60 GHz wirless receiver frontend on 90 nm CMOS

Download Report

Transcript 60 GHz wirless receiver frontend on 90 nm CMOS

A 1.2V, 60-GHz radio receiver
with on-chip transformers and
inductors in 90-nm CMOS
28th IEEE Compound
Semiconductor Symposium
David Alldred
Brian Cousins
Sorin Voinigescu
University of Toronto
Overview
1.
2.
3.
4.
5.
6.
Motivation
Circuit blocks and schematics
Passive element design
Layout and fabrication
Measurement results
Conclusion
Motivation
Why 60-GHz radio?
• Enough bandwidth for 2+Gb/s data transmission
• Simpler radio architecture
• Smaller die (lower cost?) than at 2-10 GHz.
Why CMOS?
• 90-nm MOSFETs have fMAX > 200 GHz at 1V supply and 0.2mW/mm
• 90-nm n-MOSFETs have similar NFMIN as 200-GHz SiGe HBTs
• For large volume applications CMOS can be the lowest cost solution
ISCAS-07
Goal
RFIC-06
CSICS-06
Wireless Transceiver System
• Design and test receiver subsystem in digital CMOS
• Operating frequencies: RF: 60GHz, LO: 55GHz, IF:5 GHz
• Maximize Receiver Figure of Merit:
f * G * IIP3
FoM 
( F  1) * P
Downconverter Block Diagram
• Fully differential core except mm-wave interfaces
• Two on-chip 60-GHz baluns for SE-Diff. conversion
Circuit Blocks: LNA
•
•
•
•
Two-stage cascode with middle inductor for NFMIN, S21
Biasing at 0.15 mA/mm ensures minimum noise in LNA
LVT devices to ensure operation from 1.2 V supply
Ported from 1.5V design in 90nm RF CMOS by T.Yao et
al. RFIC-2006
Circuit Blocks: Mixer
• Double-balanced Gilbert
cell for LO-RF leakage
• Design strategy:
minimize noise,
maximize linearity
• CM inductor for noise
and headroom
• Middle inductors
increase gain and
reduce noise
• Interdigitated diff-pair
layout sharing well
Circuit Blocks: LO buffer, IF buffer
LO Buffer
Output Buffer
• Differential CS topology for simplicity
• LO buffer designed as CML gate for fast switching
• IF buffer @ 0.3mA/mm for maximum linearity (determines
receiver linearity and drives 50-Ohm loads)
Passive Element Design
Inductors
2π inductor model
Vertically-stacked inductor
• Y-parameters from field solver (ASITIC)
• Broadband 2π-model extracted from simulated Y-params
• Skin effect included for 55GHz and 60GHz inductors
Passive Element Design
Balanced single-to-differential converter
Balun model schematic
Layout of verticallycoupled, two-layer balun
• Balun model adapted from
inductor π-model
• Mutual capacitances distributed
for balanced operation
Fabrication
•
•
•
•
90nm GP CMOS (STM)
7M digital back-end
LVT transistors
fT/fMAX = 125/240 GHz
600x475 μm including all pads
Receiver Measurement Results
•
•
•
•
RF = 60 GHz, IF = 5 GHz
VGS for min. NF50 is slightly below peak gain (fMAX) bias
Confirms Jopt =0.15 - 0.2 mA/mm even at 60 GHz
IF bandwidth is > 2 GHz centered at 4.5 GHz
Conversion Gain, S11 vs. RF freq.
•
•
•
•
Peak gain of 21 dB at 52 GHz
Coincides with center frequency of S11
Input LNA inductor LG was 10% too high
Re(Zin) = 50 W @ 60 GHz.
Supply variation and LO power
• Operation verified over 1.1V to 1.5V range
• Noise Figure and Gain saturate for LO power > 2dBm
Linearity and LO-RF leakage
Performance Summary
Simulation Results
Parameter
Measured
Simulated
Frequency
60 GHz
60 GHz
Power Gain
16-20 dB
21 dB
NFMIN
5.5-7 dB
7.4 dB
P1dB
-21 dBm
-20 dBm
IIP3
-
-11.3 dB
Pdiss
60 mW
62.5 mW
Isolation (LO-RF)
< -94 dB
-120 dB
Area
0.6x0.48 mm2
0.6x0.48 mm2
• Results are very close to simulated value
• 6 dB lower NF than similar design in 150GHz SiGe
BiCMOS [M. Gordon, SiRF-06] but no on-chip VCO
Conclusion
• 60 GHz wireless receiver is feasible in 90 nm GP CMOS
• LVT devices critical for low-voltage/low-power
• Thick copper back-end was NOT critical
• Similar NF and P1dB to SiGe HBT technology but lower power
• Mixer and LNA consume only 20 mA (24 mW)
• Power could be further reduced with integration (buffers)
• Ultra small area of 0.28mm2 because of inductors
Acknowledgments
• CMC for fabrication and CAD tools
• Keith Tang and KenYau for help with
measurements
• OIT, CFI and ECTI for equipment
Thank you for your time.