reliability and performance scaling of very high speed sige hbts
TRANSCRIPT
Microelectronics Reliability 44 (2004) 397–410
www.elsevier.com/locate/microrel
Introductory Invited Paper
Reliability and performance scaling ofvery high speed SiGe HBTs
Greg Freeman *, Jae-Sung Rieh, Zhijian Yang, Fernando Guarin
IBM Microelectronics Semiconductor Research and Development Center, 2070 Rt. 52,
Hopewell Junction, New York, NY 12533, USA
Received 5 November 2003
Abstract
We discuss the SiGe HBT structural changes required for very high performance. The increase in collector con-
centration, affecting current density and avalanche current, appears to be the most fundamental concern for reliability.
In device design, a narrow emitter and reduced poly–single-crystal interfacial oxide are important elements in mini-
mizing device parameter shifts. From the application point of view, avalanche hot-carriers appear to present new
constraints, which may be managed through limiting voltage (to 1.5·–2· BVCEO), or through circuit designs robust to
base current parameter shifts.
� 2003 Elsevier Ltd. All rights reserved.
1. Introduction
SiGe HBT speed and noise performance continue to
improve, with demonstrated results among the highest
speed and lowest noise of any device in any material
system. Recent performance results exceed 350 GHz fT[1] and NFMIN results demonstrated below 0.4 dB at 10
GHz [2]. Compared to III–V devices, only the smallest
dimension HEMT devices [3] or laboratory InP HBT
devices [4] have demonstrated such high fT or low
NFMIN, and silicon NFETs are expected to achieve
comparable fT performance at the 25 nm node. Largely
responsible for the silicon-based HBT advancement are
material innovations, the first of which is 10–25% Ge
incorporation into the silicon lattice, which provides
bandgap modifications and drift-fields to enhance per-
formance above non-Ge containing devices. Another
more recent material innovation is the additional
incorporation of <1% C into the SiGe epitaxy, which
dramatically reduces the base layer boron diffusion
through subsequent thermal processing and thus reduces
transit time by retaining narrow base layers.
* Corresponding author. Tel.: +1-845-892-4690.
E-mail address: [email protected] (G. Freeman).
0026-2714/$ - see front matter � 2003 Elsevier Ltd. All rights reserv
doi:10.1016/j.microrel.2003.11.003
Structural innovations have also contributed to the
improved performance. Because performance is a func-
tion of parasitic capacitance and resistance as well as
transit times, the reduction in parasitic elements is
also needed. Small feature sizes, shallow junctions, self-
aligned silicides, trench isolation, and other silicon
process features have been employed to maximize device
performance. A survey of the various SiGe HBT devices
reported will find a wide range of structures, reflecting
the baseline silicon process capability at different facili-
ties, as well as differences in innovation (e.g., see [5]).
Thus these material and structural capabilities well make
up for the lower electron mobility in silicon to boost
SiGe HBT performance above most InP and GaAs
based devices.
Device construction to achieve such high perfor-
mance typically results in higher electric fields, higher
temperature operation, or higher current densities, each
of which may result in increased device degradation if
not appropriately accounted for in device design. The
material system and the associated fabrication capability
determine the tools and the innovations available to
adjust for the negative effects of scaling in order to
construct a highly reliable device.
The scaling experience and reliability expectations
from III to V devices are commonly applied to the
ed.
398 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
silicon system. This may be because the same engineers
who have historically designed and assessed telecom-
munication, space, and military applications with III–V
devices are using SiGe HBTs and more recently silicon
FET devices. Experience defines the expectation of
failure signatures and the methods for assuring reliable
devices. III–V device reliability has historically exhibited
failure modes within the crystal structure that exhibit
abrupt and catastrophic end-of-life characteristics.
There are also failure modes driven by the integrity of
ohmic contacts and by the mesa ledge and its passiv-
ation. Silicon device reliability has historically exhibited
failure modes within device non-crystal elements, such
as dielectrics and thin metal lines. Gradual degradation
is observed in the dielectric interfaces, and catastrophic
degradation may be found in such structures as the
metal lines when overstressed due to improper metal
layout (detailed design guidelines are provided to avoid
these situations).
Regarding the Si or SiGe crystal structure, experience
indicates that, in contrast to III–V systems, the Si/SiGe
crystal structure is extremely robust. From the material
science to the practical fabrication of the devices, there
are many important differences that indicate the
robustness of the SiGe HBT device. First, we note that
the fabrication temperatures used are substantially
higher with silicon and SiGe systems. During the wafer
fabrication, thermal oxidation and anneal process con-
ditions many times exceed 900 �C and at times exceed
1000 �C with well-behaved effects. III–V fabrication
processes generally restrict temperatures to below 500
�C, utilizing low-temperature epitaxial growth and
ohmic metal contact formations as well as mesa struc-
tures for isolations. One reason for the low-temperature
processing in III–V is the difference in diffusion prop-
erties of dopants. Comparing the diffusion constants of
typical p-type dopants Be in GaAs to B in Si (chosen
since they affect base stability in HBTs) one finds that Be
in GaAs exhibits higher diffusion constant of approxi-
mately six orders of magnitude at a moderately high
processing temperature of 800 �C (e.g., see [6]; we note
that due to the complexity of diffusion in III–V hosts,
reported diffusion constants vary widely). As one should
expect, this temperature tolerance difference results in
significant guard banding in normal 125 �C device
operation, where the diffusion constant difference is ex-
pected to be even larger, extrapolating to over 20 orders
of magnitude in favor of the Si system. Because the
implication of the diffusion constant difference is difficult
to understand in the device operation and reliability,
consider the use of these diffusion characteristics for
approximating the temperature required for the dopant
to move a distance of 10 nm in a period of 10 years
(driven only by a concentration gradient). In this extrap-
olation, we estimate that the Be dopant would require a
temperature of 390 �C, compared to B in Si which would
require a temperature of 871 �C for that amount of
movement.
This difference in thermal response manifests itself in
the well-known device failure mechanisms of III–V
systems [7–9], and the observation that such failure
mechanisms have not been reported in silicon or SiGe
systems. Common failure mechanisms in III–V devices
are movement of dopants or host atoms. As a result,
location of the dopant relative to the band offset is
found to change through the life of the device, eventu-
ally resulting in significant rapid degradation once the
dopant moves beyond the band discontinuity. The dif-
ference in gradient-driven diffusion of base dopants
discussed above, while not the same mechanism as takes
place during device operation, still provides insight into
the different electrical mechanisms observed. Addition-
ally, defect clusters may grow within the device, resulting
in increasing charge storage and carrier recombination,
and thus device performance degradation. Or crystal
material movement may cause threading defect migra-
tion in the device, causing sudden shorts or leakage
between device terminals. Heat and current density are
often responsible for accelerating these mechanisms, and
without changes in the fundamental device structure,
increasing these effects with device performance scaling
can impose serious limitations in achieving reliable
performance in III–V devices.
