Advanced SiGe BiCMOS Process
Technologies for mm-Wave
Applications
IEEE Phoenix Waves and Devices Chapter
April 27, 2012 Technical Workshop
edward.preisler@towerjazz.com
TowerJazz RF/HPA Business Unit
Newport Beach, CA
Overview
 SiGe Production History
 Comparison of SiGe BiCMOS to other competing
technologies
 Summary of key mm-Wave components
 Circuit Examples
Historical View of SiGe BiCMOS RF Applications
350
Peak FT (GHz)
300
250
200
150
100
2000
2005
2010
2015
2020
Year of Production
 mm-Wave applications are expected to make up a good
portion of the market for SiGe BiCMOS technologies over
the next decade
Historical Volume of Wafer Production
 Last 6 years of production
history for the SBC18 family of
processes
SBC18 Wafers
120
110
100
80
 Roughly 20K Wfr/Yr Run Rate
70
60
50
40
30
20
10
1/1/2012
7/1/2011
1/1/2011
7/1/2010
1/1/2010
7/1/2009
1/1/2009
7/1/2008
1/1/2008
7/1/2007
1/1/2007
7/1/2006
0
1/1/2006
1000s of Wafers
90
 Almost none of this is mmWave but it shows the
experience with producing
wafers on a technology
capable for mm-Wave
applications
Comparisons with RFCMOS
ITRS Roadmap Data
RF CMOS
SiGe HBT
350
500
450
300
Next Gen.
250
200
SBC18H3
45nm
SBC18H2
65nm
150
100
0.5
400
SBC18H
1.0
1.5
2.0
(GHz)
FPEAK
T
FPEAK
(GHz)
T
32nm
RF CMOS
SiGe HBT
350
300
250
2.5
Max VDD / BVCEO (V)
 SiGe HBTs have at least a 0.5V
advantage in usable supply voltage but
usually it’s quite a bit more since
devices are often operated past BVCEO
200
10
100
Minimum Critical Dimension (nm)
 At the device level, RF CMOS can achieve
similar RF performance to SiGe HBTs, but at
much more advanced nodes
 For the moment, SiGe BiCMOS has a
distinct cost advantage over the equivalent
RF CMOS node
Comparisons with InP
InP Data from Northrop Grumman
Foundry Services Website
300
Next Gen.
H3
(GHz)
FPEAK
T
250
200
H2
150
H
SiGe HBT
InP HBT
100
50
0
1
2
3
4
5
6
BVCEO (V)
7
8
9
10
 InP-based devices can
achieve similar RF
performance to SiGe HBTs
with much higher breakdown
voltage but at several times
the cost per die
 InP technologies also offer
a much lower level of
integration
Overall mm-Wave Technology Comparison Table
For technologies currently in production
Technology
FT (GHz)
FMAX (GHz)
Supply
Voltage (V)
Level of
Integration
Quality of
Passives
Cost
45nm
RFCMOS1
240
280
1.1
High
Low
Medium
SiGe
BiCMOS2
240
280
>1.6
Medium
Medium
Low
InP HBT3
250
300
>4
Low
High
High
1.
2.
3.
ITRS Tables
TowerJazz SBC18H3
Northrop Grumman 0.6um InP HBT Technology
 SiGe BiCMOS offers a kind of “sweet-spot” for mm-wave applications
due to its combination of RF performance with low cost and adequate
levels of integration and quality of passives.
Key mm-Wave Components
24/79GHz Dual-Band LNA
Jain et al., IEEE JSSC 2009 p. 3469
SiGe HBT
Capacitors
24 GHz VCO
Jain et al., IEEE JSSC 2009 p. 2100
Transmission
Lines
Inductors
KU-Band Phase Shifter
Wang et al., IEEE µwave & Wireless
Comp. Lett. 2010 p.37
Varactors
p-i-n Diodes
RF Ground
 All of these components need to be stable and well-characterized out to very high freq.
 Ideally all of these components could be integrated onto a single chip
~7.5um
SiGe HBT Device: SBC18H3
~9um
130nm
Minimum footprint device
 TowerJazz’ 3rd generation fully self-aligned 0.18um SiGe BiCMOS
process technology
 CMOS and back end are exact replicas of mature SBC18
technology family (>1 decade, >150,000 wafers)
Advanced SiGe HBTs: What matters to mmWave Designers?
 Noise (RF and 1/f)
 FT vs. FMAX
 Gain at low current (low VBE)
 Gain in saturation (low VCE)
 Transconductance (GM / Y21)
 Short-emitter devices
 Wide Emitter devices
 BVCER
 Linearity
 ….
