Design of Analog Integrated Circuits

Chap. 5 Passive and Active Current Mirrors

1
 Overview

Reading
– Textbook Chapter 5.

Introduction
In analog circuits current sources acts as a large resistor without
consuming excessive voltage headroom. This lecture deals with the
design of current mirrors as both bias elements and signal processing
components. Following a review of basic current mirrors, we study
cascode mirror operation. Next, we analyze active current mirrors and
describe the properties of differential pairs using such circuits as loads.

2
 Basic current sources

Application

Definition of current by resistive divider

Vb = VDD ⇒ Vgs =
R2 R2
R1 + R2 R1 + R2
VDD

⇒ I out = µ n Cox (Vgs − VTH )

1 W 2
Vb 2 L
 R2  
2

⇒ I out = µ n Cox 
  − TH 
1 W
DD 
 R1 + R2
V V
2 L  

3
 Basic current mirrors

Basic current mirror

Neglecting channel-length modulation, we can write

W 
I REF = µ nCox   (VGS − VTH )
1 2

2  L 1
W 
I out = µ n Cox   (VGS − VTH )
1 2

2  L 2
obtaining
(W / L )2 I
(W / L )1 REF
I out =

4
 Basic current mirrors (cont’d)

Currents mirrors used to bias a differential amplifier

0.5IREF
2IREF 2IREF 2IREF
5IREF

Current mirrors usually employ the same length for all of the transistors so as
to minimize error due to the side-diffusion of the source and drain areas (LD).
Furthermore, the threshold voltage of short-channel devices exhibits some
dependence on the channel length. Thus, current ratioing is achieving by only
scaling the width of transistors.

5
 Basic current mirrors (cont’d)

Consider channel length modulation (λ≠0)

W 
I D1 = I REF = µ nCox   (VGS 1 − VTH ) (1 + λVDS 1 )
1 2

2  L 1
W 
I D 2 = I out = µ n Cox   (VGS 2 − VTH ) (1 + λVDS 2 )
1 2

2  L 2
obtaining
I out I D 2 (W / L )2 1 + λVDS 2
= = ⋅
I REF I D1 (W / L )1 1 + λVDS1
While VDS1 = VGS1 =VGS2, VDS2 may not equal VGS2 because that
the circuitry fed by M2 will affect the output voltage. That is,
M2 may be controlled in saturation or in triode region.
Voltages are not same.

We wish to obtain VDS1=VDS2 to delete the effect of channel length

6
 Example 1

Calculate the current gain?

I out (W / L )2 (W / L )4
I REF (W / L )1 (W / L )3
= i

Calculate the small-signal gain?

(W / L )3 V (W / L )3
(W / L ) 2 (W / L )2
Vout = g m1Vin RL ⇒ out = g m1 RL
Vin
gmVin

7
 Cascode current source

Scheme – suppress the effect of channel-length modulation

VGS0+VX=Vb

How do we generate Vb in (a) to ensure VY =VX ? (suppress channel-length effect)

Proper choice the dimensions of M0 with respective to those of M3 yields
VX + VGS 0 = VY + VGS 3 ⇒ VGS 0 = VGS 3
Thus, if the aspect ratio is made as
ID2 (W L )2 (V GS 2 − V TH )2 1 + λ V DS (W L )2
(W L )1 (V GS 1 − V TH )2 (W L )1
= × 2
=
I D1 1 + λ V DS 1

I out (W L )3 (V GS − V TH )2
1 + λ V DS (W L )3
= × =
(W L )0 (V GS )2 (W L )0
3 3

I REF − V TH 1 + λ V DS

(W / L )3 (W / L )2 ,
0 0

I out I D 2
(W / L )0 (W / L )1
⇒ = ⇒ =
I REF I D1
then VGS2 = VGS1 ( VX = VY ), VGS3 = VGS0 , VDS2 = VDS1 and VDS3 = VDS0.
8
 Cascode current source (cont’d)

Voltage headroom consumed by a cascode mirror
VP , min = 2(VGS − VTH ) + VTH
VP ,min = 2(VGS − VTH )

VX ≠ VY VX = VY

The minimum allowable voltage at node P for (b) is equal to

VP,min = VN − VTH = VGS0 + VGS1 – VTH
= (VGS0 – VTH) + (VGS1 – VTH) + VTH
In (a), Vb is chosen to allow the lowest possible value of VP but the output
current does not accurately track IREF because of VDS1 ≠ VDS2. In (b), higher
accuracy is achieved but the minimum level at P is higher by one threshold
voltage. [(a) has lower headroom than (b)]

9
 Cascode current source (cont’d)

Operation of cascode current mirror

M3 Triode Triode Sat.

