Lecture 20 - Diff-Amp Anal. I: Metrics, Max. Gain -... Announcements Review -

advertisement
6.012 - Microelectronic Devices and Circuits
Lecture 20 - Diff-Amp Anal. I: Metrics, Max. Gain - Outline
• Announcements
Announcements - D.P.: No Early effect in large signal analysis; just LECs.
Lec. 21 foils useful; Sp 06 DP foils, too (on Stellar)
Do PS #10: free points while working on D.P.
• Review - Differential Amplifier Basics
Difference- and common-mode signals: vID =vIN1-vIN 2, vIC =(vIN1+vIN2)/2
Half-circuits: half of original with wires shorted or cut (familiar, easy analyses)
• Performance metrics - specific to diff. amps.
Difference- and common-mode gains
Common-mode rejection ratio
Input and output voltage swings
• Non-linear loads
The limitation of resistive loads: Gain limited by voltage supply
Non-linear loads: High incremental resistance/small voltage drop
• Active loads
Lee load
Current mirror load
Clif Fonstad, 11/19/09
Lecture 20 - Slide 1
Differential Amplifiers - overview of features and properties
Intrinsic advantages and features:
- large difference mode gain
- small common mode gain
- easy to cascade stages; no coupling capacitors
- no emitter/source capacitors in CS/CE stages
Performance metrics:
- difference mode voltage gain, Avd
- common mode voltage gain, Avc
- input resistance, Rin
- output resistance, Rout
- common mode input voltage range
- output voltage swing
- DC offset on output
- Power dissipation
Clif Fonstad, 11/19/09
Today
Today
∞
Lec 21
Today
Today
Lec 21
Lec 18
Lecture 20 - Slide 2
Differential Amplifiers - common-mode input range
(VC,min ≤ vC ≤ VC,max)
vGS stays constant
We have said the output
changes very little for
common-mode inputs.
This is true as long as the
vC doesn't push the
transistors out of
saturation.
There are a minimum and
maxiumum vC:
VC, max: As vC increases,
vDS8 and vDS9 decrease
until Q8 and Q9 are no
longer in saturation.
Q4 + Q5
vC
up
vC
down
v DS gets
smaller
+
Q8
vDS
+
vGS
+
vDS
-
+
-
-
+
vGS stays constant
B
Q8, Q9 forced
out of
saturation
if v C too high
v
Q9 upC
-
-
VC, min: As vC decreases,
vDS10 decreases until Q10
is no longer in saturation.
Clif Fonstad, 11/19/09
+ 1.5 V
+Q
7
Q6
-
v DS gets
smaller
vDS
Q10
+
-
-
vGS
vc
down
Q10 forced out
of saturation
if v c too low
- 1.5 V
The node between Q8/Q9 and Q10 moves up
and down with vC.
Lecture 20 - Slide 3
Differential Amplifiers - output voltage range
+ 1.5 V
Fixed
+
+ +
v
v SG16
SG12
Q12
v SD12
-
+ Q13
! 0.6V
- +
v DS15
+
v GS15
-
Q15
-
AQ16
(VOUT,min ≤ vOUT ≤ VOUT,max)
+
v SD12 and v SD16 decrease as
vOUT goes up, so Q 12 and/or Q 16
may be forced out of
v SD16
saturation if v OUT is too high
-
+ Q20 vOUT
+
! 0.6V
up
! 0.6V
- Q
Q18
+
+
vOUT
17
! 0.6V
! 0.6V
- Q down
21
Q19 +
v DS19
B+
v DS15 and v DS19 decrease
Fixed
v GS19
- - 1.5 V
as v OUT goes down, so
Q15 and/or Q 19 may be
forced out of saturation if
vOUT is too low
As vOUT goes down, Q15 and/or Q19 may go out of saturation;
as vOUT goes up, the same may happen to Q12 and/or Q16.
Clif Fonstad, 11/19/09
Lecture 20 - Slide 4
Differential Amplifier Analysis incremental analysis exploiting symmetry and superposition
+
vid
-
a LEHC:
one half
of sym.
LEC
+
vod
-
+
vic
-
No voltage on
common links, so
incrementally they
are grounded.
a LEHC:
one half
of sym.
LEC
+
voc
Clif Fonstad, 11/19/09
a LEHC:
one half
of sym.
LEC
+
vid
-
+
-vod
-
a LEHC:
one half
of sym.
LEC
No current in
common links, so
incrementally they
are open.
+
-vid
-
+
voc
-
a LEHC:
one half
of sym.
LEC
+
vod = A vdvid
-
+
vic
-
+
vic
-
a LEHC:
one half
of sym.
