Wide band RF CMOS circuit design techniques SSCS DLP, Fort Collins Domine Leenaerts Semiconductors, The Netherlands
Wide band RF CMOS circuit design techniques
SSCS DLP, Fort Collins
design techniques
Domine Leenaerts
Semiconductors, The Netherlands
Content
• Why wide band RF design?• How to make a wide band LNA?• What about the wide band Rx
architecture?architecture?• What about the technology?• Conclusions
History: SpeakEasy project (1992)
• Large scale military software defined radio
• motivated in large part by the communications interoperability problems that resulted from different branches of the military services having different branches of the military services having dissimilar (non-interoperable) radio systems
• ten different radio waveforms in software on a single platform operating in frequency bands between 2 and 2000 MHz
• Operated initially at the TMS320C40 processor
RJ Lackey and DW Upmal, "Speakeasy: The Military So ftware Radio," IEEE Communications Magazine, May 199 5
History: SpeakEasy project (1992)
LNA ADC
PA DAC
µPLNA ADC
PA DAC
µP
TM
S32
0C40
TM
S32
0C40
Final result in 1994:
several hunderd processors and the ‘radio’ filled the back of a truck
Today: 10 different standards?
There is a convergence of standards ongoing, but st ill…
Today: 10 different standards?
Frequency (GHz)
Today in industry
WCDMA/HSDPA/HSUPA/EGPRS, Skyworks
130nm cmos
ISSCC 2009
Today in industry
WCDMA/HSPA/EGPRS+ GPS and Rx diversity
WCDMA:4-LNA Primary Rx (2 SAW-less)
WCDMA:3-LNA diversity
Quad-Band GSM/EDGE (800-900-1800-1900)
0.18um CMOS ISSCC 2009
Quad-Band GSM/EDGE (800-900-1800-1900)
Qualcomm
Today in industry
HSUPA+BT+FM in 65nm CMOSSource: www.broadcom.com
Today in industry
• Industrial focus is on– Multi-band / multi-mode– Selectivity at RF
• Front-end module by means of antenna filters• Front-end module by means of antenna filters• Selective LNA’s• LO generation per mode or band
Software-defined radio (SDR)
• Based on signal processing– A/D requirements– Digital processing
LNA ADC
PA DAC
µPLNA ADC
PA DAC
µP
Use Reconfigurable radio instead
• Analog selectivity– A/D requirements more relaxed– Wide-band RF front-end– Digitally assisted radio / FE module– Digitally assisted radio / FE module
IQ-DAC
IQ-ADC
TX-filter
RX-filter
LO generation XO
radio
CAL
switc
h
FE module
BB
-MA
C p
roce
ssor
Dig
ital b
acke
ndIQ-DAC
IQ-ADC
TX-filter
RX-filter
LO generation XO
radio
CAL
switc
h
FE module
BB
-MA
C p
roce
ssor
Dig
ital b
acke
nd
Need of Wide-band RF front-ends
• First attempts were made by the development of UWB chipsets– 3 – 10 GHz– Started in 2003 with LNA design (see ISSCC – Started in 2003 with LNA design (see ISSCC
2004)
• Now focus on reconfigurable front-ends for 0 – 5 GHz operation– Started around 2005 (see ISSCC 2006)
Wide band LNA: What’s the problem?
Impedance match
Z
deliavsis PPZZ ,,* =⇒=
Z
Noise matchopts ZZ =
ZZiZs
Zi is mainly capacitive
Zopt
Im(Zopt ) is inductive in nature
How to obtain power match and noise match over wide bandwidth?
What is the problem?
