Modeling and characterization of SiGe HBT lowfrequency noise figuresofmerit for RFIC applications
ABSTRACT We present the first systematic experimental and modeling results of noise corner frequency (f_{C}) and noise corner frequency to cutoff frequency ratio (f_{C}/f_{T}) for SiGe heterojunction bipolar transistors (HBTs) in a commercial SiGe RF technology. The f_{C} and f_{C}/f_{T} ratio are investigated as a function of operating collector current density, SiGe profile, breakdown voltage, and transistor geometry. We demonstrate that both the f_{C} and f_{C}/f_{T} ratio can be significantly reduced by careful SiGe profile optimization. A comparison of the f_{C} and f_{C}/f_{T} ratio for high breakdown and standard breakdown voltage devices is made. Geometrical scaling data show that the SiGe HBT with A_{E}=0.5×2.5 μm^{2} has the lowest f_{C} and f_{C}/f_{T} ratio compared to other device geometries. An f_{C} reduction of nearly 50% can be achieved by choosing this device as the unit cell in RF integratedcircuit design.

Conference Paper: Analysis and design of 20 GHz VCOs using crosscoupled differential pair and balanced Colpitts topologies in SiGe:C BiCMOS technology
[Show abstract] [Hide abstract]
ABSTRACT: The design and analysis of fully integrated voltage controlled oscillators (VCO) for 20 GHz low cost and low power communication system is presented in this paper. Two differential topologies have been studied: balanced Colpitts VCO and LCVCO using a crosscoupled differential pair. We have focussed on oscillation frequency, tuning range, phase noise, and output power optimization and buffer stage specifications. SiGe:C heterojunction bipolar transistors of f<sub>T</sub>=55 GHz have been used and produced with a monolithic BiCMOS technology.Design and Technology of Integrated Systems in Nanoscale Era, 2008. DTIS 2008. 3rd International Conference on; 04/2008 
Conference Paper: Design and Optimization of 20 GHz LCVCOs in SiGe:C BiCMOS Technology
[Show abstract] [Hide abstract]
ABSTRACT: The design and analysis of fully integrated voltage controlled oscillators (VCO) for 20 GHz low cost and low power communication system is presented in this paper. Two differential topologies have been studied: balanced Colpitts VCO and LCVCO using a crosscoupled differential pair. We have focussed on oscillation frequency, tuning range, phase noise, and output power optimization and buffer stage specifications. SiGe:C heterojunction bipolar transistors of f<sub>T</sub>=55 GHz have been used and produced with a monolithic BiCMOS technology.Circuits and Systems for Communications, 2008. ICCSC 2008. 4th IEEE International Conference on; 06/2008  [Show abstract] [Hide abstract]
ABSTRACT: The design and analysis of fully integrated 20 GHz voltage controlled oscillators (VCOs) for low cost and low power communication system are presented in this paper. Two differential topographies have been studied: balanced Colpitts VCO and LCVCO using a crosscoupled differential pair. We have focused on oscillation frequency, tuning range, phase noise, output power optimization and buffer stage specifications. SiGe:C heterojunction bipolar transistors of a 52 GHz cutoff frequency have been used and produced via a monolithic BiCMOS technology.Microelectronics Journal 01/2010; · 0.91 Impact Factor
Page 1
IEEE
noise corner frequency [2], thereby making
Proof
ables. Emitter length, for example, can be optimized for active
noise matching in lownoise amplifier (LNA) design [7]. The
same total device area required is typically obtained by parallel
connection of small area unit cells. An array of SiGe HBT unit
cells with different emitter areas are characterized over a wide
range of biasing current to facilitate the optimum choice of de
vice geometry, biasing current, and unit cell size that minimizes
the lowfrequency noise.
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 11, NOVEMBER 20021
Modeling and Characterization of SiGe HBT
LowFrequency Noise FiguresofMerit
for RFIC Applications
Jin Tang, Student Member, IEEE, Guofu Niu, Senior Member, IEEE, Zhenrong Jin, Student Member, IEEE,
John D. Cressler, Fellow, IEEE, Shiming Zhang, Student Member, IEEE, Alvin J. Joseph, Member, IEEE, and
David L. Harame, Senior Member, IEEE
Abstract—We present the first systematic experimental and
modeling results of noise corner frequency ?
frequency to cutoff frequency ratio ?
junction bipolar transistors (HBTs) in a commercial SiGe RF
technology. The
andratio are investigated as a function
of operating collector current density, SiGe profile, breakdown
voltage, and transistor geometry. We demonstrate that both the
andratio can be significantly reduced by careful SiGe
profile optimization. A comparison of the
for high breakdown and standard breakdown voltage devices is
made. Geometrical scaling data show that the SiGe HBT with
? ? ?? ?
m?has the lowest
compared to other device geometries. An
50% can be achieved by choosing this device as the unit cell in RF
integratedcircuit design.