M2 Triode Sat. Sat.

M 3 = Sat . ⇒ V DS 3 ≥ VGS 3 − VTH 3

⇒ V X ≥ V N − VTH 3
Sat. 1
=
∆ I X
(1) ∆ V X
∵
R
Triode (2) (1 ) R ≈ ∞
(2 ) R ≈ M Ω
(3) (3 ) R ≈ k Ω

M 2 = Sat . ⇒ VDS 2 ≥ VGS 2 − VTH 2

⇒ VB ≥ VA − VTH 2 ⇒ V X ≥ VDS 3 + V A − VTH 2

10
 Example 2

If IREF requires 0.5V to operate as a
+ current source, what is its maximum
0.5V

value?
(W / L)1 = (W / L) 2 , (W / L)0 = (W / L)3

VX = VY ≈ 2 I REF /[ µ nCox (W / L)1 ] + VTH 1

2 I REF  W  W 
VN = VGS 0 + VGS 1 = [   +   ] + VTH 0 + VTH 1
µ nCox  L  0  L1

2 I REF  W  W 
VDD − ( [   +   ] + VTH 0 + VTH 1 ) = 0.5V
µ n Cox  L  0  L1

µ n Cox (VDD − 0.5 − VTH 0 − VTH 1 ) 2

I REF ,max =
2 ( (W / L) 0 + (W / L)1 ) 2

11
 Low-voltage cascode mirror

Modification of cascode mirror for low-voltage operation
VP, min = VGS4 +V GS3−VTH 4 − VTH 3
= 2(VGS − VTH )

M1, M2 are in saturation: If Vb = VGS2 + ( VGS1 − VTH1) = VGS4 + ( VGS3 − VTH3), then the
cascode current source M3-M4 consumes minimum headroom
M2: Vb − VTH2 ≤ VX (= VGS1)
while M1 and M3 sustain equal drain-source voltages, allowing
M1: VGS1 − VTH1 ≤ VA (= Vb − VGS2) accurate copying of IREF. That is
⇒ VGS2 + (VGS1 − VTH1) ≤ Vb ≤ VGS1 + VTH2 VDS 4 ≥ VGS 4 − VTH 4 ⇒ VP ≥ Vb − VTH 4
⇒ VGS2 − VTH2 ≤ VTH1⇒ VGS2 ≤ VTH1+ VTH2. ⇒ VP ,min = VGS 4 + VGS 3 − VTH 3 − VTH 4
( Lower than simple cascade current mirror about VTH)

12
 Low-voltage cascode mirror (cont’d)

Generation of gate voltage Vb for cascode mirror

Iout M1, M2 are in saturation:

Vb ,min = VDS 6 + VGS 5 = VGS 2 + (VGS1 − VTH 1 )
Select: VGS5 ≈ VGS2, VDS6 = VGS6 − RbI1 ≈ VGS1 − VTH1.

Owing to body effect, VGS5
VGS2 and Iout
I1

Important Concept!

( L) (V
M7: large (W/L)7, so that VGS7 ≈ VTH7
Iout [where I D 7 = 1 µ nCox W
2 7
GS 7 − V )2
TH 7 ]

Then
VGS6 ≈ VDS6 + VTH 7 ⇒ VDS6 ≈ VGS6 − VTH 7
⇒ Vb − Vs6 = Vb − VGS5 ≈ VGS6 − VTH 7
⇒ Vb ≈ VGS5 + VGS6 − VTH 7
Owing to body effect, VGS5
VGS2 and Iout
I1
13
 Low-voltage cascode mirror (cont’d)

Low-voltage cascode using a source follower level shifter

If MS is biased at a very low current density, ID /(W/L), then
VGSS ≈ VTHS ⇒ VN − VN ' ≈ VTHS ⇒ VN ' = VN − VTHS
and
VN = VGS 0 + VGS1
VN ' = VGS 3 + VB
Then
VB = VGS0 + VGS1 − VTHS − VGS3 = VGS1 − VTHS
implying that M2 is at the edge of the triode region.

In this topology, however,
I out (W L )2 1 + λVDS 2
 VDS 2 ≠ VDS1 ⇒ = ×
I REF (W L )1 1 + λVDS1
 If the body effect is considered for M0 , MS and M3, it is
diffifficult to guarantee that M2 operates in saturation.
 The cascode structure also provides a higher output
resistance.
14
Low-voltage cascode mirror (cont’d)

Overdrive voltage Voverdrive = VGS − VTH = VON

I D = kVON 2 For a constant ID, VON ~1/
k

15
 Active current mirrors

Current mirror processing a signal

M1 and M2 are identical: Iout = Iin (for λ = 0)

Active: Current mirrors can also process signals,

operated as “ Active” elements.

