The CRC Handbook of Thermal Engineering

Kreith F., Timmerhaus K., Lior N., Shaw H., Shah R.K., Bell K. J., etal..“Applications.
”
The CRC Handbook of Thermal Engineering.
Ed. Frank Kreith
Boca Raton: CRC Press LLC, 2000
4-458
4.19 Thermal Conduction in Electronic Microstructures

Kenneth E. Goodson
Introduction
Submicrometer dimensions are the hallmark of micromachined transistors, semiconductor lasers, sensors,
and actuators. For many of these devices, heat transfer has a large impact on performance and reliability.
The small spatial and temporal scales of the heat transfer processes can render inappropriate much of
the theory and experimental technology common for larger systems, such as continuum heat-diffusion
theory and infrared imaging. This problem has motivated much research in the last decade on thermal
engineering at small scales, much of which has focused on heat conduction theory, thin-film thermal
conductivity measurements, and the thermometry of microstructures. This section summarizes research
progress on these topics with a focus on the simplest and most effective theoretical relationships and
experimental techniques.
There is a wealth of microdevices for which heat conduction is important. Figure 4.19.1 shows a
scanning electron micrograph (SEM) of a circuit interconnect structure, in which the temperature rise
is governed by the distribution of Joule heating in the metal bridges and thermal conduction in the
surrounding dielectric layers. The thermal engineering of interconnect structures must account for the
increasing number of metal layers and the continuous introduction of novel dielectric materials, whose
thermal properties are generally not known. Figure 4.19.2 is a transmission electron micrograph of a
metal-silicon contact region in an integrated circuit, which has failed because of the temperature rise
FIGURE 4.19.1 Scanning electron microscopy (SEM) image of the six-layer copper interconnect structure for
integrated electronic circuits. (Courtesy of the IBM Corporation.)
© 2000 by CRC Press LLC

4-459
FIGURE 4.19.2 Transmission electron microscopy (TEM) image of a metal-silicon contact, which has failed
during a brief electrical pulse. The heat generated in the tungsten plug raises the temperature and induces severe
atomic diffusion. (From Banerjee, K. et al., Proc. Int. Reliab. Phys. Symp., 1996.)
during a brief pulse of electrical current. Such current pulses are caused by electrostatic-discharge (ESD),
which can occur during chip manufacture and packaging. Figure 4.19.3 shows a cross-sectional micro-
graph of a high-voltage transistor for a smart-power circuit. High-voltage and high-power transistors are
found in smart-power circuits for vehicles and are particularly susceptible to thermal failure because of
the large rates of heat generation and the electrical noise they must withstand. In this particular device,
heat diffuses laterally in a single-crystal silicon layer of thickness as low as 200 nm. Figure 4.19.4 shows
an electron micrograph of a silicon cantilever, in which the heat is conducted by a suspended, single-
crystal silicon layer. This device is used for high-density thermomechanical data storage, during which
Joule heating in the cantilever softens the surface of an organic substrate and atomic-scale forces form
a data bit of diameter below 50 nm. The bit dimensions and writing rate are governed by conduction
along the cantilever and the tip.
FIGURE 4.19.3 Cross-sectional SEM image of a high-voltage silicon-on-insulator transistor. Electrons flow from
the source to the drain and generate heat within the silicon layer, which can be as thin as 200 nm. This device,
which is called a lateral diffusion metal-oxide-semiconductor (LDMOS) transistor, is used to block hundreds of
volts for smart-power electronic circuits. (From Leung, Y.K. et al., IEEE Electron Device Lett., 18, 414-416, 1997.)

4-460
FIGURE 4.19.4 SEM images of the silicon cantilever used for high-density thermomechanical data storage by
Chui et al. (1998). An electrical bias current along the cantilever legs induces Joule heating, which causes the tip to
locally indent an organic substrate. The inset is a scanning electron micrograph of a substrate with several of these
indentations. This data storage technology promises bit diameters below 50 nm.
The small scales of micromachined electronic devices complicate the simulation of heat conduction.
Heat conduction is particularly important in these devices and is influenced by the close proximity of
interfaces and associated changes in material stoichiometry and structural quality. Continuum heat-
diffusion simulations fail when the dimensions of the device are comparable with or smaller than the
mean free path of the heat carriers. This criterion is based on the electron mean free path in metal films
and devices, which is approximately 5 to 50 nm at room temperature. For dielectric and semiconducting
materials, heat is carried predominantly by the coupled vibrations of atoms, whose energy quanta are
phonons. The phonon mean free path in crystalline materials can approach 300 nm at room temperature.
Electron and phonon scattering at material interfaces impedes heat conduction and causes anisotropy
and a size dependence of the observed effective thermal conductivities within a device. Even in devices
with dimensions much larger than the relevant mean free path, the material stoichiometry and structural
quality can differ vastly from those found in bulk materials. This situation necessitates independent
measurements of material properties in micromachined structures closely resembling the targeted devices.
The measurement of temperature fields in devices must offer exceptionally high spatial and temporal
resolution, which is inaccessible using conventional infrared imaging and thermocouple thermometry.
The relevance of the thermal engineering of microstructures can best be appreciated from a perspective
that includes all of the heat transfer processes in electronic systems. These include convection from and
conduction within the packaging for the device and microscale conduction within or very near the
micromachined transistor, sensor, or actuator. These heat transfer processes occur with timescales and
within regions of dimensions varying by orders of magnitude, as indicated by the thermal circuit in
Figure 4.19.5. The peak temperatures typically occur in electronic microstructures, such as semiconduc-
tor devices and interconnects, where heat is generated. The thermal resistance Rmicro/nano is governed by
thermal conduction over a distance between 0.5 and 10 µm, and Cdevice or Cmetal is the total effective heat
capacity of the region heated significantly above the substrate temperature. The timescale for device or
interconnect cooling, which can be estimated as the product of the relevant resistance and capacitance,
varies from 10 nanosecond (ns) to 5 microseconds (µs).

4-461
FIGURE 4.19.5 Approximate thermal circuit showing the hierarchy of thermal resistances and capacitances in an
electronic system. The temperature rise is governed by the flow of heat power (W) through resistances (K W–1) and
into capacitances (J K–1).
Heat conduction within electronic microstructures is important if Rmicro/nano is comparable with or

larger than the other resistances in the diagram. There are important examples where the micro/nano
resistance is especially large, including silicon-on-insulator (SOI) circuits (e.g., Peters, 1993), in which
a buried silicon dioxide layer strongly impedes conduction cooling of transistors (Goodson et al., 1995a).
The buried silicon dioxide offers important benefits for both VLSI and power circuits because it
diminishes the contribution of the substrate to the electrical capacitance of transistors and interconnects.
SOI technology is also attractive for smart-power circuits, such as those containing the high-voltage
transistor in Figure 4.19.3, because it allows near-perfect dielectric isolation of power and logic devices.
The microscopic resistance is also important in cantilever sensors and actuators, such as the thermome-
chanical indentor shown in Figure 4.19.4. The cantilever structure provides a small area for conduction
cooling, which augments the resistance. Another example is multilevel interconnect structures such as
that shown in Figure 4.19.1, in which Rmicro/nano is augmented by the low-thermal-conductivity passive
material (Hunter, 1997a, 1997b). The microscale resistance can be quite large in multilayer semicon-
ducting structures, including those based on GaAs/AlGaAs or Si/Ge, due to the close proximity and
large number of interfaces in these materials. This is particularly important for photonic devices (Chen,
1996) and high-frequency amplifiers, whose figures of merit include the maximum power-handling
capability.
Figure 4.19.5 indicates that microscale conduction can be very important when interconnects and
devices are subjected to brief, intense heating pulses. If a large quantity of heat is generated during a

4-462
time less than the time constant of device or interconnect cooling to the substrate, then heat travels only
micrometers or less from the surface of the chip during the heating pulse. The energy, therefore, must
be absorbed by a very small volume, which dramatically increases the temperature rise. Such brief
heating phenomena are often the result of electrical overstress (EOS), which is a major reliability concern
for both high-power circuits and compact low-power circuits in computers. Electrostatic discharge (ESD)
can induce a current pulse into the circuit terminals of a duration between about 1 and 150 ns and of a
magnitude as high as 5 A (e.g., Amerasekera et al., 1992).
This section summarizes the theory and experiments that are used to study heat conduction in electronic
microstructures. The following subsection describes simulation approaches for the temperature fields in
microstructures, including the basics of solid-state theory. After that, theory and measurement approaches
for the effective thermal conductivities of thin electronic films are presented. Finally, this section provides
an overview of the available metrology for determining thermal conducitivities of thin films and tem-
perature distributions in microstructures.
Simulation Hierarchy for Solid-Phase Heat Conduction

Heat conduction governs the temperature distributions in many micromachined electronic structures. For
many macroscopic heat conduction problems, it is possible to obtain accurate predictions without
knowing the mechanism responsible for heat transport. For isotropic materials, the heat flux is related
to the temperature gradient by Fourier’s law,
q′′ = − k∇T (4.19.1)
where k is the thermal conductivity. Equation 4.19.1 can be used to accurately predict the temperature
distribution in macroscopic structures without knowledge of what is carrying the heat, as long as the
appropriate values of the thermal conductivity are taken from experimental data.
The value of the conductivity can be interpreted by considering the microscopic mechanism of
transport, which depends on the medium. Molecular motion carries heat in gases, for example, for which
the kinetic theory yields (e.g., Vincenti and Kruger, 1986; Rohsenow and Choi, 1961)
1
k = CνΛ (4.19.2)
3
The conductivity is proportional to the heat capacity per unit volume C [J m–3 K–1], the mean molecular
speed ν, and the mean free path Λ of molecules between consecutive collisions. In solid metals and
many metallic alloys, heat conduction is dominated by the motion of electrons. In dielectric and
semiconducting materials, the coupled vibrations of atoms are responsible for heat conduction. Quantum
theory accounts for atomic vibrational energy in discrete units called phonons, which can be visualized
as packets of energy traveling at speeds near that of sound in the material. Conduction in solids can be
simulated using the kinetic theory of gases with expressions analogous to Equation 4.19.2 for electrons
and phonons.
Accurate simulations of microstructures must consider the heat conduction mechanism. Figure 4.19.6
summarizes the hierarchy of simulation approaches for the semiconducting and dielectric regions in
electronic microstructures. The regimes of applicability are distinguished using the mean free path Λ
and the wavelength λ of phonons, which are the dominant heat carriers in these materials. In semicon-
ducting crystals such as silicon, the phonon mean free path and wavelength near room temperature are
approximately 100 and 1 nm, respectively. Many simulations can effectively use the continuum heat
diffusion theory based on Fourier’s law, Equation 4.19.1, if care is taken to use accurate values of the
thermal conductivities and interface resistances of the constituent films. Phonon transport theory becomes
important when the dimensions of a structure are comparable with the phonon mean free path. For this

4-463
FIGURE 4.19.6 Hierarchy of thermal simulation methods for dielectric and semiconducting regions in electronic
microstructures. Most of the heat generated by electrical currents in microstructures is released in semiconducting
materials. Disordered dielectric materials, which serve for electrical isolation, can in many cases dominate the thermal
resistance.
regime, solutions to the Boltzmann transport equation for phonons account for scattering on interfaces.
Nonlocal diffusion theory incorporates solutions to the Boltzmann equation for phonons into the classical
heat diffusion equation through effective thermal conductivities. When device dimensions are comparable
with the phonon wavelength, it becomes necessary to account for the interference and diffraction of
atomic vibrational waves. Because the phonon wavelength is not much larger than the interatomic
separation, rigorous treatment of conduction at these scales is inherently atomistic in nature. Atomistic
calculations determine the impact of small dimensions on the dispersion relationships of atomic vibra-
tional waves.
The following subsections present the basic behavior and thermal properties of phonons and electrons.
Next, we present formal transport theory including the Boltzmann equation, which is used to derive
constitutive laws for heat conduction in solids. This is followed by a more detailed discussion of the
application of these modeling approaches to simulations of electronic microstructures. This section is
not intended to provide a comprehensive overview of the thermal physics of solids, for which the reader
should refer to one of several texts (e.g., Kittel, 1986; Ashcroft and Mermin, 1976; Ziman, 1960). Rather,
this section summarizes the concepts and equations used most commonly in microscale thermal con-
duction simulations.
Basic Properties of Phonons
Atomistic simulations yield information about the behavior of lattice vibrational waves and the resulting
thermal properties of crystalline solids. The primary goal is to determine the dispersion relationships
for these waves, which relate the vibrational frequency to the wavelength and direction of propagation.
In their simplest form, atomistic calculations consider a one-dimensional linear chain of atoms with
mass M separated by the distance a, as depicted in Figure 4.19.7a. The ionic bonds between neighboring
atoms are modeled by a linear spring with constant K. This is a large simplification of the situation in

4-464
FIGURE 4.19.7 The simplest atomistic simulations involve a linear chain of atoms with mass M. The atoms are
linked by springs with constant K as shown in (A). A slightly more detailed simulation uses two atoms within a unit
cell, linked alternatively by spring constants K1 and K2 as shown in (B).
a real crystal, in which atoms are bonded with several neighbors in a complex three-dimensional lattice.
However, much can be learned from a linear chain of atoms. The acceleration of atom n is related to
its position xn through the forces exerted by the nearest-neighbor atoms,
d 2 xn
M = K ( xn+1 − xn ) + K ( xn−1 − xn ) (4.19.3)
dt 2
Equation 4.19.3 is solved by
xn = A exp(inkw a − iωt ) (4.19.4)
where A is the amplitude of vibrations and a is the interatomic separation. Equation 4.19.4 is a propa-
gating wave for the displacement of atoms from their equilibrium positions. The wave is described by
the wavenumber kw = 2π/λ, where λ is the wavelength, and angular frequency ω. The wave dispersion
relationship satisfying Equation 4.19.3 is
2K
[
1 − cos(kw a) ]
12
ω= (4.19.5)
M
Figure 4.19.8 plots this dispersion relationship. The phase velocity of any lattice wave, which describes
the rate of propagation of a given length of atomic displacement, is ω/kw . For the purposes of energy
transport by a packet of such waves, clustered about a given wavelength, we are more interested in the
group velocity
∂ω
ν= (4.19.6)
∂kw
For wavelengths long compared to the interatomic separation a, the group velocity becomes indepen-
dent of the wavenumber
K
ν= a ; a << λ (4.19.7)
M

4-465
Figure 4.19.8 Dispersion relationship and group velocity for the linear chain of atoms in Figure 4.19.7a.
In a three-dimensional crystal, the group velocity for long wavelengths governs the propagation of sound.
For wavelengths comparable with the interatomic spacing a, Equations 4.19.5 and 4.19.6 show that the
group velocity approaches zero because the vibrations of neighboring atoms oppose each other and are
out of phase by 180°. For wavelengths shorter than 2a, it can be shown that the atomic displacements
are redundant with those for a solution with wavelength longer than 2a (e.g., Kittel, 1986). The
wavenumbers kw = ±π/a therefore bound the useful dispersion relationship, which lies within the first
Brillouin zone.
The next level of complexity for atomistic calculations connects atoms with springs of two different
constants, K1 and K2. The atoms lie along a linear chain as in Figure 4.19.7b. This is a more realistic
model of the situation within a three-dimensional crystal, in which the basic repeating unit, called the
unit cell, often includes two or more atoms. In this case, the first Brillouin zone is bounded by kw =
±π/2a and contains two solutions. The two frequencies and group velocities for the long wavelength
limit (kwa << 1) are
2( K1 + K2 )
ω= ; ν=0 optical branch (4.19.8)
M
ω=
( K1K2 ) k a ; ν=
( K1K2 ) a acoustic branch (4.19.9)
2 M ( K1 + K2 ) w 2 M ( K1 + K2 )
The vibrations with finite frequency in the long-wavelength limit have zero group velocity. This branch
of solutions accounts for the motion of two atoms within a given unit cell with respect to each other,
approximately 180° out of phase. Because these vibrations can be caused by oscillating electric fields
they belong to the optical branch of solutions. Equation 4.19.9 presents a limit not dissimilar to that in
Equation 4.19.5 and yields a finite group velocity. Because these vibrational waves carry sound in solids,
they belong to the acoustic branch.

