6_OpAmp
Transcript of 6_OpAmp
-
7/29/2019 6_OpAmp
1/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 1 Ki
1. MOS Differential Pair with Resistive Loads
A very important amplifier stage is the differential pair. It consists ofa current source (sink) biasing a pair of matched input transistors, thatis, (W/L)1 = (W/L)2, and each transistor drives a load. The inputtransistors can either be NMOS or PMOS transistors, and twoimplementations withresistive loads(R1 = R2 = R) are shown below.
Let us consider the implementation with NMOS input transistors for
discussion. The transistor Mb biased with a gate voltage Vb serves asthe current sink Ib. To bias the differential pair properly, a dc voltageis applied to both V i+ and Vi to keep M1 and M2 working in theactive region. Because this voltage is common to both input nodes, itis called thecommon mode voltageVcm.
1M
ddV
1R 2R
2M
bMbV
1M
ddV
1R 2R
iV
oV oV
iV2M
bMbV
iV iV
oVoV
xV
xV
1M
ddV
1R 2R
2M
bMbV
iV iV
oVoV
cmV
xV
bI
I I
-
7/29/2019 6_OpAmp
2/21
ELEC 3400 VI. Operational Amplifiers
Since (W/L)1 = (W/L)2, by symmetry, the drain currents of M1 andM2 are equal:
Id1 = Id2 = I = Ib/2.
Both Vo+ and Vo are then equal, and the output common modevoltage Vocm is
Vo+= Vo = Vocm= Vdd IbR.
In general, all transistors should operate in the active region. Forsimplicity, let all transistors have the same gate overdrive voltage Vov.Hence, for the above equations to hold, the lowest Vcm is
Vcm = Vgs1 + Vov = Vtn + 2Vov
For Vcm+= Vdd, Vds1 = Vov when Vg1 = Vdd. Now, Vx = Vg1 Vgs1 =Vdd Vtn Vov. For M1 to remain in the active region, we require
IbR < Vdd Vov Vx = Vtn,
or R < 2Vtn/Ib.
The input common mode range is defined as the extreme inputvoltages such that all transistors in the amplifier are still operating in
the active region, and is thus Vcm to Vcm+.
Example:Given nCox=50A/V
2, Vtn=0.8V, (W/L)b=80, (W/L)1=(W/L)2=40,n=0/V. Also, Vdd=5V, Vb=1V and Vcm=2V. Find R such that theoutput swing is the largest. With this R, what is Vcm+and Vcm?
Solution:
Ib = 5080(1 0.8)2 = 80A
I = Ib/2 = 40A
With (W/L)1 + (W/L)2 = (W/L)b, we have Vov1 = Vov2 = Vovb = 0.2V.Given Vcm= 2V, we have
Vx = Vg1 Vgs1 = 2 1 = 1V.
To maximize the output swing, Vocm= 1 + (51)/2 = 3V, and
R = (Vdd Vocm)/I = 2/40 = 50k.
HKUST 2011 Fall 6 2 Ki
-
7/29/2019 6_OpAmp
3/21
ELEC 3400 VI. Operational Amplifiers
With Vocm = 3V, the highest Vx to keep M1 and M2 in the activeregion is 3 0.2 = 2.8V. The corresponding Vcm+is then
Vcm+ = Vx(highest) + Vgs1 = 2.8 + 1 = 3.8V.
Also, Vcm = Vx(lowest) + Vgs1 = 0.2 + 1 = 1.2V.The input common mode range is then 1.2V to 3.8V.
2. Diff. Pair with Resistive Loads: Large Signal Analysis
Suppose Vi stays constant and Vi+ increases. Then I1 increases and I2decreases (because I1 + I2 = Ib, a constant), driving the drain of M1down and the drain of M2 up. Therefore, Vd1 is the negative output
node Vo, and Vd2 is the positive output node Vo+.
