op-amp open-loop GBW simulation problems

Hi,

I have problems with simulation of fully-differential op-amp gain-bandwidth. I use spectre cmdmprobe (and stb analysis) to break differential+common mode loops at output of op-amp (folded cascode with source follower outputs).

Simulating by shorting output to input (ie. unity gain configuration), gain-bandwidth is OK. But then, by when adding Rfb between output and input, and also adding input resistor Rin, open-loop gain-bandwidth reported by stb decreases as Rfb increases.

I keep Rin=1k, then change Rfb and Vin so that Rfb*Vin=constant - so that outp is always max and outn is always min, so that internal node voltages (and gds/rout etc...) are the same and I'm always simulating in the worst case situation.

So why does open-loop gain-bandwidth change? Dominant pole position doesn't change much, only DC gain seems to change (so also reducing the unity-gain-bandwidth). Manually calculating Av0=gm*Rout gives the correct value for the unity-gain feedback configuration, but as Rfb is increased, the simulation diverges further and further from this value (ie. smaller and smaller Av0 in simulation) as Rfb is increased. Can anyone help explain?

To give an idea of what is happening, here are measured results from simulation:

If GBW is determined by gm/(2*pi*Cl), and Cl is not changing (same devices, same circuit), and gm is not changing (same current through it), then why should GBW change? Also, Rout decreases slightly as Rfb increases but this not enough to account for the large change in Av0 seen. What is happening here?

i guess what the stb simulation does is for the loop-gain, which will include the feedback factor. so the amplitude response in the bode plot will actually move up or down according the feedback factor, and the unity gain bandwidth changes..

Hi ansu-s,

perhaps due to my limited knowledge of the english language - but I don't really understand what you are doing.
Are you interested in the GBW of the opamp with or without feedback?
It would be helpful to show us a simplified diagram of your test arrangement.

Hi, please see attached op-amp testbench.

I need 10MHz GBW, so I have got this in unity gain configuration (ie. output shorted to input, Rfb=0). I used a +/- 1.0V DC sources in place of the Rfb to make sure the output is at maximum slew, where because of reduced Vds over current sources, they have lower Rout (than when at common mode output of 1.65V). This way, I think I can simulate worst-case open-loop gain-bandwidth of amplifier.

But when using amplifier in feedback configuration, eg, with Rin=1k and Rfb=1k, amplifier is still in unity gain mode but GBW drops in half, and Av0 drops by 10dB! Why?! Then, I try increasing Rfb (but I must also decrease Vin, ie. Vin_new=Vin*(Rin/Rfb), else output will saturate), and as Rfb increases, GBW and Av0 keep decreasing. This is still all open-loop simulation, ie. results from stb analysis using cmdmprobe to break all loops.

windknows points out that I need to include feedback factor - I kept doubling Rfb so the feedback factor keeps halfing ie. 1, 0.5, 0.25, 0.125. So should be -6dB/step decrease in Av0 as Rfb is increased 1k/2k/4k/8k. This doesn't match table above though...

I try simulating closed loop gain-bandwidth as well, by using AC sim with AC=0.5 source in series with INP and AC=-0.5 source in series with INN, but do not understand results: Closed loop UGF increases with increasing Rfb! Have added this to the table in second post, please see last column for closed loop gain-bandwidth (GBW_CL).

for this inverting OPAMP, the feedback factor is Rin/(Rin+Rfb), actually the Vo/Vin=-A*B1/(1+A*B2)
A is the gain of your OPAMP,
B1=Rfb/(Rin+Rfb),
and B2=Rin/(Rin+Rfb)

and if you chang the A to A(s) as a one pole system A/(1+s/po), you will see the close loop gain is
Vo/Vin=A*B1/((1+A*B2)+s/po), still a one pole system.
the -3dB bandwidth is (1+A*B2)po, and DC gain is A*B1/(1+A*B2)
if we continue assume UGF close to the GBW (this might have problem since now f-3db is much close to UGF for the feedback system) and is
(1+A*B2)*po*A*B1/(1+A*B2) =po*A*B1,
po and A won't change, but B1 will increase as Rfb increase, so UGF will increase