4-466
The situation in a three-dimensional crystal is far more complicated because of the varying atomic
positions along different directions within the lattice. The situation is also complicated by the long-range
interactions of atoms and the fact that the spring constants can be different for compression (longitudinal)
and shear (transverse) waves. Figure 4.19.9 plots the dispersion relationship for silicon along a single
crystallographic direction, which can be measured using neutron scattering. One conclusion that may
be drawn from this plot is that optical phonons have relatively small group velocities, and therefore can
be expected to contribute little to the conduction of heat according to Equation 4.19.2. Another conclusion
is that a large fraction of the transverse acoustic branch in silicon also yields relatively small group
velocities. The dispersion relationships can be interpreted using atomistic calculations like those pre-
sented here, which estimate spring constants through comparison with the data.
FIGURE 4.19.9 Dispersion relationship along the (100) direction in silicon. The atomistic calculations are com-
pared with neutron-scattering data silicon. (Adapted from M. Asheghi, Stanford University.)
The atomistic simulation approach in Figure 4.19.5 is yielding helpful information about dispersion
relationships for vibrational waves, which are needed for heat transport theory as will be described later.
In addition, atomistic simulations can be used to directly predict rates of heat transfer in solids. The
atomistic approach is influencing our understanding of phonon-interface interactions, for example, which
have a large impact on thermal conduction. Atomistic simulations examined the role of inelastic phonon
scattering on heat transport across diamond-metal interfaces (Stoner and Maris, 1993). Atomistic sim-
ulations were used to study phonon dispersion and modified anharmonic coupling in superlattices (e.g.,
Srivastava, 1994), which are important when the superlattice period is comparable with the wavelengths
of a significant fraction of thermal phonons. Related work is approaching the dispersion relationship
from the perspective of wave theory by predicting the impact of phonon interference and confinement
in compact electronic structures. For most micromachined structures, phonon interference and modified
dispersion phenomena are far less important than phonon scattering on imperfections and thin-film
interfaces. These can be considered using the transport theory introduced later.
Everything discussed before this point about lattice vibrational waves can be interpreted using the
classical laws of physics. The quantum mechanical behavior of the lattice is observed through the finite

4-467
number of vibrational states available in a crystal, which are distinguished by their wavevector kw . The
wavevector defines the wave direction as well as the wavelength, because its magnitude is the wave-
number kw = 2π/λ. Another important implication of quantum theory is that each vibrational state contains
energy in discrete increments according to
u =  n +  hω
1
(4.19.10)
 2
where u is the total energy in the state, h = 1.05459 × 10–34 Js is Planck’s constant divided by 2π, and
n is the occupation number. The occupation numbers of the vibrational states, therefore, dictate the total
energy density in the solid. When the solid is in equilibrium at a given temperature, the average occupation
number obeys the Planck distribution function
1
n = (4.19.11)
 hω 
exp  −1
 kB T 
where kB = 1.308 × 10–23 J K–1 is the Boltzmann constant. Equation 4.19.11 also governs the occupation
of states available to electromagnetic radiation by its photon energy quanta and is the basis of the more
common Planck distribution function used in thermal radiation theory. The total energy density per unit
volume in the atomic lattice is the sum over all available states of the energy per state, Equation 4.19.10,
with values of the occupation number evaluated for the temperature using Equation 4.19.11. This yields
∑ ∑  n +  hω k ≅
 ω ( kw ) 2  ( w ) ∑ ∫  n +  hω D(ω )dω
1 1 1
U= (4.19.12)
V polarizations k w polarizations o
ω
2
where the first summation accounts for the three phonon polarizations (two transverse, one longitudinal)
that can possess a given value of the wavevector kw . The second expression approximates the wavevector
sum using an integral over angular frequency and the density of phonon states D(ω) per unit volume
and frequency. The product D(ω) dω has units (m–3) and is the number of states per unit volume with
frequency between ω and ω + dω. The total volume of the crystal is V.
For heat conduction calculations we are interested in the heat capacity per unit volume of the crystal,
which plays a central role in the kinetic theory according to Equation 4.19.2. The specific heat per unit
volume is
∞
∑ ∫ ∂∂Tn hω D(ω)dω
∂U
C= = (4.19.13)
∂T polarizations o
Equation 4.19.13 assumes that the dispersion relationship does not vary with temperature, which enforces
a constant volume condition on this calculation of the heat capacity per unit volume. The density of
states can be calculated from the detailed dispersion relationships in the crystal, such as that shown for
silicon in Figure 4.19.9 (e.g., Kittel, 1986). However, many useful calculations can be performed using
simple closed-form models.
The Einstein model assumes that all vibrations have the same angular frequency ωE. Given the three
polarization states (two transverse and one longitudinal) associated with each frequency, the density of
states can then be written
D(ω ) = 3 Nδ(ω − ω E ) (4.19.14)

4-468
where N is the density per unit volume of unit cells in the atomic lattice. In the Einstein model,
Equation 4.19.13 collapses to
 θE 
2
exp(θ E T )
C = 3 NkB (4.19.15)
T
[exp(θ T ) − 1]
2
E
where the Einstein temperature is θE = hωE /kB . The high-temperature limit is
C = 3 NkB ; T >> θ E (4.19.16)
which is consistent with experimental data for the volumetric heat capacity and the law of Dulong and
Petit (e.g., Kittel, 1986). However, at low temperatures Equation 4.19.15 yields a specific heat that varies
too rapidly with decreasing temperature and strongly underpredicts the experimental data. The problem
is that many vibrational states exist with frequencies below ωE and are more easily accessed and occupied
at low temperatures.
The Debye model uses a more complicated expression for the density of states, which solves this
problem. The model assumes an isotropic linear dispersion relationship, ω = ν kw , where ν is equal to
both the group and phase velocities of the vibrational waves. The density of states with a given
polarization is
ω2
D(ω ) = ; ω ≤ ωD (4.19.17)
2 π 2 ν3
D(ω ) = 0; ω > ω D (4.19.18)
The maximum frequency ωD is dictated by the total number of states to be assigned within a unit volume.
If the number density of lattice unit cells is N, then the cutoff frequency
ω D = ν 3 6π2 N (4.19.19)
sets the total number of states equal to the number of unit cells. Because there can be multiple atoms
per unit cell, the cutoff frequency is often calculated using the total number of atoms per unit volume,
Na rather than N. This approach yields the correct total number of states in the crystal and is reasonably
effective at predicting the measured heat capacity. There is a characteristic Debye temperature associated
with this cutoff frequency,
hω D hν 3 2
θD = = 6 π Na (4.19.20)
kB kB
Table 4.19.1 provides Debye temperatures, group velocities, unit-cell number densities, and atomic
number densities for several solids. The heat capacity is calculated by combining Equations 4.19.13 and
4.19.17 to 4.19.20, which yield
3 θD T
 T x 4 exp( x )
C = 9 Na k B  
 θD  ∫
0
[exp( x ) − 1] 2
dx (4.19.21)
where x serves as an integration variable here. Equation 4.19.21 is compared with experimental data for
diamond in Figure 4.19.10. The high and low temperature limits of the heat capacity are

4-469
TABLE 4.19.1 Lattice Properties of Solids

Atomic Mass Longitudinal Transverse Debye
Density Density Phonon Velocity Phonon Velocity Temperature
Solid Na (1028 m–3) (103 kg m–3) ν (103 m s–1) ν (103 m s–1) θD (K)
Si 5.00 2.33 8.97 5.33 645

C (diamond) 17.6 3.51 17.5 12.8 2230
SiO2 6.62 2.66 6.09 4.10 492
GaAs — 5.32 5.24 2.48 345
Al 6.02 2.70 6.24 3.04 428
Cu 8.45 8.93 4.91 2.50 343
Au 5.90 19.3 3.39 1.29 165
Ag 5.85 10.5 3.78 1.74 225
Source: Data adapted from Swartz and Pohl, 1989 and Kittel, 1986.
FIGURE 4.19.10 Volumetric heat capacity data for diamond compared with the predictions of the Debye model.
3
12 π 4  T
C≅ Na kB   ; T << θ D (4.19.22)
5  θD 
C ≅ 3 Na kB ; T >> θ D (4.19.23)
The high-temperature limit is consistent with the Einstein model and the law of Dulong and Petit.
The Debye model is somewhat less successful at predicting thermal conduction, because it does not
properly distinguish between the group velocities of the acoustic and optical branches. One remedy for
this problem is to assign the Debye cutoff based on the number density of unit cells, N, and to account
for the optical phonons using the Einstein model with an appropriately averaged characteristic frequency.

4-470
FIGURE 4.19.11 Contributions of the phonon branches to the volumetric heat capacity of silicon. (Adapted from
M. Asheghi, Stanford University.)
Another approach is to calculate the density of states explicitly from the dispersion relationships, such
that D(ω) varies continuously with the wavevector and is different for each branch. This approach yields
the greatest precision in calculating the heat capacity. Figure 4.19.11 plots the result of this approach
for silicon, showing the relative contributions of the branches to the specific heat.
Basic Properties of Electrons in Metals
Electron transport in solids is modeled by integrating the Schroedinger wave equation subjected to a
periodic potential function representing the atomic lattice (e.g., Ashcroft and Mermin, 1976). The
solutions yield the electronic band structure in the material, which describes the variation of the electron
energy with the electron wavevector. The wavevector description of electrons recognizes that they behave
as waves while traveling through the atomic lattice. The electronic band structure is analogous to the
dispersion relationship for phonons, because it relates the energy and group velocity of electron waves
to the direction of propagation and wavelength. A detailed discussion of band structure and electron
properties is beyond the scope of this treatment and can be found in basic texts on the solid state.
Here, we are most concerned with electron conduction in metals, in which electrons behave much as
free charged particles in a vacuum. In contrast to phonons, which can multiply occupy a given state, the
Pauli exclusion principle restricts occupation of electron states to two, with opposite spin directions. At
zero temperature the free electrons fill the states with the lowest energy in the system up to the Fermi
energy
h2
( )
23
EF = 3π 2 Ne (4.19.24)
2m
where Ne is the number density of free electrons in the metal and m = 9.10956 × 10–31 kg is the free
electron mass. Because of the Pauli exclusion principle, only those electron states with energy close to
the Fermi level can contribute significantly to heat conduction. Electrons occupying states with lower
energies travel within the lattice but yield no net charge or heat current because the states with opposite
direction are occupied as well. This makes the velocity of electrons near the Fermi level very important,

4-471
TABLE 4.19.2 Electron Properties in Common Metals

Thermal Electron Mean
Electron Fermi Fermi Sommerfeld Electrical Conductivity, Free Path,
Concentration Energy Velocity Parameter Conductivity, 300K 300 K 300 K
Metal Ne (1028 m–3) EF (10–19 J) ν (106 m s–1) γ (J m–3 K–2) σ (105 Ω–1 m–1) k (W m–1 K–1) Λ (nm)
Rb 1.15 2.96 0.81 46.0 0.80 58 16

Ag 5.85 8.78 1.39 62.8 6.21 429 49
Au 5.90 8.83 1.39 71.4 4.55 317 32
Cu 8.45 11.2 1.57 97.5 5.88 401 26
In 11.49 13.8 1.74 322 1.14 82 1.5
Pb 13.20 15.0 1.82 653 0.48 35 0.3
Al 18.06 18.6 2.02 405 3.65 237 2.9
Note: The Fermi energy and velocity are calculated using equations in the text. The Sommerfeld parameter and the electrical
and thermal conductivities are from experimental data. The mean free path is calculated using k = CνL/3 = γTνΛ/3.
Source: Data derived from Kittel, 1986.
12
 2 EF 
(
=   3π 2 Ne
h
)
13
νF = (4.19.25)
 m   m
Table 4.19.2 provides the properties of electrons in a variety of metals. The volumetric heat capacity
of electrons is calculated using
∫ (E − E ) dT D(E)dE
df
C= F (4.19.26)
0
where the density of electron states is
32
1  2m 
D( E ) = E1 2 (4.19.27)
2π2  h2 
Electrons follow the Fermi-Dirac distribution function
1
f = (4.19.28)
 E − µ
exp  +1
 kBT 
where the chemical potential µ is very nearly equal to the Fermi energy EF at practical temperatures.
The chemical potential is calculated in a manner that conserves the number density of electrons in the
metal. This number conservation relates the chemical potential to the temperature according to
µ  1  πkBT  
2
= 1 −  
EF  3  2 EF  
(4.19.29)
 
At room temperature in most metals, the ratio kBT/EF is much smaller than unity. The distribution function
for electrons in aluminum near the Fermi energy is plotted using Equations 4.19.28 and 4.19.29 in
Figure 4.19.12. As the temperature increases, electrons are excited to higher energy states within a band
above and below the Fermi energy that is comparable with kBT. Because the Fermi energy is much larger

4-472
FIGURE 4.19.12 Fermi-Dirac distribution function for the occupation of electron states in aluminum, plotted
using Equations 4.19.28 and 4.19.29.
than kBT, only a small fraction of the electrons participate in thermal excitation and, therefore, contribute
to the specific heat.
The volumetric heat capacity of electrons increases linearly with temperature in a metal. In a very
qualitative interpretation, the number of electrons available for excitation is proportional to the width of
the energy band kBT. The energy difference they experience due to the excitation is comparable to kBT.
As a result, the total electron energy increases according to a term proportional to the square of the
temperature, and the heat capacity, therefore, is linearly proportional to temperature. The exact expression
follows from Equations 4.19.25 through 4.19.29:
π 2 kB2 NT
C= ≡ γT (4.19.30)
2 EF
which defines the Sommerfeld parameter γ. Table 4.19.2 includes experimental values of the Sommerfeld
parameter for a variety of metals, which agree qualitatively with the predictions of Equation 4.19.30
(Kittel, 1986).
Transport Theory for Phonons and Electrons
The preceding two subsections describe the basic thermal physics of the equilibrium states of lattice
vibrational waves and electrons. Heat conduction is analyzed by considering the relative motion of these
particles in conditions departing from equilibrium. Because a detailed simulation of the motion of these
particles often is computationally infeasible, many practical calculations use distribution functions ƒ. In
this approach, ƒ denotes the number of electrons or phonons in a given state denoted by wavevector kw
and position r. The evolution of the distribution function with time and space is described using the
Boltzmann equation
∂f ∂k  ∂f 
+ v ⋅ ∇r f + w ⋅ ∇k f =   (4.19.31)
∂t ∂t  ∂t  Scat