Additional well-known advantages of the silicon
system are the higher thermal conductivity of the silicon
substrate and the narrow line patterning capability,
which together ensures lower device temperature and the
reliable operation of device components that are sensi-
tive to temperature (such as metal migration). At 1.48
W/cmK, the thermal conductivity of Si is approximately
2–3 times higher than InP or GaAs at 0.74 and 0.45
W/cmK respectively. The narrow device dimensions
achievable in the silicon anisotropic dry etches enable
overall lower power (through smaller devices) as well as
better spreading of heat at the device (from a more
narrow emitter stripe), further improving the thermal
properties and reliability of the integrated circuit. Thus,
as compared to III–V systems, the silicon material and
processing capabilities have proven to be robust to his-
toric failure modes in high performance III–V device
reliability. We conclude that, based on fundamental
material properties as well as consistent lack of evidence
that similar mechanisms exist, that the reliability expe-
rience in III–V devices should not be directly applied to
Si devices.
To better understand the reliability of the SiGe HBT
to ever increasing performance, we explore the topic
through a systematic approach focusing on historic sili-
con bipolar device reliability issues. In this paper, we
relate the trends in SiGe HBT device structure and
electrical properties to the known degradation mecha-
nisms in silicon and SiGe bipolar devices. To accomplish
CollectorBase
Ge ramp
Con
cent
ratio
n (l
og s
cale
)
G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410 399
this, we discuss device scaling in the next section as
background to enumerate the required structure and
electrical modifications to enhance performance. This is
followed by a review of the device degradation mecha-
nisms reported for Si and SiGe devices, relating these
mechanisms to the previously discussed device scaling
trends. In this section, we report the status of high-speed
SiGe HBT device reliability assessment for 200 GHz
HBTs and comment on trends observed. The results and
trends may be extrapolated to the 350 GHz devices and
beyond.
Depth
Emitter Base Collector
Structure Replace Arsenic with Phos.
Reduce interface layer
Narrow baseReduce base doseIncrease Ge rampAdd CarbonImprove extrinsic
structure
Increase concentration
Improve extrinsic structure
Electrical Lower RE
Higher IB
Lower VBE
Reduce total RB
Higher avalancheHigher peak fT JC
Higher CCB
Fig. 1. SiGe HBT device scaling.
2. Device scaling
Device performance is improved through reduction
in transit time and parasitic resistances and capaci-
tances. fT improvement is achieved through improve-
ment in each of the delay components, related to fT in
the well-established approximation:
1
2pfT¼ sEC ¼ sE þ sC þ sB þ sCSCL
� kTqIC
CEB þ kTqIC
�þ RC þ RE
�CCB þ W 2
B
cDnþ WCSCL
2vSAT
;
where CEB and CCB are emitter–base and base–collector
capacitance, RC and RE are collector and emitter resis-
tance, WB and WCSCL are neutral base and base–collector
space charge layer width, respectively, k is Boltzmann
constant, T is temperature, q is unit electron charge, Dn
is the electron diffusivity, c is field factor, and vSAT is the
electron saturation velocity.
Due to its unambiguous definition and straightfor-
ward extraction, fT is probably the most common figure
of merit for an RF transistor. However, in many cases
the fMAX figure of merit is more predictive of circuit
performance [10]. This figure of merit emphasizes RB and
CCB due to their importance in driving power to a load,
as in the following approximation:
fMAX �
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffifT
8pRBCCB
s:
The device design generally focuses separately on the
emitter, the base, and the collector. Fig. 1 illustrates and
itemizes the scaling aspects, and these are discussed be-
low.
2.1. Emitter design
The goal of emitter design in high-speed SiGe HBTs
is to provide an optimal junction depth for maximum fTperformance, to provide low emitter resistance with tight
parametric control, to minimize the adverse effects on
base dopant diffusion during processing, and to provide
for a reliable device as will be discussed later in this
article. Reduced drive-in temperature requirements have
established phosphorous over arsenic as the preferred
emitter dopant in the highest performance SiGe HBTs
[1,11,12]. This lower temperature reduces the base dop-
ant diffusion and thus the electron transit time. On the
other hand, arsenic emitter junction drive-in requires a
higher temperature, which is more compatible with
NFET and PFET device activation anneals. This has
influenced a majority of production BiCMOS processes
to adopt such an emitter dopant. Carbon doping in the
base layer is enabling the use of arsenic in high perfor-
mance BiCMOS processes by minimizing the base
dopant diffusion during these higher thermal cycles
[13,14]. Further, the higher current densities in high
performance SiGe HBTs (due to the collector design to
be discussed) have required the compensation of higher
currents with lower emitter resistance to maintain a low
RE � IC product. This has caused some manufacturers to
adopt an in situ-doped re-aligned emitter with minimal
interfacial oxide [15,16].
2.2. Base design
The goal in base design is the reduction in base tran-
sit time sB without severe degradation in base resistance
and fMAX. Base transit time is typically reduced by some
combination of the following methods: (1) thinning the
base film as deposited; (2) reducing thermal cycles to
minimize the base dopant diffusion; (3) adding carbon to
the SiGe epitaxy to reduce boron diffusion and (4)
increasing the Ge ramp to accelerate electrons more
400 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
quickly across the neutral base. In some cases, the total
boron dose (i.e. the base Gummel number) is reduced as
a method to define a thinner base, and in other cases, the
reduced diffusion from temperature and/or the carbon
effect on diffusion leads to a thinner base with the same
total dose. In the former case, the lower boron dose
leads to higher base resistance, which negatively impacts
device fMAX figure of merit, as well as NFMIN, and makes
the device more susceptible to punch-through at higher
collector–base voltages. In addition, the higher base
resistance may have an impact on operating voltage
limits as will be discussed later. The incorporation of
carbon has little effect on the electrical characteristics,
and thus it appears to have little negative effect on device
degradation. Instead, it may produce a positive effect,
from constraining boron diffusion and thus improving
the field at the silicon surface, as will be discussed in the
next section. Further, because the increase in Ge ramp
rate is accompanied by a reduction in base width, the
higher Ge concentration is over a lesser depth in the
device, and thus the effective strain is reduced going to
higher performance devices.
Total base resistance must also be reduced, as the
figure of merit fMAX is highly dependent on the base
resistance, and fMAX need also be increased with fT for
improved circuit performance. The extrinsic base resis-
tance must be reduced. This may be accomplished by a
variety of methods, including raised extrinsic base [11]
and self-aligned silicides to the emitter [17].
