10.06.2016 Views

eldo_user

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Monte Carlo Analysis<br />

Importance Sampling Monte Carlo Examples<br />

Importance Sampling Monte Carlo Examples<br />

The sample netlists monte_carlo_sram_bitcell.cir and monte_carlo_vco_lc_tank.cir can be<br />

used to demonstrate the advantages of the IMSC method over the standard Monte Carlo<br />

method.<br />

SRAM Bit Cell Circuit<br />

Netlist file: monte_carlo_sram_bitcell.cir<br />

The sample circuit is a classic six-transistor SRAM bit cell designed in CMOS 45nm. A subset<br />

of the device model parameters has variability information, in the form of a DEV/GAUSS<br />

specification. The total number of statistical parameters is 36 (6 per device). Although the<br />

IMSC method is insensitive to the number of statistical parameters, it is always preferable to<br />

keep this number “reasonable”.<br />

The cell is configured for measuring the write delay, and the purpose of the Monte Carlo<br />

simulation is to estimate the probability of the write delay exceeding a certain threshold. This<br />

threshold could be used to define a “fail” behavior for yield estimation. Formally, the problem is<br />

to estimate π = Pr{Y > y} and the associated confidence interval on (the empirical estimator<br />

of π), where y is the threshold.<br />

The following .EXTRACT MC … MCROB() statement is given in the netlist:<br />

.EXTRACT MC LABEL=prob_plus_6 MCPROB(write_delay_qx, GE, 53.12)<br />

This statement illustrates how to use the .EXTRACT and MCPROB commands to specify this<br />

kind of probability estimation problem: refer to (usually) another .EXTRACT or .MEAS<br />

statement, specify a right tail (upper threshold) or a left tail (lower threshold), then specify the<br />

threshold. The following results are obtained:<br />

Lower Pr. Value Upper CV(%) #Runs Speedup<br />

Tail / Complementary<br />

-----------------------------------------------------------------------------<br />

8.89e-06 9.65e-06/1.00e+00 1.04e-05 7.87 8060 8.0e+03<br />

This kind of low probability estimation is very difficult (if not impossible) using the standard<br />

Monte Carlo method because a very large number of runs is required to obtain a decent<br />

accuracy (in terms of the confidence interval on the probability). In this example, the ISMC<br />

method estimates a probability of 9.6×10 −6 to within +/-7.8% (with a 95% confidence level) in<br />

roughly 8,000 runs, whereas the standard Monte Carlo method would require roughly<br />

64,000,000 runs (see Table 11-1).<br />

The estimation of the speedup of the ISMC method over the standard Monte Carlo method<br />

comes from the binomial law. The random variable:<br />

528<br />

Eldo® User's Manual, 15.3

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!