4-473
The gradient operators yield derivatives in physical space, described using the subscript r, and wavevector
space, described using the subscript kw . The group velocity v of particles usually depends on the
wavevector kw and has magnitude ν. The time rate of change of the distribution function (first term on
left) is caused by spatial gradients in the distribution function (second term on left), external forces that
change the wavevector (third term on left), and by scattering (first term on right). The third term on the
left is only relevant for electrons, for which an electric field vector E induces changes in the electron
state according to
∂k w e
= E (4.19.32)
∂t h
where e = 1.60219 × 10–19 C is the proton charge. The scattering term on the right of Equation 4.19.31
can be expressed as an integral over initial and final states for the electrons or phonons involving the
probabilities for the transitions and the distribution functions (e.g., Zimon, 1960). This integral is often
modeled using the equilibrium distribution function and the relaxation time approximation,
 ∂f  f − f − fDE
  = EQ ≡ (4.19.33)
 ∂t  Scat τ τ
The equilibrium distribution function ƒEQ at the local temperature is calculated using either
Equation 4.19.11 for phonons or Equation 4.19.28 for electrons. It must be noted that for electrons, the
value of ƒ must lie between zero and unity. For phonons, the occupation number can be arbitrarily large
and ƒ is positive and unbounded. The distribution function fDE = ƒ – fEQ is the departure of the particle
system from equilibrium. The relaxation time τ describes the rate at which the number of particles in
the state would return to equilibrium in the absence of spatial gradients. Because there are a variety of
scattering mechanisms, it is helpful to break the scattering rate τ –1 into components using Matthiessen’s
rule
τ −1 = ∑τ
j
−1
j (4.19.34)
The sum in Equation 4.19.34 is over all of the significant scattering mechanisms. For phonons, the
summation accounts for phonon-phonon scattering, phonon scattering on imperfections or impurities,
and phonon-electron scattering (Ziman, 1960). For electrons, the summation is dominated by scattering
on phonons and imperfections. Equation 4.19.34 can be augmented to account for scattering on thin-
film and grain boundaries using τ ~ d/ν, where d is the characteristic film or grain size and ν is the
velocity magnitude. Subsequent sections of this review will show that more detailed models are available
for handling scattering due to film and superlattice interfaces and grain boundaries. For superconducting
materials, Equation 4.19.34 can be augmented to account for phonon scattering with the fraction of the
electron gas remaining in the normal state at a given temperature (Uher, 1990).
The Boltzmann equation can be integrated most simply for the steady state and when spatial derivatives
of the departure from equilibrium, fDE, are negligible. The second condition is satisfied when the
boundaries do not significantly perturb the distribution function, which essentially requires that the
medium is large compared to the particle free path. Under these circumstances, we may write
∂fEQ
v ⋅ ∇r f ≅ v ⋅ ∇r T (4.19.35)
∂T
and

4-474
∂k w ∂f
⋅ ∇ k f ≅ −ev ⋅ E EQ (4.19.36)
∂t ∂E
where Equation 4.19.36 is relevant only for electrons. Under these approximations, the Boltzmann
equation reduces to
∂fEQ ∂fEQ
fDE = v ⋅ ∇ r T τ − ev ⋅ E τ (4.19.37)
∂T ∂E
For electrons, the electrical current density is calculated using
∑ −evEf
1
j= DE (4.19.38)
V all states in V
For electrons and phonons, the heat flux vector is
q′′ =
1
V ∑ vEf
all states in V
DE (4.19.39)
where E is the particle energy at a given state. The summations over all available states are usually
performed using integrals in wavevector space and an appropriate density of particle states, as discussed
in the two preceding subsections. For the electron system, Equations 4.19.38 and 4.19.39 yield the
Onsager relations governing thermoelectricity in metals (Ziman, 1960; Rohsenow and Choi, 1961), with
j = LEE E + LET ∇T (4.19.40)
q′′ = LTE E + LTT ∇T (4.19.41)
A helpful feature of electron transport in metals, as was discussed above, is that transport is dominated
by electrons with energy very nearly equal to the Fermi energy. As a result, the coefficients in
Equations 4.19.40 and 4.19.41 are dominated by the properties of electrons near the Fermi energy. The
electrical conductivity is calculated in the absence of temperature gradients,
σ = j E = LEE = Ne e2 τ m * (4.19.42)
The thermoelectric power is P = –LET /LEE and the Peltier coefficient is LTE /LEE . Although a detailed
discussion of thermoelectricity is beyond the scope of this treatment, the reader is referred to a helpful
text on the subject (Pollock, 1991).
The thermal conductivity of a metal is calculated for the situation when no electrical currents are
allowed by setting Equation 4.19.40 to zero. This yields
 L L  1 1
k = − LTT − TE ET  ≅ Cν F2 τ = Cν F Λ (4.19.43)
 LEE  3 3
where vF is the Fermi velocity, C is the heat capacity per unit volume of electrons (excluding the lattice
contribution), and Λ = vFτ is the electron mean free path. A direct consequence of the Onsager relations
for metals, when interpreted with the help of kinetic theory and Equation 4.19.37, is that the ratio of
the thermal and electrical conductivities can be simply related if the relaxation times for both processes

4-475
are assumed to be identical. The resulting relationship is independent of material and called the Wiede-
mann-Franz-Lorenz (WFL) law (e.g., Kittel, 1986)
k
= L0T (4.19.44)
σ
where the Lorenz number is
π 2 kB2
L0 = = 2.45 × 10 −8 W Ω K −2 (4.19.45)
3e2
At room temperature in nearly pure aluminum, copper, gold, and tungsten, the Lorenz number has been
observed experimentally to be 2.20, 2.23, 2.35, and 3.04 × 10–8 W Ω K–2, respectively. The WFL law
is theoretically sound at temperatures above about 50% of the Debye temperature in metals. At lower
temperatures, scattering on phonons more strongly reduces the electron relaxation time for charge
transport than it reduces the electron relaxation time for heat transport, such that the Lorenz number
decreases compared to its room-temperature value. This problem disappears for films with a large defect
density, films that are very thin and, in general, below about 10 K, where phonon scattering is over-
whelmed by scattering on static imperfections and interfaces. These scattering mechanisms preserve the
equality of the relaxation times for charge and heat transport. Because the electrical conductivity is
usually more easily measured than the thermal conductivity, the WFL law is very powerful for studying
thermal conduction by metal regions in microstructures.
The thermal conductivity by phonons, which dominates in dielectric and semiconducting materials,
is calculated using Equations 4.19.37 and 4.19.39. The term accounting for the electric field is not used
here. Phonons over a broad range of frequencies can contribute to heat conduction, such that the final
expression for the thermal conductivity involves an integral over the dimensionless phonon frequency
(e.g., Berman, 1976)
θD T
∑ 3∫
1
k= Cx ν2 τdx (4.19.46)
0
polarizations
where x = hω/kB T is the dimensionless angular phonon frequency and Cx is a frequency-dependent

weighting function for the heat capacity per unit volume. When Debye theory is used and the temperature
is substantially below the Debye temperature in the solid, the weight contributed by the specific heat
function is strongest for phonons with energy near 4 kBT. More detailed calculations can account for
phonon dispersion through more detailed models for the frequency dependence of Cx and ν. The
summation is over the two transverse and the single longitudinal phonon polarizations, for which the
frequency-dependent specific heat function and Debye temperatures θD can be calculated from the
properties of shear and longitudinal acoustic waves, respectively. Equation 4.19.46 neglects the direc-
tional dependence of the acoustic properties in a solid.
For highly disordered materials, a large number of the phonons carrying heat near room temperature
are not well defined because of their short relaxation times. For this reason, the phonon model is not
well suited for linking diffusion theory with atomistic calculations, as depicted in Figure 4.19.6. At
temperatures above the thermal-conductivity plateau region, several heat conduction mechanisms have
been proposed to be active in addition to that associated with the propagating modes accommodated by
Equation 4.19.46. Some studies considered the hopping of localized vibrational excitations, which occurs
via anharmonic interactions with phonons. Other research focused on heat transport by diffusive as
opposed to localized vibrational excitations. A recent model considered the random walk of thermal
energy between localized oscillators, a mechanism that may be considered as equivalent to diffusive
transport. This model has been particularly useful for predicting the minimum thermal conductivity of

4-476
highly disordered crystals, and yields a reasonable estimate for the thermal conductivity (Cahill et al.,
1992)
2 θD T
 T x 3 exp( x )
π
∑
13
=   kB Na2 3
kmin
 6
polarizations
ν 
 θD  ∫0
[exp( x ) − 1] 2
dx (4.19.47)
Equation 4.19.47 yields predictions of 1.04 and 0.99 W m–1 K–1 for amorphous silicon dioxide and
silicon, which are both to be compared with the experimental data of 1.4 and 1.05 W m–1 K–1 for those
materials in bulk form.
Moments of the Boltzmann Equations
The moments of the Boltzmann equation are especially relevant for simulating semiconductor devices.
Figure 4.19.13 shows the interaction of energy carriers in semiconductor devices. Energy is generally
introduced into the material by an electric field, which accelerates electrons. In semiconductors, charge
transport must be interpreted by considering two distinct electron energy bands. At zero temperature,
electrons fully occupy the valence band and are excited across the band energy gap and into the
conduction band at finite temperature. The maximum energy in the valence band and the minimum
energy in the conduction band are separated by the energy gap, Eg, which at room temperature is about
1.1 and 1.4 eV in silicon and gallium arsenide, respectively. The presence of electrons in the conduction
band contributes free electrons to the system, and their absence from the valence band yield holes in
the occupation of the available states. The net motion of electrons in the valence band can be modeled
using positively charged holes (e.g., Ashcroft and Mermin, 1976). The electrons and holes gain energy
from the electric field, scatter with the atomic lattice, and generate phonons. The recombination and
generation of electron and hole pairs releases and absorbs energy in an amount comparable to the gap
energy of the semiconductor. The energy can be released as electromagnetic radiation, whose energy
quanta are photons, which is common in semiconductor lasers.
FIGURE 4.19.13 Energy carriers and exchange mechanisms in semiconductor devices. (From Goodson, K.E.
et al., Microscale Energy Transport, Taylor & Francis, New York, 1998. With permission.)

4-477
The distinction between the acoustic and optical phonon branches lies mainly in their group velocities.
As discussed previously, the group velocity of acoustic phonons is comparable with the speed of sound
and much larger than that of optical phonons. Although both types of phonons scatter with electrons
and holes, only acoustic phonons contribute strongly to thermal conduction. Acoustic phonons, therefore,
are essential for removing the heat generated during the operation of semiconductor devices.
A detailed simulation of a semiconductor device would solve Boltzmann equations for the electron,
hole, and phonon distribution functions. This approach maintains a complete description of the distri-
bution functions of these carriers and their evolution in time and space. A more practical approach is to
solve conservation equations based on the Boltzmann equation. These are moments of the transport
equations that enforce conservation of charge, momentum, and energy without requiring detailed knowl-
edge of the distribution functions. It is often assumed that electrons and holes each satisfy equilibrium
distribution functions, but with temperatures that are not necessarily equal to that of the phonons. This
accounts for the severe departure from equilibrium that can be induced among the various energy carriers
in compact semiconductor devices. For the case of electrons, these moments are often called the
hydrodynamic equations (e.g., Bloetekjaer, 1970; Lai and Majumdar, 1995)
∂n
+ ∇ ⋅ (nv n ) = rg − rr (4.19.48)
∂t
∂pn nm*v
+ ∇ ⋅ (v n pn ) = −enE − ∇(nkBTn ) − n n (4.19.49)
∂t τ n, M
∂Wn
+ ∇ ⋅ (v n Wn ) = −env n ⋅ E − ∇ ⋅ (v n nkBTn ) + ∇ ⋅ (kn∇Tn )
∂t
 W − 3 nk T  (4.19.50)
 n 2 B s W
− − rr n − rg Eimp
τ n− s,E n
Equation 4.19.48 is the number conservation equation for electrons in the semiconducting material,
whose number density per unit volume is n. The electron temperature is Tn and the phonon temperature
is Ts. An analogous set of equations can be derived for holes. The average electron velocity vector vn is
related to the electron current density through jn = –en vn. Equations 4.19.49 and 4.19.50 are momentum
and energy balances in which the electron momentum and energy densities are pn = n mn* vn and Wn,
respectively. The first two terms on the right of the momentum balance in Equation 4.19.49 account for
acceleration due to an applied field and a net accumulation of electrons of higher momentum per unit
volume due to a gradient in the electron temperature. The final term on the right accounts for momentum
loss due to scattering processes and is governed by the electron momentum relaxation time, τn,M . Terms
with similar explanations on the right of the energy balance in Equation 4.19.50 are augmented by an
energy diffusion term proportional to the electron conductivity, kn, which can be approximately calculated
for the semiconductor from its electrical conductivity using the Wiedemann-Franz-Lorenz law. The
energy balance also accounts for the net loss of energy due to electron-hole recombination at the rate
rr and impact ionization at the rate rg. Impact ionization results from electron-ion collisions, which lift
an electron to a higher energy state. Each ionization event absorbs energy from the electron gas in the
amount Eimp. Further expansion of the recombination terms would distinguish between Auger processes,
which transmit energy to the electron gas, and processes involving phonon and photon emission (e.g.,
Sze, 1981).

