A stochastic model in Verilog A for Monte Carlo simulation

You need to use the cds_inherited_parameter attribute on a parameter which you are going to have a mismatch definition of in the statistics block. For example, this VerilogA module:

`include "disciplines.vams"

module res(vp, vn);

inout vp, vn;

electrical vp, vn;

(* cds_inherited_parameter *) parameter real monteres = 0;

parameter real r = 1k;

localparam real r_effective = r + monteres; // nominal resistance plus

// monte-carlo mismatch effect

analog

V(vp, vn) <+ (r_effective)*I(vp, vn);

endmodule

And I'm simulating this from this netlist:

//
parameters monteres=0

statistics {
  mismatch {
    vary monteres dist=gauss std=5
  }
}

R1 (n1 0) res
I1 (n1 0) isource dc=1m

ahdl_include "monteres.va"

monte montecarlo numruns=10 variations=mismatch {
   dc dc
   export vdc=oceanEval("v(\"n1\" ?result 'dc)")
}

Regards,

Andrew.

 Thank you very much, I will try it.

 Hello Andrew,

Thank you for your answer. How can we do the simulation with the second code? Is it written in a file".scs" ? And then how do we execute the simulation Monte Carlo? Thank you in advance for your answer.

Best regards,

UU

You can run it with "spectre monteres.scs" for example, and the results will be in monteres.raw - including an mcdata file showing the variation of the points.

Or you could create a veriloga view in virtuoso for the model, place an instance of it in a testbench similar to what I've done here, and then create a model file (called stats.scs, say) containing just the parameters and statistics block, and then include that as a model file in ADE - and then setup the monte carlo as you would usually in ADE.

I just created a simple text-only example to show the principle without needing a database.

Regards,

Andrew.

Hello Andrew,

Thank you for your guide. I have done as you said but it doesn't work, perhaps I have made some errors. I have created two files :

1 source.va 

// VerilogA for source
`include "constants.vams"
`include "disciplines.vams"

module source(a,b,c);
inout a;
     inout b,c;
     electrical a,b,c;
     integer seed;
     (*cds_inherited_parameter*)parameter real seedin = 0;
     analog begin
     seed=seedin;
     V(b) <+ cos(seed*100000000);
     V(c) <+ sin($abstime*300000000);
     $strobe("hello,I'm glad to meet you",seedin);
     end
endmodule

2 source.scs

simulator lang=spectre
ahdl_include "source.va"
parameters seedin=0
statistics {
  mismatch {
    vary seedin dist=gauss std=5
  }
}

.  Do you see any errors? Thank you.

Best regards,

UU

Not sure what you mean by "it doesn't work". I created a simple circuit containing an instance of source, and then ran a monte carlo analysis:

include "source.scs"

I1 (a b c) source
r1 (a 0) resistor r=1k
r2 (b 0) resistor r=1k
r3 (c 0) resistor r=1k

monte montecarlo numruns=10 variations=mismatch savefamilyplots=yes {
   tran tran stop=10n
}

This outputs a lot of stuff, but in particular it says (for each iteration of the monte carlo):

 hello,I'm glad to meet you -3.38873
hello,I'm glad to meet you -3.38873
hello,I'm glad to meet you -3.38873
.hello,I'm glad to meet you -3.38873
..hello,I'm glad to meet you -3.38873
...hello,I'm glad to meet you -3.38873

and then later:

     gmin = 1 pS                                           


hello,I'm glad to meet you 2.42183
hello,I'm glad to meet you 2.42183
hello,I'm glad to meet you 2.42183
.hello,I'm glad to meet you 2.42183
..hello,I'm glad to meet you 2.42183

(you're strobing on each timestep iteration, so because I ran a transient, it strobes multiple times).

Because I ran with saving family plots, you can see your signal b in the monte carlo (see attached picture).

Andrew.

 Hello Andrew,

I created the file "source.scs" and "source.va" and then executed the simulation Monte Carlo with "source.scs" imported as library using a simple circuit, and the attached error came.

Best regards,

UU