Zs Zs
Wide band impedance match
Zs
Narrow band impedance match
Zs Zs
smRgF
142
αγ+≥
NF >> 3dB
s
s
QR
LF
ω3
21+≥
Zs
Ls
gss
gs
smin sC
sLC
LgZ
1++≈
Possible solution
Zs
filter
Wide band impedance match using filter as impedance converter
Zs
Ls
The first attempts
ISSCC2004: Ismail
Band-pass LC matching network
ISSCC2004: Bevilacqua
Band-pass LC matching network
Both are bulky due to # integrated inductorsand single-ended input/output
NF: 5.5dB (3-8GHz) NF: 4.5dB (2-10GHz)
Other possible solution
ZZs
min g
Z1≈
Zs
Ls
( )CGOQR
LF
s
s ++≥ ω3
21
Note: inherent single-to-differential conversion
min g
Z1≈
‘Second’ Generation
90nm CMOS
NF: 4.5dB (0.8-5GHz)
ISSCC 2006
decouple the noise and input impedance matchbut still bulky
Next generation: starting point
Simultaneous:• Single-to-Differential• Balanced Gain• Broadband input
vout+ -• Broadband input
match ( ~1/gmCG )• Noise Canceling• Distortion Canceling
[Blaakmeer et al. ESSCIRC `07]
RS
vsvin
+
-
CG CS
IBias
LNA: noise canceling
200Ω 50Ω
Noise canceled, signal amplified at differential output
vin
+
-Ibias
in
50Ω
20mS
80mS
vs
LNA: limited bandwidth
→ CLoad < 80 fF4 @ Ω200 ==
⋅=
CGV,
SCGV,CG
A
RAR
• Limited bandwidth at output CG-stage:
• Inductive peaking required– Contrasts aim: no on-chip inductors
Solution: don’t make voltage gain @ RF!
→ CLoad < 80 fF
GHz 10 2
1 =⋅⋅⋅
= π LoadCG3dB- CR
f
Solution: insert mixer
IF: Capacitanceno problem
Mixer: Low-Z input
Z4·Z
RS
vs
RF: no high-Z nodes
Mixer: Low-Z input
Low-Z
LO
gm
4·gm
IBias
Solution: insert mixer
WIF+ IF-
Z4·Z
R
3·R
3·C
C
gm·vrf4·gm·vrf
vs
RS
W4·WLO+ LO-
vrf+
-
‘CG’
‘CS’
Solution: noise cancellation @ IF
WIF+ IF-
Z4·Z
R
3·R
3·C
C
gm·vrf4·gm·vrf
vs
RS
W4·WLO+ LO-
vrf+
-
‘CG’
‘CS’
BLixer: IQ balanced
LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+
IF I-IF I+
I-Mixer Q-Mixer
LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+
IF I-IF I+
I-Mixer Q-Mixer
BalunLNAMixer →Blixer
4·i
LO Q+
vs
RS
IBias
i 4·i
LO Q+
vs
RS
IBias
i
BLixer: 50% duty cycle
LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+ LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+ LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+ LO Q-
4·i
LO I+ LO I- LO Q+
vs
RS
IBias
i
LO Q-
4·i
LO I+ LO I- LO Q+
vs
RS
IBias
i
LO Q-
4·i
LO I+ LO I- LO Q+
vs
RS
IBias
iLO Q+
LO I+
LO Q-
LO I-
BLixer: 25% duty cycle
LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+ LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+ LO Q-
IF Q+ IF Q-
4·ZZ
LO I+ LO I- LO Q+ LO Q-
4·i
LO I+ LO I- LO Q+
vs
RS
IBias
i
LO Q-
4·i
LO I+ LO I- LO Q+
vs
RS
IBias
i
LO Q-
4·i
LO I+ LO I- LO Q+
vs
RS
IBias
iLO Q+
LO I+
LO Q-
LO I-
BLixer: silicon
Baseline65nm CMOS1.2V supply
OSC
IN IF OutLbias
OSC
IF I+LO gen + buf
1.4 mm
Active area< 0.02 mm2IN
IF I+IF I-
IF Q+IF Q-
Supply & Biasing
LO gen + buf
Core + IF-bufBiasing
1.4 mm
[Blaakmeer et al. ISSCC ‘08]
BLixer: measured results[d
B]
-20
-10
0
10
20GC
NF
20GC
Wide band behavior at RF
RX-frequency [GHz]1 10
-30
-20S11
fIF [MHz]10 100
[dB
]
0
5
10
15
500
GC
NF… and at IF/BB
BLixer: measured results
• Linearity:– IIP3 @ RF: -3 dBm ( 5.2 & 5.7GHz, fLO = 4.6 GHz )
– IIP2 @ RF: +20 dBm ( 2.4 & 5.7GHz, fLO = 3.2 GHz )
– IIP2 (Mix-leak.) >+40 dBm (5.7 & 5.8GHz, fLO:0.5 - 7 GHz)LO
• Quadrature accuracy:– Phase error < 3°– Gain error < 1dB
• LO leakage < -50 dBm
What about inverters?
