Germanium profile design options for SiGe LEC HBTs Michael Schroter a, * , Hung Tran a , Wolfgang Kraus b a Department of Electron devices and Integrated Circuits, University of Technology Dresden, Mommsenstr. 13, D-01062 Dresden, Germany b Atmel Germany, Theresienstr. 2, D-7100 Heilbronn, Germany Received 1 June 2003; received in revised form 1 September 2003 The review of this paper was arranged by Prof. S. Cristoloveanu Abstract Silicon–Germanium (SiGe) heterojunction bipolar transistors (HBTs) with a low emitter doping concentration (LEC) are investigated with respect to their electrical characteristics in dependence of the Germanium profile shape. The study is based on one-dimensional (1D) device simulation using a realistic doping and Ge profile as baseline. While keeping the doping profile unchanged the Ge profile is modified to evaluate its impact on major electrical figure of merits such as transit frequency, ideality factor, early voltage, noise figure, as well as on process control monitors (PCMs) such as internal base sheet resistance and area specific depletion capacitances. The variations in electrical characteristics and PCMs are briefly explained on a theoretical basis with regard to the impact on compact and sta- tistical modeling. Ó 2003 Elsevier Ltd. All rights reserved. 1. Introduction SiGe/SiGeC HBTs are becoming available in many high-performance BiCMOS processes thus entering mainstream applications. Presently, two types of vertical device designs are being pursued. The first and by far most popular of those contains a ‘‘conventional’’ base– emitter doping profile with a moderately doped base and a highly-doped emitter as well as a triangular-shaped Ge profile. The latter creates an aiding drift field for the electrons traversing the base, while the bandgap differ- ence at the BE junction is fairly small. The second type is similar to III–V HBTs and contains a fairly high base doping concentration and a low emitter concentration (LEC). Sufficient current gain is realized here by a sig- nificant bandgap difference at the BE junction that is created through a significant step of the Ge contents. Although LEC HBTs are––from their basic con- cept––in many respects electrically advantageous over conventional designs, their performance seems to be lagging behind. Reasons for this are that (i) aggressive vertical profile optimization as well as lateral scaling through advanced CMOS lithography have not been applied yet to LEC HBTs and (ii) just a single foundry so far has made such a process manufacturable [1]. The present process version contains a box-shaped base doping and Ge profile and achieves a peak transit fre- quency f T of about 45 GHz [1]. In this paper, therefore, possible improvements in the vertical device design for the next generation (f T 90 GHz) of LEC HBTs are investigated. An important goal is also to provide a feeling for the sensitivity of the electrical transistor per- formance with respect to the Ge profile shape. This is of high interest both for compact transistor model devel- opment (such as HICUM [8]), which requires to make simplifying assumptions based on the relative impor- tance of various influence factors, and for statistical modeling, which requires knowledge about the impact of Ge profile variations on key electrical characteristics as well as on process control monitors (PCMs). * Corresponding author. Tel.: +49-351-463-37686; fax: +49- 351-463-37260. E-mail address: [email protected](M. Schroter). 0038-1101/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.sse.2003.12.004 Solid-State Electronics 48 (2004) 1133–1146 www.elsevier.com/locate/sse
14
Embed
Germanium profile design options for SiGe LEC HBTs
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Solid-State Electronics 48 (2004) 1133–1146
www.elsevier.com/locate/sse
Germanium profile design options for SiGe LEC HBTs
Michael Schroter a,*, Hung Tran a, Wolfgang Kraus b
a Department of Electron devices and Integrated Circuits, University of Technology Dresden, Mommsenstr. 13,
1134 M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146
So far, device design studies for LEC HBTs have
received relatively little attention. There has been quite a
variety of work dealing with analytical studies regarding
the ‘‘optimum’’ Ge profile shape that minimizes the base
transit time at low current densities, often resulting in
impractical profiles from a manufacturability point of
view. In a more adequate study [2], the transit and
maximum oscillation frequency were investigated for a
triangular and box Ge profile shape. In [3], a device
simulation study for optimizing the Ge profile with re-
spect to minimizing the noise figure was presented, that
was based on a conventional, i.e. not a LEC, doping
profile. So far, the most comprehensive study on the
dependence of a variety of figures of merits (FoMs),
including ideality, early effect and transit frequency, in
LEC HBTs can be found in [4]. The study dealt with the
Ge (box-)profile variation at the BE junction and was
based on a very early version of the process in [1], that
since then has been significantly modified to make it
manufacturable. In general, for application-specific
process optimization there are many aspects of impor-
tance, that include (i) a variety of FoMs such as transit
frequency, ideality, early effect, noise; (ii) the bias
dependence and high-current behavior; and (iii) the
spatial alignment and variation of the Ge profile with
respect to both junctions. Therefore, this paper focuses
on the impact of the Ge profile design on the above
mentioned aspects, including the relevant PCMs. In
addition, explanations for the observed electrical varia-
tions are provided based on theory, which has to be
somewhat limited though due to the lack of space.