2.3. Collector design
Finally, the collector design for high performance
HBTs addresses multiple performance bottlenecks.
First, the collector–base space-charge layer contributes a
significant portion of the total transit time, and must be
reduced by an increase in collector concentration. Sec-
ond, because the charging rate of the collector–base and
emitter–base capacitance is influenced in part by the
small-signal transconductance of the device, an increase
M-1
at V
CB=1
.5V
0
5
10
15
20
25
0 100 200 300 400
Peak fT (GHz)
J C a
t pea
k f T
[mA
/m2 ]
(a) (b
Fig. 2. Effect of collector design on: (a) peak fT current density an
collector–base space-charge region.
in the collector current, and thus the transconductance
is needed. This higher transconductance is achieved
through delaying the onset of the Kirk effect, again with
a higher collector concentration.
Increasing collector concentration negatively impacts
the capacitance of the device and thus partially reduces
the performance gains obtained. Because the increase in
concentration is constrained to the region under and
slightly surrounding the emitter of the device through a
mask and implant step (or self-aligned implant), the
impact is minimized. Further, because the mask overlay
control is improved between technology generations, the
surrounding extrinsic parasitic capacitance may be re-
duced. Also, at the option of the circuit designer, the
mask may produce a complete block to the implant,
resulting in an alternate device with lower collector
concentration that may be optimal for purposes where
the bias current is not required to be as high.
The straightforward increase in collector concentra-
tion has probably the greatest impact on the concerns
for device degradation as will be discussed in Section 3.
The electrical effects are shown in Fig. 2. Fig. 2a shows
the increased peak fT current density JCP resulting
from the delayed Kirk effect as a function of peak fT for
various device designs. This has implications to device
self-heating and other degradation if device sizes are not
adjusted to take such effects into account. Fig. 2b shows
the increased collector current avalanche multipli-
cation M-1, also as a function of peak fT value. Since
this avalanche process is a result of high kinetic energy
carriers and high electric fields, it results in significant
hot-carrier generation at some bias conditions, as will
also be discussed.
Extrinsic collector resistance reduction is also a key
element to performance enhancement. The higher col-
lector dose just described, as well as modifications to the
subcollector layer and the elimination of shallow-trench
to reduce the resistive path length [18] may have an
impact. Because only majority carriers are involved, and
no significant degradation has been identified relating to
0
0.005
0.01
0.015
0.02
0 100 200 300 400
Peak fT (GHz))
d (b) excess collector and base current from avalanche in the
G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410 401
the collector–base diode, these modifications are not
expected to result in significant reliability implications.
3. Silicon bipolar transistor reliability
The various bipolar degradation signatures and
mechanisms are described next. We relate the mecha-
nisms to the device scaling just described. The degra-
dation mechanisms are organized into the regions of
device operation as shown in Table 1.
3.1. Reverse emitter–base
Reverse emitter–base stress, with the emitter–base
diode in reverse bias and the collector terminal either
open or shorted to the base, not only measures the
degradation of the device in reverse operation, but also
characterizes the emitter–base passivation for suscepti-
bility to interface trap generation [19,20]. Poor degra-
dation is typically a result of a combination of the
electric field at the emitter–base perimeter as well as the
quality of the passivation to silicon interface. Worsened
degradation from this stress is not expected to accom-
pany performance improvements in the device since
vertical scaling fundamentally does not involve changes
in these structural elements. Further, little or no degra-
dation is expected from the reverse bias condition in
actual operation on most circuits. This is due to the fact
that the energy required to form the interface traps is not
likely to occur in typical ECL circuits, or in other con-
figurations where this is more likely, the excursions may
be limited by certain circuit topologies [21].
To illustrate that little change in degradation is ex-
pected from reverse operation for high performance
Table 1
Bipolar reliability overview and performance trend related issues
Operation
region
Stress condition Typical
signature
Mechanism
VBE VBC
Reverse ) Low bias IBnon-ideality
E–B oxide
interface
states
Saturation + + Low bias IBnon-ideality
E–B oxide
interface
states
Forward
mode
+ ) Low bias IBnon-ideality,
parallel IBshift, RE shift,
self-heating
E–B oxide
interface
states, poly–
sc interface
oxide, metal
migration
Avalanche + ) Low bias IBnon-ideality,
self-heating
E–B oxide
interface
states
devices in comparison to lesser performance devices,
we show in Fig. 3 the degradation in base and collec-
tor current for very different emitter structures, with the
arsenic emitter of an fT ¼ 120 GHz device and the
phosphorous emitter of the fT ¼ 200 GHz device. No
degradation is observed at a reverse bias of 2 V in both
cases, confirming the earlier point that degradation in
this operating regime does not present a concern for
continued scaling to higher performance.
3.2. Saturation
Operation of the device in saturation, where both the
emitter–base and collector–base junctions are forward
biased, is similarly of limited concern in device scaling.
Again, the degradation occurs in the emitter–base pas-
sivation and occurs principally with significant minority
carrier injection from the collector–base diode into the
neutral base [22]. Like the reverse bias condition, the
properties of the device structure relevant to this deg-
radation change little with increased performance de-
vices. Also, circuits designed to take advantage of high
performance devices are not driven into saturation as
this affects the propagation delay of the output signal.
Furthermore, the need to approach saturation during
operation is alleviated by the smaller voltage swings seen
in these circuits. Thus this presents little concern for
scaling.
3.3. Forward mode
Forward bias operation is a concern for device scal-
ing, since the operation current density increases be-
tween generations, of the order shown in Fig. 2. Several
Stress condition
in typical
circuit?
Trend with
scaling
Remedy Refs.
Limited Neutral Limit reverse
voltage in design
[19–21]
Limited Neutral Limit saturation
voltage in design
[22]
Yes Worse Min. interface
oxide, min. e-
fields at surface,
narrow WE
[22–30]
Yes Worse Design robust to
IB variations,
limit voltage in
design
[33–35]
1.E-15
1.E-13
1.E-11
1.E-09
1.E-07
1.E-05
1.E-03
0.2 0.4 0.6 0.8
VBE (V)
IC initialIC finalIB initialIB final
1.E-15
1.E-13
1.E-11
1.E-09
1.E-07
1.E-05
1.E-03
0.2 0.4 0.6 0.8
VBE (V)
Ib, I
c (A
)
IC initialIC finalIB initialIB final
fT = 120 GHz fT = 200 GHz
Fig. 3. Reverse emitter–base junction degradation for 120 and 200 GHz devices with As and P emitters, respectively. Stress condition is
VBE ¼ �2 V, collector open, for 104 s and TAMB ¼ 30 �C.
402 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
degradation signatures are relevant and will be discussed
here.
