6.

976
High Speed Communication Circuits and Systems
Lecture 6
Enhancement Techniques for Broadband Amplifiers,
Narrowband Amplifiers
Michael Perrott
Massachusetts Institute of Technology

Copyright © 2003 by Michael H. Perrott

 Resistor Loaded Amplifier (Unsilicided Poly)
Vdd vout
RL vin
Id vout
vin slope =
Cfixed gm1RL -20 dB/dec
Vbias M1 M2
1 f
gm1
Ctot = Cdb1+CRL/2 + Cgs2+KCov2 + Cfixed
2πCtot
(+Cov1)
1
Miller multiplication factor
2πRLCtot

ƒ We decided this was the fastest non-enhanced amplifier

- Can we go faster? (i.e., can we enhance its bandwidth?)
ƒ We will look at the following
- Reduction of Miller effect on C
- Shunt, series, and zero peaking
gd

- Distributed amplification
M.H. Perrott MIT OCW
 Miller Effect on Cgd Is Significant

RL
Id
vout
Zin C
gd

M1 CL
vin Rs
Cgs
Vbias

ƒ Cgd is quite significant compared to Cgs

- In 0.18µ CMOS, C gd is about 45% the value of Cgs
ƒ Input capacitance calculation

- For 0.18µ CMOS, gain of 3, input cap is almost tripled

over Cgs!

M.H. Perrott MIT OCW

 Reduction of Cgd Impact Using a Cascode Device

RL vout
Vbias2
M2
Zin C CL
gd

M1
vin Rs
Cgs
Vbias

ƒ The cascode device lowers the gain seen by Cgd of M1

- For 0.18m CMOS and gain of 3, impact of C gd is reduced

by 30%:

ƒ Issue: cascoding lowers achievable voltage swing

M.H. Perrott MIT OCW
 Source-Coupled Amplifier

RL
vout
Zin C Cgd
gd

M1 M2
vin Rs

2Ibias
Vbias

ƒ Remove impact of Miller effect by sending signal

through source node rather than drain node
-C gd not Miller multiplied AND impact of Cgs cut in half!

ƒ The bad news

- Signal has to go through source node (C significant)
-
sb
Power consumption doubled
M.H. Perrott MIT OCW
 Neutralization

CN RL
-1 Id
vout
Zin Cgd

M1 CL
vin Rs
Cgs
Vbias

ƒ Consider canceling the effect of Cgd

- Choose C = C
- Charging of C
N gd

gd now provided by CN
ƒ Benefit: Impact of Cgd removed
ƒ Issues:
- How do we create the inverting amplifier?
- What happens if C is not precisely matched to C
N gd?

M.H. Perrott MIT OCW

 Practical Implementation of Neutralization

CN RL RL
CN

Cgd Cgd
Rs
M1 M2
vin Rs
-vin
2Ibias
Vbias

ƒ Leverage differential signaling to create an inverted signal

ƒ Only issue left is matching CN to Cgd
- Often use lateral metal caps for C (or CMOS transistor)
- If C too low, residual influence of C
N

- If C too high, input impedance has inductive component

N gd

N
ƒ Causes peaking in frequency response
ƒ Often evaluate acceptable level of peaking using eye diagrams
M.H. Perrott MIT OCW
 Shunt-peaked Amplifier
Vdd
Ld

RL
Id vout
vin
Cfixed
Vbias M1 M2

ƒ Use inductor in load to extend bandwidth

- Often implemented as a spiral inductor
ƒ We can view impact of inductor in both time and
frequency
- In frequency: peaking of frequency response
- In time: delay of changing current in R
L
ƒ Issue – can we extend bandwidth without significant
peaking?
M.H. Perrott MIT OCW
 Shunt-peaked Amplifier - Analysis
Vdd Small Signal Model
Ld Zout
vout

RL
Id Ld
vout
vin iin=gmvin Ctot
Cfixed
Vbias M1 M2 RL

ƒ Expression for gain

ƒ Parameterize with

- Corresponds to ratio of RC to LR time constants

M.H. Perrott MIT OCW
 The Impact of Choosing Different Values of m – Part 1
Small Signal Model
ƒ Parameterized gain expression
Zout
vout