Differential pair with current source load

16
 Active current mirrors (cont’d)

Differential pair with current source load (cont’d)

Calculate Gm Calculate Rout

 
Assuming ro = 0
  1 
Rout1 = ro2 +1+ × gm1 
ro2
   
Iout
( + )
1
 mb2 
Gm = g g
Vin V
= ro2 + (1+ gm2ro2 )(1/ gm1)
m2
out =0

= 2ro2 +1/ gm1

+

gm1Vin / 2
=
Vin ≈ 2ro2
gm1 Thus,
=
2 Rout ≈ (2ro2)| | ro4
1 Av = Gm Rout and
V in
2 Av = − Gm × Rout

≈
g m1
[(2ro2 ) ro4 ]
2
17
 Active current mirrors (cont’d)

Calculate Vp/Vin
Calculate Vout/Vp
From eq. (3.110)
1  ro 4 
From eq. (3.104)
Req ≈ + o4 = 1 + 
1 r
g m 2 g m 2 ro 2 g m 2  ro 2  Vout 1 + g m 2 ro 2
=
VP r
ro 4 1 + o2
1+ ro 4
VP Req ro 2
= ≈ g m 2 ro 2
Vin R + 1 r
2 + o4 ≈
eq
g m1 ro 2 r
1 + o2
ro 4
Note: if ro4 → 0, Vp/Vin → 1/2, and if ro4 → ∞, Vp/Vin → 1.

Calculate Vout/Vin
ro 4
1+
Vout VP Vout
= ⋅ =
ro 2 g m 2 ro 2 g m 2 ro 2 ro 4 g m 2
ro 4
⋅ =
ro 2 2ro 2 + ro 4
= [(2ro 2 ) ro 4 ]
Vin Vin VP 2+ 1+ 2
ro 2 ro 4
Regardless with M1
The drain current of M1 is “wasted”.

18
 Differential pair with active current mirrors

Concept of combining the drain currents of M1 and M2

M3 and M4 are identical.

−g
= ( m 2 − m1 )  ro 2 ro 4 
Vout g
Vin 2 2
Vout
⇒ ro 2 = ro 4
Vin max

It converters a differential input into a single-ended one.

19
 Differential pair with active current mirrors
Large-signal analysis

Large-signal analysis

Operation:
 If Vin1 << Vin2, M1 is off and so are M3 and M4.
M2 and M5 operate in deep triode region, carrying
zero current. Thus, Vout = 0.
 As Vin1 approaches Vin2 for a small difference, M2 and
M4 are saturated, providing a high gain.
 As Vin1 becomes more positive than Vin2, ID1, |ID3|,
|ID4|, and |VGS4| increase and ID2 decreases, eventually
driving M4 into the triode region.
 If Vin1 >> Vin2, M2 turns off, M4 operates in deep triode
region with zero current, and Vout = VDD.

The choice of the input common-mode voltage:
For M2 to be saturated, Vout ≥ Vin,CM − VTH. Thus, to allow
maximum output swings, the input CM level must be as
low as possible, with Vin,CM, min = VGS1,2 + VDS5, min.

20
 Differential pair with active current mirrors
Small-signal analysis

Small-signal analysis

Asymmetric swings in a differential

pair with active current mirror

Calculate Gm

Node P can be viewed as a virtual ground.


= = =
g m1Vin
 D1
I I I

D3 D4
2
 I D 2 = − g m 2Vin
|ID4|


ID1
ID2
2
⇒ I out = I D 2 − I D 4 = − g m1,2Vin
∴ Gm = g m1,2

21
 Differential pair with active current mirrors
Small-signal analysis (cont’d)

Calculate Rout

VX V
IX ≈ 2 + X
1
2ro1,2 + // ro3 ro 4
g m3
where the factor 2 accounts for current copying action of M3 and M4.
For 2ro1,2 >> (1/gm3)//ro3, we have
Rout ≈ ro2 // ro4

Calculate Av
| Av | = GmRout = gm1,2 (ro2 // ro4)

22
 Differential pair with active current mirrors
Small-signal analysis (cont’d)

Substitution of the input differential pair by a Thevenin equivalent

Calculate Veq and Req

Veq = gm1,2 ro1,2 Vin
Req = 2 ro1,2 +
  − Vin   r
VDS 2 = −  g m 2 × 
  2   ( o 2 )

= g m 2 × in ×
V
ro 2
2
23
 Differential pair with active current mirrors
Small-signal analysis (cont’d)