Low-bias base current non-ideality caused by for-
ward mode stress has been observed for some devices
[22–25] and not observed for other devices [26]. Again, it
has been proposed that the forward operation may cause
the creation of surface states at the emitter–base
perimeter passivation silicon-oxide interface. Because a
carrier traversing the emitter–base space charge region
does not have sufficient energy to create such a state, it
has been proposed that such energy may be created
through Auger recombination [27]. Because such deg-
radation has not been observed in some devices, it ap-
pears that the degradation is sensitive to the specific
device structure. In particular the low-bias base current
non-ideality appears to be less common in SiGe devices.
The epitaxy process may have a significant advantage in
containing the boron to a deeper region of the device
compared to an implanted base process, thus reducing
the electric fields at the emitter–base interface and
reducing many of the degradation mechanisms associ-
ated with the emitter–base passivation [28]. The trend
toward reducing the boron diffusion in high-perfor-
mance devices maintains the low electric fields at the
passivation interface, and thus is expected to maintain
low degradation in the device.
A second signature is a parallel shift in the base
current when plotted on the logðIBÞ versus VBE plot.
Once again, device structure appears to affect how a
device behaves with regard to this signature. Both an
increase in IB [25,26] and a decrease in IB [23–26,29] have
been observed. In both cases, the change has been
attributed to the modification of the very thin oxide
layer between the single-crystal to polycrystalline emitter
regions of the device with high current stress. In the case
of increasing IB, it has been hypothesized that the gen-
eration or de-passivation of traps with high current
density enhances the trap-assisted tunneling through the
oxide material, causing the increase in IB [25,26]. This
has been reported to accompany a reduction in emitter
resistance RE due to the same mechanism [26]. In the
case of a parallel reduction in IB, again the interfacial
oxide is implicated, with a hypothesis that the hydrogen
passivation of the interface states cause reduced hole
recombination velocity at this interface [23,24,26,29].
Although the mechanism has been reported to pro-
ject a relatively small shift in base current over the life-
time for a device with fT ¼ 120 GHz, a strong current
sensitivity was also reported [25,26], suggesting this
interfacial film may pose significant degradation issues
with higher performance devices. We show here a similar
current density sensitivity in the degradation of the
fT ¼ 200 GHz device. Fig. 4 shows the time evolution of
beta degradation for an fT ¼ 200 GHz device, illustrat-
ing the difference in degradation between two samples.
The stress of these samples differ in both stress current
density JSTRESS and junction temperature TJ. In Fig. 5,
we show the degradation of a variety of devices with
different junction temperatures, but with constant
JSTRESS. Evident from these two plots is that JSTRESS has
a greater impact accelerating the device degradation
than TJ. As described in both [25] and [26], the 2· cur-
rent stress has the effect of accelerating the device deg-
radation time by a factor of more than 100·. In the
referenced work, the temperature played a small role in
acceleration, as is also demonstrated in Fig. 5 in this
work. With this assumption, the 2X stress degradation
in Fig. 4 may be considered compressed on the time
1.E-111.E-101.E-091.E-081.E-071.E-061.E-051.E-041.E-031.E-02
0.4 0.5 0.6 0.7 0.8 0.9 1VBE
IB, I
C (A
)
-50%-25%
0%25%50%
0.6 0.7 0.8 0.9 1VBE
∆IB
∆IC∆β
Fig. 6. Gummel plot of 1· stress after 1140 h at 160 �Cambient. Solid lines are the pre-stress currents and dotted lines
after stress. Inset is the change in post-stress current compared
to the pre-stress current. Measurement is at 25 �C ambient.
-10-8-6
-4-2024
68
10
1 10 100 1000 10000Stress time (Hrs)
% B
eta
shift
AE=0.12x2.5µm2
Jstress=11mA/µm2 (1x IC at peak fT)Tamb=160C, Tj =192C
VCE=1.5VAE=0.12x2.0µm2
Jstress=22mA/µm2 (2x IC at peak fT)Tamb=180C, Tj=239C
Fig. 4. Current gain shift under forward-mode stress, com-
paring stress current of 1· and 2· of IC at peak fT. Measure-
ment and stress condition are the same.
-10
0
10
1 10 100 1000Stress time (Hrs)
% B
eta
shift
AE=0.12x2.0µm2, TJ=239C, TAMB=180CAE=0.12x0.64µm2, TJ=214C, TAMB=180CAE=0.12x2.0µm2, TJ=159C, TAMB=100C
Fig. 5. Current gain shift under forward current stress com-
paring two different size devices and two ambient temperatures,
all with the same stress current density of 2· peak fT JC ¼ 22
mA/lm2. Measurement is at 1· peak fT JC and at the stress
temperature.
G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410 403
scale compared to the actual expected degradation at 1·operation. By shifting the 2· degradation characteristics
on the log time scale to approximately 100· later in time,
one may approximate the expected degradation to
approximately 3 · 104 h. It is worth noting that base
current degradation is gradual, indicating a lack of any
catastrophic failure.
Fig. 6 shows the parameter shift bias dependence at
the end of the stress, for the 1· stress device of Fig. 5
now measured at 25 �C. Note that the fractional shift in
base current for the 200 GHz device is similar across the
entire low-VBE bias range, diminishing above VBE ¼ 0:85V where it is increasingly dominated by emitter resis-
tance. The small collector current shift (and similar
amount of base current shift) is likely a result of a slight
(i.e. 1–2 �C) ambient temperature condition difference
present during the module test outside of the stress
chamber.
For the forward current degradation just described,
we compare the direction of parameter shift and the time
dependence to the referenced work. In contrast to Figs.
4 and 5 of this work, where the current gain monoton-
ically increases with time, Refs. [25,26] report an initial
decrease in current gain followed by an increase occur-
ring (either measured or projected) between 103–105 h of
operation. The stress on the 200 GHz device here shows
a result more similar to the effect reported in [23], where
a decrease in current gain is not observed, and current
gain only increases. Each report ascribes the shift in
current gain principally to a base current shift, as we
have shown in Fig. 6. Each report also indicates an
acceleration resulting principally from high current
density (rather than temperature) as we also have ob-
served here.
As described above, the commonly understood
mechanism responsible for this shift in base current is
the generation or passivation of the traps at the emitter
single-crystal to polysilicon interface. The increase of
base current is a result of the generation of traps, and
the decrease of base current is a result of the passivation
of traps. In the reports of [25,26] with an increase fol-
lowed by a decrease in IB, competing mechanisms are
suggested with trap generation followed by passivation.
