ECE695F
RFIC
Prof. S. Mohammadi
Amplifiers
 Gain
 BW
 Noise
 Stability
 Linearity
…
Highfrequency Linear Amplifier
Highfreq linear amps
widebandwidth Amps
Tuned Amps (Narrow band)
gain
f
BW
3dB
gain
f
3dB
BW
* Characteristics of wideband Amps
 useful for
* optoelectronic receivers
constant gain vs frequency
* instrumentation amplifiers
network analyzer
oscilloscope
input
constant gain
+constant delay
vs frequency
linear phase
vs frequency
no phase distortion


167
ECE695F
RFIC
Prof. S. Mohammadi
 low efficiency
(high power supply consumption)
T
f
Bandwidth
gain
=
×

Therefore
low gain
True only for
first order systems
(one pole)
 high input noise power
f
kTR
v
n
∆
=
4
2
(
)
f
v
n
∆
∝
2
* Characteristics of narrowband Amps
 high gain
(
)
T
f
bandwidth
gain
=
×
⇐
 high efficiency (lower power consumption)
 useful for low noise mobile receivers
good for mobile
application
Stability is of concern in both wideband and turned Amps,
but more importantly in wideband Amps
typical RF transistor


168
ECE695F
RFIC
Prof. S. Mohammadi
potentially
unstable
f
gain
MSG
MA
G
unconditionally
stable region
you need positive
feedback to make
the transistor oscillate
transistor can oscillate
without any external
positive feedback
So for a good RF transistor
we always have too much gain
at very lowfrequency + feedback
through C
gd
is always there
(how?)
Internal feedback through C
gd
+ high gain
oscillator
* So for widebandwidth Amps
since we have gain at lower frequency,
the chance of instability
at lower frequency is higher
f
gain
*
for tuned Amps,
we generally do not have high gain
at lower frequencies
as far as the amplifier is stable
at around the Bandwidth we are fine
gain
f
BW
0dB
freq. range to
look for
unstable condition


169
ECE695F
RFIC
Prof. S. Mohammadi
 How do we design / analyze
linear amps.
* let
s say you have an RF transistor and you want to design
a linear amp with this transistor.
How to proceed?
’
1st measure transistor for its RF performance
(Sparm)
use
network
analyzer
Design
using
Sparm of the transistor
22
21
12
11
S
S
S
S
f
GHz
1
.
0
1
2
3
Try to model
the transistor
SmallSignal Model
LargeSignal Model
Spice model
BSIM model
g
r
gd
C
o
r
gs
m
V
g
gs
V
gs
C
* only need Sparm measurement
+ insight to the physics of device
for initial values
need DC
d
g
I
V
−
ds
d
V
I
−
+ RF
V
C
−
parm
S
−
+insight to device physics
older technique
that measures Caps
vs.
supplied bias
need Sparm
for different biases
So largesignal modeling is a difficult task
often your transistor supplier provides that


170
ECE695F
RFIC
Prof. S. Mohammadi
For integrated circuit technologies
* you cannot possibly measure every given geometery
you measure certain geometery (RF +DC)
generate a scalable largesignal model
that fits all possible geometeries
for instant
geometery transistors
g
r
W
(
)
min
L
geometery
you only model 4 transistors
then you curvefit
(
)
W
f
your model r
g
for different
that you measured
 Scalable models are not very accurate,
gain may be off by a couple of dB
but you cannot afford not knowing the stability situation
that is why fab foundries often supply
Sparm data along with the model
But raw Sparm data is useless
when you deal with an integrated transistor


171