NATIONAL RADIO ASTRONOMY OBSERVATORY

Charlottesville, Virginia 22903

Electronics Division Internal Report No. 295

(also distributed as mmA memo No. 103)

Some Fundamental and Practical Limits on

Broadband Matching to Capacitive Devices
and the Implications for SIS Mixer Design
A. R. Kerr

September 30, 1993

Number of Cop es: 150

CONTENTS

I.

Introduction . .

II.

The Capacitive Device

III.

The Capacitive Device with Series Inductance

rv.

The Inductively Tuned Device with Series Inductance . .

V.

Results and Discussion'.

VI.

.
Application to Mixers
.
Negative Output Resistance in SIS Mixers

VII.

Practical Constraints on the Series Inductance .

VIII. An Example .

..

Conclusion .

X.

Acknowledgements .

XI.

References .

..

. ..

..

.. 7
. . . . 8

...

...

...
.

. 11
. 13

..

. 14

. 17

.
..

..

..

..

EX.

..

..

19

..

. ' . 20

...

. . 20

Appendix: Details of the "Typical" SIS Mixer Used in the Simulations

Some Fundamental and Practical Limits on

Broadband Matching to Capacitive Devices
and the Implications for SIS Mixer Design
A. R. Kerr

ABSTRACT

In a given frequency band, the achiev able match between a capacitive

microwave or millimeter-wave device and a resistive source is limited by the
capacitance of the device and its series inductance. The fundamental limit on
the match bandwidth is examined for three circuits: (i) parallel RC,
(ii) parallel RC with series L, and (iii) parallel RCL with series L. The
broadband matching theories of Bode (1945) and Fano (1950) are used, the
latter modified to avoid the standard low-pass to band-pass mapping in case

(ii) because the terminals of the capacitance are not generally accessible for
connection of the r e quisite parallel inductor. The results are fundamental to
the design of broadband mixers, multipliers, switches, and detectors using
Schottky diodes or SIS junctions.

Practical limitations imposed by the minimum realizable dimensions of

millimeter-wave integrated circuits fabricated by standard photolithography
are discussed in the context of SIS mixers with series arrays of junctions,
and an example of a coplanar SIS mixer design is given. For a series array of
N devices with a given total resistance, it is shown that there is an upper
limit to N, below which the theoretical match bandwidth depends only on the RC
product of the devices and not on the series inductance of the array.
I.

Introduction

Many semiconductor and superconductor devices are inherently capacitive

and can be well described by the parallel RC circuit shown in Fig. 1(a). This
is true of Schottky diodes and SIS superconducting quasi-particle tunnel
junctions at frequencies up to several hundred GHz. In many practical cases
the geometry of the device adds an unavoidable series inductance, as indicated
in Fig. 1(b). In either case, the bandwidth over which it is possible to
couple power efficiently to the device from a resistive source can be
increased, often considerably, by inserting a lossless matching network
between the source and the device. In some cases, it is possible to connect
an inductive tuning element directly across the terminals of the intrinsic
device, as shown in Fig. 1(c). The purpose of the present work is to
determine the fundamental match-bandwidth limits of these circuits.
Specifically, in sections II-V, we determine the lowest upper bound on the
magnitude of the reflection coefficient that can be achieved over a given
frequency band. This is illustrated in Fig. 2, where it is required that
I A :5_ P. within the frequency band co l :5_ w (4, 2 , but can have any value outside
that frequency range. The resulting minimum possible value of p a is of
fundamental importance in the design of broadband circuits incorporating Sis
unctions or semiconductor diodes.

(b)

(a)

Fig. 1 . Equivalent circuits of: (a) a capacitive device, (b) a capacitive device with series inductance, and
(c) an inductively tuned device with series inductance.

Ipi

Fig. 2. 1 p J is required to lie below the solid curve. The minimum possible value of p a , in the given
frequency band co and CO , is to be determined.
i

This work uses the well-known broadband matching theory, published in

1945 and 1950, of Bode [1] and Fano [2]. In their work, the conditions of
physical realizabiiity were used to derive limits on the voltage reflection
coefficient of a given load with an arbitrary lossless matching network. In
section II of this paper, Bode's match-bandwidth limit is applied to the
circuit of Fig 1(a). In sections III and IV, the theory of Fano is used to
derive limits for the circuits of Fig. 1(b) and (c). For the case of
Fig. 1(b), this requires a slight modification to the theory to cover the case
of band-pass matching without the use of the usual low-pass to band-pass
mapping. Section V presents the theoretical bandwidth limits for the three
circuits of Fig. 1 in graphical form. It is apparent that the series
inductance of a device limits the bandwidth of a circuit only when it exceeds
a certain value which depends an R and C.
Section VI discuses the implications of the match-bandwidth limits for
practical double -sideband mixers, using broadband SIS mixers as examples.
Sections VII and VIII examine the practical constraints on the series
inductance of devices and series arrays of devices in coplanar waveguide

circuits, and give an example of a broadband coplanar SIS mixer with an array
of junctions. For circuits using series arrays of devices, it is shown that
there is no trade-off between the attainable bandwidth and the number of
devices in the array so long as the number of devices does not exceed a
critical value.
II.

The Capacitive Device

In the case of the simple circuit of Fig. 1(a), connected via a lossless
matching
network to a resistive source, Bode [1] showed that the reflection
coeff ient
p is constrained by the integral equation:
ic
(1)
Inspection of (1) indicates that the lowest value of the upper bound of hol
(02 is achieved when Ipl - pa
(p. in Fig. 2) within the frequency band w i
within that band , and IPI
1 at all other frequencies. The optimum Ip( )1
therefore coincides with the solid curve in Fig. 2. The integral in (1) is
then simply evaluated, giving
r/RC(w2 wi)
It is apparent that p a,min depends on the desired bandwidth
on the location of the band along the frequency axis.
in(l/Pa,min)

III.