Fig. 1. Schematic cross-section of th
The paper is organized as follows: First, the investi-
gated process is briefly characterized. Next, the selected
FoMs and PCMs are defined. Since for this investigation
the vertical profiles under the emitter are of main
interest, and the emitter width of the process under
consideration is relatively large, one-dimensional (1D)
device simulation is employed. The simulation results
are presented for different groups of Ge profile changes,
followed by a discussion and conclusions regarding a
vertical Ge profile design compromise and consequences
for compact modeling.
2. Investigated device structures
Fig. 1 shows a schematic cross-section of the inves-
tigated LEC transistor fabricated with non-selective epi
growth. More detailed information about the process is
provided in [1].
The corresponding doping profile of the ‘‘reference’’
transistor is shown in Fig. 2a. The metallurgical widths
of the base and lightly-doped emitter region are wBm ¼0:027 lm and wE ¼ 0:025 lm, respectively, while the
width of the collector is wC ¼ 0:25 lm. The doping
concentrations of emitter, base and collector are NE ¼3 · 1018 cm�3, NB ¼ 6 · 1019 cm�3, and NC ¼ 2 · 1017cm�3. The zero-bias internal base sheet resistance is 1470
X/square and varies only insignificantly with the Ge
profile variation and contents.
Fig. 2b contains a sketch of the shapes of the inves-
tigated Ge profiles. The box profile (1) is also used in the
e investigated SiGe LEC HBT.
0 0.1 0.2 0.3 0.4 0.5 0.6
1017
1018
1019
1020
D [c
m−3
]
x [µm]
xxjc
∆VG
∆VGmax
x e0
2
3
x je
1
∆VGp
(a)
(b)
Fig. 2. (a) Doping profile (along the dashed line under the emitter in Fig. 1) of the investigated reference transistor; (b) band gap
changes of generic Ge profile shapes. xje and xjc are the junction depths, and xe0 is the BE SCR edge at VBE ¼ 0.
M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146 1135
associated production process. The drift profile (2),
which was already discussed for transistors with con-
ventional BE doping profiles in, e.g. [2,3], does not look
favorable for the LEC transistor investigated here and,
thus will not be pursued further. In contrast, the ‘‘pla-
teau’’ profile (3) appears to yield promising results for
conventional doping profiles regarding the noise per-
formance (cf. [3]) and hence is investigated in this paper
in more detail. In all cases, the peak Ge mole fraction is
28% which does not cause any film stability issues due to
the thin base layer. The reference device (named ‘‘Ref’’
in this paper) represents a realistic profile design for a
high-speed transistor version of the process, with a peak
fT of 92 GHz. Further data will be shown later.
