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

2000 2005 2010 2015 2020

100

150

200

250

300

350

Peak

FT (

GHz

)

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

1/1/

2006

7/1/

2006

1/1/

2007

7/1/

2007

1/1/

2008

7/1/

2008

1/1/

2009

7/1/

2009

1/1/

2010

7/1/

2010

1/1/

2011

7/1/

2011

1/1/

2012

0102030405060708090

100110120

1000

s of

Waf

ers

SBC18 Wafers Last 6 years of production history for the SBC18 family of processes

Roughly 20K Wfr/Yr Run Rate

Almost none of this is mm-Wave but it shows the experience with producing wafers on a technology capable for mm-Wave applications

Comparisons with RFCMOS

0.5 1.0 1.5 2.0 2.5100

150

200

250

300

350 RF CMOS SiGe HBT

Next Gen.

SBC18H

SBC18H265nm

45nm

FPEAK

T (G

Hz)

Max VDD / BVCEO (V)

32nmSBC18H3

10 100200

250

300

350

400

450

500

RF CMOS SiGe HBT

FPEAK

T (G

Hz)

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

ITRS Roadmap Data

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

Comparisons with InP

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

0 1 2 3 4 5 6 7 8 9 1050

100

150

200

250

300

SiGe HBT InP HBT

Next Gen.

H

H2

FPEAK

T (G

Hz)

BVCEO (V)

H3

InP Data from Northrop Grumman Foundry Services Website

Overall mm-Wave Technology Comparison Table

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

For technologies currently in production

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.

1. ITRS Tables 2. TowerJazz SBC18H3 3. Northrop Grumman 0.6um InP HBT Technology

Key mm-Wave Components

24/79GHz Dual-Band LNA Jain et al., IEEE JSSC 2009 p. 3469

24 GHz VCO Jain et al., IEEE JSSC 2009 p. 2100

KU-Band Phase Shifter Wang et al., IEEE µwave & Wireless

Comp. Lett. 2010 p.37

Capacitors SiGe HBT Transmission Lines

Inductors 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

SiGe HBT Device: SBC18H3

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)

130nm

~9um

~7.5

um

Minimum footprint device

Advanced SiGe HBTs: What matters to mm-Wave 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

1E-4 1E-3 0.010

50

100

150

200

250

300

FT

FMAX

F T / F

MAX

(GHz

)

Current Density (mA/µm2)

Data is from 15 sites on a typical SBC18H3 wafer Shaded areas show 25-75 percentile data spread

1 100

2

4

6

8

10

|h21| UG|h

21| /

UG a

t 100

GHz

(dB)

DC Power Density (mW/µm2)

A different way of looking at the same data: power consumption for a given gain at 100GHz

RF Performance with Process Variation

0.95 0.96 0.97 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.050.5

125

102030405060708090959899

99.5

Slow Corner (+10% Base Doping)

Nom Fast Corner

(-10% Base Doping)

Cum

ulat

ive P

roba

bility

(%)

Normalized FPEAKT (GHz)

-1.0 -0.5 0.0 0.5 1.00.950.960.970.980.991.001.011.021.031.041.05

Rework

Norm

alize

d FPE

AKT

(GHz

)

Relative Collector Implant Misalignment, 1=rework limit

Rework

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

Minimum noise figure is substantially improved with each succeeding process generation along with FT / FMAX

0 10 20 30 40 50 60 70 80 90 1000123456789

10

SBC18H2 SBC18H3

NFM

IN (d

B)

Frequency (GHz)

-1.3dB

0 2 4 6 8 10 12 14 16 180.0

0.5

1.0

1.5

2.0

2.5

3.00.13x20 µm Single Emitter, Dual Base, Dual Collector

40 GHz

32 GHz

20 GHz

NFM

IN (d

B)

IC (mA)

8 GHz

Ideally the NF is very flat across bias since the NFMIN never coincides with the peak gain condition

Gain vs. Noise

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.

0 10 20 30 40 50 60 70 80 90 1000

2

4

6

8

10

12

14

NF/A

ssoc

iate

d G

ain

(dB)

Frequency (GHz)

NF @ 6mA/µm2

Peak Unilateral

Gain @ NFMIN

InP HEMT

Varactors

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

n++ Buried layer

n++ cathode Sinker

CMOS NWell n+ n+

Hyper-Abrupt Junction Varactor Nwell-MOS Varactor

p+ anode Hyper-abrupt n-implant

Gate poly anode Source/Drain Cathode

1 10 100

10

100

Junction Varactor MOS Varactor

Q a

t VG=1

(MO

SVar

) V A=

-0.5

(Jun

ctio

n Va

r.)

Frequency (GHz)

1 10 100 1000 10000

10

100

1000

10000

Data Model

Capa

cita

nce

(fF)

Capacitor Area (µm2)

2fF/µm2 MIM CapacitorOther 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

=

p-i-n Diodes

n++ Buried layer

n++ cathode Sinker

p+ anode

RF in

+ - DC Switch

Voltage

RF out

+ - Cathode

Bias

p-i-n Diode

• 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

N- Epi

1 10 100-5.0-4.5-4.0-3.5-3.0-2.5-2.0-1.5-1.0-0.50.0

2x2um 4x4um 6x6um

Inse

rtion

Los

s (d

B)

Frequency (GHz)

• 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

1 10 100-55-50-45-40-35-30-25-20-15-10-50

2x2um 4x4um 6x6um

Isol

atio

n (d

B)

Frequency (GHz)

p-i-n Diodes

RF Ground Solutions

Metal 1

Backside Metallization

p++ handle wafer

p- Epi

•Deep Silicon Vias

emitter collector

Metal 1

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

•Through-Silicon Vias

• Extremely “localized” grounding. DSVs can be placed within several µms of active devices. • <5pH/via. < 50 W/via • In production now

Metal 6 Inverted Ground Shield

RF Back End

• 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

Metal 1

Metal 2

Metal 3

Metal 4

Metal 5 Top MIM Top MIM

Metal 6 3u

m

11um

Bottom MIM Bottom MIM

Top MIM

MIM Stacked MIM TiN Resistor

M6

Metal 5

M6

M6

Metal 1

M6 Inductor with slotted V5 underpass and M1 ground shield

Bump Bond

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 HS NPN 240 GHz FT / 280 GHz FMAX

STD NPN 55GHz FT / 3.2V BVCEO

LPNP β=35

Resistors Poly 235 Ω/sq and 1000 Ω/sq

Metal 25 Ω/sq TiN on M3

Capacitors Single MIM 2 or 2.8 fF/µm2

Stacked MIM 4 or 5.6 fF/µm2

Varactors 1.8V MOS Q @ 20GHz = 20

Hyper-abrupt junction Q @ 20GHz =15, Tuning Ratio = 21%

RF Diodes p-i-n Isolation <-15dB, Insertion loss > -3.5dB at 50GHz

Schottky FC > 800 GHz

Roadmap

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) 1 10

0

100

200

300

SBC18H2 SBC18H3 SBC18H4 (prototype)

FPEAK

MAX

(GHz

)JC (mA/µm2)

VCB = 0.5VFMEAS=20GHz

3µm Emitter 122 Device

RF Modeling

NF

at 4

0GH

z

P-i-n Diode s12 vs. Freq.

Gm (Y21) 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

05

101520253035

75 76 77 78 79 80 81 82

Gai

n &

NF

(dB

)

Freqency (GHz)

80GHz RX+SX Test Results

NF

Gain

80GHz LNA built in SBC18H2 Technology (Courtesy Sabertek Inc.)

W-band 5-Stage LNA built in SBC18H2 Technology (UCI)

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