(12)

The Capacitive Device with Series Inductance

Fano [2] extended the work of Bode to include a more general load
equivalent circuit for the case of low-pass matching (i.e., for w i .... 0 in
Fig. 2). This can be applied to the case of band-pass matching by using the
standard low-pass to band-pass mapping, ca (ax(c0/(0.-(.0x/(0), which transforms
all inductors and capacitors in the circuit into series or parallel LC
resonators. However, this transformation is not applicable to the device
described by Fig. 1(b), as the terminals of the capacitor are not accessible
for connection of a parallel inductor. The procedure of Fano is applied
below, modified slightly to avoid using a low-pass to band-pass
transformation, to determine the lowest possible value of the upper bound of
the magnitude of the reflection coefficient within a given frequency band, for
the circuit of Fig. 1(b).
The given device and its external matching circuit are depicted in
Fig. 3. Following Fano, the best possible match over a specified frequency
band is deduced from consideration of the function F(s)
111(1/p 1 (s)), where
S a + jco is the complex frequency variable. Expanding F(s) as a Taylor
series in lis:
F(s)

1n(1/ 12 1 )

jAl + A 1 (1/s) + A3 (1/ ) 3 + .

GWEN DEVICE
N'
rwoliO

REACTIVE

REACTIVE MATCHING

ELEMENTS

NETWORK

Source

J
1

Fig. 3. The load R and network N' are part of the given device. Network N" contains reactive matching
elements.

The coefficient Al is equal to 0 or r, while the other A, must all be real,

since Ip i j is an even function of w and arg(p i ) is an odd function of w. Fano
showed (2, pp. 62-63] that since the in transmission zeros of N' must also be
transmission zeros of the whole circuit (N' + N"), it follows that except in
the degenerate case discussed below, p i and its first (2m-1) derivatives,
evaluated at the transmission zeros of N', are independent of the matching
network N". The same is true of ln(l/p i ) and its derivatives. The first 2m
Taylor coefficients, A, (1/n1)eF(s)/d(1/s)n, can therefore be evaluated
from a knowledge only of the equivalent circuit of the given device.
If the left-hand element of N" is of the same type CL or C) and
orientation (series or parallel) as the right-hand element of N', the circuit
is said to be degenerate. N' and N" then have a common transmission zero and
only (2m-2) derivatives of p i are independent of N" (2, pp. 71-731.
Now consider the circuit of Fig. 1(b), terminated on the right by an
arbitrary resistance, as shown in Fig. 4. For convenience, the values of the
circuit elements are impedance-scaled to R 1 ohm (the impedance level has no
effect on reflection coefficient or matching bandwidth).c The network N' has
two zeros of transmission, due to C and L, both at s o, so m 2. Then,
Z'
and

Z'1-4Z' - 1

In terms of the variable

sL + r
+ srC

4)

s2LC + s(L-i-rC)
szLC s(L-rC)

(l+r)
(1-r)

(5)

us,

inf
111( 1
Pu

e (l+r)

Li.TC) A- LC/

ea-r) M-rC) LC

(6)

Fig. 4. The network N ', for the circuit of Fig. 1(b), arbitrarily terminated on the right. Values of the
circuit elements are impedance-scaled to R = 1 ohm.

The first eva

2m Taylor coefficients Au are obtained from the derivatives of
in(l/p / )
luated at e 0, and are independent of the arbitrary resistance r
as expected:
(7)

When the network N' in Fig. 3 has two transmission zeros at s

showed that, in the non-degenerate case, physical realizability of the
matching network N" requires the following integral equations to apply:
r*1n11--:Idw
JO
p

(A1 - 2 Z An)

ro)

co

2i

1 -1,
P1

- -2. (A - 32 E A3)
r
3

where An
a ri
jwri are the zeros of p i that lie in the right half of the splane, and depend on both networks N' and N" in Fig. 3. In the degenerate
case, eq. (8) alone is required.
The upper bound on lp i l is minimized if the matching network N" is
so that within the pass band, w i w w2 , Ip i l is equal to the upper
bound p a , while outside the pass band hod 1, as depicted by the solid curve
in Fig. 2. With this form of p(c), the integrals in (8) and (9) can be
evaluated, giving
chosen

0,02

and

(u12 3

co l ) in 1/p a l

c) l )1n11/p. I

(r/2)

( A1 -

2 Z

-(r/2)(3A1 - 2 E A ri 3 ) .

( 1 0)

(11)

These band-pass equations are similar to Fano's law-pass equations (2,

eqs. (25) and (26)]. In the degenerate case, eq. (10) alone is required.
The remaining step is to find the right half-plane zeros A ri of p i in
(10) and (11) that maximize lnilip a l (i.e., minimize p.) for a given pass
band, w i to wt . Following the argument used by Fano for the non-degenerate
law-pass case (w i 0), it is evident that to maximize lnilip a l the matching
network must be selected so that E A ri 3 is as large as possible while keeping
E A ri as small as possible. This is accomplished by using a single real zero,
a, [2, p. 72]. Then, from (10) and (11),
042

and

(w2

-(r/2)(3A1

3)1n11/Pal

(12)

ar )

(ir/2)(A1

w 1 )11111/Pal

Eliminating a between these equations and putting K (2//01n(1/p

gives
K(w2 3 - w i 3 )

(13)

2c3)
a4a.in )

3A3 - (1/ 4 ) [A l - K(w

(14)

which is a cubic equation in K.