? and noise corner
? for SiGe hetero
andratio
and
reduction of nearly
ratio
Index Terms—Breakdown voltage, corner frequency, cutoff
frequency, device modeling, flicker noise, heterojunction bipolar
transistor (HBT), lowfrequency noise, phase noise, RF integrated
circuit (RFIC), SiGe.
I. INTRODUCTION
S
both wired and wireless telecommunications applications be
cause of its superior analog and RF performance, together with
its CMOS integration capability [1]. By employing bandgap en
gineering, SiGe HBTs outperform Si bipolar junction transis
tors (BJTs) in nearly every important performance metric and,
in several areas, provide improved performance over the III–V
HBTs. One of the areas in which SiGe HBTs exceed GaAs
HBTs is in low
iGe heterojunctionbipolartransistor(HBT) technologyhas
come of age as an important semiconductor technology for
Manuscript received April 4, 2002. This work was supported by the National
Science Foundation under Grant ECS0119623 and Grant ECS0112923, by
the Semiconductor Research Corporation under Grant SRC 2001NJ937 and
Grant 2000HJ769, by IBM under a Faculty Partnership Research Award, and
by the Alabama Microelectronics Science and Technology Center.
J. Tang, G. Niu, and S. Zhang are with the Alabama Microelectronics Sci
enceandTechnologyCenter,ElectricalandComputerEngineeringDepartment,
Auburn University,Auburn, AL36849 USA(email:tangjin@eng.auburn.edu).
<Z. Jin affiliation must match current biography affiliation. Pls. cor
rect.> Z. Jin and J. D. Cressler were with the Alabama Microelectronics Sci
enceandTechnologyCenter,ElectricalandComputerEngineeringDepartment,
Auburn University, Auburn, AL 36849 USA. They are now with the School of
Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta,
GA 303320250 USA.
A. J. Joseph and D. L. Harame are with IBM Microelectronics, Essex
Junction, VT 05452 USA.
Publisher Item Identifier 10.1109/TMTT.2002.804519.
them an excellent choice for lownoise amplifiers, oscillators
[3], and power amplifiers.
The traditional figureofmerit for lowfrequency noise, the
noise corner frequency
, accounts for only
circuit design, however, the speed of the transistors is also a
critical design issue. Si BJTs typically have low
have sufficient gain to sustain oscillation at RF and microwave
frequencies because of their limited
, buttypically havehighand, hence, generate largerphase
noisewhenusedinoscillators.SiGeHBTs,however,provide
comparable to GaAs HBTs and lower
them an attractive choice for ultralow phasenoise oscillators
[3].Abetterfigureofmerittocharacterizelowfrequencynoise
for these applications is the
in [5], because it takes into account highfrequency response
through
.
This paper presents modeling and experimental results of the
lowfrequency noise figuresofmerit
mercial SiGe RF technology. Lowfrequency noise spectra and
highfrequency parameters were measured, from which
and are extracted. Four SiGe HBT wafers featuring different
SiGe profile designs were used to examine the impact of SiGe
profile shape on
and
sults are then used to derive a new figureofmerit for SiGe
profile design. Many wireless systems operate at frequencies
below 50 GHz, the peak
achieved in this SiGe technology.
The excess
can be traded for higher breakdown voltage to
allow larger signal power, which reduces additive phase noise
in voltagecontrolled oscillators [6]. We will examine how the
fabrication process modifications required for increasing break
down voltage affects lowfrequency noise, which then deter
mines modulative phase noise in oscillators.
To facilitate optimum choice of biasing current in circuit de
sign,
andweremeasuredacrossawiderangeofoperating
current.OneoftheleversinRFintegratedcircuit(RFIC)design
is that the device size and layout can be used as design vari
noise. In
, but do not
. GaAs HBTs have high
than Si BJTs, making
ratio [4], originally defined
andin a com
. The profile comparison re
00189480/02$17.00 © 2002 IEEE
Page 2
IEEE
designs have a higher Ge content and a larger Ge gradient in the
neutral base to achieve higher
and higher, but less Ge ret
rograding into the collector to keep the total Ge content within
the thermal stability limit.
