The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> PSS analysis for a clocked comparator
https://designers-guide.org/forum/YaBB.pl?num=1457299317

Message started by ULPAnalog on Mar 6^th, 2016, 1:21pm

Title: PSS analysis for a clocked comparator
Post by ULPAnalog on Mar 6^th, 2016, 1:21pm

Dear experts

I have been investigating noise properties of clocked comparators (schematic attached) and in that process happened to encounter the following issue. The inputs of the comparator are set to a nominal common mode voltage (600mV) and transient analysis was performed (results attached). In the absence of transient noise/mismatch, both the outputs (Vout+, Vout-) of the comparator are identical, which I would expect to be the case.
Now, I perform a PSS analysis and look at the output obtained after the initial transient analysis that is performed as part of PSS Shooting method. To my surprise, I find the output of transient analysis done as part of PSS analysis to be different from the transient analysis performed in the above step.
Conservative settings was used for errpreset in both the cases.
Could you please give some insight on what might be going on here.

Thanks and regards

Title: Re: PSS analysis for a clocked comparator
Post by Ken Kundert on Mar 6^th, 2016, 3:06pm

You are rolling a ball along a knife edge expecting it not to fall off. You get lucky with transient, but with PSS the ball falls off. It should not be too much of a surprise. A comparator is designed to be very sensitive. Indeed you would never expect to see the transient results on a real circuit.

-Ken

Title: Re: PSS analysis for a clocked comparator
Post by ULPAnalog on Mar 6^th, 2016, 3:12pm

Hi Ken

Thank you for the reply. I agree that the behavior shown by transient would never be seen in real world. But what causes the ball to roll off in PSS analysis and not in transient? Has it got something to do with the simulation noise injection? If such is the case, why does it not appear i transient analysis.

Best regards

Title: Re: PSS analysis for a clocked comparator
Post by Ken Kundert on Mar 6^th, 2016, 6:34pm

You might think, because the circuit is perfectly symmetric, that the ball would just roll perfectly balanced along the knife edge. And indeed it does in transient. But the simulator introduces very tiny amounts of asymmetry, if for no other reason that the simulator is being run on a von Neuman machine, and so things must be done serially. For example, the nodes are placed one at a time into the matrix and then the corresponding rows are factored, so a node from one side of the circuit will be processed before its pair on the other side. This can inject small amounts of asymmetry on the order of machine precision (1 part in 10^15) into the calculations. This asymmetry can be seen by the comparator, which might cause it to latch one way or the other. The PSS does more computation than the transient, and perhaps is more sensitive to these small errors than transient.

The point is that in transient the ball is slowly rolling down the entire edge of the knife without falling off. That is a spectacularly unlikely thing to happen. You should not be asking why it happened to fall off during PSS, you should be asking how it could have possibly made it all the way across the knife without falling off in transient. The answer to that question ...

Luck.

-Ken

Title: Re: PSS analysis for a clocked comparator
Post by ULPAnalog on Mar 6^th, 2016, 6:48pm

Quote:

The PSS does more computation than the transient, and perhaps is more sensitive to these small errors than transient.

This is something that I was not aware of. You are right about my question being asked from wrong perspective but I got your point.

Quote:

The answer to that question ...

Luck.

Not something that I rely on when it comes to circuits !

So let me think this through. For the comparator, PSS analysis could take the state one way or the other under 'seemingly' perfectly balanced conditions. If such is the case, can PSS reliably converge given that the periodic operation is now dependent on other things including machine precision?

Thank you again for your posts. Appreciate that always.

Best regards