3. Investigated figures of merit
In this study the simulator DEVICE [6] was em-
ployed which uses physical models very similar to the
simulator used in, e.g. [5]. According to the available
literature (e.g. [5]) and also to the experience of the au-
thors with a variety of Si and SiGe processes, a drift-
diffusion formulation that includes the relevant physical
models as a function of Ge mole fraction is sufficient for
this task. A number of DC and small-signal simulations
have been performed to evaluate the sensitivity of the
electrical characteristics with respect to changes in
the Ge profile; in all cases, the doping profile was exactly
the same. If not noted otherwise, all characteristics
shown in this paper were obtained valid for VCE ¼ 0:8 V,and a unit emitter area of 1 lm2 is assumed. From the
electrical characteristics selected FoMs are determined,
which are defined next.
The transit frequency is calculated for each bias point
(IC; VCE) from a single frequency f [9]:
fT ¼ f
Imy11y21
� �����IC ;VCE
; ð3:1Þ
1136 M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146
with y21 and y11 as transconductance and input admit-
tance in common-emitter configuration, respectively.
Note, that fT can also be calculated as 1=ð2pdQp=dICÞwith Qp as hole charge entering the base. The maximum
value fT ;p has been selected as the first FoM; in addition,
the corresponding current, IC;p, is also determined, since
it provides useful information about the current drive
capability and power consumption (at high speed oper-
ation) of a process. Since fT is a compound variable, the
transit time sf 0 at low current densities is calculated
from fT using the same method as for measurements
(e.g. [11]). This provides more detailed information on
the cause of variations in fT around the peak.
The ideality of the ICðVBEÞ characteristics determines,e.g., the maximum achievable transconductance, and
depends strongly on the Ge profile at the BE junction as
will be explained later. Thus, a useful FoM is the ideality
factor
mCf ¼ ICVT gm
����ðIC0 ;VCE0Þ
; ð3:2Þ
with the low-frequency transconductance gm and the
thermal voltage VT . Since the ideality factor is slightly biasdependent at low current densities, it is calculated at a
given bias point ðIC0; VCE0Þ ¼ ð0:1 mA; 0:8 VÞ. Note thatdue to the exponential transfer characteristics relatively
small changes in mCf can cause large variations in IC .Another important FoM for analog circuit design is
the early voltage VA, defined as
VA ¼ ICgo
����ðIC0 ;VCE0Þ
ð3:3Þ
with go as output conductance. Since VA is bias depen-
dent, it is defined here at the bias point given before.
For RF circuit design, noise plays an important role.
For the purpose of this work, the high-frequency noise
factor was calculated from small-signal simulations
Fig. 9. Minimum noise figure as a function of collector current for selected Ge profiles and at two different frequencies f /GHz¼ 2, 6.
1142 M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146
M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146 1143
5. Discussion
Below, the impact of Ge profile variations on elec-
trical characteristics is discussed in view of consequences
for compact and statistical modeling. An important
point related to the latter is to explain the observed
changes in FoMs due to the Ge profile variation using
existing compact model expressions.
5.1. Ideality and transconductance
The best ideality factor (i.e. mCf close to 1) is obtained
for those transistors that have the smallest Ge change
(i.e. bandgap difference) within the BE space charge re-
gion on the base side. For instance, in transistor Ref the
Ge increases almost entirely across the BE SCR, while in
transistor Hr9 that increase has been moved into the
emitter, and in Hrp0 the Ge concentration does not
change at all within the BE SCR.