It is convenient at this point to change the frequency variables from wl
and w t to the center frequency wo and fractional bandwidth b. Substituting
w i wo(l b/2) and (4 2 (4 0 (1 b/2) in (14) gives
3 3 3

W0 b K

2 2AiK2
3Wo-b-

(,) b

[c,0 2b 2

1.2W02

Substituting the values for A i and

a0 C, gives
Bc
6
bBC

A3

3Al21

K + [12A 3 -A1 3 1

(15)

from (7), and putting XL 0L and

2B + 12 + b

(16)

which can be solved for K.

In the degenerate case, the first element of the optimum matching
network is inductive, thereby augmenting L. Equation (10) alone is then
required, for which lnil/p a l is maximized when p i has no right-half-plane
zeros; i.e., E A
0. Using A i 2/C from (7), (10) becomes identical to
the Bode limit of eq. (2). In terms of K, b, and Bc, the Bode limit is given
by:
(16a)
bB
C

The Inductively Tuned Device with Series Inductance

For an inductively tuned device with series inductance, as in Fig. 1(c),
the standard
low-pass to band-pass mapping allows the fundamental limit on the
match bandw idth to be deduced from the low-pass case. The low-pass prototype
is simply the circuit of Fig. 1(b). Equations (12) and (13) are used, with
0, whence
(17)
3
(42 K

(18)

-3A3 ar

Eliminating o between these equations, and substituting for A i and A3 from

(7), gives

which can be solved for K(

We now use the low-pass to band-pass mapping
w
co [
x

(20)

where w. 1/./(LI C) is the resonant frequency of all the LC resonators in the

transformed (band-pass) circuit. Low-pass and band-pass circuits related by
(20) have equal bandwidths. Since the low-pass circuit under consideration
has a bandwidth equal to w 2 (i.e., w i 0 in Fig. 2), the fractional bandwidth
of the band-pass circuit is ,8 wilwx . Substituting w 2 Max in (19) gives
3
K

6 v2

12

fl2

2c

24

41K -

3 3 21,
w

(21)

The fractional bandwidth 0 used here differs in definition from the fractional
bandwidth b used in Section III. If the band edge frequencies are c a and us
then 0 = ( wb-w.)/wx, while b (wb-w.)/(40, where the mean frequencies
wx j(wwb ) and wo (w.+wb )/2. To facilitate comparison between results for
the different equivalent circuits in the next section, it is necessary
to
2
/
change the variables
A and wx , in (21) to b and wo using A - b/.4 (1 b /4) and
wx woj(1 - b 2/4). Writing Bc woC, and X L woL gives
3

K - bB K + [_12
+ 4jK b 2B2
C
C

24

B c2:XL

In the degenerate case, when the first element of the optimum matching
network is inductive, the Bode limit applies and is given by the low-pass to

band-pass mapping of eq. (2) according to eq. (20). The result is identical
to eq. (16a), viz.
2
bi3C

(22a)

Results and Discussion

It is not usually known in advance whether a particular circuit of the

form of Fig. 1(b) or (c), connected to its optimum matching network, will be
degenerate or not If degenerate, the series inductance L does not exceed
some limiting value, Ls, below which the match bandwidth is limited only by C.
The Bode limit then is attainable, and eq. (16a) or (22a) is applicable. If
on the other hand, the optimally matched circuit is non-degenerate, then
L > Ls, and the match bandwidth is limited by C and L; eq. (16) or (22) is
then appropriate. In the present work, however, this uncertainty does not
present an obstacle; the equations for both cases are solved, the correct
solution being the more restrictive one as is clear from the graphical results
below.
The parameters X L - woL and Bc co 0 C in eqs. (16), (16a), (22), and (22a)
apply to the circuits of Fig. 1(b) with impedances scaled to R - 1. For the
unscaled circuit, they become:
woL

xL '11

and

co RC

(23)

Equations (16), (I6a), (22), and (22a) have been solved for a range of
values of XL and Bc. As the present work was done in the context of SIS mixer
design, values of XL and Bc appropriate to SIS mixers are used in the examples
below. It should be noted that Bc is not the familiar woR N C J product of the
SIS junction(s), but is woRuC j , where Rsy is the small signal RE' input
impedance of the mixer. Likewise, XL - cooLIRRy. (For a well designed SIS
mixer, Ptilp usually lies within a factor of -3 of P14; i.e., 4/3  RRT 3RN
(3, 4].)
Corresponding to a range of different devices, we have chosen
woRC
2, 4, 8, and values of XL - woL/R in the range 0 to 10. The values
C
of the plowest upper bounds, lo a,min 1 on the reflection coefficient, and
1/(1 I a4.1.1 2 ) on the reflection loss, versus fractional bandwidth b, are
shown in Figs. 5 and .6 for the circuits of Figs. 1(b) and 1(c), respectively.
The Bode limit for le 0 (i.e., the circuit of Fig. 1(a)) is indicated by the
dashed curves. The dotted curves, shown for comparison, are for a parallel RC
device tuned by a parallel inductor and connected to a source resistance R,
L 0).
but with no additional matching (i.e., the circuit of Fig. 1(c) with
2 at
and 1 1
of
The short horizontal lines indicate the values
/( -iP1 ]
IPI
frequency coo for a parallel RC device connected to a source resistance R with
no matching elements (i.e., the circuit of Fig. 1(a)).
B

0.9
0.8

i:

' .

:
::

.P.

7
: ,:4I'
Y /

:
..4..

0.5

ro 0.5

../:
......

0.4

0.2

...................-..........,

ia

0.3

1.4

.. . .

1 1-

....................,....4.4,..

A
I
4.../... ........ ... .. ................

...........