RF Performance
10
300
200
FT
FMAX
|h21| / UG at 100GHz (dB)
FT / FMAX (GHz)
250
150
100
50
0
1E-4
8
6
4
|h21|
UG
2
0
1E-3
0.01
Current Density (mA/µm2)
 Data is from 15 sites on a typical
SBC18H3 wafer
 Shaded areas show 25-75 percentile
data spread
1
10
DC Power Density (mW/µm2)
 A different way of looking at the same
data: power consumption for a given
gain at 100GHz
99.5
99
98
95
90
Slow Corner
(+10% Base Doping)
Nom
Fast Corner
(-10% Base Doping)
80
70
60
50
40
30
20
10
5
2
1
0.5
0.95 0.96 0.97 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05
(GHz)
Normalized FPEAK
T
Normalized FPEAK
(GHz)
T
Cumulative Probability (%)
RF Performance with Process Variation
1.05
1.04
1.03 Rework
1.02
1.01
1.00
0.99
0.98
0.97
0.96
0.95
-1.0
Rework
-0.5
0.0
0.5
1.0
Relative Collector Implant
Misalignment, 1=rework limit
 SBC18H3 has been designed for process insensitivity
 +/- 5s variation in base doping only leads to about +/- 1% in FT
 Even beyond rework limits, most challenging mask alignments will lead to
only +/- 3% in FT
RF Noise
0.13x20 µm Single Emitter, Dual Base, Dual Collector
10
3.0
9
SBC18H2
SBC18H3
8
2.5
2.0
6
NFMIN (dB)
NFMIN (dB)
7
5
4
3
-1.3dB
2
40 GHz
1.5
32 GHz
1.0
20 GHz
8 GHz
0.5
1
0.0
0
0
10 20 30 40
50 60 70 80 90 100
Frequency (GHz)
 Minimum noise figure is
substantially improved with
each succeeding process
generation along with FT / FMAX
0
2
4
6
8
10
12
14
16
18
IC (mA)
 Ideally the NF is very flat
across bias since the NFMIN never
coincides with the peak gain
condition
NF/Associated Gain (dB)
Gain vs. Noise
14
12
10
8
Pe
ak
Ga Unil
a
in
@ tera
l
NF
MI
N
6
4
2
0
2
A/µm
m
6
NF @
InP
HEMT
0 10 20 30 40 50 60 70 80 90 100
Frequency (GHz)
 If we benchmark against an
InP HEMT (from ITRS tables)
the gain at mm-Wave is
comparable
 Noise floor is still inferior to
III-V technologies but the gap
is closing.
Varactors
Nwell-MOS Varactor
Hyper-Abrupt Junction Varactor
Gate poly anode
Hyper-abrupt
n-implant
Source/Drain
Cathode
n+
CMOS NWell
n+
n++ Buried layer
 BiCMOS technologies offer two types:
 Hyper-abrupt p-n junctions for
linearity
 MOS for high Q
 Varactor Q is often the limiting factor in
the loss of the VCO circuit.
 At mm-wave frequencies the Q of both
devices starts to look similar
 Frequency synthesis at ~100GHz usually
uses harmonic generation so Q at 50 or
even 33 GHz might be most important
100
Q at VG=1 (MOSVar)
VA=-0.5 (Junction Var.)
n++
cathode
Sinker
p+ anode
10
Junction Varactor
MOS Varactor
1
10
Frequency (GHz)
100
Other VCO Topologies
• “Digital” varactors have become
common in high frequency VCOs
• MOSFETs are used to switch MIM
capacitors in or out of the circuit.
These are often used in parallel
with a traditional varactor for
ultra-linear fine-tuning
•A key enabling feature for these
devices is accurate modeling of
ultra-small MIM capacitors
10000
Capacitance (fF)
=
2fF/µm2 MIM Capacitor
1000
100
Data
Model
10
1
10
100
1000 10000
Capacitor Area (µm2)
p-i-n Diodes
+
-
+
-
DC Switch
Voltage
Cathode
Bias
p-i-n Diode
RF in
• p-i-n diodes can be used as RF
switches when surrounded by a bias tee
• Off-State capacitance is very low due
to low-doped n- intrinsic region
inherent in BiCMOS technologies
RF out
p+ anode
n++
cathode
Sinker
N- Epi
n++ Buried layer
0.0
-0.5
-1.0
-1.5
-2.0
-2.5
-3.0
-3.5
-4.0
-4.5
-5.0
Isolation (dB)
Insertion Loss (dB)
p-i-n Diodes
2x2um
4x4um
6x6um
1
10
Frequency (GHz)
100
0
-5
-10
-15
-20
-25
-30
-35
-40
-45
-50
-55
2x2um
4x4um
6x6um
1
10
Frequency (GHz)
100
• Smaller devices exhibit lowest best isolation due to lower total capacitance but
suffer somewhat in insertion loss due to higher series resistance.