The passivation process is described in [23] and is sup-
ported with further evidence in [25]. In this model,
hydrogen bound to metal and polysilicon grains is re-
leased by the high electron flux. These hydrogen atoms
migrate to the region of the device where minority car-
rier (hole) diffusion length affects the base current in the
device. With the hydrogenation of silicon dangling
bonds within the polysilicon–single-crystal interface re-
gion, the recombination velocity and thus the base
current is reduced. The absence of the initial increase
of base current observed in this work implies that no
404 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
significant trap generation occurred in the early phase of
the stress on the 200 GHz devices. Instead, it indicates
that the parameter shift was dominated by the passiv-
ation of the existing traps.
The time progression shown in Figs. 4 and 5 are
measured at high current densities, where device resis-
tances cause the parameter shifts to be less obvious
compared to the low-bias region of operation. This bias
region is more representative of device operation con-
ditions, but the lower bias region provides more insight
into the origin of the parameter shift and a better
comparison to published work. In Fig. 7, we compare
the time progression of degradation measured at low
bias for the device described here and three published
devices, each stressed at similar current densities in the
vicinity of 10 mA/lm2. Note that even though the
comparison is at similar current densities, the current
density for each device to attain its peak fT performance
may be much different between these devices. Thus the
degradation shown here is not a comparison of degra-
dation at operating conditions for all devices.
Note again the difference in sign of degradation be-
tween the different devices. This suggests that the trap
generation mechanism, indicated by the reduction of
beta (i.e. an increase in IB) is minimized in the device
design and fabrication of the 200 GHz device. Also by
minimizing the presence of hydrogen and optimizing the
interface properties, it has been possible to minimize the
magnitude of increasing beta in the 200 GHz device
compared to other devices shown. The comparison
indicates that optimized process conditions and device
design may achieve reduced parameter degradation
required for the higher current density operation.
-15
0
15
0.01 0.1 1 10 100 1000Sress time (Hrs)
% B
eta
or -%
IB s
hift
[23] 150C 10mA/µm2 0.5V[25] 125C 11mA/µm2 0.8V[26] 140C 8.5mA/µm2 0.7VThis work 160C 11mA/µm2 0.7V
Fig. 7. Comparison of beta shifts across published work mea-
sured at low bias, each stressed in the vicinity of the current
density used in this work. In cases [23,26], �IB shifts are shown
and IC shift is indicated to be small. The legend lists the stress
and measurement ambient temperature and stress current
density, with measurement bias point.
The degradation so far described, and which the de-
signer to some degree should expect, is regarding base
current and emitter resistance. For ECL and CML cir-
cuits, which are the most common high-speed circuit
configuration, the variation in base current in itself is
not much of a concern since it does not alter the oper-
ating points. In other configurations that are more
sensitive to base current degradation, an understanding
and modeling of the degradation will enable the designer
to avoid the degradation through the device bias, or to
design the circuit for tolerance to the degradation. The
reduction in RE, while it affects operating point, gain,
and noise margin, must be once again included in the
device model and thus worst case scenarios can be
evaluated during the design process.
What remains to be discussed are the electromigra-
tion concerns in delivering the current to the device and
managing the device self-heating. These topics are clo-
sely tied, since the self-heating negatively impacts the
metal migration, and the metal migration in turn limits
the useable current through the device. An approach to
view this is through the simultaneous solution of the
current–temperature relations for both maximum cur-
rent allowed through a metal configuration and device
self-heating, conservatively assuming the metal temper-
ature is the same as the junction temperature. This ap-
proach is applied to a 200 GHz device as shown in Fig.
8, and provides a boundary for the steady-state bias of
the device to maintain reliable operation. The goal is to
enlarge the useable operation range as much as possible
through device design and reliability analysis. Modifi-
cations to device thermal resistance or metal current
carrying capability from layout or material system
0
5
10
15
20
0 0.5 1 1.5 2 2.5 3
VCE [V]
JC [m
A/µ m
2 ]
1 um2.5 um5 um10 um
Region of safe operation for electromigration
Region of concern for electromigration
Boundaries
Fig. 8. Boundary conditions by emitter stripe length, obtained
for each bias point through simultaneous consideration of both
maximum current through a metal line and device self-heating.
Curves represent 200 GHz device with a certain layout.
G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410 405
improvements (such as the improvements from copper
interconnect) strongly affect this boundary.
Approaching the layout from the view of improving
electromigration, the target is to obtain constant or re-
duced collector and emitter current per length of emitter
with increasing peak fT current density JCP. Employing
the technique known as constant current scaling requires
reducing the emitter stripe width WE in proportion to
1/JCP, such that the current through the metal system is
constant and the electromigration issues are the same
between generations.
Thermal resistance may also be maintained with this
technique. Because designers typically choose the device
size to achieve a certain peak fT current, it is illustrative
to compare the self-heating effect of different scaling
methods from a fixed current perspective. To maintain
fixed current, the emitter area between device genera-
tions is proportional to 1/JCP. With a peak fT current of
2 mA and utilizing a structure-based analytic thermal
model as described in [30], we compare in Fig. 9 the
approximate self-heating of devices from fT ¼ 47 to 350
GHz. The emitter scaling methods compared are to (1)
fix WE and vary the length with 1=JCP, or (2) fix the
emitter length and vary WE with 1/JCP. The case with
fixed WE results in more concentrated power dissipation
as well as substantial deep-trench enclosure area reduc-
tion, and consequently a large increase in thermal
resistance and self-heating for the higher current density
devices. In contrast, the emitter width scaled as 1=JCP
maintains a near constant device perimeter and main-
tains only a slight decrease in the deep trench enclosed
area. Thus the self-heating is managed with an appro-
priate reduction in stripe width. With the 0.20 lm device
as a baseline, we find that WE of 0.12 and 0.075 lm for
200 and 350 GHz devices, respectively, are required for
the device self-heating properties shown in Fig. 9.
Thus we find that the two most fundamental con-
cerns regarding increasing current density, e.g., provid-
ing current to the device through the metal system and
0
5
10
15
20
25
30
0 100 200 300 400Peak fT [GHz]
Dev
ice
self-
heat
ing
∆TJ
[K] We=0.2um fixed
We scaled with 1/Jc
Fig. 9. Self-heating versus fT for length compared to width
scaling.
the self-heating of the device, critically depend upon the
ability to spread the current over the length of the device
and maintain a near constant current per unit length of
the device. With emitter width reduction in proportion
to inverse peak fT current density, we expect that the
device may be scaled for reliable operation beyond peak
fT values of 350 GHz and JCP of 20 mA/lm2.
3.4. Avalanche
It is well known that lower breakdowns accom-
pany higher fT performance [31]. With the collector–
base breakdown BVCBO and off-state collector–emitter
breakdown BVCER in the range of 5–6 V for devices with
fT of 350 GHz, designers may find considerable latitude
in voltages that they may apply to the device. However,
the on-state breakdown is considerably more complex
since the carriers that traverse the collector–base space-
charge layer generate significant additional carriers
through a process of avalanche multiplication. To
illustrate the magnitude and the scaling, we show in Fig.
