Xyce with Glade Design Environment

100 views
Skip to first unread message

Murat Eskiyerli

unread,
Apr 17, 2020, 6:44:56 AM4/17/20
to xyce-users
Dear all,

As a weekend project, I combined Xyce with Glade Design environment and FreePDK15 PDK and associated models for 14nm Finfet process. I summarised my experience at a Linkedin post. If anybody is interested you could read it at:
https://www.linkedin.com/feed/update/urn:li:activity:6655836899007057920

Any feedback is welcome.

Regards,

Murat Eskiyerli

xyce-users

unread,
Apr 17, 2020, 11:23:30 AM4/17/20
to xyce-users

Nice!   One small correction.  Xyce actually does have Monte Carlo in it, as part of the UQ capabilities you mention.  We haven't advertised it very much because it is pretty new, and is (so far) executed sequentially.    Here is a really simple netlist that uses the feature:

-----------
Regression test for simple normal distribution sampling

c1 1 0 1uF IC=1
R1 1 2 1K
v1 2 0 0V
.print tran v(1)
.tran 0 5ms
.options timeint reltol=1e-6 abstol=1e-6

* 100 normally distributed samples, mean=3K; std deviation=1K
*.SAMPLING NORMAL R1 3K 1K 10
.SAMPLING
+ param=R1
+ type=normal
+ means=3K
+ std_deviations=1K

.options SAMPLES numsamples=100

.end


xyce-users

unread,
Apr 17, 2020, 11:26:16 AM4/17/20
to xyce-users

For sampling, there are two types:  Monte Carlo and Latin Hypercube Sampling (LHS).    LHS is generally a better option as it will get better statistics in the tail of a distribution, so it is the default.   You can manually choose which of these is used via the sample_type parameter.

.options samples  sample_type=mc

.options samples  sample_type=lhs

xyce-users

unread,
Apr 17, 2020, 11:30:15 AM4/17/20
to xyce-users

Like I mentioned, this is a very new capability, and it isn't as parallel as it could be, in that it runs the samples sequentially.  If the individual sample simulations are big enough to run in parallel, that type of parallelism will still happen.

Part of why we didn't implement this feature until recently is that most of our users who want to do sampling do it using Dakota.  Dakota is an open source UQ toolkit that you can also get from Sandia.  It is very mature, and handles doing the sampling in parallel very well.  You can read more about Dakota (and download it if interested) here:


thanks,
The Xyce Team
Reply all
Reply to author
Forward
0 new messages