• At high frequencies it seems as if RS is no longer the limiting factor for IL.
• 2x2um devices project to at least -10dB of isolation with better than -3dB of IL at
100GHz
RF Ground Solutions
•Deep Silicon Vias
emitter
collector
Metal 1
•Through-Silicon Vias
Metal 1
p- Epi
p++ handle wafer
Backside Metallization
• Extremely “localized” grounding. DSVs
can be placed within several µms of
active devices.
• <5pH/via. < 50 W/via
• In production now
Backside Metallization
• Through-Silicon Vias for low inductance /
low resistance emitter ground leads
• 1000 µm2 Pad can produce 22pH
inductance to ground with less than 1W/via
• In prototype now
RF Back End
11um
3um
Bump Bond
Metal 6 Inverted Ground Shield
Metal 6
Metal 5
Top MIM
Metal 4
Bottom MIM
Metal 3
Top MIM
M6
M
6
M
6
Metal 5
Bottom MIM
Top MIM
Metal 2
Metal 1
Metal 1
MIM
Stacked MIM
M6 Inductor with
TiN Resistor slotted V5
underpass and
M1 ground shield
• Top 3um metal used for inductors
• 11um separation between M6 and silicon
• Slotted vias available for inductor underpasses
• Can use M6 ground shield combined w/Bump bonding for uninterrupted ground plane
Complete SBC18H3 Device Roster
Family
Device
Characteristics
CMOS
1.8V CMOS
Model-exact copy of all other TJ 0.18um CMOS
3.3V CMOS
Bipolar
Resistors
Capacitors
Varactors
RF Diodes
HS NPN
240 GHz FT / 280 GHz FMAX
STD NPN
55GHz FT / 3.2V BVCEO
LPNP
β=35
Poly
235 Ω/sq and 1000 Ω/sq
Metal
25 Ω/sq TiN on M3
Single MIM
2 or 2.8 fF/µm2
Stacked MIM
4 or 5.6 fF/µm2
1.8V MOS
Q @ 20GHz = 20
Hyper-abrupt junction
Q @ 20GHz =15, Tuning Ratio = 21%
p-i-n
Isolation <-15dB, Insertion loss > -3.5dB at 50GHz
Schottky
FC > 800 GHz
Roadmap
300
FPEAK
(GHz)
MAX
 Prototype devices for SBC18H4 have
been built but require some special
processing steps that are not ready for
manufacturing yet.
 Rev. 0 model available now
(otherwise compatible with H3 kit)
 Tentative date for PDK and first
allowed tape in is July 2012
 Advantages of SBC18H4 will be along
the same lines as H3 over H2 (higher
FMAX, lower NFMIN)
3µm Emitter 122 Device
VCB = 0.5V
FMEAS=20GHz
200
100
SBC18H2
SBC18H3
SBC18H4 (prototype)
0
1
10
2
JC (mA/µm )
NF at 40GHz
RF Modeling
Gm (Y21) Vs. Freq.
P-i-n Diode s12 vs. Freq.
 Accurate RF models are almost more critical than the process they are trying to
model!
 Challenges with calibration and de-embedding at mm-Wave frequencies make RF
modeling a complex science.
Circuit examples from past technology generations
Gain & NF (dB)
80GHz RX+SX Test Results
35
30
25
20
15
10
5
0
Gain
NF
75 76 77 78 79 80 81 82
 W-band 5-Stage LNA built in SBC18H2
Technology (UCI)
Freqency (GHz)
 80GHz LNA built in SBC18H2 Technology (Courtesy
Sabertek Inc.)
 Simulated data shows accuracy of models out to mm-wave frequencies
 Even past generation devices seem adequate to create reasonable circuits
out to 90GHz
 New generations push past 100GHz and lower power consumption for
circuits at lower frequencies
Conclusion
 SiGe BiCMOS technologies capable of producing
practical circuits operating up to at least 100GHz are
currently available
 These technologies are based on a background of
nearly a decade of high-volume processing
 Newer generations increase design margin and
reduce power consumption at mm-Wave
frequencies, making them more suitable for
commercial manufacturing