Note 5: Current Mirrors

Rohit J. Braganza

Abstract This will be a quick note-sheet in our march towards the operational am-
plifier. There are many good resources online and in textbooks on current mirrors;
we will therefore focus more on the practical considerations, especially for BJTs.

1 The Diode-Connected Transistor

A transistor in the diode configuration becomes a two terminal device; in the case of
the BJT, it is actually a PN junction diode (convince yourself this is the case). BJTs
came before MOSFETs so we still call the MOSFET version diode connected, even
though the “diode” mosfet is a quadratic device, not exponential.

1.1 The Diode-Connected BJT

A diode connected BJT is essentially just a PN junction diode.

Fig. 1 The diode-connected

BJT. It provides a nice diode
drop across itself.

Rohit Julian Braganza

email: [email protected]

1
2 Rohit J. Braganza

This configuration is very handy for DC biasing, as it provides a diode drop (0.6-
0.9V), equivalent to a Vbe , across itself.

1.2 The Diode-Connected MOSFET

The MOSFET version is set up essentially the same way:

Fig. 2 The diode-connected

MOSFET. Note that it is
not an actual diode, since
the MOSFET’s Vgs has only
a quadratic dependence (at
best) with Id .

The MOSFET version is also a two terminal device, but not actually a PN diode.
It too is used often for DC biasing purposes, though it is a bit more tricky than the
BJT version. To find the output voltage (note it is the same as Vgs here, in Fig. 3),

Fig. 3 The diode-connected MOSFET, except used as a voltage source/biasing method, assuming
Id is held fixed.

using the square law assumption:

W µnCox
Id = Ire f + Iout = (VGS −VT n )2 (1)
L 2s
Ire f + Iout
Vout = VGS = VT n + 1 W (2)
2 L µnCox
Note 5: Current Mirrors 3

1.3 Output/Input Resistance of the Diode-Connected Transistor

Luckily the analysis is quick and easy in this case. We take the output to be the gate
or base of the transistor (the same node as the source/collector). Fig. 4 shows the
setup for the output impedance (same as the input). By observation:

Rout = Rs = 1/gm k ro ≈ 1/gm (3)

Notice that it has a low impedance- this is a good thing (as we will see later). This

Fig. 4 Output resistance

setup of the diode Connected
transistor.

holds true for both the BJT and the MOSFET.

2 The MOSFET Current Mirror

Fig. 5 The basic MOSFET

current mirror. Iout = (W /L)2
(W /L) Iin
1

Consider Fig. 5; if we apply some input current Iin to M1, then this current will
flow through the source of M1 to ground because the gate of M2 has infinite resis-
tance. This current will produce a suitable gate voltage (relative to ground) on M1,
to satisfy the I-V square law relationship of the MOSFET. Notice then that the gate
voltage of M1 is shared with the gate of M2. Therefore, M2 will have the same Vgs
drop, and therefore the same current (if its W/L is the same).
4 Rohit J. Braganza

(W /L)2
Iout = Iin (4)
(W /L)1

Notice as well that the input impedance is low: 1/gm1 , which is perfect for a current
input. Likewise notice that the output impedance is high: ro , which is great for a
current source.
Sources of Error It is important to note that for MOSFETs there is that pesky λVds
term. Without this (meaning infinite output resistance) we would have a perfect
current mirror. This factor however usually adds roughly 10-20% error. One way
to limit this is to increase the output impedance of the circuit (notice right now
it is ro ); this can be done for instance with a source resistor, but more commonly
for MOSFET another set of transistors is used (cascode current mirror). The only
drawback is the decreased voltage headroom allowed at the output (two pairs of
transistors, each with at least a Vov needed).
The Improved Cascode Current Mirror A simple cascode has limited voltage
headroom; I will therefore show as well the more popular improved version, which
can be used in low voltage supply applications.

Fig. 6 A: the cascode current mirror. B: the low voltage cascode version.

First note that the output resistance has been improved to approximately gm ro2 , while
keeping the same input impedance of 1/gm , for both of the two designs. The differ-
ence in comes in the minimum output voltage allowed.
For the basic cascode mirror in Fig. 6 (A), it can be shown (and is a great exercise
for the reader to do):
Vout,min = Vov1 +Vov2 +Vtn (5)
The threshold requirement is removed by the use of Fig 6 (B), which allows for:
Note 5: Current Mirrors 5

Vout,min = Vov3 +Vov4 (6)

This works so long as Vb is chosen appropriately to put M2 at the edge of saturation

while keeping M1 in saturation as well. This occurs if:

VGS1 + (VGS2 −Vtn2 ) ≤ Vb ≤ VGS2 +Vtn1 (7)

and Vov1  Vtn2 (8)

A fantastic discussion on the cascode current mirror and its improved version can
be found in [1]. It is a highly recommended read.

