Is dcmatch analysis in Cadence useful?

Hello,

I been reading the User manual of Spectre and came across dcmatch analysis. I tried using the dcmatch analysis on one of the circuit and straight off wasnt really sure how to use it and what purpose it might serve. Tried to read the manual further but didnt get anywhere.

Anybody out there who can give some tips on the usefulness of dcmatch? Does it have any potential use?

Cheers
Anil

Hi,
yes it helps sometime, Its the same as Monte carlo mismatch. But dc match runs on all corners and provide the voltage variation at any node. While Monte Carlo Mismatch we run for typical.

varunkant2k said:

Hi,
yes it helps sometime, Its the same as Monte carlo mismatch. But dc match runs on all corners and provide the voltage variation at any node. While Monte Carlo Mismatch we run for typical.

Click to expand...

Thanks Varunkant2k for your reply. Does that mean DCmatch analysis is quicker and limited way of doing Process+Mismatch option of Montecarlo analysis? Because from what I understand the Process+mismatch option gives mismatch variation for matching transistors at various statistical Process points( i.e statistical Corner points as against particular Corner points like SS, SF, FS etc ). Also I think we can save any node in Montecarlo simulation as well but before we start the run. Is that what you meant? Am I getting it right?

Appreciate your help!

With Monte Carlo you might have to run 30-100 simulations before you got a good estimate of the standard deviation (SD) due to random mismatch. On the other hand, DC match is a sensitivity analysis that is much like a noise analysis. It will give you the SD in a single simulation, making it very useful for large circuits. The downside is that you only get results for DC parameters like offset voltage. If you want to investigate things like transient settling times you would need to use monte carlo.

So... dc match is the best way to get the SD for Dc parameters, but it won't work for other things.

rg

Monte Carlo is Industry proven and shows very accurate results and good information with Hystogram. dc match is offset variation in SD as RobG said, and you do not get much information.

Thank RobG and varunkant2k for your replies.

It seems very clear now that dcmatch is a subset of Montecarlo analysis looking at specific parameters. Also I found out that you need specific dcmatch models to perform this simulation. I have access to ST micro and Austria Micro Systems model files and doesnt look like they write models for dcmatch. Also while setting up the simulation it seems that if the user wants to sweep a model parameter he needs to know the range to sweep against. Probably there is a way to find out the variation for specific parameters? Any knowledge how?

Cheers!

DC match does not need any model anyway. IN MC you need mismatch model.

My statement was based from the discussion in this forum. .

anilmutha said:

Thank RobG and varunkant2k for your replies.

It seems very clear now that dcmatch is a subset of Montecarlo analysis looking at specific parameters. Also I found out that you need specific dcmatch models to perform this simulation. I have access to ST micro and Austria Micro Systems model files and doesnt look like they write models for dcmatch. Also while setting up the simulation it seems that if the user wants to sweep a model parameter he needs to know the range to sweep against. Probably there is a way to find out the variation for specific parameters? Any knowledge how?

Cheers!

Click to expand...

Subset of Monte Carlo probably isn't the right term since they are different analyses. DC match is like a noise analysis. I even think mismatch "sources" are added for each device similar to what is done with a noise analysis. DC match is *far* better for anything that can be measured with a DC analysis because you can get the information in 1 simulation. Furthermore, it is "exact" in the sense that it deterministically determines the standard deviation of the measured parameter based on sensitivity analysis. A Monte Carlo would take hundreds of simulation to converge to the same result (although much fewer are generally needed to estimate the spread). Even if you only do 10-20 Monte Carlo simulations to estimate spread, the simulation will take an order of magnitude longer and you won't know how each component is affecting the result so it is impossible to optimize the areas of each device for minimum mismatch.

As I mentioned, the analysis itself is similar to a noise analysis in that mismatch "sources" are added to each device. This has the added benefit of allowing you to determine how much each component contributes to the overall variation. Thus you can target the worst offenders and optimize the design. You can't do that with Monte Carlo.

However, the mismatch is a "small signal" measurement so if a mismatch would cause a device to change operating point (e.g. enter the linear region) DC match would miss that, but Monte Carlo *might* cause the device to enter the region. Therefore Monte Carlo is good for catching poorly designed unstable circuits, like a poorly designed startup circuit or naively balancing leakage currents or balancing a flip-flop, etc. An analogous poor design would be a pencil standing on end. Small signal simulations will show it balanced, but Monte Carlo will show that it will never work.

You most certainly need to have the model parameters set for it to work! Same thing with Monte Carlo. It can't guess at how much the Vt, etc vary in your particular process!

Thanks RobG for bring out the subtle differences.

Is there a way to use the MonteCarlo model parameter (or any other model parameter) for dcmatch sim? I fear extracting MonteCarlo parameter for dcmatch would be more time consuming then using trial and error method with MonteCarlo sims to optimize design. But some one might have a easier solution.

Happy New Year to all

In theory you should be able to do it, but the model has gotten so complex I don't even try to understand it anymore.

Sorry I can't help you there,
Rob