1M
ddV
1R 2R
2MiV iV
oVoV
cmV
xV
bI
1I 2I
cmV
2/vid 2/vid
For normal operation, a differential input signal vid is applied acrossVi+and Vi, such that
Vg1 = Vi+= Vcm+ 2
vid
Vg2 = Vi = Vcm2
vid
and Vi+ Vi = vid
Clearly,
Vo+ = Vdd I2R
and Vo = Vdd I1R
HKUST 2011 Fall 6 3 Ki
-
7/29/2019 6_OpAmp
4/21
ELEC 3400 VI. Operational Amplifiers
The differential output signal is
Vo+ Vo = vod = (I1 I2)R.
At this point, we skip the computation details and only state the
results of I1 and I2 as follows:
I1 =2
Ib +2
Ib2
ov
id
V2
v1
ov
id
V
v
and I2 =2
Ib 2
Ib2
ov
id
V2
v1
ov
id
V
v
The differential current is
id = I1 I2 =
2
ov
id
V2
v1
ov
id
V
vIb
A typical plot is shown below.
0 5.0 20.1ovid V/v
b2,1 I/I
5.00.1
5.0
0.11I
2I
2
Observe that the transconductance of M1 and M2 is
gm1 = gm2 =tnGS
b
VV
)2/I(2
=
ov
b
V
)2/I(2
With vid
-
7/29/2019 6_OpAmp
5/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 5 Ki
and I2 =2
Ib gm12
vid
The differential current is thus
I1 I2 = id = gm1vid
Recall
vod = Vo+ Vo = (I1 I2)R
= gm1vidR
Adm =id
od
v
v= gm1R.
Hence, a linear amplifier is obtained. To increase the linear range ofthe input signal, Vov should be as large as possible.
3. Diff. Pair with Resistive Loads: Small Signal Analysis
The small signal model of the differential pair is shown below.
@vo: gm1(vi+ vx) +1ds
xo
r
vv +R
vo = 0 (1)
@vo+: gm1(vi vx) +1ds
xo
r
vv +R
vo = 0 (2)
@vx:R
vo +R
vo =dsb
x
r
v(3)
dsbr
)vv(g
xi
1m
R R
iv
ov ov
1dsr
)vv(g
xi
1m xv
iv
1dsr
-
7/29/2019 6_OpAmp
6/21
ELEC 3400 VI. Operational Amplifiers
(1) gm1vi+=1ds
o
r||R
v +1ds1
x
r||)g/1(
v(4)
(2) gm1vi =1ds
o
r||R
v +1ds1
x
r||)g/1(
v(5)
(4) (5) gm1(vi+ vi) =1ds
oo
r||R
vv
With vid = vi+ vi and vod = vo+ vo, the differential mode gain is
Adm =id
od
v
v=
ii
oo
vv
vv= gm1(R||rds1)
(3) in (4)+(5) gm1(vi++ vi) =1dsr||R
R
dsb
x
r
v+
1ds1
x
r||)g/1(
v2(6)
Note that Vi+and Vi are applied differentially, with Vi+= Vcm + vid/2and Vi = Vcm vid/2. With vi+= vid/2 and vi = vid/2, and (6) gives
vx = 0 (7)
and vo+ = gm1(R||rds1)2
vid (8)
vo = gm1(R||rds1)
2
vid . (9)
Equation (7) means that vx can be considered as the virtual ground.As such, the differential pair with resistive load can be analyzed usingahalf circuit.
where vo = 2
vod = gm1(R||rds1)2
vid
R
2
vv idi
ov
1dsri1vg
gndvx
Adm =id
od
v
v= gm1(R||rds1)
The input and output impedance can easily be found to be:Ri =
Ro = 2(R||rds1)
HKUST 2011 Fall 6 6 Ki
-
7/29/2019 6_OpAmp
7/21
ELEC 3400 VI. Operational Amplifiers
4. Differential Pair with Active Loads: Small Signal Analysis
The resistive loads can be replaced by PMOS transistors biased in theactive region. The gain will thus be enhanced. A useful structure is toreplace the resistive loads by a PMOS current mirror, and produce adifferential pair with a single-ended output. Note that Mb is relabeledas M5, and Vx as V1.