10 the avalanche current multiplication factor M-1 ver-
sus VCB for various device designs. Note that the M-1
value for fixed VCB is plotted in Fig. 2b and shows an
approximate linear relation to device peak fT. Looked at
with an eye to maintaining a constant avalanche multi-
plication M-1, it is interesting to note that the voltage
reaches a near minimum (e.g. at approximately VCB ¼1:4 V for M-1¼ 0.01). This has implications on the
voltage limits with continued device scaling as will be
apparent with the discussion that follows.
The bias voltage at which the avalanche current in
the device becomes significant (i.e. the same as the base
current) is approximately BVCEO, which is reported to be
1.4 V for the 350 GHz device. Many designers have
viewed the value of BVCEO as a limit to avoid due to the
additional complications it represents. With ever-higher
Fig. 10. Avalanche multiplication for SiGe HBTs with increas-
ing fT performance.
0
50
100
150
200
250
0.0001 0.001 0.01Ic (A)
fT (G
Hz)
1.5
2.0
2.5
VCB
Fig. 11. fT performance variation with increasing avalanche, at
VCB values listed in legend. VCE values range from approxi-
mately 2.4–3.4 V at peak fT, compared to BVCEO ¼ 1:8 V.
Emitter size is 0.12· 1.0 lm2.
406 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
fT and ever-lower BVCEO devices, the design boundary is
now required to go beyond BVCEO, and the operation
and degradation mechanisms of the device need to be
better understood in this operation regime.
First, we demonstrate that performance does not
significantly degrade due to avalanche. Shown in Fig. 11
are fT vs. IC characteristics with various VCB bias points
for the fT ¼ 200 GHz SiGe HBT. It is apparent that the
avalanche conditions have little effect on the AC per-
formance of the device. The small fT reduction might be
ascribed to the increasing junction temperature with
increasing VCB. Circuit results also demonstrate the
performance capability of the SiGe HBT operating over
BVCEO. For instance, an electro-absorption modulator
driver operating to 48 Gb/s with an fT ¼ 120 GHz SiGe
HBT has been reported, where a device with BVCEO of
1.9 V drives an output of 3.5 V peak–peak, resulting in a
device which sustains VCE of approximately 4.5V [32].
This again demonstrates the lack of performance loss
operating well over BVCEO.
With the understanding that BVCEO is not a hard bias
limit and high-speed operation can take place well above
BVCEO, we wish to explore what limits the maximum
voltage, and how this varies with device scaling. The
answer is complex, and in part depends on the applica-
tion. First, voltage driven self-heating needs to be con-
sidered. This is mostly independent of avalanche, but
imposes a practical limit to the current–voltage product.
Also, increased avalanche impacts noise, and device bias
voltage must be kept low to minimize avalanche and
maintain low noise. Fortunately, the circuit designer
typically has flexibility in choosing a collector through
the previously described mask-selectable collector op-
tion in the device layout. Since low collector current is
typically chosen as a bias point for the lowest noise
conditions, the higher collector concentrations are not a
requirement for the operation of the low noise device
since the peak fT is generally not utilized. An alternate
lower concentration collector, typically available on the
same chip, provides an alternative to the designer with
little loss of performance at lower bias conditions but at
an incrementally higher fabrication cost.
More fundamental are the effects of avalanche on
device instability and on device degradation. These ef-
fects are discussed in [33] and [34,35], respectively, and
are described here in the context of device scaling.
Instability results from the avalanche current flowing
out of the base, causing a reversal in base current and a
potential drop across the intrinsic base. A significant
potential drop across the intrinsic base results from a
large amount of avalanche current, and can cause a
device operation discontinuity, or ‘‘pinch-in’’ of the
device operation into the center of the device. While
degradation has not been associated with this phenom-
enon, it appears to represent a device operation dis-
continuity that should be avoided. Shown in Fig. 12 is a
diagram of three regions of device operation: low,
moderate, and high avalanche, with varying VCB. The
third region of operation illustrates the just described
‘‘pinch-in’’ condition, where the avalanche causes the
current to focus into the center of the device. The pinch-
in voltage is both a function of avalanche and of base
resistance. In order to compare different device designs,
one may compare the pinch-in voltage according to the
technique described in [33]. This pinch-in is observed as
the voltage at which VBE drops abruptly on a common
base, forced IE measurement, as shown in Fig. 13 for
various device designs. This set of characteristics show
that the lower performance devices do not demonstrate
pinch-in below 5 V. The remaining devices, at fT ¼ 120,
195, and 210 GHz, show little trend in the pinch-in
voltage with increasing avalanche and fT. This is because
of a dramatic decrease in base resistance from the 120
GHz device to the 195 and 210 GHz devices. Thus the
increased avalanche does not fundamentally reduce the
instability voltage, since base resistance scaling, which is
required for fMAX optimization in device design, com-
pensates the negative effect and maintains an acceptable
pinch-in voltage.
Increased avalanche in high-speed devices under
forward current operation also affects device degrada-
tion as described in [34,35]. Hot carriers generated in the
collector–base space-charge region are said to traverse
to and damage the perimeter emitter–base dielectric and
the base–collector dielectric. The location of the hot
carrier generation and the damage are shown in Fig. 14.
The degradation signature of most concern appears to
relate to the damage created in the emitter–base peri-
meter dielectric, and is observed in the base current
degradation at low bias, in a similar fashion described
earlier for degradation from different operating regimes.
An example of this degradation is shown in Fig. 15.
h+ h+
EB(a) EB
h+h+
(b)
EB
h+h+
(c)
Fig. 12. Avalanche effect on device operation: (a) low VCB; (b) moderate avalanche and (c) ‘‘pinch-in’’ resulting from high voltage drop
across base resistance.
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0 1 2 3 4 5Vcb (V)
Vbe
(V) a
t Ie=
1e-5
210
195
120
90
50
Fig. 13. Measured VBE vs. VCB for fixed IE ¼ �10�5 A. Pinch-in
is apparent in the voltage drop across the base resistance,
exhibited as a sudden drop in VBE. The legend lists the peak fTvalue for the device in GHz.
Fig. 14. Hot carrier generation from avalanche and trapping in
dielectric layers.
1.E-13
1.E-11
1.E-09
1.E-07
1.E-05
1.E-03
0.1 0.3 0.5 0.7 0.9
I C, I
B [A
]
T0
T1
VBE [V]
Fig. 15. Avalanche generated base current degradation for 200
GHz SiGe HBT. Emitter area is 0.12· 2.0 lm2. Stress is
VCB ¼ 2:5 V, IE ¼ 1:92 mA. T0 is the initial device characteristic
and T1 is after 8.3 h under the specified stress.
