A stochastic model in Verilog A for Monte Carlo simulation

That's the IC version; as I mentioned earlier, it's the MMSIM version that makes the difference in terms of parsing the Verilog and creating the CDF which controls how it is netlisted. You're using MMSIM10.1 and this was fixed in MMSIM11.1 ISR17.

Regards,

Andrew.

 Hello Andrew,

When I did the MC simulation with ten points, every time the ten numbers I got are always the same(for example always 203,184,151,91,66,46,-1,-28,-86,-90). Is it normal? When I used the library of CMOS to do the MC simulation it's the same case. For one time, the parameter is random, but for 10 times, always the same random numbers. Thank you.

Best ragards,

UU

Yes, it's normal. The seed in ADE XL is set (by default) to 12345 - and this means that simulation results are repeatable. That's nearly always what people want - if I don't change anything, I should get the same answer. You can get a different sequence by specifying a seed on the monte carlo options form, but this isn't any  more random - it's an equally random, different sequence of random numbers. So I always think it's of limited usefulness.

Kind Regards,

Andrew.

 Ok, thank you. For the statistic blocs, I used gauss distribution as the majority of people used. If I want to use other distributions, like exponential, uniform etc. what should I write? Can you recommend me a document to learn more about this? Because in the Verilogaref there is just VerilogA file, no other things. Thank you.

Best regards,

UU

If you run <MMSIMinstDir>/tools/bin/cdnshelp you can access the entire documentation for the simulator. Look in the spectre reference and user guides for more detail - or search for "statistics" or "montecarlo".

Or you could type:

spectre -h montecarlo

from the UNIX command line. This will describe the statistics block used by  monte carlo too.

Regards,

Andrew.

 Hello Andrew,

I have found that when I execute the simulation in a computer of 32 bits, "seedin"(defined as cds_inherited parameter) is regarded as a component parameter, and the Monte Carlo simulation goes well, but in a computer of 64 bits, "seedin" is automatically eliminated from the list of component parameters(not in the CDF),  at the same time Monte Carlo simulation doesn't work(always one result with no matter how many points of simulation of Monte Carlo). Is there any difference between the use of this kind of parameter in the two systems? Thank you for your response.

Best regards,

UU 

That seems a bit unlikely - are you sure you're using the same version of the tools on the different machines? I've never seen this behaviour before. On very rare occasions there can be issues which only show up when running spectre in 32 bit or 64 mode (and I'm talking about the bitness of the application here, not the bitness of the OS), but I've not seen any issues between the OS bitness for many years. Even then it was not a difference in functionality, as you're describing here.

I was certainly running on a 64 bit OS when I tried this out, because in fact nowadays (since MMSIM11.1 if my memory is correct) we only support 64-bit OS. I was using MMSIM13.1. You were (I think) using MMSIM10.1 - and it's probably wise to update to something more recent (and supported) anyway.

Regards,

Andrew.

 Thank you, I will try to find the real problem.

Best regards,

UU