Cesd
VSS
INV1X6RF
VBDVBD10
12
14
gain
, NF
, S11
[dB
]GaNFS11
VSS
VDD
outin
VBD
VBS
VSS
VDD
outin
VBD
VBS
-2
0
2
4
6
8
10
0 5 10 15 20 25
frequency [GHz]
gain
, NF
, S11
[dB
]
What about inverters?
Add package:
HVQFN
2
4
6
8
10
12
14
gain
, NF
, S11
[dB
]
GaNFS11Ga_packageNF_packageS11_package
Cesd
VSS
INV1X6HVQFNRF LNA
-2
0
2
0 5 10 15 20 25
frequency [GHz]
What about inverters?
2
4
6
8
10
12
14
gain
, NF
, S11
[dB
]
Ga_packageNF_packageS11_packageGA_PANF_PAS11_PA
Cesd
VSS
INV1X6HVQFNRF
INV
1X10
PA
LNA
-2
0
2
0 5 10 15 20 25
frequency [GHz]
‘Inverter-only’ LNA
INV
1X10
PA INV1X0.5
No inductors!
Add additional stage and feedback in LNA
Cesd
VSS
INV1X6HVQFNRF
LNA
INV1X4
‘Inverter-only’ LNA
5
10
15
20
gain
, NF
, S11
[dB
]
GA_PANF_PAS11_PAGA_FBNF_FBS11_FB
-20
-15
-10
-5
0
0 5 10 15 20 25
frequency [GHz]
gain
, NF
, S11
[dB
]
‘Inverter-only’ PA
INV
1X6
PALNA
No inductors!
Cesd
VSS
INV1X4 HVQFN
RFINV1X4 INV1X10
Add additional stages and feedback in PA
‘Inverter-only’ PA
5
10
15
20ga
in, S
22 [d
B]
GpS22
-20
-15
-10
-5
0
0 5 10 15 20 25
frequency [GHz]
gain
, S22
[dB
]
RF front-end only with inverters!RF I/O
Noi
se F
igur
e [d
B]
Measured NF of front-end in BG1
Measured NF of front-end in BG6
Noi
se F
igur
e [d
B]
Measured NF of front-end in BG1
Measured NF of front-end in BG6
radio65nm (LP) CMOS
Frequency [Hz]
Noi
se F
igur
e [d
B]
Simulated NF of LNA only
Measured NF of LNA only
Frequency [Hz]
Noi
se F
igur
e [d
B]
Simulated NF of LNA only
Measured NF of LNA only
[Leenaerts et al. ISSCC ‘09]
• We can make wide band LNA’s, but what about the receiver?
What about the RX architecture?
• Commodity is to use passive mixers– Inherently wide band nature as they act as
switch only– Can be made very linear– Can be made very linear
• Followed by analogue wide band BB filters– Tunable in filter order, bandwidth
‘Classical’ ZIF architecture
ISSCC 2009: IMEC, SDR in 45nm CMOS
... but how to handle interferers?
Nonlinear !!
Wide band at RF give rise to interference issuesISSCC 2009: Ru, SDR in 65nm CMOS
... but how to handle interferers?
and ‘square-wave’ LO results in harmonic mixing
... but how to handle interferers?
• One solution is to use band-pass filters– Need tunable BP, difficult at RF
• Other solution is to NOT make (too much) gain at RFgain at RF
ESSCIRC 2001: Leenaerts, 180nm CMOS
… but harmonic mixing remains
harmonic rejection mixer removes 3*LO and 5*LO
Amplitude weighting
Phase shifting
Emulate sine-wave LO
ISSCC 2001: Weldon
… but how to make accurate?2
ISSCC 2009: Ru
41=2x5 + 3x7 +2x529=3x5 + 2x7
41/29 = 1.4138 and = 1.4142, so error is 0.03%2
2-stage poly phase network
And the result …
65nm CMOS
… but what about other interferers?
