Lecture 3

Common Source Amplifier Performance

Figures of Merit for Transistors

Boris Murmann
Stanford University
murmann@stanford.edu
Copyright © 2004 by Boris Murmann

B. Murmann EE 214 Lecture 3 (HO#5) 1

 Overview
• Reading
– 1.6.8 (Transit Frequency)
• Introduction
– Having established some basic modeling tools, we will now
look at the achievable performance of our common source
stage. We'll derive first order expressions for the amplifier's
bandwidth and power dissipation. As we will see, these
metrics are proportionally related to more fundamental
performance measures of the MOS device itself: Transit
Frequency and Transconductor Efficiency.

B. Murmann EE 214 Lecture 3 (HO#5) 2

 Common Source Amplifier Revisited

• Consider a common source stage driven by a "transducer"

• Ideally, what we want to achieve is
– A certain voltage gain
– Large bandwidth, low power dissipation
VDD

RL
VO +vo + +
Transducer Ri
vi vgs Cgs gmvgs ro RL vo
ID+id - -

vi Ri

VI

B. Murmann EE 214 Lecture 3 (HO#5) 3

 Transfer Function

vo ( s ) 1
H ( s) = = g m (ro || RL ) ⋅
vi ( s ) 1 + sRiC gs

DC gain Frequency
Dependence

• For simplicity, let's assume that RL is fixed, and RL<< ro

• The desired DC gain (ADC) then dictates our choice for gm
ADC
gm =
RL
• Now recall that
2I D
gm =
VOV
1
I D = g mVOV
2

B. Murmann EE 214 Lecture 3 (HO#5) 4

 Power Dissipation

• Power dissipation is given by

P = VDD ⋅ I D

• Substituting ID and gm using the expressions on the previous

slide yields

1 VDD
P= ⋅ ADC ⋅ VOV
2 RL

• Clearly, minimizing VOV will also minimize power dissipation

– Actually, how about making VOV ≅ 0?
• More later…
• Let's now look at the bandwidth of this circuit

B. Murmann EE 214 Lecture 3 (HO#5) 5

 Bandwidth

• The –3dB bandwidth of our circuit is

1 2
ω−3dB = with : C gs = WLCox
RiC gs 3

• In order to make this expression more insightful, we can

eliminate Cox using
W
g m = µCox VOV
L
• It then follows that
3 RL 1 µ
ω−3dB = ⋅ ⋅ ⋅ VOV
2 Ri ADC L2

B. Murmann EE 214 Lecture 3 (HO#5) 6

 Discussion

1 VDD 3 RL 1 µ
P= ⋅ ADC ⋅ VOV ω−3dB = ⋅ ⋅ 2 ⋅ VOV
2 RL 2 Ri ADC L

• With the assumed fixed design specifications, and a given

technology (µ, Lmin) , both power and bandwidth of our circuit are
completely determined by the choice of VOV
• Bad news: Making VOV small to save power also means that we
lose bandwidth
– This makes intuitive sense since
W gm
=
L µCoxVOV

• With gm and L fixed, smaller VOV translates into a bigger (wider)

device, and thus larger Cgs

B. Murmann EE 214 Lecture 3 (HO#5) 7

 Generalization (1)
• What we really want from our MOS transistor
– Lots of gm without investing much current
– Lots of gm without having large Cgs
• To quantify how good of a job our transistor does, we can
therefore define the following "figures of merit"
gm gm
and
ID C gs
• Using long channel MOS equations, we find
gm 2 g m 3 µVOV
= and =
I D VOV C gs 2 L2

• As expected, these equations reflect the same tradeoff that

we've observed in the above circuit example

B. Murmann EE 214 Lecture 3 (HO#5) 8

 Generalization (2)

gm/I D

gm/Cgs

VOV

• Part of your job as a designer is to choose VOV such that

– You get sufficient bandwidth
– And use as little power as possible to accomplish this
• Even though we've come to this graph using a very simple
example, the observed tradeoff tends to hold in general
– Of course, second order considerations will factor in as you
learn more about circuit design…

B. Murmann EE 214 Lecture 3 (HO#5) 9

 Product

• In cases where we want to get the "best of both worlds", it is

interesting to look at the product of our two figures of merit
g m g m 3µ
⋅ =
I D C gs L2

gm/I D

gm/I D*gm/Cgs

gm/Cgs

VOV
• While this result looks boring, it shows that using smaller
channel lengths improves circuit performance
– Either or both speed and power efficiency
B. Murmann EE 214 Lecture 3 (HO#5) 10
 Scaling Impact
• Thanks to "Moore's Law" feature sizes and thus the available
minimum channel lengths have been shrinking continuously
– Lmin decreases roughly 2x every 5 years
– Lmin=10µm in 1970, Lmin=90nm in 2004
• From the above discussion, it is clear that we can exploit
technology scaling in different ways
– Build faster circuits (gm/Cgs), while keeping power efficiency
constant (gm/ID)
• E.g. A/D converter for a disk drive - want to maximize
bandwidth/throughput
– Build more power efficient circuits (gm/ID), while keeping the
bandwidth constant (gm/Cgs)
• E.g. A/D converter for video signals - bandwidth fixed by a
certain standard

