computer-Aided VLSI
Design, IEEE Trans. on Fuzzy Systems, Vol. 6, No 1, February 1998, pp. 163-172.
Thompson, A.; Harvey, I. and Husbands, P. (1996). Unconstrained evolution and hard
consequences. In E. Sanchez and M. Tomassini, editors, Toward Evolvable
Hardware: The Evolutionary Engineering Approach (Lecture Notes in Computer
Science, Vol. 1062), pages 136--165, Heidelberg, Germany, Springer-Verlag.
Torres, A. (2010). Metaheurísticas Evolutivas en el Diseño de Circuitos Analógicos. Tesis
Doctoral. Universidad Autónoma de Aguascalientes.
Torres, A.; Ponce de León, E.; Hernández, A.; Torres, M.D. and Díaz, E. (2010). A Robust
Evolvable System for the Synthesis of Analog Circuits. Computación y Sistemas,
Revista Iberoamericana de Computación. April-June, Vol. 13, No.4, pp. 295-312.
Torres, A.; Ponce de León, E.; Torres, M. D.; Díaz E. and Padilla, F. (2009). “Comparison of
Two Evolvable Systems in the Automated Analog Circuit Synthesis”. Artificial
Intelligence, MICAI 2009. Eighth Mexican International Conference on, vol.1, pp 3-
8, 8-13.
Yao, X. and Higuchi, T. (1999) “Promises and Challenges of Evolvable Hardware”, IEEE
Transactions on Systems, Man and Cybernetics_Part C. Applications and Reviews,
Vol 29, No.1.
Zebulum, R.; Pacheco M. and Vellasco M. (1996). Evolvable Systems in Hardware Design:
Taxonomy, Survey and Applications, Proceedings of The First International
Conference on Evolvable Systems: From Biology to Hardware (ICES'96), Lecture
Notes in Computer Science 1259, pp. 344-358, Tsukuba, Japan, October 7-8.
Zebulum, R.; Pacheco M. and Vellasco, M. (1998). Comparison of different evolutionary

methodologies applied to electronic filter design. In Proc. Of IEEE. Inttl. Conf. On
Evolutionary Computation. May.
Zebulum, R.; Vellasco, M. and Pacheco, M. (2000) Variable length representation in
evolutionary electronics. Evol. Comput., vol. 8, no. 1, pp. 93–120.

Advances in Analog Circuits PDF

Uploaded by

Copyright:

Available Formats

Advances in Analog Circuits PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advances in Analog Circuits PDF

Uploaded by

Copyright:

Available Formats

ADVANCES IN

Copyright © 2011 InTech

Publishing Process Manager Katarina Lovrecic

First published February, 2011

A free online edition of this book is available at www.intechopen.com

Advances in Analog Circuits, Edited by Esteban Tlelo-Cuautle

Part 1 Circuit Design 1

Chapter 1 Analog CMOS Design Automation

Chapter 2 A New Approach to Biasing Design of Analog Circuits 15

Chapter 3 New Port Modeling

Chapter 4 Behavioral Modeling of Mixed-Mode Integrated Circuits 85

Part 2 Design Issues 109

Chapter 5 Parallel Preconditioned Hierarchical Harmonic

Chapter 6 Lifetime Yield Optimization of Analog Circuits Considering

Chapter 7 Linear Analog Circuits Problems

Chapter 8 Analog Design Issues

Chapter 9 Tunable Analog and Reconfigurable

Chapter 10 Statistical Analog Circuit Simulation:

Chapter 11 Advanced Statistical Methodologies

Part 3 Applications 245

Chapter 12 Analog-aware Schematic Synthesis 247

Chapter 13 An SQP and Branch-and-Bound Based Approach

Chapter 14 Analog Circuit for Motion Detection Applied

Chapter 15 Analog Circuits Implementing a Critical Temperature

Chapter 16 Evolvable Metaheuristics on Circuit Design 347

Electronic design automation of analog circuits is presented in Chapter 1 for low-pow-

Enjoy the book!

Analog CMOS Design Automation

2. Operational ampliﬁer sizing optimization

Pdiss = (VDD − VSS ) · IDD (7)

Fig. 1. Schematics of a two-stage Miller OTA.

Fig. 2. Simulation-based design ﬂow using genetic algorithms.

Fig. 3. gm/ID x ID /(W/L) curves for 0.35μm CMOS technology.

Fig. 4. Design ﬂow for the gm/ID design methodology.

Fig. 5. Cost function evolution.

5.1 Methodology 1: Simulation-based

(a) Gain-bandwidth product (b) Phase margin

(c) Low-frequency voltage gain (d) Slew-rate

Fig. 6. Evolution of 4 design speciﬁcations for Design 3 with Simulation-Based methodology.

5.2 Methodology 2: gm/ID

5.3 Comparison results

Av0 GBW PM [°] SR ICMR [V] Pdiss A gate

1.1 Circuit biasing

2. Nullators, fixators and norators

2.1 Port parameter fixing

1 A component here can represent any size two-terminal network.

Fig. 5. A diode circuit with an unspecified supply voltage V1

R2 = VR 2 / I R 2 = 1.8 / 0.5 = 3.6 KΩ

R3 = VR 3 / I R 3 = 1.08 / 1.0 = 1.08 K Ω

2.2 Rules governing fixators and norators in a circuit

3. Circuit solutions containing fixator-norator pairs

3.2 Sensitivity in fixator-norator pairs

3.3 Paring fixators and norators in a circuit

VCC V VBB V VBE V RB KΩ RBE KΩ RO KΩ β

100 KΩ 10 KΩ 1 KΩ 100 KΩ 10 KΩ 1 KΩ 100 KΩ 10 KΩ 1 KΩ

(a) (b) (c)

4. Component modeling with fixator

4.1 Complete modeling of devices

4.2 Partial modeling of devices