ƒ Comparison of new and old Ld

3 dB frequencies iin=gmvin Ctot

RL

ƒ Want to solve for w2/w1

M.H. Perrott MIT OCW
 The Impact of Choosing Different Values of m – Part 2

ƒ From previous slide, we have

ƒ After much algebra

ƒ We see that m directly sets the amount of bandwidth

extension!
- Once m is chosen, inductor value is
M.H. Perrott MIT OCW
 Plot of Bandwidth Extension Versus m
Bandwidth Extension (w /w ) Versus m
2 1
1.9

1.8

1.7

1.6
w1/w2

1.5

1.4

1.3

1.2

1.1
0 1 2 3 4 5 6 7 8 9 10
m

ƒ Highest extension: w2/w1 = 1.85 at m ≈ 1.41

- However, peaking occurs!
M.H. Perrott MIT OCW
 Plot of Transfer Function Versus m
ƒ Maximum
bandwidth: 5
Normalized Gain for Shunt Peaked Amplifier For Different m Values

m = 1.41 m=1.41
m=2.41
(extension = 1.85)
0

ƒ Maximally flat -5 m=3.1

m=infinity
response: -10

m = 2.41 Normalized Gain (dB)

-15
(extension = 1.72)
ƒ Best phase
-20

response: -25

m = 3.1 -30
(extension = 1.6)
ƒ
-35
No peaking:
m = infinity -40

ƒ Eye diagrams often -45

1/100 1/10 1 10 100
used to evaluate Normalized Frequency (Hz)

best m
M.H. Perrott MIT OCW
 To Do

ƒ Add eye diagrams

M.H. Perrott MIT OCW

 Zero-peaked Common Source Amplifier
Vdd vout
vin zero
RL
Id vout 1
2πCsRs overall
Cfixed g mR L response
M1 M2 1+gmRs
vin
f
Vbias Rs Cs 1
2πCtotRL

ƒ Inductors are expensive with respect to die area

ƒ We can instead achieve bandwidth extension with
capacitor
- Idea: degenerate gain at low frequencies, remove
degeneration at higher frequencies (i.e., create a zero)
ƒ Issues:
- Must increase R to keep same gain (lowers pole)
- Lowers achievable gate voltage bias (lowers device f )
L

t
M.H. Perrott MIT OCW
 Back to Inductors – Shunt and Series Peaking
Vdd vout
RL vin
-20 dB/dec
L1 -40 dB/dec
g mR L
Id L2 vout
vin
Cfixed f
Vbias M1 M2 1
2πCtotRL

ƒ Combine shunt peaking with a series inductor

- Bandwidth extension by converting to a second order
filter response
ƒ Can be designed for proper peaking
ƒ Increases delay of amplifier

M.H. Perrott MIT OCW

 T-Coil Bandwidth Enhancement
Vdd
RL

Vdd
RL L2
CB
L1 k L1
L3 Id L2
vout vout
vin vin
Cfixed Cfixed
Vbias M1 M2 Vbias M1 M2

ƒ Uses coupled inductors to realize T inductor network

- Works best if capacitance at drain of M is much less than
1
the capacitance being driven at the output load
ƒ See Chap. 8 of Tom Lee’s book (pp 187-191) for analysis
ƒ See S. Galal, B. Ravazi, “10 Gb/s Limiting Amplifier and
Laser/Modulator Driver in 0.18u CMOS”, ISSCC 2003, pp
188-189 and “Broadband ESD Protection …”, pp. 182-183
M.H. Perrott MIT OCW
 Bandwidth Enhancement With ft Doublers

I1 I2
ƒ A MOS transistor has ft calculated as

M1 M2
vin
ƒ ft doubler amplifiers attempt to
Vbias 2Ibias
increase the ratio of
transconductance to capacitance
ƒ We can make the argument that differential amplifiers
are ft doublers
- Capacitance seen by V for single-ended input:
- Difference in current:
in

ƒ Transconductance to Cap ratio is doubled:

M.H. Perrott MIT OCW

 Creating a Single-Ended Output

I1 I2 Io

M1 M2
M1
vin
vin M2
Vbias 2Ibias Vbias Ibias Ibias

ƒ Input voltage is again dropped across two transistors

- Ratio given by voltage divider in capacitance
ƒ Ideally is ½ of input voltage on Cgs of each device
ƒ Input voltage source sees the series combination of
the capacitances of each device
- Ideally sees ½ of the C gs of M1
ƒ Currents of each device add to ideally yield ratio:

