Bad Circuit Design 12 - Using RC Networks and Miller Compensation

As discussed in Bad Circuit Design 3 it's bad design to connect a compensation capacitor back to a high

impedance node when trying to compensate an op-amp. In other words Miller compensation is, generally,

bad design. One may think, however, that by adding some RC networks here and there in an op-amp the 

resulting additional poles and zeros can be placed so that Miller compensation can still be the better 

design choice (that is, better than indirect compensation.) Using Miller compensation, even with the 

addition of RC networks, is generally bad design. 

The step response of the op-amp seen in Fig. 24.8 with Rz = 1/gm = 6.5k and Cc reduced to 240 fF driving 

a 10 pF load is seen below. This is a simple two-stage Miller compensated design with zero-nulling resistor 

and it's bad design.

Below is the step response of the op-amp seen in Fig. 24.21 driving, again, a 10 pF load. This is a simple 

two-stage indirect compensated design. 

The layout size of each of these op-amps is roughly the same (the design above is slightly larger because 

of Rz). The power consumption is precisely the same. Making the Rz track 1/gm over process, voltage, and 

temperature (PVT) shifts is a practical concern as discussed in the book (see Fig. 24.15 and the associated 

discussion). Clearly the design seen below (indirect compensation) using a split-length current mirror (SLCM) 

load is more stable than the design above using Miller compensation and a zero nulling resistor.

For the same op-amp topology, e.g. a diff-amp followed by common-source amp, indirect compensated designs 

can be 4 to 10 times faster and smaller than the equivalent Miller compensated design. A good estimate for 

area is to look at the capacitor values in the op-amp. The stable design seen in Fig. 24.8 uses a 2.4 pF cap 

while a 240 fF cap is used in the design in Fig. 24.21.

Let's say that we want to try to make the Miller compensated design compete with the indirect 

compensation design. Note that to make a fair comparison we have to use the same op-amp topology. Below

a RC network has been added to reduce the gain of the first stage, the diff-amp, at higher frequencies to 

lower the overall gain and make the op-amp more stable. The values of the added RC were selected using 

simulations to give the best step response. There is obviously a significant increase in layout size but more 

importantly the speed doesn't compete with the indirect compensation design. Further, the resulting design 

is less stable as indicated by the size of the overshoot.

It's interesting to re-simulate this design without Rz and Cc. The result is actually more stable and slightly 

faster than what is seen below (still slower than the design using indirect compensation). However, with no 

load, or other load capacitances, the op-amp oscillates. This isn't robust or practical design. 

Again note that changes in the R and C during manufacture influence the step response something that isn't 

an issue in the indirect compensation design. While simulations are important one can fiddle with values in 

the simulations, as done here, to get the best response. However, unless it's an academic discussion, 

fabrication and test should be used to determine the merits of a design. It isn't reasonable to fiddle with 

values in a simulation using one topology and compare the simulation results to measured results of an op-amp 

in a different topology. 

Before stopping let's try adding more RC networks to try to make Miller compensation competitive with 

indirect compensation. 

Below we've added an AC path, via C2, from the output of the diff-amp to the gate of M8B to increase the 

gain of the second stage. This will increase the effective capacitance on the output of the diff-amp helping to

stabilize the op-amp. Further, for pulses or higher frequencies, the output stage now behaves like a class AB 

output (see Fig. 24.29, among other figures, for how to implement a true class AB output stage). A 50k resistor 

and an additional 1 pF decoupling are added to isolate the DC voltage, Vbias4, from the AC signal coupled to 

the gate of M8B. The layout size of this design is considerable larger than the indirect compensated design and 

it's still slower.

There may be situations where adding RC networks can be useful to improve the speed of an op-amp; however, 
it's bad design to add these networks so that Miller compensation can be used instead of indirect compensation. 

For more information see Dr. Saxena’s Master’s Thesis:
Indirect Feedback Compensation Techniques for Multi-Stage Operational Amplifiers or
the presentation here. See also Vishal Saxena OpAmps Matlab Design Kit.zip. 

Return