• Harmonic mixing helps only for interferers which are at harmonics of the used LO frequency
• How do we improve robustness against • How do we improve robustness against other interferers?– Make notch at RF
Notch at RF by translational loop
Darabi, ISSCC 2007
Notch at RF by translational loop
Darabi, ISSCC 2007
Impact on NF of LNA is 3dB, so activate filter only in presence of interferer
Some conclusions
• Wide Band CMOS design is possible– Convergence towards a receiver comprising
wide band LNA, passive mixers and BB filteringfiltering
– Additional tricks added to cope with interference are still needed
• But which technology is favorable?
What about the technology?
Does CMOS scaling improves wide band behavior?
• Technology RF performance– 180nm, 90nm, 65nm, 45nm
• UWB CMOS circuits – 90nm, 65nm, 45nm (and SiGe BiCMOS)
• UWB receiver design– 65nm, 45nm
Used technology• All experiments have been done in a low
standby power process node, i.e. LP CMOS
CMOS 90nm 65nm 45nm
Lgate [um] 0.1 0.06 0.04
Vdd [V] 1.2 1.2 1.1
Iem,M1,minW [mA@110 C]
0.21 0.11 0.07
#metal 6 + ALU 7 + ALU 7 + ALU
Technology
distance
ArearoC εε=
Some definitions
• Cut-off frequency ft– unity gain current frequency for short-circuited output
gg
mT C
gf
π2~ input capacitance
• Voltage gain bandwidth fa– Frequency for 3-dB DC voltage gain drop, calculated
for a resistive load making the DC voltage gain 2x
dddc
mA CA
gf
π2~ output capacitance
Some definitions
• 1-dB minimum Noise Figure frequency (fNF1dB)– frequency for which this NF is still reached assuming
optimum NF matching
• Finger width Wf– Folded device finger width: Wf = W/FOLD – Folded device finger width: Wf = W/FOLD
Difference schematic - layout
• Schematic– MOS model takes into account naked device
behavior.
• Layout• Layout– parasitics due to metal connections,
especially CO-M1– metal layers to fulfill current reliability
Layout has major impact in deep sub-micron
Difference schematic - layout100
GH
z
80
60
50
no interconnect
• NMOS device• L = 0.18 um• W = 32 um• contact = 2• @ fold = 8
f T G
Hz
Folding factor
1 102 3 4 6 8 20 30
50
40
30
with interconnect–– fT =70 GHzfT =70 GHz–– fT =60 GHzfT =60 GHz
gate
p+ guard-ring
gate
S S S S SD D D D
Comparison of this effect
• For both devices: single finger, double gate contact– Purple: schematic (naked
device)– Blue: layout extraction
• Serious impact of metallization
90nm W/L = 0.12um/0.1um
0
10
20
30
40
50
60
70
80
90
ft [G
Hz]
• Serious impact of metallization– 90nm: 70%– 65nm: 55%
0
0.001 0.01 0.1 1
drain current [m A]
65nm W=0.12um L=0.06um
0
10
20
30
40
50
60
70
80
90
0.001 0.01 0.1 1
drain current [mA]
ft [G
Hz]
65/45nm: ft obtained after layout
65nm L=0.06um
60
80
100
120
ft [G
Hz] W=0.12um
W=1um
45nm L=0.04um
150
200
250
ft [G
Hz] W=2um
W=10um
(Wf = 0.5 um for 1 and 10 um)
0
20
40
0,001 0,01 0,1 1 10 100
drain current [mA]
ft [G
Hz]
W=10um
65nm LP: ft naked device: 160GHz
0
50
100
0.001 0.01 0.1 1 10 100
drain current [mA]
ft [G
Hz]
W=10um
(Wf = 0.5 um for 2 and 10 um)
45nm LP: ft naked device: 270 GHz
Comparison for ft: scaling helps
W=10um, FOLD=20
150
200
250
ft [G
Hz] 90nm
0
50
100
150
0.1 1 10 100
drain current [mA]
ft [G
Hz] 90nm
65nm45nm
65nm: influence of finger-width
65nm: W=10um, L=0.06um
200.0
250.0
300.0
350.0
freq
uenc
y [G
Hz] ft
fa_2
0.0
50.0
100.0
150.0
200.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Wf = W/FOLD [um]
freq
uenc
y [G
Hz]
fa_2
fNF1dB
fmax
fcross
seems optimal layout includes only M1
RF performance incl. all layout effect
CMOS ft [GHz] fa [GHz fNF1dB [GHz] Wf opt.