.........

0.1

......

0.0

,
...........

....

' :

0.7
-

wRC

".RC 8

1.0

-....

.0 1

a. I

I(

Fractional Bandwidth b

Fig. 5. For the capacitive device with series inductance (Fig. 1(b)), the lowest upper bounds, I pa,
2
on the reflection coefficient, and 1/(1J ) on the reflection loss, vs fractional bandwidth
b, with ca oRC = 2, 4, and 8, and various values of woL/R. The Bode limit for the case L = 0 is
indicated by the dashed curves. The dotted curves are for a parallel RC device tuned by a
parallel inductor but with no additional matching. The short horizontal lines indicate the values o
p I and 1/(1- 1 p1 2 ] at frequency coo for a parallel RC device connected to a source resistance
R With no matching.

1.0

0.9
0.8 ---

444444444.4444 444.
.

0.4 ----

4,444.:4141.111.1444 44444., 444.4

0.3 ---

0.2

.44...

.. . 1"'.::`
.. .

0.1 - .. . ...

.. .

. ..

0.0
0.01

0.1
Fractional Bandwidth b
viRC 4

*RC 4

15.0

1.0
.

4.44
...... .. . .. . ..

0.8 -

....
c:3
,0 1 2.0 -

0.7 0.6 -...

_c 0.5 -...
0.4 .....
. .. 4

0.3 -...

*1

3.0-

. .

0.2

.........

0.0 ..
0.01

I T.*

1
0.1
Fractional Bandwidth b

0.0
0.01

10

15.0

0.9 ,..........

1 2.0

0.8

0.7
0.6

::
: : . .::
. :
. : :::
.....f....4.4.4.44444.1.4i .1
: i : i : : .:
. . :. : . : . : : :

0.4

0.3

,...........

0.2 ,
0.1

4 1 4 i 44.0

. .
:
" : : .

I. . .1. ...........4.....4..........
:, : .: : :. :. i
:
: ::
4 4i.44H........ 44+.4.444
..
:
.

, . i ...........i........
...... :,. : .:i. . :.v::::

. i.......

--;---:----,--..--:
: ."- i :
.....
.... . . . :..,
: :.
. .. '.... t *
: : :,:.:

0. 0
0.0 1

Fig.

to

wRC = 8

wRC = 8
1.0

0.1
Fractiond Bandwidth

..

Ural

:. : :

.......... ..: . Z..: . :.:. ..... ......:. ...

: :

.

. .

.
0 .0
0.0 1

10

0.1
Fractional Bandwidth

0.1
Fractional Bandwidth b

6. For the inductively tuned capacitive device with series inductance (Fig. I (c)), the lowest upper

bounds, I P 8,,tain 1 on the reflection coeffic ent, a d 1/(1- I a,min I ) on the reflection loss, vs
fractional bandwidth b, with woRC = 2, 4, and 8, and various values of 0) 0 1./R. The Bode limit for
the case L = 0 is indicated by the dashed curies. The dotted curves are for a parallel RC
device tuned by a parallel inductor but with no additional matching. The short horizontal lines
indicate the values of P
/ (1- I P 1 2 at frequency too for a parallel RC device connected
to a source resistance R with no matching.
i

I and 1

10

The Bode limit, indicated in Figs. 5 and 6 by the dashed curves, applies
when L is less than the quantity Ls, whose value depends on the other elements
of the equivalent circuit and the desired bandwidth. When L ..1; Ls, the match
bandwidth is not limited by L, but
only by C and R. In this case, the first
network
is a series inductance (Ls -L)
element of the optimum matching
effectively augmenting L. This is the degenerate case discussed above. When
L exceeds Ls (the non-degenerate case), it too limits the attainable match
bandwidth of the circuits, as indicated in the figures.
Application to Mixers
caution

Some
is required in interpreting the results of Figs. 5 and 6 in
the context of matching to double sideband mixers. While the Ipl curves in
these figures indicate the range of variation of the source impedance
seen by
the intrinsic mixing device (R in Flg. I) the loss (1/(1 -Ip1 2 ]) curves should
not be regarded simply as part of the conversion loss of the mixer which can
be eliminated by tuning. This is because the signal frequency input impedance
of a mixer depends in part on the image frequency embedding impedance. If the
signal and image circuits of the mixer are physically the same, tuning the
circuit changes the image termination and therefore the signal frequency input
impedance of the mixer. In effect, the device resistance R at the signal
frequency becomes a function of the matching circuit, and the minimum
conversion loss does not, in general, occur for a matched input.
The dependence of the performance of a typical 230 GHz double sideband
SIS mixer receiver on embedding admittance is indicated in Fig. 7, which shows
contour plots of the mixer and receiver parameters plotted on Smith charts of
RF embedding admittance (i.e., in the (-p)-plane) normalized to
the optimum
source c onductance l iRs,opt ( Rs,opt is defined in the A pPendix) . The receiver
includes an IF amplifier with Tyr .... 4 K, and an IF isolator at 4 K. The IF
load impedance Z ip is fixed and equal to Rs ,opt . Note that the source
(embedding) admittance is defined to include the capacitance of the
junction(s), and is assumed equal at the upper and lower sideband frequencies.
The mixer gain and receiver noise temperature are shown as single-sideband
(SSB) quantities. Further details of the mixer and the method of analysis are
given in the Appendix.
Fig. 7 clearly demonstrates three quantum characteristics of SIS mixers
that are not possible in classical diode mixers: (i) the (SSB) conversion
loss is less than the 3 dB classical limit for a DSB mixer, (ii) the contours
are asymmetrical about the real axis, an effect caused by the quantum
susceptance of the SIS junction, and (iii) the output resistance can be
negative, which is impossible in classical resistive mixers. The significance
of the negative output resistance is discussed further below.

