ECEN474/704: (Analog) VLSI Circuit Design

Fall 2016
Lecture 11: Noise

Sam Palermo
Analog & Mixed-Signal Center
Texas A&M University

Announcements
HW3 is due Oct 27
Exam 2 is on Nov 8

2:20-3:45PM (10 extra minutes)

Closed book w/ one standard note sheet
8.5x11 front & back
Bring your calculator
Covers material through lecture 12
Previous years exam 2s are posted on the
website for reference
2

Agenda & References

Noise Types
Noise Properties
Resistor Noise Model
Diode Noise Model
MOSFET Noise
Filtered Noise
OTA Noise Example

Reading
Razavi Chapter 7

Noise Significance
Why is noise important?
Sets minimum signal level for a given performance
parameter
Directly trades with power dissipation and bandwidth

Reduced supply voltages in modern technologies

degrades noise performance
Signal Power Vdd SNR Psig / Pnoise
2

Vdd

Vnoise

Noise is often proportional to kT/C

Increasing capacitance to improve noise performance
has a cost in increase power consumption for a given
bandwidth
4

Interference Noise
Interference Man-Made Noise
Deterministic signal, i.e. not truly random
Could potentially be modeled and predicted, but practically this may be
hard to do

Examples
Power supply noise
Electromagnetic interference (EMI)
Substrate coupling

Solutions
Fully differential circuits
Layout techniques

Not the focus of this lecture

Unless the deterministic noise is approximated as a random process
5

Inherent Noise
Electronic or Device Noise
Random signal
Fundamental property of the circuits
Examples
Thermal noise caused by thermally-excited random motion of carriers
Flicker (1/f) noise caused by material defects
Shot noise caused by pulses of current from individual carriers in
semiconductor junctions

Solutions
Proper circuit topology
More power!!!

Is the focus of this lecture

6

Noise Properties

Noise is random

[Johns]

Instantaneous noise value is unpredictable and the

noise must be treated statistically
Can only predict the average noise power
Model with a Gaussian amplitude distribution
Important properties: mean (average), variance,
power spectral density (noise frequency spectrum)
7

RMS Value
If we assume that the noise has zero mean
(generally valid)
RMS or sigma value is the square-root of the
noise variance over a suitable averaging time
interval, T
1

V v t dt
T

12

2
n

n rms

Indicates the normalized noise power, i.e. if

vn(t) is applied to a 1 resistor the average
power would be
Pn

Vn2rms
1

Vn2rms

Signal-to-Noise Ratio (SNR)

signal power
SNR 10 log

noise power

For a signal with normalized power of Vx2rms

Vx rms
Vx2rms
SNR 10 log 2 20 log

V
V
n rms
n rms

Quantified in units of dB

Thermal Noise Spectrum

The power spectral density (PSD) quantifies how
much power a signal carries at a given frequency
Thermal noise has a uniform or white PSD

The total average noise power Pn in a particular

frequency band is found by integrating the PSD
f2

Pn PSD f df
f1

For white noise spectrum : Pn n0 f 2 f1 n0 f

10

Thermal Noise of a Resistor

The noise PSD of a resistor is
PSD f n0 4kT
where k is the Boltzmann constant and T is the absolute temperatur e (K)

The total average power of a resistor in a

given frequency band is
f2

Pn 4kTdf 4kT f 2 f1 4kTf

f1

Example: f=1Hz Pn=4x10-21W=-174dBm

11

Resistor Noise Model

An equivalent voltage or current generator
can model the resistor thermal noise
VRn2 Pn R 4kTRf

2
I Rn

Pn 4kT

f
R
R

[Johns]

Recall the PSD is white (uniform w/

frequency)
12

Noise Summation

vno t vn1 t vn 2 t
1
Vno2 rms
T

v
t

v
t
dt
n2
n1
0

2
no rms

2
n1 rms

2
n 2 rms

2
vn1 t vn 2 t dt
T 0

Same procedure applies to noise current

summing at a node
13

Correlation
Last term describes the correlation
between the two signals, defined by the
correlation coefficient, C
T

1
vn1 t vn 2 t dt

T 0
C
Vn1rms Vn 2 rms
Vno2 rms Vn21rms Vn22 rms 2CVn1rms Vn 2rms

Correlation always satisfies -1C1

C=+1, fully-correlated in-phase (0)
C=-1, fully-correlated out-of-phase (180)
C=0, uncorrelated (90)
14

