The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> Monte Carlo Sim
https://designers-guide.org/forum/YaBB.pl?num=1126140887

Message started by river on Sep 7^th, 2005, 5:54pm

Title: Monte Carlo Sim
Post by river on Sep 7^th, 2005, 5:54pm

thank you, Ken and Andrew for your kindly reply of the question "segmentation fault" posted by me. I will contact Cadence support team.
I have another question related to monte carlo simulation.
I have a circuit which has several control pins, e.g. S<2:0>. By changing the setting
of S<2:0>, I can change the performance of the circuit to cover P.V.T.( Process,
voltage and temperature) variation.
Now I am doing Monte Carlo simulation to check the performance of that circuit.
I set S<2:0> = 100, at Run #32, I fail to meet the spec. but if I change the setting
of S<2:0> to 101, maybe I can meet the spec. The problem is, if I change the setting
of S<2:0>, the Gaussian distribution of each devices of the circuit also change, so
Run #32 is not the same between setting S<2:0> = 100 and S<2:0> = 101, then I cannot know with S<2:0> = 101, under the failed condition, the circuit can work or not.
I am not a native English speaker. I don't know whether I have expressed my question clearly or not. If any confusion, please let me know.

Best Regards,
River

Title: Re: Monte Carlo Sim
Post by Andrew Beckett on Sep 7^th, 2005, 10:12pm

You're right - if you change the circuit, the gaussian distribution could change (although I'm not sure that if you change just the parameters on a voltage source (say) that
it would alter the distribution (try making the voltage sources driving the bus controlled by design variables and then just altering those).

However, the point is really one of yield, I'd have thought. If you have sufficiently high yield with each control input setting, then you'll be OK. I guess the danger is that if you have enough different control input settings, and each one causes a failure mode which is uncorrelated with other control input settings, you may have a circuit which will always fail under one control input setting - but this is probably pretty unlikely with most circuits... (I'm thinking out loud here).

Regards,

Andrew.

Title: Re: Monte Carlo Sim
Post by achim.graupner on Sep 9^th, 2005, 2:03am

Is you input a kind of a trimming input which get programmed during wafer-test or so? This means that you have to simulatate the trimming process too. If your need help on this, let me know.
If S is a input in your device, the distribution of the random device parameters should not change.
- achim

Title: Re: Monte Carlo Sim
Post by Ken Kundert on Sep 9^th, 2005, 7:09am

Achim is right, you would have to simulate the trimming process somehow. You could either do it during the simulation, or you could run a full Monte Carlo simulation for each setting of S, and discard the results for the 7 of 8 cases that would not occur.

-Ken

Title: Re: Monte Carlo Sim
Post by river on Sep 9^th, 2005, 6:30pm

Yeah, I believe during trimming process, the random device parameters will not change. but unfortunatly, I cannot do that trimming simulation, because it takes long time and monte carlo simulation is unrealistic. So now I want to check whether my control code S<2:0> is
enough or not to cover 3-sigma process variation. I believe the trimming process has no problem. As long as the control code S<2:0> can cover the process variation, the trimming process can calibrate the circuit to the correct working point.
I believe if I don't stop the simulation, the random device parameters will not change.
but only after simulation I can check the performance of the circuit, and then change the control code (say, change the voltage of a DC voltage source which is connected to the control code). With this new control code, I re-run the simulation and I want to check the performance of the circuit with previous device parameters. I think Run# is the seed for random device parameters generation, that means if Run# is the same, the device parameters should also be the same. That's true if I don't change the control code. The monte carlo simulations show that fact. But if I change the control code, even the seed is the same, but the device parameters are changed.
Finally, my question is how to attain the same device parameters with different control code settings?

Best Regards,
River

Title: Re: Monte Carlo Sim
Post by Ken Kundert on Sep 12^th, 2005, 1:29pm

I recommend that you create a Verilog-A model that takes the various process variables as parameters and drives the control pins to the proper settings.

-Ken

Title: Re: Monte Carlo Sim
Post by Geoffrey_Coram on Sep 14^th, 2005, 4:54am

river wrote on Sep 9^th, 2005, 6:30pm:

But if I change the control code, even the seed is the same, but the device parameters are changed.

This seems odd to me. Unless the control code changes the topology of the circuit, it seems that the seed should give the same results for the parameters. (If you had more devices in your circuit, then there would be more calls to the random number generator, and the values for devices encountered after the new devices would be different in the new run.)

Title: Re: Monte Carlo Sim
Post by Andrew Beckett on Sep 19^th, 2005, 9:18pm

Geoffrey,

I thought that was odd too. I wouldn't have expected it. That's why I wondered whether the circuit was being hooked up differently for each setting of the input, rather than it being parameterised somehow (in my previous append).

Regards,

Andrew.

Title: Re: Monte Carlo Sim
Post by river on Sep 20^th, 2005, 7:13pm

you are right.
Changing those control signals will not alter the random device parameters.
I neglected the fact that vth is a function of vds(here, I mean vds, not vbs).
thank you for your help.

Best Regards,
river