4-478
A large electric field accelerates electrons into states of much higher energy than are occupied in
equilibrium at the lattice temperature. Although much of this excess energy is carried by electrons
traveling in the direction of the electric field and is described by the average electron velocity vector vn,
electron-electron scattering causes a significant fraction to be distributed without order among the
electron wavevectors. This is reflected by the definition for the electron energy density,
Wn = n  kBT + mn* v n 
3 1 2
(4.19.51)
2 2 
The first term on the right accounts for electron kinetic energy that is distributed among electron states
according to a Maxwellian distribution at the temperature Tn. The second term on the right accounts for
excess energy due to net motion of the carriers in a given direction. The electron temperature in the first
term is related to the quasi Fermi level for electrons and the electron concentration by means of
Equation 4.19.6. The hole temperature, Tp, similarly quantifies a directionally averaged departure of the
holes from equilibrium at the lattice temperature. The electron and hole temperatures strongly influence
the rate of energy transfer to phonons and the phonon thermal conductivity.
Since phonons can be created and destroyed, their number is not conserved. Because they can exchange
momentum with the crystal lattice through Umklapp scattering processes (e.g., Kittel, 1986), a momen-
tum conservation equation is of little use. Only the energy moment of the phonon Boltzmann equation
is relevant. For example, Sverdrup et al. (1999) have used the method of discrete ordinates to integrate
Equation 4.19.31 for semiconductor devices to determine the impact of interface scattering and small
dimensions of the domain of lattice heating compared to the mean free path. However, for practical
simulations it is more feasible to integrate the classical heat diffusion equation,
∂Ts
CS = ∇ ⋅ (ks∇Ts ) + q ′′′ (4.19.52)
∂t
where ks is the phonon thermal conductivity, Cs is the phonon heat capacity per unit volume, and the
heat generation term is given by
 W − 3 nk T   W − 3 pk T 
 n 2 B s  p 2 B s W W 
q ′′′ = + + rr  n + p + Eg  (4.19.53)
τ n− s,E τ p− s,E  n p 
The first and second terms in Equation 4.19.53 account for heat generation due to scattering with electrons
and holes. The second and third terms couple Equation 4.19.52 with Equation 4.19.50 and an analogous
equation, not provided here, for the hole energy density Wp. The final term on the right accounts for
phonon generation due to the recombination of carriers. This term needs to be reduced if Auger or
radiative processes are considered explicitly, because these processes transfer energy to electrons and
photons, respectively, rather than to phonons.
The solution of multiple moments of the Boltzmann equations has found broad application for bipolar
and field-effect transistors. The solution of the charge and energy moments of the electron Boltzmann
equation together with the energy moment of the phonon Boltzmann equation is available through
commercial software, such as PISCES-2ET (e.g., Yu et al., 1994; Shur, 1990). This simulation approach
has been applied to study ESD failure in silicon field-effect transistors (Beebe et al., 1994) and the
importance of self-heating effects in bipolar and SOI field-effect transistors (e.g., Apanovich et al., 1994,
1995). The analysis of Apanovich et al. (1995) is distinguished by its detailed treatment of heat generation
due to electron-hole recombination processes, which are most important for bipolar transistors. These
authors compared their predictions for the breakdown collector-base voltage in a submicrometer bipolar

4-479
transistor with predictions using drift-diffusion theory (e.g., Sze, 1981). While the current-voltage
predictions using the two different techniques agree well for low values of the collector-emitter voltage,
which corresponds to low electric fields, the breakdown collector-emitter voltage is underpredicted by
about 40% using drift-diffusion theory. This is attributed to the nonlocal nature of impact ionization,
which is better simulated by independently balancing the electron and phonon energies. Apanovich et al.
(1995) found that although self-heating of the transistor was important, temperature gradients within the
transistor were not very important. The temperature rise was dominated by the thermal resistances that
were assumed to prevail between the transistor and the environment, which are analogous to the
macroscopic resistances in Figure 4.19.5. This indicates that relatively simple simulations using the heat
diffusion equation and an isothermal lattice in the channel may be appropriate when coupled with more
rigorous simulations of energy generation and transport by the free charge carriers.
Diffusion Theory
Most simulations of electronic microstructures use heat-diffusion theory, which is based on the Fourier
heat-conduction law, together with measured or predicted effective thermal conductivity values for
constituent films. The electron system is simulated using drift diffusion theory, in which the local electron
and hole current densities are dictated by gradients in the electrostatic potential, the number density,
and the temperature. The electrostatic potential is calculated using the Poisson equation. Under these
circumstances the heat conduction problem collapses to
∂T
C = ∇ ⋅ (k ∇T ) + q ′′′ (4.19.54)
∂t
where there is now a single temperature and conductivity for the solid and the heat generation rate is
(Wachutka, 1994)
 2 jp 
2
( )[ ( )] ( )
 jn  + r − r E − E + eT P − P − T j ⋅ ∇P + j ⋅ ∇P − S (4.19.55)
q ′′′ = +
 enµ n epµ p  r g Fn Fp p n n n p p R
 
Equation 4.19.55 is strictly valid only in the steady state, but gives perspective on all of the heat generation
mechanisms that can occur in a semiconducting material. The first term on the right accounts for resistive
heating due to electron and hole scattering processes and dominates, for example, in field-effect tran-
sistors. This is the Joule heating component with separate contributions from electrons and hole current
densities, jn and jp, respectively. The electrical mobility of each carrier type is denoted here using µ with
an appropriate subscript. The second term on the right accounts for heat released due to net recombination
of electron-hole pairs and is most important for diodes and bipolar transistors. The largest component
of heat release is given by the difference in the Fermi energies of the two carrier types, EFn – EFp . These
Fermi energies are analogous to that previously described for electrons in metals. The third term on the
right is the rate of Thomson heating, which results from current flow along a gradient in the thermoelectric
power P of electrons or holes. Thomson heating is most important near gradients in the impurity
concentration, which influence Pn and Pp. A large fraction of practical device simulations neglect the
Thomson component and some can also avoid calculation of the impact of carrier recombination.
Drift-diffusion simulations have been applied for detailed studies of transistor breakdown (e.g.,
Amerasekera et al., 1993) and is also available commercially (e.g., the TRENDY software package, see
Wolbert et al., 1994). The results of Apanovich et al. (1995) place the drift diffusion theory in question
for highly nonequilibrium studies. However, since the Boltzmann equation moments and the drift
diffusion models are both phenomenological, it can be appropriate to tailor critical parameters, such as
the high field mobility and the impact ionization rate, to account for the nonequilibrium conditions. The
weakness of existing transistor simulation capabilities is the large number of unknown microscopic

4-480
parameters, such as the distinct relaxation times for electron momentum and energy transfer to the lattice.
Although a solution using more governing equations may be more rigorous in principle, the practical
relevance is often strongly degraded by the uncertainty of the many phenomenological parameters.
Lumped heater simulations treat some element of an electronic microstructure as isothermal in
solutions to Equation 4.19.54 (e.g., Goodson et al., 1995a). This approach can benefit from the wealth
of analytical techniques available for solution of the macroscopic heat diffusion equation,
Equation 4.19.55, which are documented with sufficient detail elsewhere in this handbook. Figure 4.19.14
provides an example application of this calculation approach for a silicon-on-insulator transistor, for
which conduction cooling can be modeled as occurring along extended one-dimensional regions. In this
case, the heat diffusion equation collapses to a set of simultaneous equations for extended surfaces with
linked end conditions,
d 2T T − Tsub
− =0 (4.19.56)
dx 2 LH2
where the healing length along the silicon, aluminum, and polysilicon extended surfaces, or fins, is
k fin d fin doxide

LH ≅ (4.19.57)
koxide
An exact form of Equation 4.19.57 uses a shape factor for the spreading of heat in the oxide underlying
each extended surface. Figure 4.19.14 depicts a model geometry showing the extended surfaces repre-
senting the source and drain regions, the gate, and the metallization. The temperature rise in the channel
was calculated using the simultaneous solution of Equations 4.19.56 written for the three types of
extended surfaces. The resulting temperature rise can be used to modify parameters in current-voltage
models for the transistors (e.g., Su et al., 1994). Lumped device models are very useful for circuit-level
simulations. Their accuracy depends very sensitively, however, on the accuracy with which thin-film
thermal conductivities can be determined. The determination of thin-film thermal conductivities is the
subject of the next subsection.
Thermal Conduction Properties of Electronic Films

Most electronic microstructures consist of lithographically patterned films of thickness much smaller
than their in-plane dimensions. The temperature distributions in these structures can be predicted with
reasonable accuracy using the classical heat diffusion equation if care is taken to determine the thermal
conductivities of the films and the thermal resistances at their mutual interfaces. For this reason, the
problem of simulating temperature fields in electronic microstructures can, in many cases, be reduced
to that of determining the thermal conductivities and interface resistances of their constituent semicon-
ducting, electrically insulating, and metal films.
Although many electronic materials can be produced and thermally characterized in bulk form, in
thin-film form their thermal properties can be dramatically different. This is illustrated in Figure 4.19.15
for nearly pure, single-crystal silicon films, in which phonon-interface scattering reduces the in-plane
conductivity. While the data in Figure 4.19.15 show the most significant reduction at low temperatures,
more recent data illustrated that phonon-interface scattering causes a conductivity reduction by approx-
imately a factor of two for layers of thickness near 100 nm at room temperature and above (Ju and
Goodson, 1999a). The thermal conductivity is not a size-independent parameter for films of thickness
comparable with the phonon mean free path, in which phonon-interface scattering is important. This
problem is discussed in greater detail with the definition of effective thin-film conductivities in the
following subsection.

4-481
FIGURE 4.19.14 Cross-sectional schematic (a) and model geometry (b) for a silicon-on-insulator transistor. The temperature distribution in calculated through the
simultaneous solution of several one-dimensional differential equations for extended surfaces. (From Goodson, K.E. et al., ASME J. Heat Transfer, 117, 574, 1995a. With
permission.)
4-482
FIGURE 4.19.15 Temperature-dependent thermal conductivity data along silicon films. (Adapted from Asheghi,
M. et al., ASME J. Heat Transfer, 120, 30, 1998.) (The predictions are based on the thermal conductivity integral
for phonons, Equation 4.19.46, together with a solution to the Boltzmann equation accounting for phonon-interface
scattering, Equation 4.19.60.)
The discrepancy between bulk and film properties can be large at room temperature for films whose
fabrication process introduces imperfections, impurities, or structural modifications. The local thermal
conductivity within the film can be strongly nonhomogeneous and anisotropic, even for materials that
are isotropic in bulk form. This is illustrated for fully amorphous polyimide and polycrystalline diamond
films in Figure 4.19.16. The orientation of grain boundaries in the polycrystalline film favors out-of-
plane conduction, and the variation of grain size within the layer yields nonhomogeneity. For the organic
FIGURE 4.19.16 Thickness-dependent thermal conductivity data for diamond and polyimide films. (Data from
Touzelbaev and Goodson, 1998; Verhoeven et al., 1997; Kurabayashi et al., 1999.)

4-483
film, molecular strand orientation favors in-plane conduction and yields substantial anisotropy that varies
little with film thickness.
This section summarizes experimental data and analytical tools available for determining the thermal
conductivities and interface resistances of electronic films of thickness between 50 nm and 5 µm. For
additional detail on the thermal transport properties of electronic films, the readers are directed to several
recent overview articles (Goodson and Ju, 1998; Cahill, 1997), as well as articles focusing on thin-film
interface resistances (Swartz and Pohl, 1989), experimental methods (Goodson and Flik, 1994), diamond
films (Graebner, 1993; Touzelbaev and Goodson, 1998), superlattices (Hyldgaard and Mahan, 1997;
Chen, 1998), and organic films (Kurabayashi et al., 1999; Bauer and Dereggi, 1996; Rogers et al., 1994).
Definitions of the Effective Thermal Conductivities in Films
The thermal conductivity measured in films is not necessarily a property of the film material. Because
of electron and phonon scattering at interfaces and interface thermal resistances, the apparent film
conductivities often depend on the direction of heat flow (even for isotropic materials), on the film
thickness, and on the properties of the film boundaries. For this reason, measurements yield effective
thermal conductivities for films, which are applicable only for heat propagation in a given direction for
films of a given thickness fabricated using a given set of processing conditions. The measurement of
effective thermal conductivities is therefore a very important and frequent task.
The effective conductivities are defined here using the geometry in Figure 4.19.17. The effective in-
plane thermal conductivity, for conduction along the film, is
qx  dT  −1
ka = − (4.19.58)
d L  dx 
where x is the coordinate along the film, qx is the heat power conducted along the film in the x direction,
d is the film thickness, and L (not shown in Figure 4.19.17) is the depth of the film normal to the plane
illustrated in Figure 4.19.17. The effective out-of-plane thermal conductivity, for conduction normal to
the film, is defined to include the volume resistance of the film as well as the thermal resistance between
the layer and the bounding media,
qy d
kn =
w L (T0 − T1 )
(4.19.59)
where qy is the heat power conducted through the film width w (not shown in Figure 4.19.17) and the
temperatures T0 and T1 are those of the media just outside of the film interfaces.
FIGURE 4.19.17 Geometry for defining the effective thermal conductivities of thin films.

4-484
Equations 4.19.58 and 4.19.59 define properties of practical relevance for device simulations.
Equation 4.19.58 acknowledges that it is exceedingly difficult to extract film volume and interface
resistances independently. In films of thickness comparable with the heat carrier mean free path, the
distinction between volume and interface resistances is no longer meaningful in the macroscopic sense.
This is usually not a problem because the temperature distribution in microstructures is far more sensitive
to the total out-of-plane thermal resistance than to the distribution of resistance within the film. For high-
conductivity layers, the in-plane conductivity is most important because these layers govern lateral
conduction in multilayer structures. This is the case for the silicon layer in the SOI transistor in
Figure 4.19.3, for example. For films with low thermal conductivities, such as amorphous oxides and
organic films, the out-of-plane conductivity is often more important. These films dominate the thermal
resistance in the y direction. Membrane thermal isolation structures, such as those used for infrared
bolometers (Richards, 1994), are an important exception to this rule. For these devices, the in-plane
conductivity of a suspended amorphous oxide film governs the sensitivity and time constant.
Single-Crystal Semiconducting Films and Superlattices
The in-plane conductivities of semiconducting monolayers are of practical relevance, because they tend
to be much larger than the conductivities of surrounding passive regions in micromachined structures.
In-plane conduction was most extensively investigated for silicon films of thickness between 70 nm and
about 1 µm in silicon-on-insulator (SOI) wafers. SOI technology offers complete dielectric isolation and
promises enhanced performance for semiconductor devices in both high-power and low-power circuits.
Phonon-interface scattering has a dramatic impact on heat conduction in SOI thin films below room
temperature.
Figure 4.19.15 showed the predictions and data of Asheghi et al. (1998), who measured the thermal
conductivities of SOI layers at temperatures between 20 and 300 K. The predictions are based on the
model of Holland (1963) for conduction in silicon, which is very similar to Equation 4.19.46, and account
for phonon-interface scattering. The relaxation time for the film is reduced by a solution to the Boltzmann
transport equation for the in-plane transport direction (Sondheimer, 1952):
τ film 3(1 − p) ∞
1 1  1 − exp( −δξ)
τ
= 1−
2δ ∫1
 3 − 5
ξ ξ  1 − p exp( −δξ)
dξ (4.19.60)
where δ = d/Λ = d/ντ, d is the layer thickness, and p is the fraction of phonons that are specularly
reflected by the film boundaries. Equation 4.19.60 with p = 1 yields the ideal limit of completely specular
reflection, for which no reduction in the conductivity is expected. The use of p = 0 yields completely
diffuse scattering, which minimizes the conductivity for a given film thickness. While there is reasonable
agreement between the predictions and data for the thickest layer, with d = 1.5 µm, the disagreement is
substantial for the thinner layers. One possible explanation for the discrepancy is larger concentrations
of imperfections in the film than in the bulk material. The general trend of the conductivity with
temperature is predicted with a reasonable degree of accuracy, which indicates that Equations 4.19.46
and 4.19.60 describe the dominant physics of the interface scattering problem in a single film.
It can be helpful to examine the impact of interface scattering directly on the phonon relaxation time,
independent of the thermal conductivity integral in Equation 4.19.46. For computational purposes it can
be useful to note that for completely diffuse scattering at the interfaces, p = 0, Equation 4.19.60 collapses
to
τ film 3 1
= 1− − ε (δ ) + ε 5 (δ ) (4.19.61)
τ 2δ  4 3 
where the exponential integral is defined as