Calculate Av = Vout / Vin

The current through Req is
Vout − Veq Vout − g m1, 2 ro1, 2Vin
IX1 = =
  || r 
Req +   1  o3  2ro1, 2 +
1
 m3  
// ro 3
g g m3

IX2 The fraction of this current that flows through 1/gm3 is mirrored into M4
with unity gain. That is,
IX1 Vout − g m1,2 ro1,2Vin Vout − g m1,2 ro1,2Vin 1 V
+ × ( || ro 3 ) g m 4 + out = 0
1 1 gm3 ro 4
2ro1,2 + || ro 3 2ro1,2 + || ro3
g m3 g m3
Assuming 2ro1,2 >> (1/gm3,4) // ro3,4 and ro3,4>> (1/gm3,4), we obtain

= g m1, 2 (ro1, 2 // ro3, 4 )

Vout g m1, 2 ro3, 4 ro1, 2
=
Vin ro1, 2 + ro3, 4

24
 Example 3

= g m1,2 ( ro1,2 // ro3,4 ) = − g m 2 ( ro1 // ro 2 )

Vout Vout
Vin Vin

For given device dimensions, it require half of the

bias current compared to a differential pair.

However, the differential operation often outweighs
the power penalty.

25
 Differential pair with active current mirrors
Common-mode properties

Differential pair with active current mirror sensing a common-mode change

The CM gain is defined in terms of the single-ended

output component produced by the input CM change:
∆Vout
ACM =
∆Vin ,CM

26
 Differential pair with active current mirrors
Common-mode properties (cont’d)

Simplified circuit of CM circuit

1 r
// o3,4
RD 2g 2
ACM = − ≈ − m3,4
( gm1 + g m2 )
1 + RSS 1
+ RSS
2 g m1,2
1 g
=− ⋅ m1,2
1 + 2 g m1,2 RSS g m 3,4
where we have assumed 1/(2gm3,4) <<
ro3,4 and neglected the effect of ro1,2/2.

Even with perfect symmetry, the output signal is corrupted by input CM variations,
a drawback that does not exist in the fully differential circuits.
RSS affects ACM. As the capacitance shunting the tail current source exhibits a
lower impedance ACM increases and degrade the performance.

CMRR g (1 + 2 g m1, 2 RSS )

= g m1, 2 (ro1, 2 // ro3, 4 ) m3, 4 = (1 + 2 g m1, 2 RSS )g m 3, 4 (ro1, 2 // ro3, 4 )
ADM
CMRR =
ACM g m1, 2
27
 Differential pair with active current mirrors
Mismatch

Differential pair with gm mismatch
Considering M 1 and M 2 as a single transistor with g m = g m1 + g m2,
RSS
∆ V P = ∆ Vin ,CM
1
R SS +
g m1 + g m 2
where body effect is neglected. The change of ID1 and ID2 are given by
∆ Vin ,CM
∆ I D1 = g m1∆ VGS 1 = g m1 (∆ Vin ,CM − ∆ VP ) =
g m1
1 g m1 + g m 2
RSS +
g m1 + g m 2
∆ Vin ,CM
∆ I D 2 = g m 2 ∆ VGS 2 = g m 2 (∆ Vin ,CM − ∆ VP ) =
gm2
1 g m1 + g m 2
RSS +
g m1 + g m 2
And
 1 
∆ I D 4 = g m 4 ∆ VGS 4 = g m 4  // ro 3  ∆ I D1
 g m3 

28
Differential pair with active current mirrors
Mismatch (cont’d)

Neglecting the effect of ro1 and ro2, and gm3=gm4:

∆Vout = (∆I D 4 − ∆I D 2 )ro 4
 
 g m1∆Vin ,CM g m 2 ∆Vin ,CM 
= −  ro 4
ro 3
 1 + ( g + g )R SS r +
1 1 + ( g + g ) RSS 
 
m1 m2 m1 m2
o3
gm3
∆Vin ,CM ( g m1 − g m 2 )ro 3 − g m 2 / g m3 r
=
1 + ( g m1 + g m 2 )RSS ro 3 +
o4
1
g m3

If ro3 >> 1/gm3, we have

∆ Vout ( g − g m 2 )ro3 − g m 2 / g m 3
≈ m1
∆Vin ,CM 1 + ( g m1 + g m 2 )RSS
(gm1 − gm2)ro3 reveals the effect of the transconductance mismatch on
the common-mode gain.

29
 HW#6

P. 5.6
P. 5.8
P. 5.9 for Fig. 5.33(d)
P. 5.10 for Fig. 5.34 (a)
P.5.14
P. 5.15

30