Uncorrelated Signals
For two uncorrelated signals, the meansquared sum is given by
Vno2 rms Vn21rms Vn22 rms
Add as though they were vectors at right angles

For two fully correlated signals, the meansquared sum is given by

Vno2 rms Vn1rms Vn 2 rms

Sign is determined by phase relationship

RMS values add linearly (aligned vectors)

15

Noise Example #1: Two Series Resistors

vn2rms vn21rms vn22rms 2Cvn1rms vn 2 rms

The noise of the two resistors is uncorrelated or

statistically independent, so C=0
vn2rms vn21rms vn22 rms 4kT R1 R2 f

Always add independent noise sources using

mean squared values
Never add RMS values of independent sources
16

TAMU-ELEN-474 2009

Jose Silva-Martinez

Noise Example #2: Voltage Divider

Lets compute the output voltage: Apply superposition (noise sources
are small signals, you can use small signal models)!

vn1 R
1
+
vin
-

v0
vn2

R2

R2
R1
R2
v0
v n1
v n 2
v in
R1 R 2
R1 R 2
R1 R 2
Above is what you do for deterministic signals,
but we cannot do this for the resistor noise

But noise is a random variable, power noise density has to be used rather than voltage; then
the output referred noise density (noise in a bandwidth of 1 Hz) becomes
2

R2
R1
2
2
2
v0n
v
v

n1
n2
R1 R 2
R1 R 2
2

v0n

R1
R2
4
kTR
1

4kTR 2

R1 R 2- 17
R1 R 2

General Case :
2
on ,T

f H x s
x

2
s j 2f

v x2 f

Diode Noise Model

Shot noise in diodes is caused by pulses of current
from individual carriers in semiconductor junctions
White spectral density

[Johns]

Where q=1.6x10-19C and ID is the diode DC current

18

TAMU-ELEN-474 2008

Jose Silva-Martinez

Thermal Noise
=> Spectral Density of the thermal noise drain current (CMOS
transistor biased @ linear region)
B

N+
substrate

P+

id2

P+
N

Transistor

Channel

i d2

Resistor
in12=4kT/R1
R DS

R1
- 19 -

4kT

R DS
1

C ox

W
VGS VT VDS
L

TAMU-ELEN-474 2008

Jose Silva-Martinez

White Noise
i d2

@ Triode region
W
i 4kTC ox VGS VT VDS
L

4kT
R DS

2
d

f
Low current noise => W/L
@ Saturation
2
id

8
kTg m
3

go

=> gm or go

1
2
gm
R DS
3

W
8kT
i
C ox VGS VT
L
3
2
d

- 20 -

MOSFET 1/f (Flicker) Noise

Caused by traps near Si/SiO2 interface that
randomly capture and release carriers
K F g m2
i f
WLCox f
2
d

KF is strongly dependent on the technology

21

1/f Noise Corner Frequency

This is the frequency at
which the flicker noise
density equals the
thermal noise density
K F g m2
4kTg m
WLCox f co
f co

gm
KF
K F 1 g m I D

4kTCox WL 4kTCox L I D W

For a given gm/ID (which sets ID/W), the

only way to reduce fco is to use longer
channel devices
22

TAMU-ELEN-474 2008

Jose Silva-Martinez

Output and input referred noise

Current noise is the real one

i d g m Vgs

Thermal Noise

i d2 g 2m Vgs2

Voltage noise representation is an

artifact to facilitate system analysis

Flicker Noise

Veq2

2
Vgseq

8 kT

3 gm

K F g m2
i
WLCox f
2
d

id2

Referred to the input

K F g m2 1
v
WLCox f g m2
2
eq

veq2
- 23 -

KF
Cox

1 1

WL f

TAMU-ELEN-474 2008

Jose Silva-Martinez

Equivalent input referred voltage noise

2
Veq

V
2
eq

Equivalent input referred

noise voltage means that all
current noise sources are
accounted as drain
current and represented
by an equivalent noise
voltage at transistor gate

i
2
gm

2
dth

2
df

8 kT K F 1 1
V

3 g m Cox WL f
2
eq

Veq total RMS

- 24 -

2
v
eq (f )df
BW

TAMU-ELEN-474 2008

Jose Silva-Martinez

NOISE COMPONENTS (values provided are for a 0.8 m technology)

id

Noise density (V2/Hz)

Spice model

v eq v th v1/ f
KF
8 kT
2
v eq
df
df
WLCOX f
3 gm
2

veq2
M1

KF
9.8x10 9 V 2 / m Hz( NMOS)
C OX
0.5x10 9 V 2 / m Hz(PMOS)

FOR LOW-FREQUENCY APPLICATIONS,

WHEREIN 1/F NOISE IS DOMINANT,
PMOS DEVICES MUST BE USED.