4-485
FIGURE 4.19.18 Influence of interface scattering on the relaxation time in planar thin films of thickness d and a
circular wire with diameter d. The interfaces reduce the relaxation time τ and the mean free path Λ = ντ.
∫
ε n ≡ ζ − n exp( −ζδ ) d ζ
1
(4.19.62)
Figure 4.19.19 plots Equation 4.19.60 together with a closed-form approximation provided by Flik
and Tien (1990) also for completely diffuse scattering. For p = 0, Equation 4.19.60 agrees within 20%
with
τ film 2
= 1− ; δ >1 (4.19.63)
τ 3πδ
τ film
= 1−
(
2 1 − S3 ) + 2δ ln  1 + δ + S  − 2 arccos(δ) ; δ <1 (4.19.64)
τ 3πδ π 1+ δ − S π
where S = (1 – δ2)0.5. Figure 4.19.18 shows that the reduction in the mean free path of a film decreases
dramatically within decreasing film thickness. The predictions for a thin film Equations 4.19.60, 4.19.63,
and 4.19.64, yield a weaker reduction in the relaxation time than can be expected for conduction along
bridges with comparable thickness and width. Scattering on the side boundaries substantially reduces
the relaxation time for this situation. This can be appreciated through an approximate result for a circular
wire (e.g., Ziman, 1960)
τ film 1
=
( )
(4.19.65)
τ 1 + δ −1

4-486
FIGURE 4.19.19 Room-temperature effective thermal conductivities along thin silicon films. (Adapted from Ju,
Y.S. and Goodson, K.E., Appl. Phys. Lett., 1999a.)
where δ in this case is the ratio of the wire diameter and the mean free path. Equation 4.19.65 approx-
imates within 5% the exact solution to the Boltzmann equation for conduction along a wire with diffuse
scattering. It should be noted that Equation 4.19.65 can be directly derived from Matthiessen’s rule,
Equation 4.19.34, when a relaxation time of ν/d is substituted, where ƒ in this case is the wire diameter.
Conduction modeling is most poorly understood at elevated temperatures, where the complexity and
anisotropy of the phonon dispersion relationship more strongly influence transport. The complications
become substantial near about 100 K for GaAs, Ge, and AlAs, and near about 200 K for Si due to its
higher Debye temperature. Figure 4.19.19 plots room-temperature in-plane thermal conductivity data
for silicon layers in SOI substrates as thin as 74 nm, after the work of Ju and Goodson (1999a). The
thermal conductivity is smaller by as much as 50% than the bulk value and decreases slowly with
decreasing film thickness. Figure 4.19.19 also includes two simple model calculations based on
Equation 4.19.60 that assume totally diffuse scattering, p = 0. A graybody approximation overpredicts
the data because it neglects the spectral dependence of phonon scattering. The second model assumes
that either longitudinal or transverse phonons dominate conduction in the bulk material. While either
assumption can be used to interpret the thermal conductivity data of bulk samples, the data for thin films
lend some support to the hypothesis that longitudinal phonons dominate. This hypothesis also receives
support from the highly dispersive nature of high-frequency transverse acoustic phonons in silicon, which
reduces their group velocities.
The thin-film boundary conditions for phonon transport analysis warrant careful consideration.
Equation 4.19.60 neglects participation by the bounding materials, which are amorphous silicon dioxide
for the case of the SOI film studies described here. Using the theory of radiation reflection with interfaces,
an expression for the specularity factor can be derived which decreases exponentially with the square
of the ratio of the characteristic interface roughness, η, to the phonon wavelength. It can be argued that
because phonons are scattered so strongly in the silicon dioxide, this material acts as a diffuse black
phonon emitter and absorber with p = 0. When the mean free paths in neighboring films are comparable,
which is the case in many superlattices, this approximation and Equation 4.19.4 are no longer valid.
Epitaxially grown high-T c superconducting films, such as YBa 2 Cu 3 O 7 , EuBa 2 Cu 3 O 7 , and
BiSr2Ca1Cu2O8, exhibit highly anisotropic thermal conduction properties due to their orthorhombic unit
cell and the large density of oriented imperfections (Uher, 1990). These films are promising for low-
loss interconnects and Josephson junctions in hybrid superconductor/semiconductor circuits, as well as

4-487
for thermal radiation detectors. While there are few if any data available for the thermal conductivity
along thin superconducting films, there have been several measurements of the thermal interface resis-
tance between films and their underlying substrate materials (Prasher and Phelan, 1997). The bulk
properties and anisotropy have been well documented, and models have been developed for the conduc-
tivity reduction in films with the c axis oriented normal to the substrate (Richardson et al., 1992; Goodson
and Flik, 1993). These calculations needed to consider the simultaneous contribution of electrons and
phonons, because both carriers are significant at temperatures above a few tens of Kelvin. The electron
contribution in YBa2Cu3O7 was found to be approximately 50% of the total conductivity at the transition
temperature, 90 K. The reductions in electron and phonon transport were estimated independently using
Equations 4.19.46, 4.19.63, and 4.19.64. Figure 4.19.20 plots the temperature dependence of the thermal
conductivity for films of varying thickness. This calculation neglected the anisotropy of the acoustic
properties of the sample, which could be expected to diminish the size effect, and the spectral dependence
of the phonon mean free path, which will underestimate the size effect. However, this model illustrates
the anticipated physics. The reduction is substantial for films thinner than about 100 nm and is dominated
by the impact of interfaces on the phonon contribution. The electron thermal conductivity and associated
thin-film size effect is strongly reduced at low temperatures due to the increasing concentration of Cooper
pairs, which do not contribute to heat conduction.
FIGURE 4.19.20 Data for the thermal conductivity in a bulk high-temperature superconducting sample and
predictions for the conductivity in thin films. (Adapted from Goodson, K.E. and Flik, M.I., ASME J. Heat Transfer,
115, 17, 1993.)
Expressions accounting for lateral transport along crystalline superlattices based on the Boltzmann
transport equation consider the degree of specularity in phonon exchange at the periodic interfaces (e.g.,
Chen, 1998). For completely specular reflection at the interfaces of materials with similar elastic
constants, the thermal conductivity approaches that of parallel films of the two materials without interface
scattering, which is simply an average of the two bulk conductivities. This is analogous to the predictions
of Equation 4.19.60 for a single film with p = 1. Completely specular reflection is highly unlikely at
room temperature given the relatively short wavelengths of phonons, even for the best superlattices, such
that predictions for partially or fully diffuse reflection are expected to be more appropriate. Another

4-488
complicating phenomenon at superlattice interfaces is the possibility of the confinement in one layer of
high-frequency phonons, whose energies lie beyond the frequency maximum of the acoustic branches
in the bounding solid.
Although heat conduction normal to single-crystal films is being investigated most extensively for
superlattices, there remains a very large gap between phonon transport predictions and data. This problem
owes much to large concentrations of imperfections, which occur during the growth process due to the
mismatch of lattice constants in the constitutent films. Superlattices also highlight any problems with
the assumed phonon-interface scattering behavior. The key distinctions are between specular and diffuse
reflection and transmission at an interface, which has already been discussed, and between elastic and
inelastic scattering. Elastic scattering assumes that phonons striking an interface must depart with the
same frequency, such that there is no energy transfer among different modes. Fully elastic scattering
underpredicts the rate of heat conduction by neglecting the anharmonic interfacial coupling among
different modes, which can be simulated using atomistic calculations. Fully inelastic scattering overpre-
dicts the rate of heat conduction because it neglects the confinement of phonons in layers with a higher
Debye temperature. Chen (1998) has illustrated the impact of these boundary conditions by solving the
Boltzmann transport equation. The simplest limiting result neglects scattering within the layers and
assumes fully diffuse, fully inelastic scattering, yielding
 Cν C ν d +d 
kn =  1 1 2 2  1 2 (4.19.66)
 C1ν1 + C2 ν2   2 
where the subscripts 1 and 2 denote parameters of the two repeating films in the superlattice.
Equation 4.19.66 is very approximate because it neglects the frequency-dependence of scattering and
does not explicitly distinguish between the transport characteristics of optical and acoustic phonons.
Equations 4.19.66 and more rigorous calculations are compared with the data of Lee et al. (1997) for
pure Si/Ge superlattices in Figure 4.19.21. The model is consistent with the data for the smaller superlattice
period, although the results for thicker films disagree substantially. This may result from larger concen-
trations of imperfections within the layer than are present in the bulk materials, which may occur when
FIGURE 4.19.21 Predictions and data for the out-of-plane thermal conductivity in superlattices. (Adapted from
Chen, G. and Neagu, M., Appl. Phys. Lett., 71, 2761, 1997.)

4-489
the superlattice period is too large to allow the material to sustain lattice-mismatch strain. Figure 4.19.21
supports the conclusion that interface scattering is predominantly diffuse. The data also suggest that the
highly strained superlattices based on pure films contains a much higher concentration of imperfections
than superlattices containing alloys. More modeling and measurements must accompany material fab-
rication work to resolve the role of the lattice imperfections in reducing the thermal conductivity.
Polycrystalline Dielectric and Semiconducting Films
In polycrystalline films, internal scattering can overwhelm scattering on material boundaries. Increased
rates of internal scattering result from the grain boundaries and, in some cases, from lattice imperfections
and impurities that occur with high densities near grain boundaries. The polycrystalline thin-film material
whose thermal properties have received the greatest attention is chemical vapor-deposited diamond
(Graebner, 1993; Touzelbaev and Goodson, 1998). The orientation and minimum size of diamond grains
are governed by the details of the deposition process, in particular the nucleation method, the substrate
temperature, and the composition of the process gases. This subsection uses the data for these films for
a representative study of the impact of enhanced phonon scattering in polycrystalline dielectric films.
The simplest approach to modeling phonon scattering on grain boundaries is to augment the scattering
rate in Equation 4.19.46 by
ν
τ G−1 = B (4.19.67)
dG
where B is a dimensionless parameter that increases with the phonon reflection coefficient at grain
boundaries and varies with the grain shape. The mean free path calculated using Equation 4.19.67 is
Λ = dG /B, where dG is the characteristic grain dimension. For phonons with wavelengths that are long
compared to the strained region around the grain boundary, the reflection coefficient is independent of
the phonon wavelength and comparable with (∆ν/ν)2, where ∆ν is the difference in acoustic velocities
in the crystal directions for the incident and refracted paths (Ziman, 1960). The minimum grain size
dG0, which varies from a few tens of nanometers to about a micrometer, is approximately given by the
inverse square root of the nucleation density. For films deposited from the gas phase, this density is
controlled by the substrate material and the nucleation method.
Equation 4.19.67 is overly simplistic for relatively defective polycrystalline films. For example,
Equation 4.19.67 is questionable for diamond films because there is no corresponding signature in the
measured temperature dependence of the low-temperature thermal conductivity. At temperatures above
a few degrees Kelvin, the rate of phonon scattering on distinct grain boundaries, which is independent
of the phonon frequency, was found to be much smaller than the rate of scattering on other types of
defects, e.g., impurities, which depends strongly on the phonon frequency. The densities of these
imperfections are closely coupled to the size and orientation of grain boundaries in the films. If these
imperfections occur predominantly near grain boundaries, the phonon scattering rate can still be coupled
to the grain size dG . This concept was framed mathematically by Goodson (1996), who defined the
dimensionless grain-boundary scattering strength,
ηG = ∑σ n
j
j j (4.19.68)
The frequency-dependent cross section and number density per unit grain-boundary area of various
imperfections j are denoted as σj and nj, respectively. The scattering rate for randomly-oriented grains
is given by Equation 4.19.67 with Β = 2ηG , which assumes that defects associated with a given grain
boundary are distributed uniformly within grains. This simple form overpredicts the scattering rate for
transport normal to films with columnar grains, for which imperfections near grain boundaries can have

4-490
a relatively small influence on conduction. An average of phonon free paths over all directions in a
columnar grain is well approximated by
2ν   π2  
τ G−1 = 1 − exp − ηG   (4.19.69)
πdG   4 
which yields a scattering rate with an upper bound dictated by the grain size.
Figure 4.19.22 compares the data for the vertical thermal resistance of thin layers with calculations
using Equations 4.19.46 and 4.19.69. The scattering strength is governed by point and extended defects
with cross sections taken from research on thicker diamond films (Graebner, 1993). The calculations
are reasonably successful for the data included in the plot, which were measured for films deposited at
high temperatures. However, agreement for films deposited at lower temperatures requires the use of an
internal scattering term, which is independent of the grain size (Touzelbaev and Goodson, 1998). There
is only one study, to our knowledge, reporting both the in-plane and out-of-plane thermal conductivities
of a given set of diamond films. Verhoeven et al. (1997) observed a particularly large degree of anisotropy
for films with predominantly heteroepitaxial grains, with the in-plane conductivity smaller by about one
order of magnitude than the out-of-plane conductivity.
FIGURE 4.19.22 Predictions and data for the out-of-plane thermal conductivity of polycrystalline diamond films.
(From Touzelbaev, M.N. and Goodson, K.E., Diamond Relat. Mater., 7, 1, 1998, and Verhoeven, H. et al., Diamond
Relat. Mater., 6, 298, 1997.)
Amorphous Oxide and Organic Films

Process-dependent material structure and stoichiometry influence the thermal conductivities and volu-
metric heat capacities of highly disordered films, including those of amorphous glasses and organic
materials. The best representative data available at present are for silicon-dioxide and polyimide films.
Figure 4.19.16 shows that polyimide films exhibit anisotropic conductivities due to the partial alignment
of molecular strands in the film plane, which is sensitive to the spin-coating parameters. The conduction
of atomic vibrational energy is more effective by means of the electronic bonds coupling atoms along
a molecule than by the forces acting between neighboring molecules. The volumetric heat capacity in

4-491
silicon dioxide, which influences transient heat conduction, can depend strongly on the processing
conditions through their impact on the final concentrations of impurities such as water and silanol. The
modeling of heat conduction in disordered films is far more approximate than that of crystalline films,
and is limited to estimates of the minimum conductivity (Cahill et al., 1992) and simple models for the
anisotropy organic films (Kurabayashi and Goodson, 1999).
Dielectric films are often deposited at low temperatures to avoid diffusion or other failure mechanisms
in micromachined structures. The low deposition temperature can lead to microstructures or stoichiometry
very different from those found in bulk samples. The influence of these differences on thermal transport
properties is important not only for process optimization but also for study of the minimum thermal
conductivity. There have been numerous studies of the out-of-plane thermal conductivities of silicon-
dioxide films prepared using varying methods. These studies reported reductions in the conductivity by
as much as 50% from the value in bulk fused silica. Figure 4.19.23 summarizes representative data on
silicon dioxide films at room temperature, for which density data are also available. Shown as a dotted
line is the prediction of the effective medium theory (e.g., Ju and Goodson, 1999b), that models the
density deficit and thermal conductivity reduction using spherical microvoids uniformly distributed
within a fully dense matrix. The thermal conductivities of the silicon dioxide films cannot be explained
by the porosity alone. One explanation is that the density deficit occurs at an atomic scale and is not
well modeled by a theory for macroscopic pores. Recent measurements indicated that, when corrected
for impurity and density deficit, the volumetric heat capacity of LPCVD silicon dioxide films does not
depend strongly on processing history (Ju and Goodson, 1999b). Interesting observations have been
made about the impact of high-temperature annealing on the local atomic structure and thermal conduc-
tivity of CVD silicon dioxide films (Goodson et al., 1993). Following a high-temperature anneal, the
root-mean-square deviations of bond lengths, which can serve as a measure of the degree of disorder,
diminish considerably and approach those of thermally grown silicon dioxide films. This structural
change is usually accompanied by a corresponding increase in the thermal conductivity.
The thermal properties of organic films are expected to be highly sensitive to their chemical compo-
sition and structural configuration. These materials can be semicrystalline, such that the conductivity
should depend on the degree of orientation of crystalline regions and of the molecular strands in the
FIGURE 4.19.23 Data for the out-of-plane thermal conductivity in silicon dioxide films for which density data
are also available (Cahill et al., 1994; Cahill and Allen, 1994; Lee and Cahill, 1997). The data are compared with
predictions and bulk data for porous samples. (Data from Cahill et al., 1990 and Einarsrud et al., 1993.) (Graph
adapted from Ju, Y.S. and Goodson, K.E., J. Appl. Phys., in press.)