G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410 407
Device degradation on the 200 GHz SiGe HBT in
avalanche (i.e. operation above BVCEO) has been re-
ported and a model presented in [35]. The key data upon
which the model is built is shown in Fig. 16. As observed
in this plot, degradation exhibits a power-law function
of avalanche charge (which is principally a function of
collector–base voltage VCB and collector current) and an
exponential function of VCB. The unit perimeter base
current degradation DIB=PE is a function of the accu-
mulated avalanche charge normalized to device perim-
eter QB=PE, as exhibited in the following empirical
relation:
DIB=PE ¼ A expð�V0=VCBÞðQB=PEÞn
where A, V0 and n are fitting parameters for a given VBE
bias. The model does not account for possible saturation
1E-3 0.01 0.1 1
1E-3
0.01
0.1
∆I B/
PE,
(nA/
µm)
QB/PE, (C/µm)
VCBstr=3.0 V VCBstr=2.5 V VCBstr=1.5 V
Fig. 16. Avalanche base current degradation normalized to
device perimeter versus avalanche charge normalized to device
perimeter, for different collector–base voltage VCB.
408 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
effects as observed in Fig. 16 above 0.1 C/lm, and so it is
conservative. Further investigation is required to
understand the saturation of the degradation with
increasing avalanche charge, and the degradation recov-
ery during forward non-avalanche bias.
Conservatively neglecting the saturation and recov-
ery effects, the model predicts degradation that is
acceptable for most applications with typical voltage
operation above BVCEO. For instance, in a digital
application with a load line on the 200 GHz device
traversing VCE from 2.0 to 2.5 V, assuming the operation
in the avalanche regime is approximately 1% of the total
operation time, degradation after 100 K power-on hours
is expected to be approximately 5% in IB at a 10% peak
fT bias point [35]. Again, the allowable degradation is
highly application dependent, and many applications are
likely to tolerate much higher degradation (e.g. see [36]
which reports X–Ku band 1W HBT power MMIC,
where 40% DC gain loss at the operating point causes a
relatively small <0.5 dB power loss), and thus be able to
operate with higher VCB, with little loss of performance.
Again, we address how scaling is expected to influ-
ence the degradation just described. At a given collec-
tor–base voltage, the degradation is expected to become
worse between generations according to the M-1 in-
creases as shown in Fig. 2b. However, because degra-
dation is a sensitive function of voltage due to the
exponential collector voltage dependence of both ava-
lanche carrier generation and of interface state creation,
the negative effects of avalanche can easily be reduced by
a small decrease in operating voltage.
In summary this avalanche-induced hot carrier deg-
radation appears to introduce a degradation mechanism
not clearly observed in lower performance devices, due
to the higher collector concentration and higher ava-
lanche current. Compared to the pinch-in voltage, this
degradation mechanism appears to take place at a lower
voltage, at approximately VCE ¼ 3 V. Because the deg-
radation only takes place when the device experiences
higher values of VCE, and because it is seen in base
current and not in the more critical collector current, it is
expected that a limited number of circuit configurations
will be affected. Those that are affected may employ
robust circuit techniques, such as common base config-
urations, to substantially reduce the performance impact
to such device parameter shifts.
4. Conclusions
We find that with the appropriate attention to device
design as well as some attention in the circuit design in
less common high and low voltage regimes, the trends in
SiGe HBT design to increase performance to over 350
GHz impose no fundamental limitations to fabricating
reliable circuits. Device design needs to focus on
reducing stripe widths for electromigration and self-
heating, and the emitter polysilicon to single-crystal
interface layer needs to be minimized to reduce base
current and emitter resistance shifts. Further, the
reduction in base resistance, which is already a key de-
sign goal in advanced HBTs, is needed to maintain a
high pinch-in voltage. Avalanche introduces a degrada-
tion mechanism, to be taken into account when
designing at the highest voltages. In general, it should be
possible to continue scaling beyond today’s 200 GHz
devices and the demonstrated 350 GHz devices, with
straightforward extension of the mechanisms and con-
cerns outlined in this paper.
Acknowledgements
The authors would like to acknowledge partial sup-
port of this work by DARPA under SPAWAR contract
number N66001-02-C-8014. The authors also would like
to thank Seshadri Subbanna, Peter Cottrell and David
Greenberg for their helpful discussions, and Dale Jadus
and Michael Longstreet for their help in the device test.
References
[1] Rieh J-S, Jagannathan B, Chen H, Schonenberg KT,
Angell D, Chinthakindi A, et al. SiGe HBTs with cut-off
frequency of 350 GHz. In: Proceedings of the International
Electron Devices Meeting; December 2002. p. 771–4.
[2] Greenberg DR, Jagannathan B, Sweeney S, Freeman G,
Ahlgren D. Noise performance of a low-base resistance 200
G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410 409
GHz SiGe technology. In: Proceedings of the International
Electron Devices Meeting; December 2002. p. 787–90.
[3] Yamashita Y, Endoh A, Shinohara K, Hikosaka K,
Matsui T, Hiyamizu S, et al. Pseudomorphic In0:52-
Al0:48As/In0:7Ga0:3As HEMTs with an ultrahigh fT of 562
GHz. IEEE Electron Device Lett 2002;23(10):573–5.
[4] Hafez W, Lai J-W, Feng M. InP/InGaAs SHBTs with 75
nm collector and fT > 500 GHz. IEE Electron Lett
2003;39(20):1475–6.
[5] Terpstra D, Deboer W, Slotboom JW. High-performance
Si–SiGe HBTs SiGe-technology development in ESPRIT
project 8001 TIBIA: an overview. Solid State Electron
1997:1493–502.
[6] Madelung O, editor. Data in science and technology:
semiconductors––group IV elements and III–V compound.
Berlin: Springer-Verlag; 1991.
[7] Henderson T. Model for degradation of GaAs/AlGaAs
HBTs under temperature and current stress. In: Proceed-
ings of IEEE International Electronic Devices Meeting;
1997. p. 811–4.
[8] Sugahara S, Nagano J, Nittono T, Ogawa K. Improved
reliability of AlGaAs/GaAs heterojunction bipolar tran-
sistors with a strain-relaxed base. In: Proceedings of the
GaAs IC Symposium; 1993. p. 115–8.
[9] Kiziloglu K, Thomas S, Williams F, Paine B. Reliability
and failure criteria for AlInAs/GaInAs/InP HBTs. In:
Proceedings of the Indium Phosphide and Related Mate-
rials Conference; 2000. p. 294–7.