Proof
noise
and inversely pro
are constant. This is in contrast to the prediction of a
independentratio in [5], which assumed
in our devices). At higher
and
determined by the
higher
, and smaller
A smaller
indicates better phase noise performance at
higher frequencies.
2IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 11, NOVEMBER 2002
Fig. 1.Schematic cross section of the SiGe HBTs used in this study.
II. DEVICE TECHNOLOGY
Fig. 1 shows a schematic cross section of the SiGe HBT used
in this study. The SiGe HBT has a planar selfaligned struc
ture with a conventional polyemitter contact, silicided extrinsic
base, and deep and shallowtrench isolation. The SiGe base
was grown using the ultrahigh vacuum/chemical vapor depo
sition (UHV/CVD) technique. Devices of two different break
down voltages were obtained on the same chip in the same fab
rication flow by selective implantation during collector forma
tion. The standard breakdown voltage (SBV) devices received
both a deep and a shallow collector implant, and have a peak
of 50 GHz (V). The high breakdown voltage
(HBV) devices received only the deep collector implant, and
have a peak
of 30 GHz (
fabrication process can be found in [8].
Four wafers with different SiGe profile designs were mea
sured,includinga10%peakSiGecontrol,a14%peaklownoise
design (
), a 18% peak lownoise design (
BJTcomparison.DetailsoftheSiGeprofiledesigncanbefound
in [9] and [10]. All of the wafers were fabricated in the same
wafer lot under identical processing conditions. The SiGe films
in all of the SiGe designs are unconditionally stable to defect
generation. Compared to the SiGe control, the
V). Details of the
), and an Si
and
III. LOWFREQUENCY NOISE FIGURESOFMERIT
It has been experimentally established that the major
noise source in these SiGe HBTs is the base current
[2], [3]. The
noise is proportional to
portional to the emitter area
as follows:
(1)
where
typicalSiGeHBTs.
is a technology dependent constant, and
correspondstotheflickernoisecon
for
stant
byequatingthe
in SPICE. The noise corner frequency
noise
is obtained
asfollows:totheshotnoise
(2)
where
below,
gain.
Equation (2) suggests that
and inversely proportional to
in [5]. The derivation of [5] showed that
biasing current density because
to mobility fluctuation. This, however, is not the case in our
devices, which all show an
close to two.
The figureofmerit for frequency response, cutoff frequency
, is related toby
is assumed for simplicity and insight, as detailed
isthecollectorcurrentdensity,andisthedccurrent
is proportional to
. This differs from that derived
is independent of
was assumed according
and,
(3)
where
voltage,
is the total junction depletion capacitance per unit area. Prior
to
, rolloff at highand
ical
range of interest to RF circuits (0.1–1.5 mA/ m ). The
ratio is obtained by combining (2) and (3) as follows:
is the forward transit time,
is the transconductance per unit area, and
is the thermal
are constant in the typ
(4)
Thus, the model suggests a linear increase of the
with operating collector current density
ratio
andprovided that
(
, whereis larger,
. Thus, theratio is
term at higher
factor are desired to reduce
. A smaller,
.
Page 3
IEEE
A. Collector Operating Current Dependence
The measured and calculated
versusare shown on the
left axis of Fig. 3 for an SBV HBT on the SiGe control wafer.
The measured
versusdependence is shown on the right
axis. The cutoff frequencyincreases with
to (3) prior to the
rolloff at high injection. The corner fre
quency
increases with, as predicted by (2). The common
practice of quoting corner frequency value without specifying
biasing current and device geometry can be misleading because
strongly depends on biasing current density, as shown by
the data in Fig. 3. The calculated
is in close agreement with
measured data. The slight deviation from a linear increase is
caused by the
dependence ofand the deviation of
two.Fig.4showsthemeasured
ratio,togetherwithmod
eling results calculated using (4). The modeling results agree
well with the measured data. The
ratio increases with
Proof
gether with measured
very close in the standard and HBV devices. As
in the HBV devices decreases because of the enhanced Kirk
effect due to the low collector doping. This
lates into an
difference. The
in the standard and HBV devices before the
injection. The
ratio becomes higher in the HBV device
after the
rolloff.
The similar lowinjection
HBV devicesindicates thatthe
TANG et al.: MODELING AND CHARACTERIZATION OF SiGe HBT LOWFREQUENCY NOISE FIGURESOFMERIT3
Fig. 2.
(?
Typical lowfrequency noise spectrum of SiGe HBT used in this study
? ??? ? ??? ?m , ?
? ? ?A).
