Bad Circuit Design 7 - Not Controlling the Current in the Output Stage

This topic is also discussed in the book on pages 817-820 but, again, it’s such a common mistake we discuss

it again here.

The unity-gain frequency of an op-amp is given by fun = gm1/2pCc where gm1 is the transconductance of the

input diff-pair and Cc is the value of the compensation capacitance. We know that the fT of the MOSFETs

should be much larger than fun (say > 10 times). We also know that the selection of the fT sets the overdrive

voltages, Vovn and Vovp (Eq. [9.55]). Finally, we know that we don’t fiddle with the devices’ overdrive

voltages with the exception of modest changes to the diff-pair (say an increase in widths up to 4) to increase

gm1. Having a design with varying fTs because we’ve fiddled with the W and L is low quality (one of the

devices will be “the weakest link” and slow down the design.)

So, how do we increase the speed of an op-amp? (Yes, this is also discussed in the book. See, for example,

the bold important on page 903.) The answer is that we reduce Cc and increase (modestly as discussed

above) gm1. It’s that simple. Ultimately, neglecting f2 (the pole associated with the output of the op-amp), we

are limited by the biasing that sets the devices’ fT.

Assuming the fT of the MOSFETs is large then what limits the value of fun? Answer: f2. As we push out fun we

also have to push out f2 (see Eq. [24.24]). So how do we push out f2? Well, if we could reduce the load

capacitance that would help but, of course, the load generally isn’t a value the op-amp designer can control.

So, to answer the question, we push out f2 by increasing gm2 (the transconductance of the output stage). We do

this by increasing the widths of the devices used in this output stage. These devices’ overdrive voltages remain

unchanged (important) but their drain currents increase and thus so does gm2. This is the practical way of

increasing the speed of an op-amp.

So, finally, the key thing to note is that we have to precisely control the current in the output buffer in order to

set gm2. There are lots of “good” op-amp designs in the book (at least the author things so ;-). There are also

“bad designs” seen in Figs. 24.58 and 24.59 (below).

Note that one of the places it’s easy to mess-up the biasing of an output stage is in a 3-stage op-amp design.