LEC
+
voc = A vc vic
Lecture 20 - Slide 5
Looking at the design problem circuit:
Lesson - Draw the difference and common mode half circuits.
roLL dm
Difference mode
half circuit:
roCM dm
Q17
+
vid
-
Q8
+
roQ 16
roLL cm
Common mode
half circuit:
Q12
vic
-
Q8
2roQ 10
vod
-
RLOAD
roCM cm
Q17
+
Q20
Q20
+
Q12
roQ 16
voc
RLOAD
-
We have reduced the transistor count from 22 to 4, and we see that our
complex amplifier is a just cascade of 4 single-transistor stages.
Clif Fonstad, 11/19/09
Lecture 20 - Slide 6
V+
What's with these active, nonlinear loads?
Why doesn't the design problem use resistors?
RSL
Linear Resistor Loads:
the limit on maximum stage gain
- with linear resistor loads we must
make a compromise between the
voltage gain and the size of the
output voltage swing.
Maximum voltage gains
g
CO
gmv gs
-
go
+
v out
-
vout
-
+
vin
IBIAS
d
+
v in = v gs
+
Q
CS
G SL
(=1/RSL )
V-
MOSFET :
Av,max = gm RSL =
2 [ ID RSL ] max
2 ID RSL
$
[VGS # VT ] [VGS # VT ] min
Bipolar :
Av,max = gm RSL =
[I R ]
qIC RSL
$ C SL max
kT
VThermal
s,b
s,b
"
What are [ICRSL]max, [IDRSL]max, and [VGS - VT]min ?
! 11/19/09
Clif Fonstad,
* For a MOSFET, gm = (2KID/α)1/2 = K(VGS - VT)/α = 2ID/(VGS - VT)
Lecture 20 - Slide 7
Resistor Loads: cont.
- What are [IDRSL]max, [ICRSL]max, and [VGS - VT]min?
[IDRSL]max, [ICRSL]max:
- [IDRSL]max and [ICRSL]max are
determined by the desired
voltage swing at the output
and/or by the commonmode input voltage range.
- The ultimate limit is the
power supply.
[VGS - VT]min:
- [VGS - VT]min is set by how close
to threshold the gate can safely
be biased before the strong
inversion, drift model fails. We
will say more about this shortly
(Slide 23).
Clif Fonstad, 11/19/09
V+
RSL
CO
Q
+
vin
-
+
vout
-
IBIAS
CS
V-
Lecture 20 - Slide 8
Current Source Loads:
Incrementally large resistance
Relatively small quiescent voltage drop
-
- transistors with a DC input voltage, i.e. set up as sources/sinks -
MOSFET:
g
+
v gs = 0
sv bs = 0
+b
+
V REF
+
V REF
-
gmv gs
=0
gmb v bs
=0
go = " I D =
Bipolar:
+
V REF
-
b
+
V REF
gmv !
=0
e
-
Clif Fonstad, 11/19/09
+
v !=
g! ! 0
-
d
d
s
go
s, b, g
c
c
e
go
e, b
go
ID
VA
go
IC
go = " IC =
VA
Lecture 20 - Slide 9
V+
Current Source Loads:
the limit on the maximum
stage gain
IStage Load
CO
- current source loads eliminate the
need to compromise between the
voltage gain and the output
voltage swing
d
+
v in = v gs
gmv gs
+
v out
-
go
-
IBIAS
"
MOSFET :
Bipolar :
with VA,eff "
V[VGS # VT ]
!Clif Fonstad, 11/19/09
Av,max
2VA,eff
2 ID
gm
=
=
$
go + gsl
ID VA,Q + ID VA ,SL
[VGS # VT ] min
Av,max
VA ,eff
gm
qIC kT
=
=
=
go + gsl
IC VA ,Q + IC VA,SL
Vt
VA ,QVA,SL
[V
CS
gsl
s,b
s,b
vout
-
+
vin
-
Maximum voltage gains
g
+
Q
A ,Q + VA,SL ]
, Vt "
kT
q
Typically VA,eff >> [ID RSL]max
* For a MOSFET, gm = (2KID/α)1/2 = K(VGS - VT)/α = 2ID/(VGS - VT)
Lecture 20 - Slide 10
!
Current Source Loads: the maximum stage gain, cont.
- the similarity in the results for BJT's
and MOSFETs operating in strong
inversion extends to MOSFETs
operating sub-threshold and in
velocity saturation, also:
g
d
+
v in = v gs
gmv gs
go
-
+
v out
-
gsl
s,b
s,b
The MOSFET LEC: the same for all.
