The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Mixed-Signal Design >> Corner Analysis
https://designers-guide.org/forum/YaBB.pl?num=1074857505

Message started by bvishwakarma on Jan 23^rd, 2004, 3:31am

Title: Corner Analysis
Post by bvishwakarma on Jan 23^rd, 2004, 3:31am

What is Corner Analysis ? how it helps in optimizing the design ?

Title: Re: Corner Analysis
Post by Paul on Jan 23^rd, 2004, 11:36am

Corner analysis means that you check whether your circuit works when you take into account manufacturing tolerances for devices, temperature range, variations of external signals. It does not optimize your design, it is thought to check whether your circuit will work in real life over the possible variations of these parameters.

Typically you consider min/max values for resistors, capacitors and temperature and 4 corners for the MOS transistors: ff, ss, fs, sf.
f stands for fast (highest drain current for given device size), s for slow (lowest drain current). The first letter is usually for the NMOS, the second for the PMOS. tt stands for typical-typical, i.e. nominal drain currents for N and P.

Have a look at the following guidelines:
http://www.vlsi.iitkgp.ernet.in/~animesh/ana_flow/analog_basic_precau.pdf

Some of the RF guys in this forum can certainly give some info about what to consider for bipolars and inductors.

Paul

Title: Re: Corner Analysis
Post by Andrew Beckett on Jan 23^rd, 2004, 11:00pm

Of course, it helps optimize your design if you have to adjust
the design to work in the corners.

There are some downsides of using process corners though.
Sometimes these corners are such that all process parameters are pushed to their extremes at the same time, and in some cases those parameters are pushed in opposite directions but are in reality correlated (often people simulate sf and fs corners for mosfets, say, but many of the process
parameters for nmos and pmos devices will be correlated).

This tends to lead to over design, as you work really hard to
make sure your circuit works in a process situation which will never (or hardly ever, at least) occur.

Also, it's possible to have designs where the worst case
is not in the process corners. You can end up missing this
with a corners type approach (luckily this doesn't
happen that often!).

A better approach is to use statistical methods (such as
monte carlo). This allows you to simulate over the
process spread in a statistical fashion, although the
downside is that it is more work for the model developers
and also you can end up doing more simulations to get
a reasonable sample over the process spread.

Just thought I'd give a bit more background.

Andrew.