Vn2

Flicker

Corner frequency

Thermal

f
- 25 -

Filtered Noise
[Johns]
2
f A j 2f vni2 f
vno
2

vni2 f

vno f A j 2f vni f

Noise output spectral density is a function only of the

magnitude of the transfer function, and not its phase
With multiple uncorrelated noise sources, combined
output is also uncorrelated
vn21 f

2
n2

2
f
vno

2
f
vno

i 1, 2 , 3

Ai j 2f vni2 f
2

vn23 f

[Johns]

26

First-Order RC Circuit Example

[Razavi]

What is the total output noise power?

27

First-Order RC Circuit Example

[Razavi]

As

vout
s 1
vR
1 sRC

2
f A j 2f vR2 f
vout

1
4kTR
2 2 2 2
1 4 f R C
To calculate Total Noise Power integrate over all frequencies
2

2
out

4kTR
2 2 2 2

f RC
1
4

Using

dx
1
x
tan

x2 1
f

2
out

2kT
2kT
tan 1 2fRC

C
C
f 0

kT
0

2
C
28

TAMU-ELEN-474 2008

Jose Silva-Martinez

Noise is generated by R but integrated noise is function of C (??)

vn R
v0

+
vin
-

1
kT

4kTR df

2
C
0 1 RC

vtotal

To get more insight, lets have a closer look on the operations!

Thermal
noise

4kTR

-2

f(log)

f(log)

1
2RC
- 29 -

Notice that:
When R increases thermal noise
increases too but the corner
frequency decreases, leading to a
constant area under the curves!

Noise Bandwidth
[Razavi]

The noise bandwidth is equal to the frequency

span of a brickwall filter having the same output
noise rms value
v B v df

2
0

2
no

For a first - order filter Bn

2
Validating with previous slides derivation :
1 kT
Total Noise Output v02 Bn 4kTR

2 2RC C

30

TAMU-ELEN-474 2008

Jose Silva-Martinez

Output referred noise: Take advantage of SYMMETRIES!

M2

M2

Output referred current noise density

Superposition: Every transistor contributes;
consider one at the time.

VX

iout
0.5id1

0.5id1
id1
M1

M1

Analysis: You can use standard circuit analysis

techniques but at the end of the day you have to
consider POWER.
Output noise density: Each noise component
represent the RMS value of random uncorrelated
noise! Then add the power noise components

IB
Noise injected into the common-source
node equally splits into the two branches

i out1

- 31 -

8
kTg m1
3

TAMU-ELEN-474 2008

Jose Silva-Martinez

Output referred noise: Take advantage of SYMMETRIES!

M2

M2

Output referred current noise density due to the Ptype devices:

id2

id2
VX

iout

iout 2 id 2
2

Id~0

M1

Left hand side transistor:

M1
IB

Noise injected into the common-source

node equally splits into the two branches

8
kTg m 2
3

Right hand side transistor

iout 2

Noise due to the current source is mainly

common-mode noise
- 32 -

8
kTg m 2
3

TAMU-ELEN-474 2008

Jose Silva-Martinez

Output and input referred noise

Output referred current noise density
M2

M2

VX

i out

iout

id1

id2

v1

Input referred noise density (V2/HZ)

v2

M1

M1
IB

8
2 kTg m1 2 kTg m 2
3

v in ,eq

8 kT g m 2
8 kT

2
2
3 g m1 g m1
3 g m1

In this case, noise due to the current source is

mainly common-mode noise
Be careful because this is not always the case!
- 33 -

TAMU-ELEN-474 2008

Jose Silva-Martinez

Integrated Input referred noise

Input referred thermal noise density (V2/Hz)
M2

M2

VX

v in ,eq

iout

8 kT 8 kT g m 2

2
2
3 g m1 3 g m1 g m1

Input referred noise level (volts)

id1

id2

v1
M1
IB

BW

v2
M1

Vin ,eq df

Noise(VRMS )

Example: for thermal noise, the noise level becomes

(assuming a single-pole system)

Noise(VRMS )

I should advise you to use: Noise( VRMS )

- 34 -

16kT
3

g
1
1 m 2
BW
g m1
g m1 2

8kT
g
1 m2
g m1
g m1

BW

4kT16x10-21 coul.V

Next Time
Three Current Mirror OTA

35