Rou /Zif

Fig. 7. Contour plots of mixer gain, receiver noise temperature, Input return loss, IF VSWR, and signal-to-image conversion gain, for a 230 Gliz
815 receiver. The contours are plotted on Smith charts of RF source admittance (i.e., in the -p plane). The dotted circles are
p I . 2 and I p 0 4. The receiver includes an IF amplifier with T it, "in 4 K i and an IF Isolator at 4 K. The mixer gain and
receiver noise temperature are shown as SSB quantities. (See text for further details.)

The table below summarizes the range of variation of the receiver

parameters in Fig. 7 when the embedding admittance falls within the circles
0.2 and 0.4 (shown dotted in the figure) over the frequency band of
interest. Analysis at other frequencies
indicates that the 230 GHz results in
S S
the table above are typical
of
receivers
of similar design using Nb
I
e
requenc
junctions at f
i s from 70 to at least 350 GHz. It is clear from the
table that even an SIS mixer whose embedding admittance is designed for jpi
0.4 should have acceptable performance across the intended frequency band for
most practical applications.
IPI -.5. 0.4

conversion gain (SSB

Receiver noise temp (ssB)
Input return loss
Signal-to-image cony . loss
Output VSWR (R t/ZIT)

-1.5 dB
15 K
> 8 dB
9 -4. 11 dB
-17

-0.5
12

+0.5 -4.0 dB
12 -+ 20 K
> 5 dB
9 -.0 13 dB
-5 +3

Negative Output Resistance in SIS Mixers

On the contour plot of R out/Z EE. in Fig. 7, the transition of the output
impedance from a large positive value to a large negative value is quite
sudden_ However, if the contours were re-plotted as output conductance, the
transition from positive, through zero, to negative would be quite smooth and
continuous, and there is no reason to expect any sudden change in receiver
performance as this occurs. Indeed, when tuning a receiver, this transition
can often be observed on the pumped I-V curve as the differential conductance
at the bias point goes from positive, through zero, to negative. However,
under certain conditions, the presence of negative output resistance can have
two possibly adverse effects: (i) it can cause (small-signal) instability and
oscillation in the RF, IF, or bias circuits, and (ii) in a series array of
junctions, negative DC differential resistance can cause instability in the
biasing of the junctions and the division of the LO voltage between them.
These will be discussed further below. A third possible concern is that noise
from the IF amplifier or isolator, incident on the IF port of the mixer, will
be reflected with gain back into the amplifier from the negative output
resistance of the mixer. This, however, is taken into account in the analysis
used to generate Fig. 7.
(i) A device with negative (differential) resistance is potentially
unstable. Depending on the device, it may be stable with a high impedance
load or with a low impedance, as can be demonstrated by considering series and
parallel LRC circuits with negative resistance. (Analysis of the time-domain
differential equations describing these circuits shows that the former is
stabilized by a high resistance load, and the latter by a low resistance
load.) In general, to determine the stability of a negative resistance device
connected to a load, the complex impedance of the device and load must be
known at all frequencies. It is possible, however, to make some general
observations about the nature of the stability of mixers with negative
resistance. With no LO power applied, an SIS junction (or array) is stable,
regardless of the embedding impedance Z,(40, and the reflection coefficient of

13

the junction
(or array) with respect tow the embedding impedance,
IP10,0 I - IZ J -Z:I/1Z 4 A-Z e 1 S 1 [ 1 ], here Z(w) is the (small signal)

impedance of the junction. As the LO power is increased, it is possible that

O.
at some frequency x
i will become infinite; i.e., Z i Ou.d +
4A x, Ip

Under this condition the circuit is on the point of oscillation, and an

impulsive excitation of the junction will cause undamped ringing at frequency
(4,2 . Because of the mixing action of the junction, this incipient instability
occurs. simultaneously
at co x, and at all the related sideband frequencies
0
4
- S n 0, at each of which Ip j l co. Clearly, as the LO power is
=ow (4
increased from zero, instability is preceded by high reflection gain and
conversion gain at a complete set of related sideband frequencies, but not
necessarily within the normal signal, image, and IF bands. In practice, one

can only ensure that, as the LO power is increased from zero to the operating
level (usually a 1), the gain does not become large within the normal IF
band, for which the signal, image, and IF embedding impedances are known. In
the region of negative IF output resistance In Fig. 7, for example, the
1.9 almost to the edge of the Smith chart, and no
magnitude of Iltout/Z IF I
instability would be expected at frequencies where the IF load impedance Zu
remains near its nominal value. Instability could occur far above the IF band
if the bias -T, circulator, or amplifier exhibited a resonance which would
allow Z 1 (c41.) Z 9 (cJ.) - 0 to be satisfied. For the broadband SIS mixers
reported to date, this has not, apparently, been a problem.
(ii) The second and least predictable effect of negative resistance has
been observed in fixed-tuned SIS mixers using series arrays of SIS junctions
[4]. It is suspected that negative DC differential resistance in a series
array can cause an unstable situation in which the individual junctions become
unequally biased and unequally driven by the LO. Ultimately, a stable dynamic
state is reached in which the junctions remain unequally biased and driven.
As this instability appears to be avoided in mechanically tunable mixers using
arrays of SIS junctions with integrated tuning circuits, similar to those used
in [4], we surmise that it can be avoided by appropriate design of the
embedding impedance as a function of frequency.
VII Practical Constraints on the Series Inductance
For integrated circuits, such as mixers, multipliers, switches, and
modulators, in the 400-400 GHz range, several basic circuit configurations
are possible. These Include waveguide mounts, micros trip with conductors on
one or both sides of the substrate, suspended stripline, and coplanar
waveguide circuits. Of these, coplanar waveguide has two major advantages:
( i ) it allows the use of a relatively thick substrate (e .g., 0.0035" for a
200-300 GHz mixer (4]), and (ii) with the center conductor and ground-plane on
the same side of the Tubstrate, transmission line dimensions can be kept small
to minimize the parasitic inductance and capacitance associated with a series
array of devices, which can ultimately limit the bandwidth of the circuit. A
broadband transducer from coplanar waveguide to rectangular waveguide can be
made using an intermediate section of suspended stripline, as described in
(51.
In practical applications, a series array of devices may be preferable
to a single device for two reasons. Firstly, for a given overall impedance
14