The observed tendencies can be analytically ex-
plained using the generalized integral-charge control
relation (GICCR) [12] which at low current densities
reads
IC ¼ cexp
VBEVT
� �Qp0 þ hjEiQjEi þ hjCiQjCi
ð5:1Þ
with Qp0 as zero-bias hole charge, c as a constant, and
QjEi and QjCi as BE and BC depletion charge, respec-
tively. For a given doping and Ge profile in the base
physical expressions can be derived for the factors hjEiand hjCi. Fig. 2b shows generic shapes of Ge profiles thatcan be used to discuss the ideality coefficient. The Ge
concentration is graded in case 2 across the BE SCR and
in case 3 across the neutral base. For a constant doping
profile, such as in LEC HBTs, the factor hjEi is
approximately given by [13]
hjEi ¼expðvÞ � 1
v; grading across BE SCR
v1� expð�vÞ ; grading across neutral base
8><>:
ð5:2Þ
with the variable
v ¼ DVGmax � DVGpVT
ð5:3Þ
that depends on the maximum Ge concentration reached
at the end of the neutral base, represented by the
bandgap difference DVGmax, and the Ge ‘‘plateau’’ con-
centration at the beginning of the base region (at
x ¼ xjE), that is represented by DVGp (cf. Fig. 2b). In the
limiting case of a box Ge profile, hjEi ¼ 1. As a conse-
quence of (5.2), grading the Ge across the BE SCR leads
to the highest hjEi values; the difference to the other casesbecomes significant above approximately v > 2.
Applying (3.2) to (5.1) yields
mCf ¼ 1
1� hjEiCjEiVTQp0 þ hjEiQjEi þ hjCiQjCi
ffi 1þ hjEiCjEiVTQp0 þ hjEiQjEi þ hjCiQjCi
ð5:4Þ
which depends directly on hjEi. Since QjEi varies less with
VBE than CjEi and in high-performance HBT processes
the change of QjEi is masked by the relatively high value
of Qp0, the influence of a hjEi variation in the denomi-
nator is small compared to that in the numerator.
Combining (5.4) with (5.2) explains the observed
dependence of the low-current ideality factor (and
transconductance) with the Ge profile in the base.
A simple qualitative explanation can be obtained also
from classical transistor theory, where IC / n2i ðxeÞ withxe as boundary between BE SCR and neutral base re-
gion. Assuming that the Ge contents starts to decrease in
the BE SCR, an increasing forward bias VBE moves xetowards the BE junction, i.e. into a region with larger
bandgap. This causes n2i ðxeÞ to become bias dependent
and to decrease with increasing VBE. As a consequence,
the differential increase of IC with VBE decreases, leadingto a more non-ideal characteristic (i.e. a larger value of
mCf ).
5.2. Early voltage
The largest Early voltage values are obtained for the
transistors with the largest grading of Ge across the
neutral base, while a spatially constant Ge concentration
does not improve the VA at all. However, since in LEC
HBTs the base doping is very high compared to the
collector doping their Early voltage is intrinsically larger
than that of BJTs.
Eq. (5.2) can also be used to easily derive the
dependence of the forward Early voltage on the Ge
profile in the base region. Applying (3.3) to (5.1) yields
VA ¼ Qp0 þ hjEiQjEi þ hjCiQjCi
hjCiCjCi� 1
hjci
Qp0
CjCi: ð5:5Þ
GICCR theory gives for the Ge profile shapes 2 and 3 in
Fig. 2b [13].
hjCi �v
expðvÞ � 1; ð5:6Þ
where DVGp in (5.3) now designates the bandgap value at
xe0, which equals DVGmax for the Ge profiles in Fig. 2b.
Again, for a box Ge profile hjCi ¼ 1. However, for a
significant grading across the base, expðvÞ 1, resulting
in hjCi � 1 and, according to (5.5), in a large increase in
VA over the corresponding BJT value Qp0=CjCi. Such
1144 M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146
increases have been observed in drift-type SiGe HBTs
with conventional doping profiles.
Note that a reverse Early voltage can be calculated
similarly and is directly related to mCf .
