[PLUG] circuit design and simulation software

[Keith Lofstrom]

On Fri, Mar 22, 2019 at 02:57:07PM -0700, Erik Lane wrote:
> The way I learned it, from an engineer who worked at Intel for many years
> doing this stuff, was that the only serious tools in this space were text
> based, as the gui ones hid too much and provided less flexibility. Yes, you
> can still use text files, but it's not their default. (I don't remember his
> other problems with them, but he had fought them for years to try to make
> them do what was needed and said that they were not adequate.)
> 
> As I remember, in his opinion all 'real' engineers used hspice when they
> had the funding for it, as it was a very expensive package that really did
> a decent job. (He was kind of in the mode of simulation being a fallback
> from testing with the real thing and having a true understanding of what
> was going on.)

When I was actively designing integrated circuits, I used
HSPICE for Redhat Linux (I still have the expired security
key dongles).  $$$$, even with a consultant's discount. 
I designed chips, and depended on process/transistor model
files from the silicon fabs.  I learned about the limited
validity of those model files; they are usually optimized
for accurate simulation of digital gates, not analog
circuits, especially at extremes (like ultra low power).

The takehome lesson for all simulators is that your results
can't be more accurate than the models you start with. 
A 1000 ohm wire-lead resistor is also an inductor, and a
capacitor, and a high frequency resonator.  It is also a
heater, a microwave antenna, and an infrared light source.  
It varies with ambient conditions, and ages over time.

HSPICE is very good within the constraints it is designed
for, but an open source simulator tool is better when you
need to modify the code to model "second order" effects,
like thermal and electromagnetic interaction with physical
environment.  Or screw dislocations in the silicon wafer
lattice, which can cause anomalous leakage in nanoamp
circuits.  If this is a problem, you must hack the models.

Hacking model code at a deep level means understanding
esoterica, such as the sparse matrix inversion that is the
core of most circuit simulators.   You can get by with
knowing what can break (and how not to break) the inputs
and outputs to the core numerical processes.  You can't
do this with black-box proprietary simulation tools.

Well, maybe Intel has the clout to get the source code.

Nothing is as accurate as actually building and measuring
a circuit, but simulation can help you choose which
circuits to build and measure.  If you are building
integrated circuit chips, with (at LEAST) six-month
million-dollar build cycles, that's pretty important.

In the end, real engineers program in solder.  The design
process should lead to easily-evaluated physical demos.

Keith

-- 
Keith Lofstrom          keithl at keithl.com