level, an array has greater

power handling capacity and dynamic range than a
single device. This c an b e an important benefit in frequency multipliers and
mixers. Secondly, the devices in a series array are
larger than a single
device of the same impedance as the array. This can substantially reduce the
difficulty of fabricationw and
result in better quality devices. The
(theoretical) match band idth of an array of devices is only lower than that
of the equivalent single device if the array i s so long that the series
inductance of the array exceeds the value Ls mentioned above; otherwise, the
Bode limit predicts equal maximum bandwidths for the single device and the
array. The SIS mixer examples below assume the simple coplanar array of
j unct i on s shown in Fi g . 8, w ith physical dimensions chosen to give the lowest
series inductance consistent with reproducible fabrication using standard
photolithography.
For the circuit of Fig 8, the series inductance depends on the number
NJ of junctions. Fig. 9 shows the normalized series reactance wL/R of the
circuit as a function of the number of junctions in the array, with frequency
as parameter. The resistance R N JR J is the small-signal RF resistance of
the array. A quartz substrate, with e r - 3.8, is assumed. At higher
frequencies, and for larger numbers of junctions, the electrical length of the
coplanar
line becomes significant, as indicated by the curvature of the upper
cu rv es in Fig. 9, and the circuit can no longer be characterized accurately as
an array of RC devices in series with a frequency-independent inductance. The
values of wL/R in Fig. 9 can be used with Fig. 5 to determine the upper
(Bode/Fano) limit of the RF match bandwidth of this circuit. The horizontal
dashed lines in Fig. 9 indicate the values of coL/R above which the inductance
of the array limits the achievable bandwidth, for wRC
wRJ C J - 2, 4, and 8.
For a given frequency of operation, the intersection of the appropriate solid
curve with the dashed curve, corresponding to the given value of ARC,
indicates the cross-over from the Bode limit (due only to R and C) to the Fano
limit. These intersections also indicate the maximum nimber of junctions that
can be used for a given (JRC without limiting the RF bandwidth. If a smaller
d, the first element of the optimum matching network
number of junctions is
is an additional inductance, and the Bode limit can, in principle, be
attained.
For the circuit of Fig. 8, with only one or two junctions, it is
possible to use inductive tuners on the individual junctions [6] to tune out
their capacitance, as indicated in Fig. 10. In this case, the bandwidth
0.2 or 0.4, the series
limits of Fig. 6 apply. It is clear that, for Ipl
inductance does not limit the achievable bandwidth, and the Bode limit applies
in most practical cases. Mixers described in [7-12] have successfully used
one or two individually tuned junctions, though not in a coplanar transmission
line circuit. The use of individual inductive tuners with more than two
junctions is not possible for this simple coplanar circuit configuration
without orienting the additional tuners perpendicular to the center conductor.
This would introduce unacceptable capacitance between the tuners and the
ground plane. To avoid this, the ground plane can be removed in the vicinity
of the tuners. Such an approach has been successfully used in SIS mixers for
75-110 GHz (5] and 200-300 GHz (4], but the equivalent circuit is much more
complicated [13], and this configuration will not be discussed further here.
15

Fig. 8. Series array of SIS junctions in coplanar waveguide, and its equivalent circuit. The dimensions
(shown in Am) are chosen to give the lowest series inductance consistent with reproducible
fabrication using standard photolithography. The physical length of such an array of N
junctions is (8N3+2)
2.0
1.8

600
...........................

1.6

...........,............. ......,.. rn..........6..........

.................1.0

...........f.

......,.....

1.4

..........r......"..."...

.............................4,...............1.,
:
:

0.6

4.........,..

6 ..

....__..........
0.4 ...........................
0.0

400
Gt+z

0.8

0.2

.i.....'''.."'..".'"...".........."." .....'...

1.2
1 0

..0..

1..

... 2
wRe
...........................
..............

...

200

1 00

.......... ....
.................
.............4,1111.0....0.1.1100:11.41011.010

4
5
3
No, of Junctions

Fig. 9. Normalized series reactance (41./R of arrays of SIS junctions in a coplanar waveguide with the
3.8. The AF resistance
dimensions shown in Fig. 8. A quartz substrate is assumed, with e r
of the array, R, is assumed to be 50 C1 The curvature of the lines for larger arrays at higher
frequencies is a result of the significant electrical length of the array. The horizontal dashed lines
indicate the values of coL/R above which the inductance of the array limits the achievable match
bandwidth, for ,c4RC - 2, 4, and 8.
16

MMW

AAPWWPWPWPWPWPWNMAMO 1lIlppr

AV AlediAPIIINIVIIMMIPAIVAMPAIWAIIP,MMIPAII

Adt

Fig. 10. A two- unction SIS mixer in cop'anar waveguide, with inductive tuners on each junction.