4-492
amorphous regions. The nature of disorder in this material renders most of the modeling concepts
discussed in this review inappropriate at room temperature. The relevant length-scales for observing size
effects, such as the anisotropy shown for polyimide films in Figure 4.19.16, are not known. While the
molecular weight and the molecular radius of gyration might be assumed to play a role, there has been
no systematic study of thermal transport in organic films as a function of these parameters. Past studies
of the conductivity focused primarily on demonstrating anisotropy and were not accompanied by detailed
spectroscopy. However some progress can be made using the extensive literature on heat conduction in
bulk organic materials, which has been reviewed (e.g., Ward, 1975).
The series model can be used to interpret thermal conductivity anisotropy data for organic films. The
molecular strands are modeled locally within the material as elongated segments with anisotropic intrinsic
conductivities. The longitudinal conductivity ka,ƒ describes the transport of atomic vibrational energy
along a strand, and the lower out-of-plane conductivity ka,ƒ is used to model the weaker coupling between
neighboring strands. While these two conductivities would be extremely difficult to observe experimen-
tally for a single strand, they can be considered equal to the strand-parallel and strand-normal conduc-
tivities in a polymer consisting of perfectly oriented molecules. The anisotropy in the material is then
given by (Henning, 1967)
1 1  1 1   1 1  
=  + − − cos2 ϕ 
kn 2  ka, f kn, f   ka, f kn, f 
 (4.19.70)
 
1  1 1  1
= −  cos2 ϕ + (4.19.71)
ka  ka, f kn, f  kn, f
The angle ϕ is between the film normal and a given strand direction. The operator < > yields the
average cosine considering all strands and is calculated from an assumed orientation distribution function
for the polymer strands and reduces to one-third for perfectly random orientation. Figure 4.19.24 shows
FIGURE 4.19.24 Predicted impact of the molecular orientation on the thermal conductivity anisotropy in organic
films. (Adapted from Kurabayashi, K. and Goodson, K.E., Proc. 5th ASME/JSME Thermal Eng. Joint Conf., San
Diego, CA, March 1999.)

4-493
that by assuming a Gaussian distribution function together with Equations 4.19.70 and 4.19.71, the
conductivity anisotropy can be predicted from the standard deviation in the angle of molecular segments
with respect to the in-plane direction (Kurabayashi and Goodson, 1999). These results will be much
more helpful when more basic modeling principles can be substituted for the intrinsic strand conduc-
tivities. Modeling is even more complicated for semicrystalline polymers, in which the properties and
volume fractions of the crystalline and amorphous regions add additional degrees of freedom.
Interface Resistance
In macroscopic systems, incomplete contact can be a very important contributor to the thermal resistance
at interfaces. For microstructures consisting of deposited films, complete contact between adjacent layers
is frequently achieved. For this reason, interface resistances, in general, are smaller and dominated by
different mechanisms: (1) phonon-interface scattering (Swartz and Pohl, 1989), and (2) near-interfacial
disorder (e.g., Touzelbaev and Goodson, 1998). Because of the very large interface-to-volume ratios in
micromachined structures, interface resistances can play a large or even dominant role in the total thermal
resistance of the system.
The resistance due to phonon-interface scattering occurs at the mutual boundaries of dielectric and
semiconducting films, since phonons are the dominant heat carriers in these materials, as well as for
interfaces of metallic films with dielectric or semiconducting materials. Although electrons carry heat
in the metal, the transport across the interface is brought about by phonons. One can anticipate an
analogous resistance at the interface of two metallic films due to electron scattering, but this resistance
is relatively small has received little attention from the research community.
Lattice vibrational waves are reflected or refracted at the interface, much as sound waves at a wall or
a boundary between two materials. The resulting resistance is often attributed to acoustic mismatch and
calculated using
∞
∂fEQ
∑ ∫ α hωD
1 1
= (ω ) dω (4.19.72)
RC 4 j
1 1, j
∂T
0
where D1,j (ω) is the density of states for phonons in material 1 of polarization j. The transmission
coefficient for phonons striking the interface from material 1 is α1.
For an atomically perfect interface, the transmission coefficients depend on the angle of incidence
and can be calculated using the equations of strain continuity at the interface and conservation of
momentum (e.g., Swartz and Pohl, 1989). This calculation is rather involved and is questionable for
interfaces with significant roughness, for which phonon reflection or transmission can be diffuse. The
diffuse mismatch model was derived for this situation. This model assumes that the transmission
coefficient is governed by the relative abilities of the two interfaces to transport phonons of a given
frequency away from the interface. The quantitative differences between the predictions of the acoustic
mismatch and diffuse mismatch models are relatively small, such that the diffuse assumption is used
most frequently for engineering calculations. The transmission coefficient for the diffuse mismatch model
is
∑ν 2, j D2, j (ω )
α1 = j
∑ν ∑ν
(4.19.73)
1, j D1, j (ω ) + 2, j D2, j (ω )
j j
The expression for the resistance can be simplified using the Debye model for the density of states in
both materials, Equations 4.19.17 and 4.19.18, which simplifies Equation 4.19.73 to

4-494
2 1
+
ν 2
ν 2
α1 = 2 ,transverse 2 ,longitudinal
(4.19.74)
 2 1   2 1 
 2 + 2 + 2 + 2 
 ν1,transverse ν1,longitudinal   ν2,transverse ν2,longitudinal 
The transmission in this model is assumed to be negligible for frequencies above the minimum Debye
frequency of the two materials.
The diffuse mismatch model yields reasonably good agreement with experimental data for a wide
variety of interfaces below 77 K. At higher temperatures, two complicating effects become important.
First, many phonons experience inelastic collisions at the interface, which can augment energy trans-
mission (e.g., Stoner and Maris, 1993). Another complication is the growing importance of near-
interfacial disorder with increasing temperature. Thin films rarely have homogeneous microstructural
quality and purity and, in general, the poorer material is concentrated near interfaces. While a detailed
simulation might attempt to resolve the varying thermal resistance in volume elements near an interface,
a practical compromise is to model the near-interfacial disorder as an interface resistance. Note that this
approach assumes that the interface resistance can be modeled independently from the resistance of the
volume of that material. If transport normal to a film is important, and the mean free path of phonons
in the film is comparable to the film thickness, it is no longer possible to isolate resistance contributions
of the interface and the volume. For this situation, it is recommended that calculations employ effective
film conductivities for a given thickness and direction of transport.
Measurement Techniques
The small dimensions of electronic microstructures make experimental data particularly important,
because the governing equations and many of the appropriate assumptions for simulations are still the
subject of research. Small dimensions often make experiments more challenging by requiring new
regimes of spatial and temporal resolution. This section provides an overview of the experimental
techniques that determine the effective thermal conductivities and interface resistances of thin electronic
films as well as the temperature distributions in electronic microstructures. A comprehensive overview
of the film thermal conductivity measurement techniques can also be obtained by reading the articles
of Graebner (1993), Goodson and Flik (1994), Cahill (1997), and Goodson and Ju (1999). More
information about microscale thermometry techniques can be obtained from the reviews of Majumdar
et al. (1996) and Goodson et al. (1998).
Thin-Film Thermal Properties
When measuring film thermal conductivities, care must be taken to obtain data that are appropriate for
simulating practical devices. Many techniques use a film structure that does not resemble the functional
device and, in many cases, requires processing that alters the purity or structural quality of the film. For
example, several techniques measure the in-plane thermal conductivity using free-standing films, which
experience a different stress history and are exposed to more contaminants than those within a device.
We focus here on techniques that are deemed to be most appropriate for use in conjunction with electronic
device design.
The techniques use either lasers or electrical currents for heating, which offer contrasting advantages.
Laser heating avoids contact with the sample film, which helps for small samples and when adhesion
is a problem. It is nearly impossible to consistently know the precise magnitude of incident laser power
absorbed by the film, which renders the absolute magnitude of the temperature response of little use.
However, the precise temporal control of intensity, which is made possible through picosecond and
nanosecond pulsed lasers and optical modulators, allows thermal properties to be extracted from the

4-495
temporal response without knowledge of the magnitude. Electrical heating is applied using currents in
conducting bridges, which are patterned on the film. The Joule heating power can be accurately measured
and controlled, in some cases with temporal resolution below 1 µs. The electrical techniques require
careful design and patterning of bridges on the film surface.
Steady-state electrical heating and thermometry in patterned bridges have been used to measure kn
for films with lower thermal conductivities than the underlying substrate. This measurement approach
was originally developed by Swartz and Pohl (1987) for determining the thermal resistance at the film-
substrate interface using the structure shown in Figure 4.19.25a. The Joule heating in one metal bridge
raises its temperature (Tbridge) and that of the underlying substrate (Tsubstrate). The interface or contact
resistance is given by
1 qy
=
( )
(4.19.75)
RC wL Tbridge − Tsubstrate
Tbridge is the temperature of the heater bridge and Tsubstrate is the temperature of the substrate immediately
below this bridge. The quantity qy is the total Joule heating power in the bridge of width w and length L.
The substrate temperature Tsubstrate is determined using electrical-resistance thermometry in the neigh-
boring bridge, which carries a very low current and generates relatively little power. Because the heating
power in the neighboring bridge is negligible, there is no significant temperature difference across the
interface of the neighboring film and the substrate. However, there is a temperature variation in the
substrate between the locations directly beneath the two bridges. This is considered using a solution to
the two-dimensional heat-diffusion equation based on the thermal conductivity of the substrate. This
technique for measuring the interface resistance can also be used to measure the out-of-plane conduc-
tivities of films (Goodson et al., 1993; 1994; Asheghi et al., 1998). The interface is replaced by an
electrically insulating film of thickness d, the measurement procedure remains essentially the same, and
the measured thermal resistance is that of the film and its interfaces. When the film thickness is not
much smaller than the width of the bridge, lateral spreading of heat in the film increases the observed
effective conductivity. This can be considered by using an approximate solution to the two-dimensional
heat equation in the film.
FIGURE 4.19.25 Representative structures for steady-state electrical measurements of (a) the interface thermal
resistance (Swartz and Pohl, 1987) and (b) the out-of-plane conductivity for thin films. (Adapted from Goodson
et al., J. Heat Transer, 116, 317, 1994.)

4-496
The steady-state parallel bridge approach illustrated in Figure 4.19.25 is most accurate when used
with high thermal conductivity substrates and when the bridge dimensions and separation can be made
very narrow. These conditions render the correction for substrate heating less important compared to the
temperature difference across the film or interface. That the measured resistance should be large compared
with that of the substrate between the two bridges,
d w
RC or >> (4.19.76)
kn ksubstrate
While the steady-state dual bridge method is relatively easy to apply, electrical techniques achieve lower
experimental uncertainty and more versatility using transient heating and thermometry. Many of these
extend the 3-omega method (Cahill, 1990; Lee and Cahill, 1997), for which a representative structure
and electrical signal flowchart are provided in Figure 4.19.26. In this method a signal generator is used
to sustain an electrical current in the bridge at angular frequency ω, which induces Joule heating at the
frequency 2ω and amplitude P. The bridge temperature, therefore, fluctuates at the frequency 2ω with
a phase shift and amplitude governed by the thermal properties of the structure. The voltage along the
line has a dominant component at the frequency ω, governed by the current, and a much smaller
component at the frequency 3ω, which is governed by the temperature-induced fluctuations in the
electrical resistance at 2ω. The 3-omega technique uses a bridge circuit and a lock-in amplifier to probe
the 3ω component, which is used to extract the amplitude and phase of temperature oscillations.
FIGURE 4.19.26 Representative structure and signal chart for transient electrical measurements of film thermal
conductivities using the 3-omega method of Cahill, 1997. (Drawing prepared by Y. S. Ju, Stanford University.)
The 3-omega method is simplest to apply for angular frequencies large enough such that heat penetrates
fully through the film during each heating period, yet small enough such that heat does not penetrate to
the bottom boundary of the substrate. This yields a requirement for the heating frequency 2ω in terms
of estimated conductivities k, volumetric heat capacities C, and thicknesses d of the substrate and film,
 k   k 
  << 2ω <<   (4.19.77)
 C d 2  substrate  C d 2  film