M.H. Perrott MIT OCW

 Creating the Bias for M2

I1 I2 Io Io

M1 M2
M1 M1
vin
vin M2 vin M2
Vbias 2Ibias Vbias Ibias Ibias Vbias M3 Ibias

ƒ Use current mirror for bias

- Inspired by bipolar circuits (see Tom Lee’s book, page 198)
ƒ Need to set Vbias such that current through M1 has the
desired current of Ibias
- The current through M 2 will ideally match that of M1
ƒ Problem: achievable bias voltage across M1 (and M2) is
severely reduced (thereby reducing effective ft of device)
- Do f doublers have an advantage in CMOS?
t

M.H. Perrott MIT OCW

 Increasing Gain-Bandwidth Product Through Cascading

vin vout
Amp Amp Amp
Cfixed Cfixed

vin A A A vout
1 + s/wo 1 + s/wo 1 + s/wo

ƒ We can significantly increase the gain of an amplifier

by cascading n stages

- Issue – bandwidth degrades, but by how much?

M.H. Perrott MIT OCW
 Analytical Derivation of Overall Bandwidth

ƒ The overall 3-db bandwidth of the amplifier is where

- w is the overall bandwidth

- A and w are the gain and bandwidth of each section
1

- Bandwidth decreases much slower than gain increases!

ƒ Overall gain bandwidth product of amp can be increased!
M.H. Perrott MIT OCW
 Transfer Function for Cascaded Sections

Normalized Transfer Function for Cascaded Sections

0
-3

-10

n=1
-20
Normalized Gain (dB)

-30

n=2
-40

-50

n=3
-60

-70

n=4
-80
1/100 1/10 1 10 100
Normalized Frequency (Hz)
M.H. Perrott MIT OCW
 Choosing the Optimal Number of Stages

ƒ To first order, there is a constant gain-bandwidth

product for each stage

- Increasing the bandwidth of each stage requires that we

lower its gain
- Can make up for lost gain by cascading more stages
ƒ We found that the overall bandwidth is calculated as

ƒ Assume that we want to achieve gain G with n stages

ƒ From this, Tom Lee finds optimum gain ≈ 1.65

- See Tom Lee’s book, pp 207-211
M.H. Perrott MIT OCW
 Achievable Bandwidth Versus G and n

Achievable Bandwidth (Normalized to f t )

Versus Gain (G) and Number of Stages (n)
ƒ Optimum gain per
0.3 stage is about 1.65
- Note than gain
0.25 per stage derived
from plot as
0.2
G=10

G=100
- Maximum is fairly
w1/wt

0.15
G=1000
soft, though
0.1
A=1.65 ƒ Can dramatically
lower power (and
0.05 A=3 improve noise) by
using larger gain
0
per stage
0 5 10 15 20 25 30
n
M.H. Perrott MIT OCW
 Motivation for Distributed Amplifiers

RL=Z0

vout
Cout Cout Cout
M1 M2 M3
Cin Cin Cin
Rs=Z0

vin

ƒ We achieve higher gain for a given load resistance by

increasing the device size (i.e., increase gm)
- Increased capacitance lowers bandwidth
ƒ We therefore get a relatively constant gain-bandwidth product
ƒ We know that transmission lines have (ideally) infinite
bandwidth, but can be modeled as LC networks
- Can we lump device capacitances into transmission line?
M.H. Perrott MIT OCW
 Distributing the Input Capacitance

RL=Z0

vout
Cout Cout Cout
M1 M2 M3
delay
Rs=Z0
Zo Zo Zo Zo
vin RL=Z0

ƒ Lump input capacitance into LC network corresponding

to a transmission line
- Signal ideally sees a real impedance rather than an RC
lowpass
- Often implemented as lumped networks such as T-coils
- We can now trade delay (rather than bandwidth) for gain
ƒ Issue: outputs are delayed from each other
M.H. Perrott MIT OCW
 Distributing the Output Capacitance
delay
RL=Z0 RL=Z0

Zo Zo Zo Zo vout

M1 M2 M3
delay
Rs=Z0
Zo Zo Zo Zo
vin RL=Z0

ƒ Delay the outputs same amount as the inputs

- Now the signals match up
- We have also distributed the output capacitance!
ƒ Benefit – high bandwidth
ƒ Negatives – high power, poorer noise performance,
expensive in terms of chip area
- Each transistor gain is adding rather than multiplying!
M.H. Perrott MIT OCW
 Narrowband Amplifiers
package

Connector die
Matching
Controlled Impedance Network
PCB trace

Driving
On-Chip
Source
Z1 L1 Vout
Delay = x Matching
Characteristic Impedance = Zo Amp
Network