180nm 41 39 10 ≈ 3
90nm 80 72 40 ≈ 1.790nm 80 72 40 ≈ 1.7
65nm 120 110 44 ≈ 1
45nm 200 130 53 ≈ 0.67
Wf/Lmin ≈ 16
How realistic is peak ft, fa?
• Peak values are normally reached for Vgs close to Vdd– happens for instance in cross-coupled diff pair
in most VCO topologiesin most VCO topologies
• But what happens in more realistic situations, e.g. Vgs = 1/2Vdd?– e.g. situation in frequency divider, LNA, …
How realistic is peak ft, fa?
90nm, W=1um, L=0.1um
60
70
80
90
freq
uenc
y [G
Hz]
38%
0
10
20
30
40
50
60
0.1 1 10
Vgs [V]
freq
uenc
y [G
Hz]
ft
faVgs=1.2V
Vgs=0.6V
38%
How realistic is peak ft, fa?
CMOS 90nm 65nm 45nm
Naked device
115 160 270
Vgs=Vdd 80 / 72 120 / 110 200 / 130
Vgs=1/2Vdd 49 / 38 51 / 41 47 / 28
x / y = ft / fa [GHz / GHz]
Scaling and wide band RF
• LNA performance• Frequency divider performance• Dual band receiver in 65nm• Dual band receiver in 45nm• Dual band receiver in 45nm
A CMOS wide band LNA
Vdd
OUT-VB
IN
0
OUT+Vss
Transformer:• Single-ended differential• Impedance matching• Voltage feedback
[ISSCC2007, Lee et al.]
Resistor:• Current feedback
A CMOS wide band LNA: 65nm
Gain: 11dB
3-dB BW: 7.5GHz
NF: 2.5dB
iIP2: +25dBm
[ISSCC2007]
A CMOS wide band LNA: 45nm
NF
Gv
S11
Gain: 8dB
3-dB BW: 11GHz
NF: 3dB
iIP2: +25dBm
S11
CMOS LNA: comparison
• Input device (for roughly same gm)– W/L 90nm: 200/0.1 um/um– W/L 65nm: 240/0.06 um/um– W/L 45nm: 200/0.04 um/um
x2
x1.25– W/L 45nm: 200/0.04 um/um
Vdd
OUT-VB
IN
OUT+Vss
0
CMOS LNA: comparison
LNA 90nm 65nm 45nm
Gv [dB] 9 11 10
BW [GHz] 7.5 7.5 11
NF [dB] 2.5 2.5 3
iIP2 [dBm] +15 +25 +25
iIP3 [dBm] 0 +12 +6
Pdiss [mW] 18 20 30
CMOS LNA versus BiCMOS
NF
S21
S11 (measured on PCB)
[dB]
S11(chip)
NF
S21
S11 (measured on PCB)
[dB]
S11(chip)
[ISSCC2005]
S12
[GHz]
S12
[GHz]
LNA 65nm SiGe 0.25um
Gv [dB] 11 20
BW [GHz] 7.5 12
NF [dB] 2.5 3
iIP2 [dBm] +25 +25
iIP3 [dBm] +12 +10
Pdiss [mW] 20 12
65nm and 45nm divider chainVDD
out0
out90
out180
out270
VDD
out0
out90
out180
out270
÷2in
out0out180
CLK
CLKb
Vbias
VSS
CLK
CLKb
Vbias
VSS
divider measurements
-10
0
10P
in [d
Bm
]
65nm
-50
-40
-30
-20
0 5 10 15 20 25
input frequency [GHz]
Pin
[dB
m]
65nm
45nm
divider measurements
CMOS 65nm 45nm
Self resonance 14GHz 18GHz
Max input freq. 18GHz 21GHz
Pin -14dBm -9dBm
Pdiss 18mW 24mW
UWB transceiver in 65nm
TIA
TIA
LNA
VtoI
VtoI
BG
1 /
BG
3 ÷2
mat
chin
g
BB
out
Ext
. LO
TIA
TIA
LNA
VtoI
VtoI
BG
1 /
BG