The choice of the number of junctions in series, N co use in an SIS

mixer depends on two main factors: the desired dynamic range (or saturation
power level), and the smallest size of junction that can be made with
sufficient quality and reproducibility. For a given embedding impedance and
critical current density Jc, the required junction area is proportional to NJ,
so several minimum size junctions in series may be required to reach the
desired embedding impedance level. The saturation power of an SIS mixer is
proportional to wiki 3 2 [14], which may therefore govern the minimum value of N.
As the series inductance of the SIS array increases with N J , it follows that
the designer may in some cases be faced with a trade-off between bandwidth and
dynamic range: using more junctions in series to increase the saturation
power will reduce the available RF bandwidth of the mixer if the series
inductance of the array exceeds the value L.
VIII. An Example

Although the question of how closely the Bode and Fano bandwidth limits
can be approached using practical circuits is outside the scope of the present
work, it is informative to examine a realistic example. We choose a 250 GHz
mixer with a coplanar array of SIS junctions, and a matching circuit of
moderate complexity. A four junction coplanar array, as shown in Fig.
connected to a matching network containing a series capacitor and four
transmission lines in series, similar, except for the capacitor, to the tuning
circuit described in (7]. The complete circuit is shown in Fig. 11 Note
that the electrical length of the array of junctions is small, and the
equivalent circuit of Fig. 1(b) is applicable. The capacitor C A was initially
adjusted to make the impedance of the SIS array real at the center frequency,
lines 1, 2, and 3 were set to a quarter wavelength, and the fourth line set to
half a wavelength. The microwave circuit design program MMICAD [15] was then
0.4 over
used to optimize the elements of the matching network to give ipl
the widest possible bandwidth. The resulting embedding admittance is shown in
Fig. 12 on an admittance Smith chart ((-p)-plane). The optimized values of
2.48 0,
the elements in the matching network are: CA 75 fF,
Z02 = 2.34 0, Z03 16.5 0, ; 4 sis 61.2 0, 1 1 0.265 A s o, 1 2 0.246
1 3 0.254 A so, and 1 4 0.516 A.
17

COPLANAR SlS ARRAY

10111.

.6W/

MOMS

.0111M

.111.11, ..1111.

.0110

MIN

MATCHING NETWORK
...Mb ...Mb +NM

+.111.

*Mb"

.10111.

t-

-J

Fig. 11. Circuit of the 250 Gliz SIS mixer used in the example. The coplanar SIS array is the same
as in Fig. 8, and has four junctions of RF small-signal resistance R j - 12.5 0, and (411C (4R3 C j = 4. The matching network contains a capacitor C A in series with four transmission
line matching sections. The source resistance Rs = 50 a

Fig. 12. Embedding admittance of the SIS mixer in Fig. 11 after optimizing the matching network to give
0.4. Admittances are normalized to the RF
the widest possible bandwidth with I p
0.4.
conductance of the array 1/R - 1/4Rj. The dotted circle is at I p I
1

.
The magnitude of the reflection coefficient Ipi s 0.4 from 206 296 GHz,
giving a fractional bandwidth b 0.36. For comparison, the Fano bandwidth
limit for the same mixer can be deduced from Fig. 5, using (oRC - 4 and
wIWR - 0.4 (from Fig. 9), and isu bp ano - 0.55. The Bode limit for SIS
junctions with (oRC - 4 is bs.t. s i 0.85. The bandwidth for an inductively
shunted junction (i.e., with a parallel tuning inductor and no other matching

18

elements and the same value of coRC, is b iai

in decreasing order in the table below:

0.22. These numbers are listed

Circuit
Bode limit for parallel RC
Fano limit with series L
Circuit of Fig 11
RC with parallel L tuning

85%
55%
36%
22%

The theoretical limit on the match bandwidth of the circuit of Fig. 1(b)
was derived by Fano for the law-pass case, and thence for the band-pass case
via the standard low-pass to band-pass mapping in which the inductors and
capacitors are transformed into series and parallel Lc resonators. When the
terminals of the intrinsic device are not accessible, it is not possible to
connect an inductor in parallel with C, and the low-pass to band-pass mapping
cannot be used. Here, Fano's analysis has been extended to determine the
theoretical limit on the match bandwidth of the circuit of Fig. 1(b) without
using the frequency mapping. For the circuit of Fig. 1(c), the theory of Fano
has been used, with the low-pass to band-pass frequency mapping, to determine
the match bandwidth. For the circuit of Fig. 1(a), the theory of Bode was
used.
For the equivalent circuits of Figs. 1(b) and (c), it is found that the
series inductance L limits the match bandwidth only when it exceeds a value
Ls dependent on the other circuit elements. For values of L less than L,
Bode's bandwidth limit is theoretically attainable, and there is no
fundamental reason to make L smaller than Ls when designing a broadband
circuit. We conclude that using a series array of N j devices imposes no
restriction on the theoretically attainable bandwidth of the circuit as long
as N J does not exceed the number for which the series inductance of the array
is equal to Ls.
In circuits using series arrays of devices, the power handling capacity
or saturation level depends on N j 2 It is possible that in designing a
broadband circuit, there will a conflict between power handling (requiring
more devices) and bandwidth (requiring fewer devices) if the series inductance
of the array exceeds L.
The relevance of the (RI) input match to the overall performance of
double sideband mixer receivers is discussed using the example of a 250 GHz
SIS mixer with typical niobium junctions. It is found that acceptable
performance results if the input reflection coefficient IPI
0.4. When the
mixer is realized in a coplanar waveguide circuit, it is shown that a useful

bandwidth of 36% (comparable to a standard waveguide bandwidth) is achievable

without the use of inductive tuning circuits on the individual junctions.
A mixer with a simple inductively tuned junction (parallel Rel.), but with no
other matching elements, would have a bandwidth of 22%. These figures are to
be compared to the maximum theoretical bandwidth of 55%.
X, Acknowledgements
The author would like to thank M. J. Feldman, J. Granlund, S.-K. Pan and
M. W. Pospieszalski for their helpful discussions and comments on the
manuscript.
XL References

(I]

H. W. Bode, "Network Analysis and Feedback Amplifier Design,' New York:

Van Nostrand, 1945.