1M
ddV
oV
2MiV iV
5MbV
3M 4M
1V
2V
Principle of operation: For simplicity in symbol, vi is used instead ofvid. Now, a voltage of Vi+ = Vcm + vi is applied at the gate of M1,and a voltage of Vi = Vcm vi is applied at the gate of M2. I1 is thenlarger than Ib/2, and I2 is smaller than Ib/2. The increase in I1 issupplied by M3, which is then mirrored to M4. Therefore, at Vo, acurrent larger than Ib/2 is sourced by M4, and a current smaller thanIb/2 is sunk by M2. The excess current that appears at Vo drives Vo
high.
To analysis the gain of the differential amplifier, the small signalmodel is drawn, as shown below. Besides the gain A = vo/vi (= Adm),we also want to compute v1/vi, to verify that v1 can actually beconsidered as a virtual ground.
HKUST 2011 Fall 6 7 Ki
-
7/29/2019 6_OpAmp
8/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 8 Ki
2dsr1dsr
4dsr
5dsr
)vv(g
1
1m
3mg/1 24vg
)vv(g
1
2m
3dsr
1v
2v ov
Assumptions: gm1 = gm2, rds1 = rds2, gm3 = gm4, rds3 = rds4, gmirdsj >> 1v+= v = vi/2
Adm = vv
vo =i
o
v
v= gm1(rds2||rds4)
Due to the diode-connected transistor of M3, the left branch is not thesame as the right branch. With this asymmetry, the validity of usingthe half-circuit technique becomes questionable, and the small signal
analysis becomes quite tedious and not much insight can be gained inthe operation of the differential pair. However, observe that v1 =[1/(2gm1rds1)]vo is very small compared to vo. We may assume indeedv1 = 0, that is, by assuming that v1 is an ac ground. With thisassumption, the analysis is much simplified, and good approximatedresult can be obtained.
-
7/29/2019 6_OpAmp
9/21
ELEC 3400 VI. Operational Amplifiers
5. Gain Analysis using Current Steering Principle
By assuming that v1 is an ac ground, a simpler way to compute the
gain of the differential pair is to use the current steering principle,which account for steering the changes in currents due to the inputs
vi+ and vi to the change in current at vo, and multiply it with the
output resistance at vo.
Let Vi+ = Vcm +2
vid
1M
ddV
ov
2M
2
vid
5MbV
3M 4M
2
vid
2
gv 1mid2gv 1mid
2
gv 1mid
Vi = Vcm 2
vid
The increase in current of M1, M3 and M4 are i/2 = vidgm1/2, while
the decrease in current of M2 is i/2 = vidgm1/2. Hence, the changein voltage at the output node is
vo = iRo = vidgm1(rds2||rds4)and
Adm =id
o
v
v= gm1(rds1||rds3).
This is the same result as obtained using the small signal model and
making appropriate approximations.
HKUST 2011 Fall 6 9 Ki
-
7/29/2019 6_OpAmp
10/21
ELEC 3400 VI. Operational Amplifiers
6. Generic Two-Stage Operational Amplifiers
In ELEC 2200, we learned that an operational amplifier (op amp) has
a differential input and a single-ended output. One may suggest using
a differential pair with active loads as the op amp, but a simple
analysis would show that the output swing is very limited, as
discussed in pp. 62. Moreover, the gain of a single-stage amplifier is
not high enough. Even a gain of 500 V/V (54dB) is difficult to get. To
achieve a gain of 10,000 V/V (80dB), a two-stage design is needed.
A generic two-stage op amp is obtained by cascading a differential-
to-single-ended amplifier (1st
stage) with an inverting amplifier (2nd
stage). Depending on the load (resistive or capacitive), a third bufferstage may be added to deliver the power needed. To ensure stability, a
compensation network is needed (to be covered in ELEC 4420).