5.3. Depletion capacitances
The larger the bandgap difference is between the
edges of the space-charge region the larger is the deple-
tion capacitance of a heterojunction; i.e. the variation in
the zero-bias depletion capacitances CjE0 and CjC0 is
caused by the variation of the bandgap at the SCR edges
on the p-and n-side of the respective junction. Thus, thearea specific zero-bias capacitance of a heterojunction
can be approximated by
C�j0 ffi Cj0
1
1� DVg;pn=VD0
� �1=z; ð5:7Þ
in which Cj0, VD0 and z, respectively, are the corre-
sponding area specific capacitance, built-in voltage, and
exponent factor, respectively, of the reference material
(e.g. Si) without intentional bandgap change but possi-
bly containing high-doping effects, while DVg;pn is the
intentional bandgap voltage difference at the depletion
region boundaries with respect to the reference material.
For instance, the higher CjE0 of transistor Hr9 compared
to transistor Hrp0 is due to the larger Ge contents at the
base-end of the BE SCR. Similarly, CjC0 of transistor
Hr6 decreases since the grading causes the bandgap at
the base-end of the BC SCR to already increase towards
the (Si) value in the collector.
5.4. Transit time and transit frequency
Two different operating regions need to be consid-
ered for discussing the variation in the transit frequency.
At low collector current densities, fT changes are causedmostly by changes of the depletion capacitances, while
at high current densities (peak and beyond) fT changes
are due to variations in the transit time. The fT peak
value is strongly correlated to the low-current transit
time sf0 which in LEC HBTs consists of a base com-
ponent sBf 0 and a BC SCR component sBC . The latter
depends very little on the Ge profile. The dependence of
sBf 0 on Ge changes can be examined using the general
equation [7]
sBf ¼Z wb
0
n2ip
Z wb
x
pVTlnn
2idx0
� �dx: ð5:8Þ
Since in LEC HBTs the base is highly doped (p ¼ NB)
the neutral base width as well as the electron mobility
are almost bias independent. According to (5.8), the
box profile (n2i ¼ constant) will not improve sBf 0 com-
pared to a BJT, so that the main reason for the higher
fT of the LEC HBT is the suppression of the emit-
ter minority charge. The variation observed for sf 0(cf. Table 1) with changing Ge slope at the BC junc-
tion is caused by the partial decrease of the Ge
concentration already in the metallurgical base region.
On the other hand, the sf 0 increase observed for
the positive Ge slope moving towards the emitter is
caused by an increasing minority charge in the base–
emitter transition region with the modified (lower)
bandgap.
The decrease of fT at high current densities is caused
by an increase in the transit time sf , once the electric
field at the BC junction has dropped to about Elim ¼ vs=lnC0, with vs as electron saturation velocity and lnC0 as
low-field mobility in the collector. Thus, the corre-
sponding critical current ICK that indicates the onset of
such high-current effects can still be determined from the
same theory as applied to BJTs (e.g. [8,14]). However, in
contrast to BJTs, the valence band barrier prevents holes
from entering the collector, but rather causes a dipole
layer to form [16] that acts as a retarding field for elec-
trons at the location where the Ge drops (e.g., at the BC
junction). The consequence is a rapid increase of the
electron density in the base, and the associated sf in-
crease can be modeled as described in, e.g. [15]. As a
computationally more simple and less physical ap-
proach, the presently existing equations in HICUM
[8,14] can still be used by assigning the sf increase pre-
dominantly to the base region rather than to the col-
lector region, i.e. setting the model parameter FTHC to
a very small value.
From both the fT and IC characteristics in Section 4,
it seems that for all transistors at high injection ICsaturates at the same current. This is due to the effects
described above. Calculating from the doping and
width of the collector the parameters entering the ICKequation used in [14] gives ICK ¼ 2:8 mA, which is a
reasonably good estimate for the onset of high current
and barrier effects (cf. Fig. 8). Compared to BJTs, fT (sf )at ICK has dropped (increased) to a smaller (larger) value
due to the additionally forming barrier for electrons,
which in HBTs causes a more rapid increase of the
electron charge with current density.
5.5. Current gain
The current gain depends strongly on both the Ge
concentration at (or before) the BE junction and the Ge
profile within the base. The first determines the barrier
for back injection of holes into the emitter, while the
latter determines the collector current. For statistical
modeling both variations need to be taken into account.