3 ÷2
mat
chin
gm
atch
ing
BB
out
Ext
. LO
A
select
VC
OV
CO
VC
O
BG
1 /
BG
3X
mat
chin
g
PA
BB
in
Ext
. LO
Bias &
Control
A
select
VC
OV
CO
VC
OV
CO
VC
OV
CO
BG
1 /
BG
3X
mat
chin
gm
atch
ing
PA
BB
in
Ext
. LO
Bias &
Control
UWB transceiver in 65nm
At mixer output, 3.2 -- 7.7 GHz< 2 dBTx output flatness
At mixer output, w/o PA4 %Tx output EVM
Two-tone: fin1=1.8 GHz, fin2=2.4 GHz+6 dBmRx IIP3
IF-input to mixer output, w/o PA+52 dBTx gain
fin= 4 GHz / 7 GHz5.0 / 5.5 dB Rx noise figure
Receiver52 mWDissipation @ 1.2V
Two-tone: fin1=2.4 GHz, fin2=5.2 GHz+25 dBmRx IIP2
Voltage gain, RF input to IF-output20 dBRx Gain
CommentValueParameter
At mixer output, 3.2 -- 7.7 GHz< 2 dBTx output flatness
At mixer output, w/o PA4 %Tx output EVM
Two-tone: fin1=1.8 GHz, fin2=2.4 GHz+6 dBmRx IIP3
IF-input to mixer output, w/o PA+52 dBTx gain
fin= 4 GHz / 7 GHz5.0 / 5.5 dB Rx noise figure
Receiver52 mWDissipation @ 1.2V
Two-tone: fin1=2.4 GHz, fin2=5.2 GHz+25 dBmRx IIP2
Voltage gain, RF input to IF-output20 dBRx Gain
CommentValueParameter
LNA/PA
ReceiverTransmitterLO generation
52 mW48 mW63 mW
Dissipation @ 1.2V ReceiverTransmitterLO generation
52 mW48 mW63 mW
Dissipation @ 1.2V
BB
filte
rs
LO
1mm
UWB transceiver in 65nm
S21
NF band1
dB
NF band9
dB
S21
NF band1
dB
NF band9
dB
S11
MHz
dB
MHz
dB
S11
MHz
dB
MHz
dB
UWB transceiver in 65nm
2nd harm.spurs
Band Group #1 Band Group #3
LOleakageleakage
UWB receiver in 45nm
RF front-end
1mm
IF stage
LO stage
Bias
UWB receiver in 45nm
10
12
14
16G
ain,
Noi
se [d
B]
-10.00
-5.00
0.00
S11
[dB
]
0
2
4
6
8
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
frequency [GHz]
Gai
n, N
oise
[dB
]
-25.00
-20.00
-15.00 S11
[dB
]
Comparison of 65nm-45nm RX
BB filter
65nm LP CMOS 45nm LP CMOS
Dies on scale : analog does not necessarily scale!
LNA
Comparison of 65nm-45nm RX
Rx performance
65nm 45nm
Gain [dB] 20 14
3-dB BW [GHz] 7 103-dB BW [GHz] 7 10
NF [dB] 5.5 7
iIP3 [dB] +5 0
iIP2 [dB] +25 +20
Pdiss [mW] 52 90
Where do we stand nowadays?
• Wide Band CMOS design is possible– Convergence towards a receiver comprising
wide band LNA, passive mixers and BB filtering– Additional tricks added to cope with interference – Additional tricks added to cope with interference
are still needed
• Impressive fT of naked device due to scaling does not tell the complete story– Layout influence becomes more and more
dominant
Hope you learned something