Differential Amplifier Design

The underlying theory about current mirrors etc is skipped but might be added at some point.

Intro to Differential Amplifier Design

Differential amplifiers amplify a voltage difference at the input and are fundamental circuits in electronics. They are used in operational amplifier circuits and when ever a differential signal need to be amplified. Differential signals do have the advantage that they are immune to crosstalk and noise as those influence have no impact on the difference of this voltage levels. see SRAM technology.

Differential Amp Resistors



The principle is that a rising will reduce the voltage at . This will reduce the voltage over and will lead to a lower current in the right path which is causing  to increase. We need a Resistor  to make sure that both Transistors stay in Saturation Region.


Key Parameters

The Overall gain consists of 2 parts, the differential gain  and the Common mode gain . Since we are interested to build up an differential amplifier, the common mode gain should be as small as possible.

The differential gain defines how much the voltage difference between  and  will be amplified whereas the common mode gain amplifies the input voltages itself (what we don't want to happen):

Common Mode Amplification

We can calculate the amplification if we connect both input pins together. As both Transistors and Resistors now work in parallel they the resistance is half of the original value whereas the transistors become twice as strong which means that we can double the Width/Length ratio.

Common Mode Amplification


As you can see this is now a common Source stage.
The gain for a CS stage is
from which we can derive the formula for our amplification.

You can see from that equation that an infinite would lead to no common mode amplification as it would be an ideal current source.

Input Common Mode Range

As it had been discussed above the amplifier just works around an operating point. The range in which the amplifier works is defined as input common mode range (ICMR). We have linear dependency between the output voltages until one of the transistors leaves saturation. The INput common mode range is ~ 50% of


Transistors are not ideal because of several reasons during manufacturing therefore an 2 equal input values must not result in equal output voltages. We therefore add an Offset Voltage to compensate this problem. is usually in the range of 5-20mV.

Real amplifiers

In real live the difference of the input voltages is amplified and a voltage which is proportional to this difference is generated.

Differential Amplifier Symbol


Note that is limited to as upper bound and  as lower bound.

We also want to be able to amplify "negative" voltages, therefore we lower  to a negative value. We can achieve the same effect by adding an offset to the input voltages. However, if we do so we generate a new reference voltage, analogue ground which is 0V and in the middle between  and .

Making it single ended

We could change the circuit from fig1 in such a way that we just ignore one of the output voltage. The Problem with that is that both output voltages "move" in the opposite direction when the input voltage difference changes. By losing one of this voltages the gain would drop by 50%.

A better solution is to use a current mirror as pull up network.

Differential Amplifier Design


Lets say   is rising:
The current in  ) is increasing while the current in ) is decreasing by the same value.
But at the same time the voltage over is decreasing which is pulling gate node of the current mirror to ground. This will cause the currents in and to get bigger which is also causing the output voltage to rise.

Also you can see that I have replaced by a NMOS. The reason is that resistors are difficult to manufacture on a chip. The bias voltage determines the "resistance" of this pull down network.


The current in is the sum of the currents that go through and


The small signal gain may be obtained from

where is the transconductance of the differential inputs.

where is determined my the and

we obtain (explanation on factors might be added in the future)

Input Voltage Range

As mentioned earlier and need to be in saturation, therefore the input voltages need to be in a certain range.

sweep the common mode voltage (inputs connected) [will be simulated later]

We have to find the highest and lowest possible input voltage for which and are in saturation.

For a single transistor we know that we need to fulfill to be in saturation region (and of course)

Maximum voltage

The maximum common mode voltage is the same as the gate voltage at . It is minus the voltage drop over Source to gate plus the threshold voltage of .

Minimum voltage

determines the minimum voltage as also has to stay in saturation.

Slew Rate

The Slew rate determines how fast the output can charge or discharge a capacitive load at the output . All internal parasitic contribute of course also to the Slew rate. However, defines this attribute


We dither a constant input voltage (which is inside the ICMR) with a AC voltage. The 3dB frequency is the frequency of the AC voltage at which the output signal of this AC is only 70% percent. It is determined by the output resistance.

Power dissipation

The power consumption is the voltage drop over the full circuit times the current in


Example Design

This example will be based on a 0.8µm Technology for which we have

 Parameter Description NMOS PMOS Unit
  Threshold Voltage      
  Channel-length-modulation factor
  Surface potential for strong
Design Targets

Supply Voltage:

Slew Rate:

3dB frequency

Input common mode range

Differential Amplification

Power Dissipation


The Slew Rate tells us how fast the load need to be charged/uncharged. As is given (5pF) we can calculate the minimum current (note that the Slew rate is given as Volts per µs and therefore needs to be multiplied)

The Power dissipation is giving us the upper bound for

If we also take the channel length modulation factors into consideration we have

And by having knowledge about the 3dB frequency we can calculate

With knowledge about   we can set a lower bound which is considering the channel length modulation factor.

We now know

To have some margin we set .

Designing the current mirror (ICMR max)

We need to determine the saturation current of (and therefore also ) to make sure that we have half of balanced on each side

we know that


Now we can determine the size of and , the current is

Note that this is an PMOS device and that 

Designing the Pull down transistors M1 and M2

From the gain specification and from the knowledge about the currents we now can calculate the W/L ratio for M1 and M2.


From the ICMR we can calculate

We first need to know

W/L ratio for

for bias current mirror



I use LTSPICE for the simulation.

Offset and transfer

First I want to check if the offset voltage and the transfer characteristics.
We set the negative input to analog 0V and sweep the positive input.


Plot 1 V(3)

From the plot you can see that the supply voltage range is used from 0,9V to 3,1V
We do have an offset but as we are in the linear region this can be fixed by adding an offset voltage of about 3,5mV to the negative input.

It also can be seen that a 0,001V change of the input voltage results in a 0,2V difference at the output, so we have met our design goal for an Amplification factor of 200.

ICMR with unity gain configuration

Not we want to check the range of operation if we connect the output to the negative input and sweep the positive input.

plot 2 ICMR

We can see that we have met our design goal of 

Bode Plot

In this configuration we want to see at which frequency the AC is not amplified anymore.

Plot 3 Bode Diagram

Plot 3 Bode Diagram

We have a -3dB frequency of around 355kHz which is better that the requirement of 200kHz so we also met this criteria.

Slew Rate

How fast can the capacitive load be charged and uncharged?


We can see that the leading and the trailing edge need ~30ns to reach the desired voltage, the slew rate therefore is (the is needed to bring it to the scale

so also the slew rate is better than our requirement.

Leave a Reply