B. Murmann EE 214 Lecture 3 (HO#5) 11

 Transit Frequency (ωT)
• The transit frequency of a transistor is defined as the frequency
where the magnitude of the common source current gain (|io/ii|)
falls to unity ii
io
(Biasing not shown)

• Ignoring extrinsic capacitance, it follows that

g 3 µVOV
ωT = m = 2
C gs 2 L
• Incidentally, this metric is identical to the figure of merit we
considered earlier in the context of a CS amplifier…
• Note that we can also express transit frequency in units of Hz
ω
fT = T
2π
B. Murmann EE 214 Lecture 3 (HO#5) 12
 Transit Frequency Interpretation

• The transit frequency is only useful as a figure of merit in the

sense that it quantifies gm/Cgs
• It does not accurately predict up to which frequency you can use
the device
– At high frequencies, many assumptions in our "lumped"
transistor model become invalid
– Rule of thumb: Lumped model is good up to about ωT/10
• At higher frequencies, device modeling becomes more
challenging and many effects depend on how exactly you layout
and connect the device
– These effects are covered in more detail in EE314
– In EE214, we will assume that we operate well below ωT/10

B. Murmann EE 214 Lecture 3 (HO#5) 13

 ωmax

• A step into the right direction for quantifying the high frequency
capability of a MOSFET is to look at its power gain with gate
sheet resistance effects included
– The quantity ωmax is defined as the frequency at which the
magnitude of the common source power gain falls to unity
– Also known as "maximum frequency of oscillation"

Cgd
• Can show that
Z
1 ωT
Pi rgate Po
ωmax =
2 rgateC gd

B. Murmann EE 214 Lecture 3 (HO#5) 14

 Gain
• Let's now return to our common source amplifier example to
gain some insight into another performance metric of interest:
Small signal DC voltage gain
VDD

RL
VO +vo + +
Transducer Ri gmvgs
vi vgs Cgs ro RL vo
ID+id - -

vi Ri

vo ( s ) 1
VI
H (s) = = g m (ro || RL ) ⋅
vi ( s ) 1 + sRiC gs

DC gain Frequency
Dependence

B. Murmann EE 214 Lecture 3 (HO#5) 15

 Case 1: RL<< ro
• In this case, the DC gain is
2 I D RL
ADC = g m RL =
VOV

• For operation in the forward active region, we definitely need

I D RL < VDD

• Therefore, we have
2VDD
ADC <
VOV

• Note that the above expression is an upper bound, and cannot

quite be achieved in practice
– Nevertheless, this is a useful equation for back of the
envelope limit considerations

B. Murmann EE 214 Lecture 3 (HO#5) 16

 Case 2: RL>> ro
• This corresponds to the case where RL is essentially replaced by a
current source
• Now we have two biasing problems
– Must precisely adjust both VI and ID to set quiescent point of Vo
– We'll solve these kinds of problems when looking at more
practical biasing approaches, later in this course….
VDD

ID
Vo

vi Ri

VI

B. Murmann EE 214 Lecture 3 (HO#5) 17

 Intrinsic Gain
• With RL→∞, the basic common source stage achieves its
maximum possible voltage gain or "intrinsic gain"
ADC = g m ro

• As mentioned before, it is nearly impossible to develop a simple

equation that yields ro as a function of the device's bias point in
modern technologies
• For the remainder of this course, we therefore regard the
combined quantity "gm·ro" as a technology dependent parameter
– Must be characterized though measurements/simulations
• Interestingly, it will turn out that the gain of other, more
complicated circuits is fundamentally linked to the intrinsic
device gain "gm·ro"

B. Murmann EE 214 Lecture 3 (HO#5) 18

 Figure of Merit Summary

• Transit Frequency
gm
ωT =
C gs

• Transconductor Efficiency
gm
ID

• Intrinsic Gain
g m ro

• Note that we can characterize any technology (MOS, BJT, …) with

respect to these quantities
– Next lecture we'll look at our "EE214 technology" (0.35µm CMOS)

B. Murmann EE 214 Lecture 3 (HO#5) 19