IV. EXPERIMENTAL RESULTS
Lowfrequency noise spectra and
sured on both standard and HBV devices for the SiGe control,
the
andlownoiseSiGedesigns,andtheSiBJTcom
parison. Lowfrequency noise was measured using an EG&G
5113 preamplifier and an HP3561A dynamic signal analyzer
controlled by a Labviewprogram. parameters were measured
from 0.5 to 40 GHz using an HP8510C vector network ana
lyzer, from which
was extracted. The forward transit time
and the depletion capacitance per unit area
mined from the intercept and slope of the linear extrapolation
of the measured
data, respectively. In the lowfre
quency noise measurements, devices were biased at collector
current densities from 0.1 to 1.5 mA/ m , the range of interest
to RF circuits for the SBV devices.
Fig. 2 shows a typical lowfrequency base current noise spec
trum
for an SBV SiGe control HBT. The noise spectrum
showsaclear
componentandthe
corner frequency
is determined from the intersection of the
componentand theshot noiselevel.Therolloffabove
10 kHz is due to the bandwidth limitation of the preamplifier
used. The measured
product was plotted as a function
of
, from which the SPICE
tracted by assuming
. The obtained
proportional to
, leading to an emitter area independent
factor of 2.010m . The measured
mately the same for all of the SiGe designs.
parameters were mea
were deter
shotnoiselevel.The
noise constantwas ex
is approximately
factor is approxi
according
from
Fig. 3.
of ?
Measured corner frequency ?
for the SBV SiGe control HBT (?
and cutoff frequency ?
? ??? ? ??? ?m ).
as a function
Fig. 4.
SiGe control HBT (?
Measured and modeled ? ??
ratio as a function of ?
for the SBV
? ??? ? ??? ?m ).
, as predicted by (4). These results suggest that in order to
reduce the
andratio, the smallest
adequate
should be used.
that provides
B. High Breakdown Versus Low Breakdown
One of the most favorable properties of SiGe HBTs is the
low
corner frequency, which makes them excellent
choices for power amplifiers and oscillators [1]. In VCOs
operating below the peak cutoff frequency of the HBV device
(30 GHz), the HBV device is a better choice than the high
device. The HBV device has the natural advantage of operating
with a larger signal power, thus reducing additive phase noise
according to Leeson’s theory [11].
A logical question is how the use of the HBV device affects
the modulative phase noise upconverted from lowfrequency
noise. Fig. 5 compares the
for standard and HBV devices on the SiGe control wafer. At the
same
, the standard and HBV devices show nearly the same
product. Due to their similar
nearly identical
at the same
shows the measured and modeled
for both devices. At lower
product as a function of
, the two devices show
, as shown in Fig. 6. Fig. 7
ratio versus, to
is
increases,
difference trans
ratio is very close
rolloff at high
noise behavior in the SBV and
noise sourcescreated bythe
Page 4
IEEE
collector implantation through the SiGe base do not contribute
significant
noise.Sincetheupconversionprocessissimilar
in the HBV and SBV devices, we expect the modulative phase
noise in the HBV device to be as low as that in the high
device for current densities lower than the peak
Proof
, were op
without sacrificing SiGe
[9], [10]. Fig. 8 shows the measured
data for the SiGe control, the two lownoise HBTs, and the
Si BJT comparison. All of the SiGe HBTs have much higher
than the Si BJT.andhave a slightly higherthan
the SiGe control. The measured
noisefactor is nearly
identical for all of the SiGe designs. We thus expect a signif
icant reduction of
, as well asin the two lownoise
SiGe designs according to (2) and (4).
4IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 11, NOVEMBER 2002
Fig. 5.
HBV SiGe control HBTs (?
Measured ?
? ? product as a function of ?
? ??? ? ??? ?m ).
for the standard and
Fig. 6.
HBV SiGe control HBTs.
Measured and modeled ?
as a function of ?
for the standard and
Fig. 7.
(righthand side) for the standard and HBV SiGe control HBTs.
Measured and modeled ? ??
ratio (lefthand side) and measured ?
SBV
.
C. SiGe Profile Dependence and Profile Design Implications
The two lownoise SiGe profiles,
timized to improve
and
film stability and peak
and
Fig. 8.
and two lownoise HBTs.
Measured ?
as a function of ?
for the SBV Si BJT, SiGe control,
Fig. 9.
and two lownoise HBTs.
Measured ?
as a function of ?
for the SBV Si BJT, SiGe control,
Fig. 10.
Si BJT, SiGe control, and two lownoise HBTs.