1A 2A 1
amplifierendedsingle
to-aldifferentistage1st
amplifiernvertingistage2nd
(optional)stagebuffer
oncompensati
V
VoV
For simplicity, let the buffer stage be left out. For each amplifier
stage, if the input impedance is zero and the output impedance
infinite, the gain is A = A1A2. In general, both the input and output
impedances are finite, and the gain is given by
Adm = vv
vo =
2i1o
2iRR
R
L2oL
RR
R
A1A2
The reduction in gain is known as the loading effect.
)vv(A1
1iR
1oR
2iR1o2vA
2oRv
v
ov1ov
LR
HKUST 2011 Fall 6 10 Ki
-
7/29/2019 6_OpAmp
11/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 11 Ki
For CMOS op amps, the input can be a pair of NMOS or PMOS
transistors. The NMOS case is shown below.
The V+ and V are referenced to the output of the whole op amp, not
the output of the differential amplifier. As an inverting amplifier is
cascaded to the differential stage, V+ is now at the gate of M2, and V
the gate of M1 (compare the connection shown in pp.6-10).
Although the output impedances of both the differential and inverting
amplifiers are finite, the input impedances are infinite, so no loading
effect occurs, and the gain is given by
Adm = vv
vo = A1A2
The compensation network with Cc and Rz is used to make the op
amp stable.
7. Transistor Sizing of 2-Stage Op Amp
Consider the 2-stage op amp with PMOS input transistors. In general,
Vtn |Vtp|, n |p|, nCoxpCox, and we define k = n/p (>1).
VV
dd
V
1M
3M 4M
5M
2M
7M
6M
zR
cC
1dsV
|V| 3gs
oV1V
bV
-
7/29/2019 6_OpAmp
12/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 12 Ki
The bias currents are related to the speed of the op amp, which will be
discussed in ELEC 4420. Let the bias current for the differential pairbe 2MI and that of the inverting amplifier be NI. The sizes of the
transistors are designed to have the same Vov. Therefore,
(W/L)3 = (W/L)4 = M
(W/L)1 = (W/L)2 = k(W/L)3 = kM
(W/L)5 = 2(W/L)1 = 2kM
(W/L)6 = N
(W/L)7 = k(W/L)6 = kN
If (W/L)4 : (W/L)7 M : N, then Vgs7 Vgs3, giving Vds3 Vds4, and
the two drain currents Ids3 and Ids4 will not be equal, thus causing an
offset voltage at the input, and a systematic error occurs.
Even with (W/L)4 : (W/L)7 = M : N, a systematic error still occurs,
because the output voltage Vo does not necessarily settle at Vdd/2
when vid = 0. The exact value of Vo can be computed approximately
by considering the bias current of the inverting amplifier NI:
NI =2
1 nCox7L
W
(Vgsn Vtn)
2(1 + nVdsn)
=2
1 pCox6L
W
(|Vgsp| |Vtp|)
2(1 + |pVdsp|)
ddV
1M 2M
3M 4M
5M
6M
7M
V V oV
bV
MI2 NI
MI NIMI
-
7/29/2019 6_OpAmp
13/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 13 Ki
With (W/L)6 = k(W/L)7 and Vgsn Vtn = |Vgsp| |Vtp| = Vov, we have,
1 + nVdsn = 1 + |pVdsp| nVo = |p|(Vdd Vo)
Vo =||
||pn
p
Vdd
The output offset voltage is thus
Vo(os) =||
||
pn
p
Vdd
2
1Vdd =
|)|(2
||
pn
np
Vdd
and the input offset voltage is
Vos =
dm
o
A
)os(V.
To finalize the exact value of L (and thus W), we may need to
consider noise issues (to be covered in ELEC 4420). A rule of thumb
for analog designs is to use 1.5 to 3 times the minimum length of the
process, for better matching accuracy, and smaller effect due to
channel length modulation. For example, if a 0.5 process is used, theminimum length is 0.5m (= 2), and we may choose L = 0.75m to1.5m.
Example: Compute the gain of the following 2-stage op amp.