A ‘‘peaky’’ shape (i.e. a visible but relatively slow drop
towards higher current densities) of the low-injection
current gain can be caused by an increased non-ideality
factor due to Ge grading across the BE SCR (cf. earlier
M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146 1145
discussion). At high current densities, the sharp current
gain drop is mostly due to the high-current effects dis-
cussed for the transit time and can be analytically
modeled as a function of the Ge contents via the GICCR
[12].
The current gain can be calculated from the ratio of
the (weighted) doping integrals of emitter and base.
Assuming that the difference of the (average) bandgap
reduction DV Ge;BE between base and emitter region, that
is due to the Ge contents, is entirely appearing in the
valence band, the bandgap dependence of the current
gain is roughly given by
B / expDV gB � DV gE
VT
� �
� expDV BGN ;B � DV BGN ;E þ DV Ge;BE
VT
� �; ð5:9Þ
where DV BGN ;B and DV BGN ;E are the average bandgap
reductions in base and emitter due to bandgap nar-
rowing. Since DV Ge;BE decreases with increasing Ge
contents in the emitter, B decreases, too, for e.g. tran-
sistors Hr8 and Hr9.
5.6. Minimum noise figure
Fig. 9 shows a comparison of the current dependent
minimum noise figure determined at 2 and 6 GHz. Note
that the calculated NFmin values in Table 1 and Fig. 9
indicate the trend and relative differences but not nec-
essarily the exact values achievable with the final pro-
cess. Overall, the Ge box profile turns out to be the best
choice, especially compared to the Ge profile with a long
plateau across the BE junction (in Hrp0). The difference
diminishes somewhat towards higher frequencies. It
seems though that for graded Ge profiles the minimum
of NFmin (like peak fT ) is shifted towards higher current
densities, where Hr11 yields slightly better results than
transistor Ref, but still with higher overall NFmin values.
Ge profiles similar to those in Ref, Hr6, Hr11, and
Hrp3_2 have also been investigated in [3] with respect
to high-frequency noise. While in our paper a similar
trend is obtained for Hr6 and Hrp3_2, the results for the
box and trapezoidal Ge profiles show the opposite ten-
dency. One reason for this observation can be the dif-
ferent doping profiles and maximum Ge mole fraction,
that influence the carrier transport and charge storage
at the end of the base and within both space-charge
regions.
6. Conclusions
SiGe HBT device designs with low-emitter doping
concentration have been investigated based on a realistic
reference doping and Ge profile. Case studies for the
sensitivity of important electrical characteristics and
typical process monitor values as a function of the Ge
profile have been performed. From a device design point
of view, the following conclusions can be drawn:
• The Ge profile design at the BC junction is of crucial
importance for transistor speed and Early voltage.
For a ‘‘useful’’ process, the Ge drop––even under
worst-case conditions––must always occur in the col-
lector. Otherwise, significant fT reductions (up to
50%) are possible. This assumption for a practically
‘‘useful’’ process allows significant simplifications in
statistical compact modeling.
• An aiding drift field in the base can significantly in-
crease the speed. For a given base–emitter doping
profile, an optimum mole fraction at the BE junction
and a maximum value of the Ge exist with respect to
speed, Early voltage, collector current ideality (i.e.
transconductance), and noise figure.
• A constant Ge contents across the BE SCR improves
the ideality (of IC and current gain) and transconduc-
tance especially at medium current densities for tran-
sistors with graded Ge profiles.
• The variation of the Ge profile slope at the BE junc-
tion has only a moderate impact on fT but can have alarge influence on the current gain.
• A reduction of a change in the maximum Ge mole
fraction has little impact on device performance, at
least for box-type profiles. There seems to exist an
optimum maximum mole fraction though.