3 BJT Current Mirrors

The BJT current mirror operates by the same method as the MOSFET version, but
with a few key differences.
First and foremost is the issue of base-drive: a current mirror supplying large
amounts of current throughout a circuit may result in unacceptable base currents.
The basic emitter degenerated BJT mirror The first major difference in BJT
mirrors is that in almost all practical implementations, they must be used with emit-
ter degeneration. The reason is simple: Vbe (think “diode turn on voltage”) vari-
ation. This is caused by a variety of factors: temperature, process variation, and
mismatches1 . Due to the BJT’s exponential response however, even a tiny change in
Vbe causes significant error in the output. Emitter degeneration mitigates this by in-
stead making the current dependent on the magnitude of the emitter resistor, which
generally is less susceptible to process variation and mismatches2 . Neglecting the
finite base impedance, we can see by simple KVL (Fig. 7):

I1 R1 +Vbe1 = I2 R2 +Vbe2 (9)

I2 R2 = I1 R1 + ∆Vbe (10)
R1
I2 ' I1 if ΔVbe is small (11)
R2
Notice how ΔVbe no longer has an exponential effect on the current, with the resis-
tors setting the output current as expected. Also note as a bonus that emitter degen-
eration increases the output impedance of the mirror, though at the cost of voltage
headroom (extra IR drop across the resistor). Emitter degeneration is not used as
commonly for MOSFETs, since it is easier to just increase the length for increased
output resistance, and because they do not have the same exponential error BJTs
have.
1 The temperature dependence of a V is actually utilized for temperature sensors and temperature
be
independent biasing. This is covered in more advanced classes (DC biasing can actually be fun!).
2 Emitter degeneration is actually a type of feedback, with the emitter resistor setting the feedback

factor. As for the temperature dependence of the resistor? Well, resistances change with tempera-
6 Rohit J. Braganza

Fig. 7 The basic BJT mirror.

Iout = I1 RR21 , ignoring base
current.

3.1 The Improved BJT Current Mirror: Base Compensation

When large currents are mirrored, the base current of the BJT becomes a problem.
Imagine for instance trying to provide a total of 10mA from an input current of
100µA. Say h f e = 200. Then IB = 10mA/200 = 50µA. This is a problem, since
this base current must come from the input current (and here we would lose half
the input current)! What Q3 does therefore is supply the required large base current
for us; essentially, it increases the equivalent “β ” of the circuit to β 2 (because Q3
requires some base current as well; it would be 50µA/β in the previous example).
This removes the error from the base current.

Fig. 8 The complete BJT

current mirror, with Base
compensation transistor Q3
and emitter degeneration.

ture, and it turns out this can be manipulated to help the analog designer (There really is a lot to
DC biasing!).
Note 5: Current Mirrors 7

3.2 BJT Cascode Version

While the same MOSFET cascode configurations can be used for BJT implemen-
tations, there are some useful tweaks made to improve the performance for BJTs.
Shown in the figure below is the Full Wilson Current Mirror.

Fig. 9 The Full (4 transistor) NPN Wilson Current Mirror [4]

It can be shown (see the Analog Devices reference, [3]) that this configuration
significantly reduces any base current mismatches that may occur. It also has a high
output impedance due to the negative feedback from Q3, through Q1, and eventually
back to the base of Q3.

3.3 The Simplest Current “Source” and Mirror

The big question of course is where does the reference current come from? Let’s
consider the simplest (and poorest) method: just sticking in a resistor3 . Notice that
in this circuit, the output current is now a function of the supply voltage. This is
actually terrible: batteries for instance change over 50% over their lifetime, before
“dying.” A customer would be very mad if his product lasted only a week! In inte-
grated circuits, supplies also have a tremendous amount of noise (from the rest of
the chip)- you don’t want this affecting your bias currents!

3 This is not even considered a current source by most textbooks since it has none of the properties
a good current source should have, besides creating a current.
8 Rohit J. Braganza

Fig. 10 Take One: Sticking in a resistor. Notice the output current is directly related to Vdd and
R1!

3.4 Practical Supply-Independent Biasing: The Base-Emitter

Referenced Current Source

For almost all circuits, Figure 10 just isn’t going to cut it. Instead, consider the
circuit shown below, in Figure 11.

Fig. 11 The Vbe Current Reference. The current depends on the Vbe of Q1. The feedback provided
by Q2 provides the additional benefits of increased output impedance and less variation in output
current from supply and mismatch variations. Instead of ground, use VSS.

To understand how this circuit works, first recognize that Q2 must supply enough
current to allow Q1 to operate; for low supply voltages therefore, this circuit will
not work. Neglecting base currents, and defining the current through R1 as IIN , we
know that:
IIN
Vbe1 = VT ln( ) (12)
IS1
Note 5: Current Mirrors 9

We can then see that the output current is set by the voltage drop across R2 , set by
Q1 to be Vbe1 .

Vbe1 VT IIN
IOUT = = ln( ) (13)
R2 R2 IS1
Vdd − 2Vbe
IIN = (14)
R1
Note that because the output current is proportional to the log of Vdd, we have
significantly decreased the affect of the supply!