V
ddV
oV
3M
1M
2M
aM b
M
bR
4M
V150 150
50
300
25
5M
6M
7M
30075
10050
A100
bI
zR
cC
LC
A50A50
25/4
cM
aR
A100
200 A 200 A
1V
-
7/29/2019 6_OpAmp
14/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 14 Ki
For the following computation, let
nCox = 100A/V2, pCox = 33.3A/V
2
n = |p| = = 0.05/V Vtn = |Vtp| = 0.8V
Vdd = 5V, Ra = 4k, Rb = 72kRecall in the active region, Vds > VgsVtn, and
Id =2
1 nCoxL
W(Vgs Vtn)
2(1 + nVds)
gm =
gs
d
V
I
=
tngs
d
VV
I2
doxn I
L
WC2
dsr
1
=ds
d
V
I
= Id
dsV1
rds dI
1
.
Bias Calculation
Remember when computing the DC biasing, the channel length
modulation parameters may be set to = 0/V, and it can be showneasily that
Ib = 50AVova = 400mV
Vovb = 200mV
and
5L
W
=6L
W
= 4c
W
L
Id5 = 450A = 200A
Id1 = Id2 = 200A/2 = 100A
gm1 = gm2 =d1
ov1
2I
V=
2 100
0.2
= 1m A/V
(Note that n = 3p, and all transistors are scaled according to thesame "current density" and should have the same Vov as Vovb).
rds1 = rds2 =1dI
1
=
1
0.05 100 = 200k
-
7/29/2019 6_OpAmp
15/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 15 Ki
DC Gain Analysis
The small signal model of the 2-stage amplifier is
Av1 =id
1
v
v
= gm1(rds2||rds4)
= 1m(200k||200k) = 100 V/V (=40dB)
Av2 =1
o
v
v= gm7(rds6||rds7)
gm6 = g m7 =d6
ov6
2I
V=
2 200
0.2
= 2m A/V
rds6 = rds7 =7dI
1
=1
0.05 200 = 100k
Av2 = 2m(100k||100k) = 100 V/V (=40dB)
Adm =id
o
v
v= Av1Av2 = 10,000 V/V (= 80dB)
The input resistance of the op amp is clearly
Rin =
and the output resistance is
Rout = rds6||rds7
id1m vg 2dsr 4dsr
1v
m6 1g v 6dsr 7dsr
ov
-
7/29/2019 6_OpAmp
16/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 16 Ki
8. Op Amp Circuits
8.1 Integrator
An op amp can be connected to do integration.
)t(vo)t(vi
R
C
)t(i
)t(vc
R
)t(vi = i(t) =dt
)t(dvC c
=dt
)t(dvC o
, vo(t) = dt)t(vRC
1i .
Hence, the output voltage is the integral of the input voltage. If vi(t)
is sinusoidal, i.e., vi(t) = Vicost, we may use phasor to analyze thecircuit.
oViV
R
sC
1
R
Vi =sC/1
V0 = sCVo
o
Vo =sCR
1 Vi
=CRj
1
Vi
In terms of transfer function, we write
H(s) =
)s(V
)s(V
i
o =
sCR
1
which can be expressed as
H(s) =o/s
1
=
s
o (o =CR
1)
or H(j) =o/j
1
= )90180( ooo
= oo 90
.
-
7/29/2019 6_OpAmp
17/21
ELEC 3400 VI. Operational Amplifiers
The corresponding Bode plot is
H
o
|H|log20
log
log
o90
dec/dB20
The magnitude plots of o/s and o/s are the same, but the phaseplots are different.
H
log
o90H
log
o90
so
so
N.B.(1) Recall in sinusoidal analysis, d/dt corresponds to j, then
written as s, i.e., d/dt s. So "s" represents differentiation.Now, 1/s can be regarded as the inverse operation of d/dt.Hence, 1/s is integration.
(2) The action of integration can be regarded as accumulating thecharge passed through R and storing it in C.
(3) In practice, the above circuit will not work since the gain at DCis infinite. Any noise at DC will drive the circuit to saturation.For the circuit to function properly, we need to put a largeresistor R2 across C to limit the gain. A common, but
erroneous, saying is that there is no DC path to ground at v.