Maximum voltage gains
MOSFET sub - threshold :
iD = IS,s"t e( vGS "VT ) nVt , gm =
VA ,eff
gm
ID nVt
=
=
=
go + gsl
ID VA ,Q + ID VA,SL
nVt
Av,max
MOSFET w. velocity saturation :
ID
(vGS " VT )
VA ,QVA,SL
[V
A ,Q
Clif Fonstad, 11/19/09
*
iD = W ssat Cox
(vGS " VT ), gm = W ssat Cox* =
VA ,eff
ID (vGS " VT )
gm
=
=
#
go + gsl
ID VA ,Q + ID VA,SL
(vGS " VT ) min
Av,max
with VA,eff "
ID
nVt
+ VA,SL ]
Lecture 20 - Slide 11
Current Source Loads: Example - biasing a source-coupled
pair differential amplifer stage
Want:
Build:
V+
V+
Q1
ILOAD
Q4
+
vIN1
-
+
vO1
-
Q3
Q2
ILOAD
+
vO2
-
Q5
+
vIN2
-
R1
Q4
+
vIN1
-
+
vO1
-
+
vO2
-
Q5
+
-
IBIAS
Q7
VNote: I LOAD = I BIAS /2
Q6
V-
Note: W1 = W2 = W3
W7 = 2W6
This is nice…can we do even better?
Yes, with active loads. Consider…
Clif Fonstad, 11/19/09
Lecture 20 - Slide 12
Active Loads:
Loads that don't just sit there and look pretty.
First example: the current mirror load
V+
Q1
Q2
Signal actively fed from
left side to right side,
and applied inputs to
"stage load" MOSFETs.
+
Q3
+
vI1
VClif Fonstad, 11/19/09
I BIAS ,
rob
Q4
+
vI2
-
vOUT
"single ended," i.e.
- Now
only one output, but it
is twice as large:
vout = 2vout1
Load self-adjusting; circuit
forces ILOAD = IBIAS/2.
Lecture 20 - Slide 13
Active Loads: The current mirror load, cont.
V+
Large differential-mode
gain, small common-mode
gain.
Also provides high gain
conversion from doubleended to single-ended
output.
The circuit is no
longer symmetrical, so
half-circuit techniques
can not be applied.
The full analysis is
found in the course
text. We find:
Q1
Q2
id
id 2id
id
id
+
vid/2
-
Q4
Q3
V-
Difference-mode inputs
Clif Fonstad, 11/19/09
v out,d
I BIAS ,
rob
+
+
-vid/2
-
vOUT
-
2gm 3
=
v id 2
( go2 + go4 + gel )
Lecture 20 - Slide 14
!
Active Loads: The current mirror load
V+
Q1
Common-mode inputs
v out,c =
gob
v ic
2gm 2
Q2
ic
ic 0
ic
ic
+
vic
-
!
With both inputs:
v out
Q4
Q3
V-
I BIAS ,
rob
+
vic
-
+
vOUT
-
v in1 " v in 2 ) gob (v in1 + v in 2 )
2gm 3
(
=
"
2
2gm 2
2
(go2 + go4 + gel )
Note: In D.P. the output goes to the base of two BJTs; gel ≠ 0 and can
be important.
Clif Fonstad, 11/19/09
Lecture 20 - Slide 15
!
What if we want an active load and yet stay differential?
Active Loads - The Lee load
A load for a fullydifferential stage
that looks like a
large resistance in
difference-mode
and small resistance in commonmode)
The conventional
schematic is drawn
here, but the coupling of the load and
what is happening
is made clearer by
redrawing the
circuit (next slide.)
Clif Fonstad, 11/19/09
Q1
Q5
+
vI1
-
V+
Q2
Q3
+
+
vO1 vO2
-
V-
I BIAS ,
rob
Q4
Q6
+
vI2
-
Normal format
Lecture 20 - Slide 16
Active Loads - The Lee load. cont.
Drawn as on the
right we see that
the load MOSFETs
on each side are
driven by both outputs. The result is
different if the two
outputs are equal
and opposite (diffmode operation) or
if they are equal
(common-mode).
The next few slides
give the results for
each mode.