Vin Transmission Line C1 C2 RL VL

ƒ For wireless systems, we are interested in

conditioning and amplifying the signal over a narrow
frequency range centered at a high frequency
- Allows us to apply narrowband transformers to create
matching networks
ƒ Can we take advantage of this fact when designing
the amplifier?
M.H. Perrott MIT OCW
 Tuned Amplifiers
Vdd
LT RL

vout

CL

M1
vin Rs

Vbias

ƒ Put inductor in parallel across RL to create bandpass

filter
- It will turn out that the gain-bandwidth product is
roughly conserved regardless of the center frequency!
ƒ Assumes that center frequency (in Hz) << ft
ƒ To see this and other design issues, we must look
closer at the parallel resonant circuit
M.H. Perrott MIT OCW
 Tuned Amp Transfer Function About Resonance

ƒ Amplifier transfer function

Cp Lp Rp vout

Ztank
ƒ Note that conductances
iin=gmvin add in parallel

ƒ Evaluate at s = jw

ƒ Look at frequencies about resonance:

M.H. Perrott MIT OCW

 Tuned Amp Transfer Function About Resonance (Cont.)

ƒ From previous slide

=0

ƒ Simplifies to RC circuit for bandwidth calculation!

vout
vin 1
R pC p
g mR p slope =
Cp Lp Rp vout
-20 dB/dec
wo
w
Ztank
iin=gmvin 1 1
Rp2Cp Rp2Cp
M.H. Perrott MIT OCW
 Gain-Bandwidth Product for Tuned Amplifiers

vout
vin 1
R pC p
g mR p slope =
Cp Lp Rp vout
-20 dB/dec
wo
w
Ztank
iin=gmvin 1 1
Rp2Cp Rp2Cp

ƒ The gain-bandwidth product:

ƒ The above expression is independent of center

frequency!
- In practice, we need to operate at a frequency less than
the ft of the device

M.H. Perrott MIT OCW

 The Issue of Q

vout
vin 1
R pC p
g mR p slope =
Cp Lp Rp vout
-20 dB/dec
wo
w
Ztank
iin=gmvin 1 1
Rp2Cp Rp2Cp

ƒ By definition

ƒ For parallel tank (see Tom Lee’s book, pp 88-89)

ƒ Comparing to above:
M.H. Perrott MIT OCW
 Design of Tuned Amplifiers

vout
vin 1
R pC p
g mR p slope =
Cp Lp Rp vout
-20 dB/dec
wo
w
Ztank
iin=gmvin 1 1
Rp2Cp Rp2Cp

ƒ Three key parameters

- Gain = g R
- Center frequency = w
m p

- Q = w /BW
o

o
ƒ Impact of high Q
- Benefit: allows achievement of high gain with low power
- Problem: makes circuit sensitive to process/temp
variations
M.H. Perrott MIT OCW
 Issue: Cgd Can Cause Undesired Oscillation
Vdd
LT RL

vout

Zin C CL
gd

M1
vin Rs
Cgs
Vbias

ƒ At frequencies below resonance, tank looks inductive

Negative
Resistance!
M.H. Perrott MIT OCW
 Use Cascode Device to Remove Impact of Cgd
Vdd
LT RL

vout
Vbias2
M2
Zin C CL
gd

M1
vin Rs
Cgs
Vbias

ƒ At frequencies above and below resonance

Purely
Capacitive!
M.H. Perrott MIT OCW
 Active Real Impedance Generator

Zin
Cf

Vin Vout
Av(s)

Av(s) = -Aoe-jΦ
ƒ Input impedance:

Resistive component!
M.H. Perrott MIT OCW
 This Principle Can Be Applied To Impedance Matching
Zin
Iout

M1
vin Rs

Vbias Ls

ƒ We will see that it’s advantageous to make Zin real

without using resistors
- For the above circuit (ignoring C gd)

Itest Vtest C vgs

gs gmvgs

Ls
Looks like series resonant circuit!
M.H. Perrott MIT OCW
 Use A Series Inductor to Tune Resonant Frequency
Zin Zin
Iout Iout
Lg
M1 M1
vin Rs Rs
vin
Vbias Ls Ls
Vbias

ƒ Calculate input impedance with added inductor

ƒ Often want purely resistive component at frequency wo

- Choose L g such that resonant frequency = wo

M.H. Perrott MIT OCW