(4) By adding R2 across C, the integrator is turned into a first order
low pass filter.
HKUST 2011 Fall 6 17 Ki
-
7/29/2019 6_OpAmp
18/21
ELEC 3400 VI. Operational Amplifiers
8.2 First Order Lowpass Filter
For the integrator with a resistor connected across the capacitor, wecan compute the corresponding transfer function.
)s(Vi)s(Vo
2R
C
1R
H(s) =)s(V
)s(V
i
o =1
sC1
2
R
||R
=2
2
1 sCR1
R
R
1
=
2ps
1
2
1
1
R
R
( p2 =
2CR
1) .
Clearly, this is a first order function, and the Bode plots are asfollows.
)(log
)(log
2p
|H|log20
H/
o90
o0
t
2p
o
180 o270
12R/R
2p1.0
2p10
t
In fact, we call it a first order lowpass function because signalbelow f2 (=p2/2) has approximately constant gain (R2/R1), while
signal at high frequencies are being attenuated. Hence, the circuitonly allows low frequency to pass through, and it is known as a 1storder lowpass filter. Also, f2 is known as the3dB bandwidth of the
circuit.
HKUST 2011 Fall 6 18 Ki
-
7/29/2019 6_OpAmp
19/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 19 Ki
Now, by changing the value of R2, we may adjust the DC gain of the
function. At the same time, the 3dB frequency will change. But if wemultiply the gain and the bandwidth together, we get
gain bandwidth = GBW =12
R
R
2CR
1=
1CR
1= t .
We call GBW thegain bandwidth product of the filter, and GBW isequal tot, theunity gain frequencyof the filter.
Observe that at high frequencies, especially for >>2 ,
H(s) 21
2
sCR
1
R
R =
1sCR
1 =
st
.
Hence, at high frequencies, the circuit functions as an integrator.
Next, refer to the transfer function of an integrator H'(s):
H'(s) =st
.
Since, H'(s) is the ratio of output over input, the correspondingmagnitude |H'(j)| can be regarded as the gain of the circuit at thefrequency (or f =/2), and we have
|H(j)| = t .or gain freq. = unity gain freq.
Hence, on the 20dB/dec line (the magnitude plot of an integrator),the product of the gain and the frequency is always a constant. Whenthe gain is unity, the corresponding frequencyt (or ft =t/2) is the
unity gain frequency. For the 1st order lowpass filter, the slopingportion is 20dB/dec, and so similar terminology is used.
)(log
s'pdifferent2
|H|log20
t
s'Rdifferent 2
1sCR1)s('H
-
7/29/2019 6_OpAmp
20/21
ELEC 3400 VI. Operational Amplifiers
HKUST 2011 Fall 6 20 Ki
8.3 Differentiator
Rearrange the position of R and Cof an integrator turns the circuitinto a differentiator:
i(t) =R
)t(v0 o =dt
)t(dvC c =
dt
)t(dvC i
, vo(t) =dt
)t(dvRC i .
If sinusoidal steady state is considered, then we should use phasor:I =
R
0 oV =)Cj/(1
iV
, Vo = jRC Vi .
In terms of transfer function, we have
H(s) =)s(V
)s(V
i
o = sRC =1p
s (p1 =
RC
1) .
The Bode plots are shown below.
The gain of this differentiator at high frequency has to be limited toavoid noise problem, and a small capacitor should be added as shown.
)t(vi )t(vo
R
C
cv
)t(i
1p1p1.0 1p10
dB20dB40
|H|log20
1p1p1.0 1p10o90
H/
)t(vi)t(vo
R1C
2C
12 CC
-
7/29/2019 6_OpAmp
21/21
ELEC 3400 VI. Operational Amplifiers
8.4 First Order Highpass Filter
The transfer function and Bode plots of a 1st order highpass filter are:
)s(Vi)s(Vo
R1C
2C
)(log
)(log
RC1
1
|H|log20
H/
o
90
o0
RC1
2
21C/C
o180
RC10
2RC1.0
2
H(s) = )s(V
)s(V
io = RsC1
RsC
21