How can I use statistical results, e.g., mean, standard deviation, from previous test as design (or global) variable of the current test?

For the first test, add an output like this:

CurStd: stddev(Current ?continuous nil ?overall t)

and set the EvalType column to "Sweeps" for this measurement. (Current is one of my existing outputs for that test).

Then in the second test, have the variable defined as calcVal("CurStd" "firstTest") . This will then pass that value to the variable for the second test and ensure that simulations are correctly sequenced to pass the value from one to the other.

Andrew

Hi, thanks! it works!
if I use ?continues nil, somehow  it does not work. I cannot put "(" after stddev.


But, it works if I use it this way:


btw, I found out that the default standard deviation shown by cadence is sample standard deviation while the one from stddev(our_signal ?overall t) command is population standard deviation.
They are almost the same for large sample though.

from Libre Office




if I use stddev(our_signal ?overall nil), I got different value and am not sure what it is.



Happy holiday!

Some new options, ?continuous and ?type, were added in the IC618/ICADVM20.1 ISR19 hotfix to allow better control of whether the data is treated as continuous (i.e. a waveform, where the x-axis is relevant) or discrete values (where the x-axis doesn't matter). The average function also now supports the ?continuous option. If the data is a waveform, the average is the integral devided by the x-range, and that's probably not what you want when computing the average or standard deviation from a set of Monte Carlo points. So the ?continuous argument can be provided as nil and then it will only look at the y values (the average is the sum divided by the number of samples, for example). Similarly this affects stddev. The ?type argument for stddev allows you to choose between population or sample standard deviation.

Note that the ?overall argument is used when you may have family data and it computes the mean (or standard deviation) across all the data. Since the values were discrete, with ?overall t  would have given you the same as ?continuous nil anyway.

Probably best to use the newer version and then you have the controls you need. Sadly it took us rather a long time to add these options to the built-in calculator functions! Much of the time the difference between sample and population standard deviation is small anyway with a big enough sample size, so usually people were not that bothered by using the population standard deviation when it was really a sample (and lots of users had forgotten the difference from high school maths - not that this is an excuse!)

Regards,

Andrew