Q1
Q3
+
vO1
-
Q5
+
vI1
-
V+
Q2
+ vO2
+
- vO1 -
+
vO1
-
V-
Q4
+
vO2
-
+
vO2
-
Q6
+
vI2
-
I BIAS ,
rob
Drawn to highlight cross-coupling
and demonstrate symmetry
Clif Fonstad, 11/19/09
Lecture 20 - Slide 17
V+
Q1
The Lee load:
analysis for
difference-mode
inputs
Q3
+
vod
+
-
-
-vod
vod
+
vod
-
Q5
+
vid
-
+
-
Q4
+
-vod
-
+
-vod
-
V-
LEHC: difference-mode
+
v id /2 = v gs5
-
Q2
Q6
+
-vid
-
I BIAS ,
rob
+
gm5v id /2
go5
go1
gm1v od /2
-gm3v od /2
go3 v od /2
-
gel
goLLd
Clif Fonstad, 11/19/09
Lecture 20 - Slide 18
The Lee load: analysis for difference-mode inputs, cont
LEHC: difference-mode
+
v id /2 = v gs5
-
+
gm5v id /2
+
v id /2 = v gs5
-
go5
-gm3v od /2
gm1v od /2
go3 v od /2
-
gel
+
gm5v id /2
g
go5
gm1
go1
-gm1
go1 v od /2
-
gel
d
+
v id /2 = v gs5
s,b
go1
+
gm5v gs5
go5
goLLd v od /2
-
gel
goLLd = 2 go1
s,b
v od
"gm 5
Avd =
=
v id ( go5 + 2go1 + ge1!)
Clif Fonstad, 11/19/09
Note: In D.P., the outputs go to MOSFET gates so gel = 0.
Lecture 20 - Slide 19
V+
Q1
Q3
Q2
Q4
The Lee load:
+
analysis for
common-mode
inputs
-
+
voc
-
Q5
+
voc
-
+
v ic
v gs5
-
voc
-
+
voc
-
V-
LEHC: common-mode
+
+
+ voc
+
- voc -
voc
Q6
+
vI2
-
I BIAS ,
rob
+
gm5v gs5
go5
go1
gm1v oc
gm3v oc
go3 v oc
gel
gob /2
Clif Fonstad, 11/19/09
-
goLLc
Lecture 20 - Slide 20
The Lee load: analysis for common-mode inputs, cont
LEHC: common-mode
+
+
+
v gs5
gm5v gs5
-
v ic
go5
gm1
go1
gm1
go1 v oc
gel
gob /2
-
g
+
v ic
d
+
v gs5
-
+
gm5v gs5
s,b
s,b
goLLc = 2 ( gm1 + go1 )
go5
goLLc
v oc
gel
" 2 gm1
gob /2
-
-
!
v oc
"gob
gob
Avc =
=
#"
v ic 2[2( gm1 + go1 ) + ge1 ]
4gm1
Clif Fonstad, 11/19/09
Note: In D.P., the outputs go to MOSFET gates so gel = 0.
Lecture 20 - Slide 21
Achieving the maximum gain: Comparing linear
resistors, current sources, and active loads
Maximum Gains
MOSFET (SI)
Bipolar-like
(BJT and Sub-VT MOS)
Linear resistor loads
"
Current source loads
"
Active loads
Difference mode
$
Common mode
$
2 [ ID RSL ] max
[vGS # VT ] min
2VA,eff
[vGS # VT ] min
2VA,eff
[vGS # VT ] min
[vGS # VT ] min
2VA ,bias
"
[IC RSL ] max
"
$
n Vt
VA,eff
n Vt
VA,eff
n Vt
n Vt
$
VA,bias
Observations:
- Non-linear (current source) loads typically yield much higher gain than
linear resistors, i.e. VA,eff >> [IDRSL]max.
- The bias point is not !
as important to BJT-type stage gain.
- An SI MOSFET should be biased just above threshold for highest gain.
- For active loads what increases Avd, decreases Avc.
Clif Fonstad, 11/19/09
Lecture 20 - Slide 22
!
Achieving the maximum gain: (vGS-VT)min = ?
For SI-MOSFETs, maximizing the voltage gain (Av or Avd) requires
minimizing (VGS-VT). What is the limit?
Sub - threshold :
Av
1
=
VA n Vt
Strong inversion :
Av
2
=
VA (VGS " VT )
Av/VA is a smooth
curve, so clearly
(VGS-VT)min > 2nVt.
?
Note: n = 1.25 was assumed.
Clif Fonstad, 11/19/09
Lecture 20 - Slide 23
6.012 - Microelectronic Devices and Circuits
Lecture 20 - Diff-Amp Analysis I - Summary
• Performance metrics - specific to diff. amps.
Difference- and common-mode gains: Avd = vod/vid, Avc = voc/vic
Common-mode rejection ratio: CMRR = Avd/Avc
Common-mode input range
• Non-linear loads
Transistors biased in their constant current regions:
MOSFETs in saturation
BJTs in their FAR
• Active loads
Current mirror load:
Achieves double- to single-ended conversion without loss of gain
Has high resistance for difference-mode signals
Has low resistance for common-mode signals
Lee Load:
Maintains differential signals
Has high resistance for difference-mode signals
Has low resistance for common-mode signals
Clif Fonstad, 11/19/09
Lecture 20 - Slide 24
MIT OpenCourseWare
http://ocw.mit.edu
6.012 Microelectronic Devices and Circuits
Fall 2009
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
Download