spectre solver for veriloga

Hi Yuanqi,

Your question is too open-ended. Of course, results certainly can be affected by the accuracy settings - that's why they are there. However, it depends on how significant those differences are.

Verilog-A can have both procedural statements as well as creating equations to be solved like other circuitry - without knowing more details on the input signal, what the Verilog-A model looks like, and the differences you are seeing in the results, it's really hard to guess what might be wrong (or even if it's wrong at all).

If you can't share that information here, please work with customer support.

Regards,

Andrew.

Hi Andrew,

The Veriloga model is a simple laplace function as below:

analog begin

vin = V(inp,inm);

vout1 = laplace_nd(vin*pow(10,dc_gain/20),{1.592e43},{1.592e43,3.247e38,3.311e33,2.191e28,1.025e23,3.468e17,8.298e11,1.288e6,1});

V(out) <+ vout1;

end

And that's why I can't understand why accuracy setting could affect the results.

I plan to sit down and debug with Marcel when he visits us next time.

Cheers,

Yuanqi

Hi Yuanqi,

One thought that occurred to me - how many points are in the PWL file, and is it a representation of a (say) pulse train - i.e. with discrete points with sharp transitions, or is it a representation of a smooth waveform? If it's a discrete (pulse train) type waveform, and there are more than 20 points, then you may need to set allbrkpts=yes on the source (on the analogLib vsource, there's a parameter labelled "Breakpoints" which has a cyclic choice of "yes" or "no").

Regards,

Andrew

Hi Andrew,

It's indeed a pulse train and more than thousands of points, we used vpwlf before but doesn't have this allbrkpts option. I will try with vsource again, before that may I ask what does this "Breakpoints" mean and what will the solver do for different settings?

Cheers,

Yuanqi

Hi Yuanqi,

For pulse sources and PWL sources with fewer than 20 points, a "breakpoint" is added at each point in the file. This tells the simulator that there's (potentially) a discontinuity at that point and so it forces the simulator to solve at the timepoint and also take care around the transition. Without breakpoints, it may miss some detail on the transitions if they are discontinuous (in this case it would be the derivative that's discontinuous, not the value). So it would be a good idea to turn on breakpoints in this case - the accuracy settings could be causing it to miss detail on the input waveform, and hence that could affect the filtered result.

Regards,

Andrew.

Hi Andrew,

I have tried with the breakpoint option, but only got even worse results. However if we specifically add more points in PWL file to make the transition time (e.g. 0v-3.3v) from 20ns to 3ns, we can get the results we want. Should the "Breakpoint" option force the simulator transient faster?

Cheers,

Yuanqi

Hi Andrew,

I have tried with the breakpoint option, but only got even worse results. However if we specifically add more points in PWL file to make the transition time (e.g. 0v-3.3v) from 20ns to 3ns, we can get the results we want. Should the "Breakpoint" option force the simulator transient faster?

Cheers,

Yuanqi

Yuanqi,

We'd have to see the example to understand what's going on - my suggestion of breakpoints was only a guess, so it's likely to be something more subtle than that. The point of using breakpoints on a large PWL file is to improve the accuracy of the simulation, not to make it faster. Clearly that's not happening in this case - so it must be something else.

Best to contact customer support as I mentioned earlier (no need to wait for one of the team to visit you - this is what customer support is for!)

Regards,

Andrew.