3.5 What about MOSFET Current Sources?

There are CMOS current sources. Replacing the BJTs in the Vbe Current Refer-
ence with MOSFETs results in the threshold voltage current reference. However,
the equations are complicated, and even worse, for MOSFETs, threshold voltages
are very imprecise and vary all over the place. Building a reference current based
on a value you have no idea about is never a good idea! Most CMOS references ac-
tually use the parasitic BJTs that are intrinsic in the MOSFET’s structure to create
V be reference circuits! MOSFET implementations of these circuits will be further
explored in later classes, in addition to the temperature and process dependencies of
biasing.

3.6 Real Current Sources: Bootstrapping!

Previously we had looked at the effect of the power supply on the input current. We
define an important term to relate the effect of the power supply on the output,
known as the PSRR, or Power Supply Rejection Ratio. For current mirrors and
sources, PSRR is defined as the change in output current due to a change in the
supply voltage (for op-amps, it is the change in output voltage). Notice that while
the Vbe referenced source mentioned previously had a fairly good PSRR, the output
current still depended slightly on the power supply due to the input resistor. To
almost fully remove the dependence on the power supply, a technique known as
self-biasing or bootstrapping is used.
In a self biased circuit, the input current depends on the output current itself (and
vice versa!), completely removing the power supply from the equation. Too see how
this works, look at figure 12, from Gray & Meyer- we can set the input current equal
to the output current by the use of a current mirror! We know the current source
will have some type of logarithmic input-output current behavior (as seen from the
Vbe reference), and we know that the current mirror demands a 1:1 input to output
ratio. As seen in figure (b), this means that the intersection of these two equations
10 Rohit J. Braganza

represents the operating point of the circuit! Note that these circuits require a special
start-up circuit. This is because a bootstrapped circuit will otherwise happily have
its input and output currents both be zero.

Fig. 12 (a): The concept of self biasing. (b) How to find the two possible operating points [2].

Below we present an example bootstrapped current source, in figure 3.6. Note

that this circuit is a bit different than the Vbe reference used in 2.4. Instead, this
is known as a VT reference. To see how it works, consider transistors Q1 and Q2.
First assume that n=1 and ignore the Wilson current mirror Q4-Q7. We know that
for a BJT Ic = Is eVbe /VT , ignoring the Early effect. Using this, (and ignoring the base
current) derive the following transcendental equation (hint: find the base voltage due
to Iin first):
Iin Is1
Iout R1 = VT ln (15)
Iout Is2
Now we can consider the effect of the mirror and the multiple emitters of Q1. In
general, multiple emitter BJTs act as a single device, except that the emitter cur-
rent is split evenly between the two emitters; basically the base drive sets the total
collector current, which is then split between both emitters. Here however the emit-
ters are all connected to R1. The trick is that adding an extra emitter increases the
reverse-saturation current (remember from diodes this current is proportional to the
area). Because we have n emitters in Q1, Is1 = nIs2 . Then, using the fact that the
mirror forces Iin = Iout , and plugging into 15 we have the final result:
VT
Iout = ln n (16)
R2
Now we will explain the startup circuit. At time equals zero, assume all nodes ex-
cept the supply have zero voltage. We will therefore have a current flowing through
R2. Some of this current will flow into the base of Q3 and turn it “on”. Q3 will
then force a current through Q7 and Q5, which is mirrored by Q6 and Q4 into Q2,
Note 5: Current Mirrors 11

Fig. 13 A basic self biased current source with a start up circuit.

turning on the entire circuit! Once on, notice that Q8 turns on as well, which pulls
current through R2 and then turns off Q3!
This circuit will turn up many times in your analog career. In later classes, you
will discover that (with some tweaks) this circuit can be used to create what is known
as a bandgap reference, which creates a reference voltage equal to the bandgap of sil-
icon across all temperatures (as a preview, notice that the thermal voltage is propor-
tional to temperature, whereas resistors are inversely proportional to temperature.
With some tweaks can theoretically tune these temperature coefficients to cancel
each other out).

4 Summary

In this notesheet we have observed some common current mirrors, which will con-
tinue to show up in many op amp examples. This notesheet was purposely short
(lest the reader become too tired of my writing) due to the many resources already
available . Some useful references are of course the ubiquitous Gray & Meyer [2],
and [3]. Note that I chose not to describe current sources and temperature effects
12 Rohit J. Braganza

since they are best left for more advanced classes and notes (despite their interest-
ing nature).

References

1. Razavi, Behzad. Design of Analog CMOS Integrated Circuits, 2nd Ed. pp. 138-146.
2. Gray, Hurst, Lewis, & Meyer (2001). Analysis and Design of Analog Integrated Circuits. New
York: Wiley.
3. Analog Device’s Educational Content https://wiki.analog.com/university/
courses/electronics/text/chapter-11
4. Wikipedia Wilson Current Mirror. https://en.wikipedia.org/wiki/Wilson_
current_mirror#/media/File:WilsonModified4.gif