Overall, for the given doping profile assumed in this
paper the optimum Ge profile seems to be close to
transistor Hrp3_2, possibly with a little lower Ge step at
the BE junction, a lower peak concentration to allow
some Ge extension into the collector for alleviating the
sharp drop in fT and current gain.
From a compact modeling point of view, the con-
clusions are
• the critical current density JCK , which indicates the
onset of high-current effects in the collector of homo-
junction bipolar transistors (BJTs) can still be used
for HBTs. However, the sf (fT ) value at JCK is for
SiGe HBTs usually larger (lower) than for BJTs
due to the barrier effect;
• the causes for the observed changes in electrical char-
acteristics can be explained from existing transistor
theory, mainly from GICCR and transit time theory;
• modeling the process tolerances due to a Ge drop al-
ready within the end of the base region should not be
relevant for statistical modeling of a practically use-
ful process;
• the simulated results are very useful for theory devel-
opment and extensions of HICUM specifically to-
wards the LEC HBT process type.
1146 M. Schroter et al. / Solid-State Electronics 48 (2004) 1133–1146
Additional transistors with ideal doping and Ge profiles
have been simulated to support and confirm the theo-
retical explanations for the observed variations in the
electrical characteristics.
The main differences between intrinsic conventional
and LEC HBTs are the shape and concentration of the
base and emitter doping. As a consequence, compared to
a conventional HBT the electrical behavior of LEC
HBTs is much more affected by the profiles within the
lightly doped emitter region. Furthermore, all presently
fabricated conventional SiGe HBTs known to the
authors have non-box base doping profiles with the
maximum of the boron concentration shifted towards
the BE junction, thus causing already a drift field for the
electrons towards the collector. Therefore, the shape of
the Ge profile within the base has a larger impact on the
intrinsic speed in LEC HBTs compared to conventional
SiGe HBTs.
Acknowledgements
The authors would like to thank J. Berntgen and L.
Kornau for providing process information and sup-
porting device simulations.
References
[1] Berntgen J et al. SiGe technology bears fruits. Mater Sci
Eng B 2002;89:13–20.
[2] Burghartz A et al. APCVD-grown self-aligned SiGe-base
HBTs. In: Proc IEEE Bipolar and BiCMOS Circuits and
Technology Meeting, 1993. p. 55–8.
[3] Ansley W, Cressler J, Richey D. Base-profile optimization
for minimum noise figure in advanced UHV/CVD SiGe
HBTs. IEEE Trans Microwave Theory Tech 1998;46:653–
60.
[4] Gruhle A. The influence of emitter–base junction design on
collector saturation current, ideality factor, early voltage
and device switching speed of Si/SiGe HBTs. IEEE Trans
Electron Dev 1994;41:198–203.
[5] Johnson J. Process and device simulation for SiGe
technology development. In: Short course at the IEEE
Bipolar and BiCMOS Circuits and Technology Meeting,
Minneapolis, 2001.
[6] Schr€oter M. Transient and small-signal high-frequency
simulation of numerical device models embedded in an
external circuit. COMPEL 1991;10(4):377–8;
see also: DEVICE––a mixed/mode device-circuit simula-
tor for DC, transient, and small-signal (hf) operation.
In: Proc NASECODE VII, Copper Mountain, USA, 1991.
p. 193–5.
[7] Kroemer H. Two-integral equations pertaining to the
electron transport through a bipolar transistor with non-
uniform energy gap in the base region. Solid-State Electron
1985;28:1101–3.
[8] Schr€oter M. HICUM, a geometry scalable physics-based
compact bipolar transistor model, December 2001. Avail-
able from: http://www.iee.et.tu-dresden.de/iee/eb.
[9] Gummel HK. On the definition of the cutoff frequency fT .Proc IEEE 1969;57:2159.
[10] Voinigescu S, Maliepaard M, Schr€oter M, Schvan P,
Harame D. A scaleable high-frequency noise model
for bipolar transistors and its applications in low-noise