Measured and modeled ? ??
ratio as a function of ?
for the SBV
The measured
tation. The noise corner frequency
and, and highest in the Si BJT. The
is the lowest in the two lownoise HBT designs because of the
much lower
and slightly higher
the same operating frequency, SiGe profiles optimized for high
, and highshould have better phase noise performance be
cause of the lower
. To achieve the same RF gain, transistors
with a higher
have the advantage to operate at a lower
which further reduces
and phase noise.
The above results suggest that the
a figureofmerit for SiGe profile optimization because
is proportional toaccording to (4). The
marily determined by the emitter structure, and independent of
the SiGe profile, as well as the collector doping profile, as evi
denced by the experimental data. An SiGe profile producing the
lowest
ratioleadstothebest
the best phase noise performance at higher frequencies.
data shown in Fig. 9 confirms this expec
is indeed the lowest in
ratio
, as shown in Fig. 10. For
,
ratio can be used as
factor is pri
ratio,andshouldhave
Page 5
IEEE
for different emitter areas. To our surprise,
m device has the lowestat the same
. Firstorder theory predicts the same ifis assumed to
be invariant among different emitter areas. This, however, is not
the case in these devices, as shown by the measured
data given in Fig. 13. The
has a higher
than the other three geometries. The higher
for this geometry is consistently observed on both the Si con
trol wafer and the three SiGe HBT wafers, and is possibly due
to the strain induced by the shallow trench isolation.
tors and
being the same, a higher leads to a lower
because is inversely proportional to current gain. On the
other hand, for a given
is independent of geometry be
fore high injection
rolloff [12]. Thedifference among dif
ferent geometries translates into an
difference. Under
the same
, them HBT has the lowest
and, therefore, it is an optimum unit cell choice for
Proof
reduction of nearly 50% can be achieved by using an optimum
unit cell size.
TANG et al.: MODELING AND CHARACTERIZATION OF SiGe HBT LOWFREQUENCY NOISE FIGURESOFMERIT5
Themodeled
, and SiGe
which were derivedusing
however, deviates from two. The deviation can be taken into
account by using
as a model parameter in the derivation of
and. The resulting equations are
andratiofortheSicomparison,SiGe
were calculated according to (2) and (4),
. Thefor SiGe control (2.24),
(5)
(6)
The
dimensions of
to calculate the modeling curve for SiGe control. The modifi
cation is necessary to achieve quantitative agreement with mea
surement for SiGe control. Equations (2) and (4), however, pro
videbetterinsightandintuitiveunderstandingofthebiasingcur
rent density dependence because of simple functional forms.
factorisnowdefinedusing,andhas
m . These modified equations were used
D. Geometrical Scaling and Optimal Transistor Sizing
In RFIC design, the device geometry and layout are often
used as design variables. The total emitter length required is
often physically realized by connecting a number of small area
unitcellsinparallel.Toinvestigatethelowfrequencynoiseper
formance of unit cells with different emitter areas, measure
ments were made on the standard SiGe control wafer for the
four device emitter areas: 1)
m ; 3)
m .
Fig. 11 compares the
base current
for the four devices. SiGe control HBTs of dif
ferentemittersizesshowthesamebasecurrentnoiseandemitter
area product
when biased at the same base current.
Fig. 12 shows the noise corner frequency
current density
the
m ; 2)
m ; and 4) and
product as a function of the
versus collector
versus
m device
fac
Fig.11.
deviceswithfourdifferentemitterareas(?
??? ? ?? ?m , and ??? ? ?? ?m ).
Measured?
??
versus?
forSiGecontrolstandardbreakdown
? ????????m ,????????m ,
Fig.12.
with four different emitter areas (?
?? ?m , and ??? ? ?? ?m ).
Measured?
versus?
forSiGecontrolstandardbreakdowndevices
? ??????? ?m , ??????? ?m , ????
Fig. 13.
with four different emitter areas (?
?? ?m , and ??? ? ?? ?m ).
Measured ? versus ?
for SiGe control standard breakdown devices
? ??????? ?m , ??????? ?m , ????
device layout. For instance, if the total effective emitter area re
quired is
m , a parallel combination of four
m HBTs should have a better lowfrequency
noise performance than ten
allel, or one
m HBTs in par
m HBT. A corner frequency
V. CONCLUSION
We have presented modeling and experimental results of
cornerfrequency
and corner frequency tocutoff frequency
ratio
in a commercial SiGe HBT technology. The
corner frequency
is proportional to the collector current
density
, and inversely proportional to . Theratio is