(2]

R. M. Fano, "Theoretical Limitations on the Broadband Matching of

Arbitrary Impedances," j. Franklin Inst., vol. 249, pp. 57-83 and 139 155, Jan. and Feb. 1950.

[3

Q. Ke and M. J. Feldman, "Optimum Source Conductance for High Frequency

Superconducting Quasi-Particle Receivers," IEEE Trans. Microwave Theory
Tech., vol. MTT-41, no 4, pp. 600-604, April 1993.

[4

A. R. Kerr, S.-K. Pan, A. W. Lichtenberger and D. M. Lea, "Progress on

Tunerless SIS Mixers for the 200-300 Gliz Band," IEEE Microwave and
Guided Wave Letters, vol. 2, no 11, pp. 454-456, Nov. 1992.

[5

A. R. Kerr, S.-K. Pan, S. Whiteley, M. Radparvar, and S. Faris, "A Fully

Integrated SIS Mixer for 75-110 GHz," IEEE Int. Microwave Symp. Digest,
pp. 851-854, May 1990.

[6]

A. R. Kerr, S.-K. Pan, and M. J. Feldman, "Integrated Tuning Elements

for SIS Mixers," Int. J. Infrared Millimeter Waves, vol. 9, no. 2,
.
PP 203-212, Feb. 1988. This paper was presented at the International
Superconductivity Electronics Conference, Tokyo, Japan, Aug. 1987.

[7]

G. Pance and M. J. Wengler, "Integrated Tuning Elements for Millimeter

and Submillime ter SIS Mixers," IEEE International Microwave Symposium
Digest, pp. 337-340, June 1992.

[8]

G. Pance and M. J. Wengler, "Broadband Quasi-Optical SIS Mixers with

Large Area Junctions," Proceedings of the Fourth International Symposium
on Space Terahertz Technology, pp. 652-660, March 1993.

[9]

F. Schafer, E. Kreysa, T. Lehnert, and K. H. Gundlach, "A Planar SIS

Receiver with Log-Periodic Antenna for Submillimeter Waves," Proceedings
of the Fourth International Symposium on Space Terahertz Technology,
pp. 661-665, March 1993.
20

A. Karpov, M. Carter, 3. Lazareff, M. Voss, D. Billan-Peron,

K. H. Gundlach, "Wideband Fixed-Tuned and Tuneabl e SIS Mixers for
230 GHz and 345 GHz Receivers," Proceedings of the Fourth International
Symposium on Space Terahertz Technology, p. 11, March 1993.
W. R. McGrath, P. Febvre, P. Batelaan, H. G. LeDuc, B. Bumble,
M. A. Frerking and J. Hernichel, "A Submillimeter Wave SIS Receiver for
547 GHz," Proceedings of the Fourth International Symposium on Space
Tarahertz Technology, pp. 50-58, March 1993.
2

J. Mees, A. Skilare, M. H. T. M. Dierichs, H. van der Stadt,

R. A. panhuyze ll , Th . de Graauw , and T. M. Klapwijk, "Double Dipole
Antenna SIS Receivers at Frequencies Above 500 GHz," Proceedings of the
Fourth International Symposium on Space Terahertz Technology, pp. 59-71,
March 1993.

13

A. R. Kerr, S. -K. Pan and N. J. Bailey, "The Embedding Impedance in SIS

Mixers and Other Circuits Using Compact and Distributed Arrays of
Identical Devices, i n preparation.

l4J M. J. Feldman, S.-K_ Pan and A. R. Kerr, "Saturation of the SIS Mixer,"
International Superconductivity Electronics Conference, Tokyo, Digest of
Technical Papers, pp. 290-292, Aug. 1987.
51 MMICAD is a microwave integrated circuit analysis and optimization
program, and is a product of Optotek, Ltd. O nta ri o , Canada K2K 2A9.
[16] A. R. Kerr and S.-K. Pan, "Some Recent Developments in the Design cif SIS
Mixers," Inc. J. Infrared Millimeter Waves, vol. 11, no. 10, pp. 1169 1187, Oct. 1990.
[17 A. R. Kerr, S.-K. Pan and S. Withington, "Embedding Impedance
Approximations in the Analysis of SIS Mixers," IEEE Trans. Microwave
Theory Tech., vol. 41, no. 4, April 1993. This paper was originally
presented at the Third International Symposium on Space Terahertz
Technology, March 1992.

APPENDIX: Details of the "Typical" SIS Mixer Used in the Simulations

The mixer is assumed to be designed according to the procedure described

in [16] modified as in [4] , in which the optimum signal and image72 source and
IF load impedances are all equal to Rs ,apt - (Rs/2.4) (100/f(GHz))- , where Rff
is the normal resistance of the junction (or array of junctions). The mixer
is analyzed using the quasi five frequency approximation (17] to Tucker's
theory , and the I -V curve , shown in Fig . A-1 , for four NbiAl -Al203/Nb
junctions in series. The pumping parameter a eAkolhuko .... 1.2, and the DC
bias voltage is at the mid-point of the first photon step. wa N C - 4 at
230 GHz.

HY430C5K11 at 4.2 K
200

I p-A

1 00

10
V rriV

15

Fig. A-1. 1-V curve used in the analysis (from [161).

22