4-497
where the quantities in parentheses are characteristic heat diffusion frequencies for the substrate and
film. The solution to the heat-diffusion equation for the temperature amplitude ∆T resulting from heat
generation at the rate P exp(2 i ωt) is
∞
∆T sin 2 ( xw 2)
∫
1 d
= dx + (4.19.78)
P L πksubstrate  2iω Csubstrate  wkn
0 ( xw 2)2  x 2 + 
 ksubstrate 
where L is the length of the measurement section between the voltage probes. Note that the film normal
conductivity kn influences only the frequency-independent component to the signal, the second term on
the right. The accuracy of the 3-omega technique improves if the film width is made as narrow as
possible, since this augments the relative influence of the film resistance on the temperature rise. When
the frequency is sufficiently large such that the diffusion length in the substrate is substantially larger
than the bridge width, specifically,
4 ksubstrate
2ω << (4.19.79)
Csubstrate w 2
then Equation 4.19.78 may be simplified to
∆T 1 1  4 k  1 iπ  d
=  ln  substrate
2
+ 0.92 − ln (2ω ) −  + (4.19.80)
P L π ksubstrate  2  Csubstrate w  2 4  wkn
This shows that the thermal resistance of the substrate contributes a frequency-dependent component to
the temperature amplitude, which can be plotted as a line on a log-linear plot. If the heat capacity of
the substrate is known with reasonable accuracy, then the substrate conductivity can be extracted using
the temperature amplitude measured at two different frequencies. Even when the condition in
Equation 4.19.79 is not strictly satisfied, Equation 4.19.80 may be applied with reasonable accuracy by
adjusting the numerical constant within the brackets. Cahill (1997) recommends 1.05.
While the 3-omega method has been used primarily to measure the out-of-plane conductivity, careful
choice of layer dimensions and operating frequencies can yield other properties in samples that are
sufficiently thick. Recent progress includes measurements of the in-plane conductivity of polyimide (Ju
et al., 1999; Kurabayashi et al., 1999) and thin silicon films (Ju and Goodson, 1999a). The measurement
on thin silicon films used a buried silicon dioxide layer beneath the silicon to promote lateral heat
conduction in the structure. When heating period is comparable with the heat diffusion time normal to
the film, it is possible to extract the volumetric heat capacity of the film (Ju and Goodson, 1999b). It
must be emphasized that these specialized applications of the 3-omega method involve substantially
more care with experimental uncertainty analysis and in general require involved numerical solutions
to the heat diffusion equation in the frequency domain. These measurements are powerful but very
difficult to implement correctly.
Pulsed laser heating and laser-reflectance thermometry have been used for noncontact measurements
of the out-of-plane conductivity, as in Figure 4.19.27. The instrumentation and analysis differs greatly
depending on the duration of the heating pulse. Measurements with picosecond-scale heating, e.g., from
a Ti:Sapphire laser, have been applied to superlattices (Capinski and Maris, 1996) and diamond-like
carbon (Morath et al., 1994). These measurements use pump-probe laser diagnostics for the temperature
and analysis must decouple the responses from the near-surface temperature rise and subsurface prop-
agation. The analysis must also consider the nondiffusive nature of heat transport in the sample films
and the disequilibrium between electrons, which can absorb much of the radiation, and the lattice.

4-498
FIGURE 4.19.27 Pump-probe measurements of thermal properties in thin films. (Drawing prepared by Y. S. Ju,
Stanford University.)
The instrumentation and extraction are simpler for measurements using nanosecond-scale heating,
e.g., from a Nd:YAG laser, during which acoustic waves are fully attenuated. Nanosecond heating and
continuous time-domain laser-reflectance thermometry yielded the out-of-plane conductivities of silicon
dioxide (Kaeding et al., 1993) and polycrystalline diamond (Goodson et al., 1995b) films of thicknesses
down to a few hundred nanometers. The drawbacks of this approach include the time-domain noise from
the detector and associated electronics. Depending on the relative thicknesses of the absorbing layer and
the underlying film, the measurements can be insensitive to the volumetric heat capacity of the sample
layer.
One benefit of laser heating is that the brief pulse duration helps to decouple the properties of the
substrate from the film. This facilitates measurements on films with much higher thermal conductivities
than the underlying substrate, such as the diamond film with kn = 600 W m–1 K–1 investigated by Verhoeven
et al. (1997) on a silicon substrate. Another advantage is that compared with the electrical methods
described in the previous paragraph, these optical methods require relatively little sample preparation.
However, the data interpretation can be more involved and the experimental uncertainties are often larger.
The lateral thermal conductivity of films on substrates has been measured using the transient thermal
grating technique. Pulsed laser radiation interferes on the sample surface, yielding a harmonic spatial
variation of energy absorption. This yields spatially varying temperature and thermomechanical displace-
ment fields, whose temporal decay is detected by the deflection of an incident probe beam. The temporal
decay is governed mainly by the lateral thermal conductivity of the film and substrate and the volumetric
heat capacity within a depth near the spatial period of energy deposition, which can be varied by altering
the difference in incident angles between the two probes.
Temperature Fields in Devices
There has been much progress in recent years on transistor and interconnect thermometry (e.g., Goodson
et al., 1998). Much of this work has used either high-resolution optical diagnostics or the scanning probe
microscope, which offer improved temporal and spatial resolution, respectively. Optical methods achieve
higher temporal resolution by probing the device without contributing to the thermal mass or the electrical
capacitance of the system. The electrical techniques are in some cases more precisely calibrated and can
offer higher spatial resolution, but in some cases require extra micromachining steps. Techniques based
on scanning probe microscopy offer unprecedented spatial resolution. However, all of the techniques
make compromises among spatial and temporal resolution and accuracy. These techniques also differ

4-499
vastly in the amount of sample preparation, such as special micromachining steps, that is required before
making the measurements. This situation motivates careful attention to the choice of the best technique
for a give problem.
A number of studies have used electrical-resistance thermometry in patterned bridges made of metals
or doped semiconducting material. This approach yields a well-calibrated temperature measurement
within a precisely defined volume. Maloney and Khurana (1985) and Banerjee et al. (1996) performed
transient electrical-resistance thermometry of interconnects subjected to current pulses of duration of
less than 1 µs. An approach for field-effect transistors is to modify the gate such that it serves as a
temperature-sensing bridge (e.g., Goodson et al., 1995a), yielding the structure shown in Figure 4.19.28.
This approach offers the benefit of accurate calibration and a precise understanding of the domain where
temperature is being measured. However, these techniques usually yield only a single temperature in
the device rather than a distribution. A related strategy is to use the electrical behavior of a device, such
as the electrical conductance of a transistor, for sensing temperature (e.g., Cain et al., 1992; Zweidlinger
et al., 1996). This approach requires minimal changes to the device and can be calibrated with reasonable
accuracy. However, the temperature obtained for the device is a weighted average over a domain that is
often difficult to define precisely.
FIGURE 4.19.28 Comparison of a field-effect transistor and an experimental structure for measuring the maximum
device temperature. (Adapted from Goodson, K.E. et al., ASME J. Heat Transfer, 117, 574, 1995a.)
Scanning probe microscopy has led to a variety of innovations in surface thermometry. These methods
scan a micromachined tip with radius of curvature as low as 5 nm across the surface of the sample and
use atomic-scale forces or an electrical tunneling current as feedback for the tip-surface separation. The
scanning thermal profiler of Williams and Wickramsinghe (1986) qualitatively mapped variations in
surface thermal properties. Because this method used the tip-surface conductance as a method for
maintaining constant separation, it has difficulty decoupling surface topographical features from variations

4-500
in the temperature. This approach was refined using a thermocouple junction at the tip of an atomic
force microscope probe by Majumdar et al. (1993; 1996). This technique maintains the tip-surface
separation using the force-feedback mechanism of the atomic force microscope, while using the ther-
mocouple signal for measurements of temperature. The data obtained using this approach for semicon-
ductor laser and transistor structures promise unrivaled spatial resolution, well below 100 nm, and have
been documented extensively in review articles (e.g., Majumdar et al., 1996). These methods are ideally
suited for obtaining images with high spatial resolution. However, thermometry methods based on
scanning probe microscopy have proven very difficult to calibrate and, therefore, are most useful for
imaging qualitative distributions of temperature.
Optical methods use radiation interaction with either the surface of a device or interconnect, in some
cases probed through transparent passivation, or with a specially deposited thin layer on the surface.
Far-field laser reflectance thermometry calibrates the temperature dependence of the surface reflectivity
of an interconnect or device structure. Because of their small radiation penetration depths in metals are
especially well suited for laser-reflectance imaging. The metal yields a precisely defined location for the
measurement and minimizes the interaction of the probe beam with the electrical currents flowing in
the circuit. This motivated Ju et al. (1997) to include an aluminum layer within the overlying passivation
of a SOI power transistor. Figure 4.19.29 shows the temperature distribution across the corner of an
aluminum interconnect on an organic passivation layer. The temperature image was captured 100 ns
after the initiation of an electrical current pulse in the interconnect, and is compared with a coupled
analysis of the temperature and potential distributions. Additional optical methods include subwavelength
near-field laser-reflectance imaging (Goodson and Asheghi, 1997), thermometry using the Raman effect
(Ostermeier et al., 1992), photothermal displacement imaging (Martin and Wickramsinghe, 1987), and
fluorescence thermal imaging (Barton and Tangyunyong, 1996), whose differing advantages are discussed
in the review of Goodson et al. (1998).
FIGURE 4.19.29 Temperature distribution across the corner of a patterned aluminum alloy interconnect subjected
to a brief electrical current pulse. (Adapted from Ju, Y.S. and Goodson, K.E., IEEE Electron Device Lett., 18, 512, 1997b.)
Summary
The decreasing dimensions of micromachined electronic structures motivate research on new regimes
of simulation and measurement tools for heat conduction. This section has provided an overview of the
simulation methods together with key elements of the solid-state physics needed to understand them.
The goal has been to emphasize the most important equations and governing relationships which are
used most frequently in practice. This chapter section has also summarized the experimental techniques
that are most effective at measuring temperature distributions, thermal conductivities, and interface
resistances in electronic microstructures. A particularly large challenge for the near future includes the
further integration of many of these techniques into the research and development infrastructure of the
semiconductor industry. It is hoped that this overview will facilitate this process.

4-501
Acknowledgments
This chapter section has benefitted strongly from the interactions of the author with many of his graduate
and postgraduate students at Stanford University, including Mehdi Asheghi, Y. Sungtaek Ju, Dr. Katsuo
Kurabayashi, and Maxat N. Touzelbaev. Financial support from the Semiconductor Research Corporation
through Contracts SJ-461, PJ-357, and MJ-653 is appreciated. The author also acknowledges support
from the ONR Young Investigator Program and the NSF CAREER Program.
References
Amerasekera, A., Chang. M.C., Seitchik, J.A., Chatterjee, A., Mayaram, K., and Chern, J.H., Self-Heating
Effects in Basic Semiconductor Structures, IEEE Trans. Electron Devices, 30, 1836-1843, 1993.
Apanovich, Y., Blakey, P., Cottle, R., Lyumkis, E., Polsky, B., and Shur, A., Numerical Simulations of
Ultra-Thin SOI Transistor using Non-Isothermal Energy-Balance Model, Proc. IEEE Int. SOI
Conf., Nantucket, MA, October 3-6, 1994, pp. 33-34.
Apanovich, Y., Blakey, P., Cottle, R., Lyumkis, E., Polsky, B., Shur, A., and Tcherniaev, A., Numerical
Simulation of Submicrometer Devices Including Coupled Nonlocal Transport and Nonisothermal
Effects, IEEE Trans. Electron Devices, 42, 890-898, 1995.
Ashcroft, N.W. and Mermin, N.D., Solid State Physics, W.B. Saunders, Orlando, FL, 1976.
Asheghi, M., Touzelbaev, M.N., Goodson, K.E., Leung, Y.K., and Wong, S.S., Temperature-Dependent
Thermal Conductivity of Single-Crystal Silicon Layers in SOI Substrates, ASME J. Heat Transfer,
120, 30-36, 1998.
Banerjee, K., Amerasekera, A., and Hu, C., Characterization of VLSI Circuit Interconnect Heating and
Failure under ESD Conditions, Proc. Int. Reliab. Phys. Symp., 1996, pp. 237-245.
Barton, D.L. and Tangyunyong, Fluorescent Microthermal Imaging-Theory and Methodology for
Achieving High Thermal Resolution Images, Microelectron. Eng., 31, 271-279, 1996.
Bauer, S. and Dereggi, A.S., Pulsed Electrothermal Technique for Measuring the Thermal-Diffusivity
of Dielectric Films on Conducting Substrates, J. Appl. Phys., 80, 6124-6128, 1996.
Berman, R., Thermal Conduction in Solids, Oxford University Press, Oxford, U.K., 1976.
Beebe, S., Rotella, F., Sahul, Z., Yergeau, D., McKenna, G., So, L., Yu, Z., Wu, K.C., Kan, E., Mcvittie,
J., and Dutton, R.W., Next Generation Stanford TCAD-PISCES 2ET and SUPREM 007, Proc.
IEEE Int. Electron Devices Meet., San Francisco, CA, December 11-14, 1994.
Bloetekjaer, K., Transport Equations for Electrons in Two-Valley Semiconductors, IEEE Trans. Electron
Devices, ED-17, 38-47, 1970.
Cahill, D.G., Thermal Conductivity Measurement from 30-K to 750-K: The 3-Omega Method, Rev. Sci.
Instrum., 61, 802-808, 1990.
Cahill, D.G., Tait, R.H., Stephens, R.B., Watson, S.K., and Pohl. R.O., Thermal Conductivity, Plenum
Press, New York, 1990, pp. 3-16.
Cahill, D.G., Watson, S.K., and Pohl, R.O., Lower Limit to the Thermal Conductivity of Disordered
Crystals, Phys. Rev. B, 36, 6131-6140, 1992.
Cahill, D.G., Kativar, M., and Abelson, J.R., Thermal Conductivity of Alpha-SiH Thin Films, Phys.
Rev. B, 50, 6077-6081, 1994.
Cahill, D.G. and Allen, T.H., Thermal Conductivity of Sputtered and Evaporated SiO2 and TiO2 Optical
Coatings, Appl. Phys. Lett., 65, 309-311, 1994.
Cahill, D.G., Heat Transport in Dielectric Thin-Films and at Solid-Solid Interfaces, Microscale Ther-
mophys. Eng., 1, 85-109, 1997.
Cain, B.M., Goud, P.A., and Englefield, C.G., Electrical Measurement of the Junction Temperature of
an RF Power Transistor, IEEE Trans. Instrum. Meas., 41, 663-665, 1992.
Capinski, W.S. and Maris, H.J., Thermal Conductivity of GaAs/AlAs Superlattices, Physica B, 220, 699-
701, 1996.