[10] Jagannathan B, Meghelli M, Chan K, Rieh J-S, Schonen-
berg K, Ahlgren D, et al. 3.9 ps SiGe HBT ring-oscillator
and transistor design for minimum gate delay. IEEE
Electron Device Lett 2003;24(5):324–6.
[11] Jagannathan B, Khater M, Pagette F, Rieh J-S, Angell D,
Chen H, et al. Self-aligned SiGe NPN transistors with
285 GHz fMAX and 207 GHz fT in a manufacturable
technology. IEEE Electron Device Lett 2002;23(5):258–
60.
[12] Kiyota Y, Hashimoto T, Udo T, Kodama A, Shimamoto
H, Hayami R, et al. 190 GHz fT, 130 GHz fMAX SiGe
HBTs with heavily doped base formed by HCL-free
selective epitaxy. In: Proceedings of the Bipolar/BiCMOS
Circuits and Technology Meeting; 2002. p. 139–42.
[13] Schuegraf K, Racanelli M, Kalburge A, et al. 0.18 lm SiGe
BiCMOS technology for wireless and 40 Gb/s com-
munication products. In: Proceedings of the Bipolar/
BiCMOS Circuits and Technology Meeting; 2001.
p. 147–50.
[14] Joseph A, Coolbaugh D, Zierak M, et al. A 0.18 lmBiCMOS technology featuring 120/100 GHz (fT=fMAX)
HBT and ASIC-compatible CMOS using copper intercon-
nect. In: Proceedings of the Bipolar/BiCMOS Circuits and
Technology Meeting; 2001. p. 143–6.
[15] Kondo M, Kobayashi T, Tamaki Y. Hetero-emitter-like
characteristics of phosphorus doped polysilicon emitter
transistors––Part I: band structure in the polysilicon
emitter obtained from electrical measurements. IEEE
Trans Electron Devices 1995;42(3):419–26.
[16] Jouan S, Planche R, Baudry H, Ribot P, Chroboczek JA,
Dutartre D, et al. A high-speed low 1/f noise SiGe HBT
technology using epitaxially-aligned polysilicon emitters.
IEEE Trans Electron Devices 1999;(July):1525–31.
[17] Washio K, Ohue E, Oda K, Tanabe M, Shimamoto H,
Onai T. A Selective-epitaxial SiGe HBT with SMI elec-
trodes featuring 9.3-ps ECL-gate delay. In: Proceedings of
the IEEE International Electron Devices Meeting; 1997. p.
795–8.
[18] Heinemann B, Rucker H, Barth R, et al. Novel collector
design for high-speed SiGe:C HBTs. In: Proceedigns of the
International Electron Devices Meeting; December 2002.
p. 775–8.
[19] Tang DD, Hackbarth E. Junction degradation in bipolar
transistors and the reliability imposed constraints to
scaling and design. IEEE Trans Electron Devices 1988;
35(12):2101–7.
[20] Neugroschel A, Sah C-T, Carroll M. Degradation of
bipolar transistor current gain by hot holes during reverse
emitter–base bias stress. IEEE Trans Electron Devices
1996;43(8):1286–90.
[21] Kizilyalli IC, McAndrew CC. Improved circuit technique
to reduce hfe degradation in bipolar output drivers. IEEE
Trans Electron Devices 1995;42(3):573–4.
[22] Tang DD, Hackbarth E, Chen TC. On the very high-
current degradations on Si n–p–n transistors. IEEE Trans
Electron Devices 1990;37(7):1698–706.
[23] Carroll MS, Neugroschel A, Sah C-T. Degradation of
silicon bipolar unction transistors at high forward current
densities. IEEE Trans Electron Devices 1997;44(1):
110–7.
[24] Chen TC, Kaya C, Ketchen MB, Ning TH. Reliability
analysis of self-aligned bipolar transistor under forward
active current stress. In: Proceedings of the International
Electron Devices Meeting; 1986. p. 650–3.
[25] Hoffmann K, Bruegmann G, Seck M. Impact of inter-
metal dielectric on the reliability of SiGe NPN HBTs after
high temperature electrical operation. In: IEEE Topical
Meeting on Si Monolithic Integrated Circuits in RF
Systems; April 2003. p. 126–9.
[26] Rieh J-S, Watson K, Guarin F, Yang Z, Wang P-C, Joseph
A, et al. Reliability of high-speed SiGe heterojunction
bipolar transistors under very high forward current den-
sity. IEEE Trans Device Mater Reliab 2003;3(2):31–8.
[27] Wachnik RA, Bucelot TJ, Li GP. Degradation of bipolar
transistors under high current stress at 300 K. J Appl Phys
1988;63(9):4734–40.
[28] Dunn J, Harame D, St Onge S, Joseph A, Feilchenfeld N,
Watson K, et al. Trends in SiGe BiCMOS integration and
reliability. Proc IRPS 2000;1:237–42.
[29] Zhao J, Li GP, Liao KY, Chin M-R, Sun JY-C, La Duca
A. Resolving the mechanisms of current gain increase
under forward current stress in poly emitter n–p–n
transistors. IEEE Electron Device Lett 1993;14(5):252–5.
[30] Rieh J-S, Johnson J, Furkay S, Greenberg D, Freeman G,
Subbanna S. Structural dependence of the thermal resis-
tance of trench-isolated bipolar transistors. In: Proceedings
of the IEEE Bipolar/BiCMOS Circuits and Technology
Meeting; 2002. p. 100–3.
[31] Johnson EO. Physical limitations on frequency and
power parameters of transistors. RCA Rev 1965;26:163–
77.
[32] Freeman G, Meghelli M, Kwark Y, et al. 40-Gb/s circuits
built from a fT ¼ 120 GHz SiGe technology. IEEE J Solid-
State Circuits 2002;37(9):1106–14.
410 G. Freeman et al. / Microelectronics Reliability 44 (2004) 397–410
[33] Rickelt M, Rein HM, Rose E. Influence of impact-
ionization-induced instabilities on the maximum usable
output voltage of Si-bipolar transistors. IEEE Trans
Electron Devices 2001;48(4):774–83.
[34] Zhang G, Cressler JD, Niu G and Joseph A. A new �mixed-
mode’ base current degradation mechanism in bipolar
transistors. In: IEEE Bipolar Circuit and Technology
Meeting; 2002. p. 32–5.
[35] Yang Z, Guarin F, Hostetter E, Freeman G. Avalanche
current induced hot carrier degradation in 200 GHz SiGe
heterojunction bipolar transistors. In: Proceedings of the
International Reliability Physics Symposium; 2003.
[36] Gupta A, Ali F, Dawson D, Smith P. Degradation of an
X-Ku Band GaAs/AlGaAs Power HBT MMIC Under
RF Stress. IEE Microwave Guided Wave Lett 1996;6(1):
43–5.