Different random variables for each device in monte-carlo simulation

[rnunes2311]

Hi, It looks like every device is getting independent/different random variables assigned in a monte-carlo simulation. I also realised that the resistors were not getting any variation. This appears to be a mistake in the parameters_res_nom.spice. For example: .param sw_sky130_fd_pr__res_high_po_rs = {325.0+corner_factor*0.0} + process_mc_factor*MC_PR_SWITCH*GAUSS(0,0.035,1)

GAUSS generates random values relative to the mean, which is 0. Therefore it always generates 0, I believe.

[RTimothyEdwards]

I'm not getting that. Gaussian random variables calculated in a .param statement should be calculated once, not for every device. And for the question above, process_mc_factor is 1 (which I think corresponds to 3 sigma variation), andMC_PR_SWITCH is 1 when the monte carlo corner is selected, so the parameter calculated should be (325 + (random number with zero mean)).

[rnunes2311]

For the resistors I see in simulation no variation at all. The ngspice manual states that the gauss function generates random numbers relative to the mean. For that parameter it is generating 325 + (random number with 0 mean and 0 sigma) = 325. For most gaussian random variables in the pdk the agauss function is used, which generates random numbers with an absolute sigma value. I see no variation at all in the resistors with a monte-carlo simulation. By modifying the expression I could get the 3.5 % sigma variation. I also see mismatch in transistors, which should not happen and when I place resistors in parallel and series to increase the area while still keeping the same equivalent resistance value the equivalent resistance variation is not 3.5 % 1 sigma, because every resistor is getting a different value assigned. This I saw after correcting the original expression.