4-502
Chen, G., Heat Transfer in Micro- and Nanoscale Photonic Devices, Annu. Rev. Heat Transfer, 7, 1-38,
1996.
Chen, G. and Neagu, M., Thermal Conductivity and Heat Transfer in Superlattices, Appl. Phys. Lett.,
71, 2761-2763, 1997.
Chen, G., Thermal-Conductivity and Ballistic-Phonon Transport in the Cross-Plane Direction of Super-
lattices, Phys. Rev. B, 57, 14958-14973, 1998.
Chui, B., Stowe, T.D., Ju, Y.S., Goodson, K.E., Mamin, H.J., Terris, B.D., Ried, R.P., and Rugar, D.,
Low-Stiffness Silicon Cantilevers with Integrated Heaters and Piezoresistive Sensors for High-
Density AFM Thermomechanical Data-Storage, J. MicroElectroMech. Syst., 7, 69-78, 1998.
Einarsrud, M.A., Haereid, S., and Wittwer, V., Some Thermal and Optical Properties of a New Transparent
Silica Xerogel Material with Low Density, Sol. Energy Mater. Sol. Cells, 31, 341-347, 1993.
Flik, M.I. and Tien, C.L., Size Effect on the Thermal Conductivity of High-Tc Thin-Film Superconductors,
ASME J. Heat Transfer, 112, 872-881, 1990.
Goodson, K.E. and Flik, M.I., Electron and Phonon Thermal Conduction in Epitaxial High-Tc Super-
conducting Films, ASME J. Heat Transfer, 115, 17-25, 1993.
Goodson, K.E., Flik, M.I., Su, L.T., and Antoniadis, D.A., Annealing-Temperature Dependence of the
Thermal Conductivity of LPCVD Silicon-Dioxide Layers, IEEE Electron Device Lett., 14,
490-492, 1993.
Goodson, K.E. and Flik, M.I., Solid-Layer Thermal Conductivity Measurement Techniques, Appl. Mech.
Rev., 47, 101-112, 1994.
Goodson, K.E, Flik, M.I., Su, L.T., and Antoniadis, D.A., Prediction and Measurement of the Thermal
Conductivity of Amorphous Dielectric Layers, J. Heat Transfer, 116, 317-324, 1994.
Goodson, K.E., Flik, M.I., Su, L.T., and Antoniadis, D.A., Prediction and Measurement of Temperature
Fields in Silicon-on-Insulator Circuits, ASME J. Heat Transfer, 117, 574-581, 1995a.
Goodson, K.E., Kaeding, O.W., Roesler, M., and Zachai, Experimental Investigation of Thermal Con-
duction Normal to Diamond-Silicon Boundaries, J. Appl. Phys., 77, 1385-1392, 1995b.
Goodson, K.E., Thermal Conduction in Nonhomogeneous CVD Diamond Layers in Electronic Micro-
structures, ASME J. Heat Transfer, 118, 279-286, 1996.
Goodson, K.E., and Asheghi, M., Near-Field Optical Thermometry, Microscale Thermophys. Eng., 1,
225-235, 1997.
Goodson, K.E., Ju, Y.S., and Asheghi, M., Thermal Phenomena in Semiconductor Devices and Inter-
connects, in Microscale Energy Transport, Tien, C.L. et al., Eds., Taylor & Francis, New York,
1998, pp. 229-293.
Goodson, K.E. and Ju, Y.S., Heat Conduction in Novel Electronic Films, Annual Review of Materials
Science, Kaufmann, E.N. et al., Eds., Annual Reviews, Palo Alto, CA, 261-293, 1999.
Graebner, J.E., Thermal Conductivity of CVD Diamond: Techniques and Results, Diamond Films
Technol., 3, 77-130, 1993.
Hagen, S.J., Wang, Z.Z., and Ong, N.P., Anisotropy of the Thermal Conductivity of YBa2Cu3O7-y, Phys.
Rev. B, 40, 9389-9392, 1989.
Henning, J., Anisotropy and Structure in Uniaxially Stretched Amorphous High Polymers, J. Polym.
Sci. C, 16, 2751-2761, 1967.
Holland, M.G., Analysis of Lattice Thermal Conductivity, Phys. Rev., 132, 2461-2471, 1963.
Hunter, W.R., Self-Consistent Solutions for Allowed Interconnect Current-Density. I. Implications for
Technology Evolution, IEEE Trans. Electron Devices, 44, 304-309, 1997a.
Hunter, W.R., Self-Consistent Solutions for Allowed Interconnect Current-Density. II. Application to
Design Guidelines, IEEE Trans. Electron Devices, 44, 304-309, 1997b.
Hyldgaard P. and Mahan, G.D., Phonon Superlattice Transport, Phys. Rev. B, 57, 14958-14973, 1997.
Ju, Y.S., Kaeding, O.W., Leung, Y.K., Wong, S.S., and Goodson, K.E., Short-Timescale Thermal Mapping
of Semiconductor Devices, IEEE Electron Device Lett., 18, 169-171, 1997a.

4-503
Ju, Y.S. and Goodson, K.E., Thermal Mapping of Interconnects Subjected to Brief Electrical Stresses,
IEEE Electron Device Lett., 18, 512-514, 1997b.
Ju, Y.S. and Goodson, K.E., Short-Time-Scale Thermal Mapping of Microdevices Using a Scanning
Thermoreflectance Technique, ASME J. Heat Transfer, 120, 306-313, 1998.
Ju, Y.S., Kurabayashi, K., and Goodson, K.E., Thermal Characterization of Anisotropic Thin Fielectric
Films Using Harmonic Joule Heating, Thin Solid Films, 339, 160-164, 1999.
Ju, Y.S. and Goodson, K.E., Phonon Scattering in Silicon Films of Thickness Below 100 nm, Appl. Phys.
Lett., 74, 3005-3007, 1999.
Ju, Y.S. and Goodson, K.E., Process-Dependent Thermal Transport Properties of Silicon Dioxide Films
Deposited Using Low-Pressure Chemical-Vapor-Deposited, J. Appl. Phys., 85, 7130-7134, 1999.
Kaeding, O.W., Skurk, H., and Goodson, K.E., Thermal Conduction in Metallized Silicon-Dioxide Layers
on Silicon Appl. Phys. Lett., 65, 1629-1631, 1993.
Kittel, C., Introduction to Solid-State Physics, John Wiley & Sons, New York, chap. 6, 1986.
Kurabayashi, K., Asheghi, M., Touzelbaev, M.N., and Goodson, K.E., Measurement of the Thermal
Conductivity Anisotropy in Polyimide Films, IEEE/ASME J. MicroElectroMech. Syst., in press.
Kurabayashi, K. and Goodson, K.E., Prediction of the Thermal Conductivity Anisotropy in Linear-Chain
Polyimide Films, Proc. 5th ASME/JSME Thermal Eng. Joint Conf., San Diego, CA, March 14-19,
1999. Submitted to J. Appl. Phys.
Lai, J.M. and Majumdar, A., Concurrent Thermal and Electrical Modeling of Sub-Micrometer Silicon
Devices, J. Appl. Phys., 79, 7353-7361, 1995.
Lee, S.M. and Cahill, D.G., Heat-Transport in Thin Dielectric Films, J. Appl. Phys., 81, 2590-2595, 1997.
Lee, S.M., Cahill, D.G., and Venkatasubramanian, R., Thermal Conductivity of Si-Ge Superlattices,
Appl. Phys. Lett., 70, 2957-2959, 1997.
Leung, Y.K., Paul, A.K., Goodson, K.E., Plummer, J.D., and Wong, S.S., Heating Mechanisms of LDMOS
and LIGBT in Ultrathin SOI, IEEE Electron Device Lett., 18, 414-416, 1997.
Majumdar, A., Carrejo, J.P., and Lai, J., Thermal Imaging Using the Atomic Force Microscope, Appl.
Phys. Lett., 62, 2501-2503, 1993.
Majumdar, A., Luo, K., Shi, Z., and Varesi, J., Scanning Thermal Microscopy at Nanometer Scales: A
New Frontier in Experimental Heat Transfer, Exp. Heat Transfer, 9, 83-103, 1996.
Maloney, T.J. and Khurana, N., Transmission line pulsing techniques for Circuit Modeling of ESD
Phenomena, Proc. EOS/ESD Symp., pp. 49-54, 1985.
Martin, Y. and Wickramsinghe, H.K., Study of Dynamic Current Distribution in Logic Circuits by Joule
Displacement Microscopy, Appl. Phys. Lett., 50, 167-168, 1987.
Morath, C.J., Maris, H.J., Cuomo, J.J., Pappas, D.L., Grill, A., Patel, V.V., Doyle, J.P., and Saenger,
K.L., Picosecond Optical Studies of Amorphous Diamond and Diamond-Like Carbon: Thermal
Conductivity and Longitudinal Sound Velocity, J. Appl. Phys., 76, 2636-2640, 1994.
Ostermeier, R., Brunner, K., Abstreiter, G., and Weber, W., Temperature Distribution in Si-MOSFET’s
Studied by Micro-Ramn Spectroscopy, IEEE Trans. Electron Devices, 39, 858-863, 1992.
Peters, L., SOI Takes Over Where Silicon Leaves Off, Semicond. Int., 16, 48-51, 1993.
Pollock, D.D., Thermocouples: Theory and Properties, CRC Press, Boca Raton, FL, 1991.
Prasher, R.S. and Phelan, P.E., Review of Thermal-Boundary Resistance of High-Temperature Super-
conductors, J. Supercond., 10, 473-484, 1997.
Richards, P.L., Bolometers for Infrared and Millimeter Waves, J. Appl. Phys., 76, 1-24, 1994.
Richardson, R.A., Peacor, S.D., Uher, C., and Nori, F., YBa2Cu3O7-δ Films: Calculation of the Thermal
Conductivity and Phonon Mean Free Path, J. Appl. Phys., 72, 4788-4791, 1992.
Rogers, J.A., Yang, Y., and Nelson, K.A., Elastic-Modulus and In-Plane Thermal Diffusivity Measure-
ments in Thin Polyimide Films Using Symmetry-Selective Real-Time Impulsive Stimulated Ther-
mal Scattering, Appl. Phys. A, 58, 523-534, 1994.
Rohsenow, W.M. and Choi, H.Y., Heat, Mass, and Momentum Transfer, Prentice-Hall, Englewood Cliffs,
NJ, 1961.

4-504
Shur, M., Physics of Semiconductor Devices, Prentice Hall, Englewood Cliffs, NJ, 1990.
Sondheimer, E.H., The Mean Free Path of Electrons in Metals, Adv. Phys., 1, 1-42, 1952.
Srivastava, G.P., The Physics of Phonons, Hilger, New York, 1994.
Stoner, R.J. and Maris, H.J., Kapitza Conductance and Heat-Flow Between Solids at Temperatures from
50 to 300 K, Phys. Rev. B, 48, 16373-16387, 1993.
Su, L.T., Antoniadis, D.A., Arora, N.D., Doyle, B.S., and Krakauer, D.B., SPICE Model and Parameters
for Fully-Depleted SOI MOSFET’s Including Self-Heating, IEEE Electron Device Lett., 15,
374-376, 1994.
Sverdrup, P.G., Ju, Y.S., and Goodson, K.E., Impact of Heat Source Localization on Conduction Cooling
of Silicon-on-Insulator Devices, paper presented at the Int. Conf. Modeling Simulation Microsys-
tems, Puerto Rico, April 19-22, 1999.
Swartz, E.T. and Pohl, R.O., Thermal Resistance at Interfaces, Appl. Phys. Lett., 51, 2200-2202, 1987.
Swartz, E.T. and Pohl, R.O., Thermal Boundary Resistance, Rev. Mod. Phys., 61, 605-668, 1989.
Sze, S.M., Physics of Semiconductor Devices, John Wiley & Sons, New York, 1981.
Touzelbaev, M.N. and Goodson, K.E., Applications of Micron-Scale Diamond Layers for the IC and
MEMS Industries, Diamond Relat. Mater., 7, 1-14, 1998.
Uher, C., Thermal Conductivity of High-Tc Superconductors, J. Supercond., 3, 337-389, 1990.
Verhoeven, H., Boettger, E., Floter, A., Reiss, H., and Zachai, R., Thermal Resistance and Electrical
Insulation of Thin Low-Temperature-Deposited Diamond Films, Diamond Relat. Mater., 6,
298-302, 1997.
Vincenti, W.G. and Kruger, C.H., Introduction to Physical Gas Dynamics, Krieger, Malabar, FL, 1986.
Yu, Z., Chen, D., So, L., and Dutton, R.W., PISCES-2ET — Two-Dimensional Device Simulation for
Silicon and Heterostructures, Stanford University Press, Stanford, CA, 1994.
Wachutka, G.K., Rigorous Thermodynamic Treatment of Heat Generation and Conduction in Semicon-
ductor Device Modeling, IEEE Trans. Computer-Aided Design, 9, 1141-1149, 1990.
Ward, I.M., Structure and Properties of Oriented Polymers, John Wiley & Sons, New York, 1975.
Williams, C.C. and Wickramsinghe, H.K., Scanning Thermal Profiler, Appl. Phys. Lett., 49, 1587-1589,
1986.
Wolbert, P.B., Wachutka, G.K., Krabbenborg, B.H., and Mouthaan, T.J., Nonisothermal Device Simu-
lation using the 2-D Numerical Process/Device Simulator TRENDY and Application to SOI-
Devices, IEEE Trans. Computer-Aided Design Integrated Circuits, 13, 293-302, 1994.
Zweidlinger, D.T., Fox, R.M., Brodsky, J.S., Jung, T., and Lee, S., Thermal Impedance Extraction for
Bipolar Transistors, IEEE Trans. Electron Devices, 43, 342-346, 1996.
Ziman, J.M., Electrons and Phonons, Oxford University Press, Oxford, U.K., 1960.

The CRC Handbook of Thermal Engineering

Uploaded by

Copyright:

Available Formats

The CRC Handbook of Thermal Engineering

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

The CRC Handbook of Thermal Engineering

Uploaded by

Copyright:

Available Formats

Kreith F., Timmerhaus K., Lior N., Shaw H., Shah R.K., Bell K. J., etal..“Applications.

4.19 Thermal Conduction in Electronic Microstructures

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

Heat conduction within electronic microstructures is important if Rmicro/nano is comparable with or

© 2000 by CRC Press LLC

Simulation Hierarchy for Solid-Phase Heat Conduction

q′′ = − k∇T (4.19.1)

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

Equation 4.19.3 is solved by

xn = A exp(inkw a − iωt ) (4.19.4)

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

D(ω ) = 3 Nδ(ω − ω E ) (4.19.14)

© 2000 by CRC Press LLC

where the Einstein temperature is θE = hωE /kB . The high-temperature limit is

C = 3 NkB ; T >> θ E (4.19.16)

D(ω ) = 0; ω > ω D (4.19.18)

© 2000 by CRC Press LLC

TABLE 4.19.1 Lattice Properties of Solids

Si 5.00 2.33 8.97 5.33 645

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

TABLE 4.19.2 Electron Properties in Common Metals

Rb 1.15 2.96 0.81 46.0 0.80 58 16

where the density of electron states is

Electrons follow the Fermi-Dirac distribution function

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

For electrons, the electrical current density is calculated using

For electrons and phonons, the heat flux vector is

j = LEE E + LET ∇T (4.19.40)

q′′ = LTE E + LTT ∇T (4.19.41)

© 2000 by CRC Press LLC

where x = hω/kB T is the dimensionless angular phonon frequency and Cx is a frequency-dependent

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

k fin d fin doxide

Thermal Conduction Properties of Electronic Films

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

where the exponential integral is defined as

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC

Amorphous Oxide and Organic Films

© 2000 by CRC Press LLC

© 2000 by CRC Press LLC