What are the tools needed to test semiconductor stability?

[rikie_rizza]

Hi All,

In order to ensure that the test program to test a semiconductor is stable, what kind of tools that you would probably use at work? I know that most statistical tools are good to predict whether a test is stable or not, but what method in statistic that can predict with high accuracy, with a small number of sampling?

Most semicon testing issue can give a headache to BE engineer because it will shown unstable symptom after some period of time.

Thanks for the suggestions,
Rikie

 

[erikl]

what method in statistic .... can predict with high accuracy, with a small number of sampling?

This is an oxymoron, a contradictio in adiecto, sorry!

 

[leo_o2]

Hope it helps:
**broken link removed**

 

[rikie_rizza]

Well, if you're on a position that must decide whether you can release the device testing or not based on a small quantity of sample, what would you do? The point to think about is the stability of the test program, device complexity, and the hardware stability. you will only have about 1K of sample from the same batch, so how can you predict that the test package is good enough to release for production?

 

[erikl]

From the results of your test you can of course extrapolate an appropriate statement on the yield (or on the quality of certain features), but such a statement is true only with a statistically caused probability of 1-1/√N , if N is the number of your test vehicles, i.e. about 96.8% in case of 1000 tested DUTs.

What would mean that your quality statement wouldn't apply to about 3 from 100 delivered lots.

 

[dick_freebird]

Assessing repeatability requires some number of repeats.
Not necessarily a large number of units, but at least
multiple touchdowns, teardowns etc. to make sure you
have a good chance of getting the same answer next
wafer or next year. You depend on your cal lab for the
equipment for the most part, and worry about the stuff
that's on you - fixturing, cabling, ambient noise, self-
calibration, control units and like that.

Sure, everybody wants ppb confidence off one unit. Too
bad about that. Usually you have to work for it.

I would think that your organization, if this is an ongoing
business, has some standards for test development, what
level of accuracy / repeatability is expected and what
proof suffices.

Once you have demonstrated that the system is short-
term and across-teardown stable, if it goes off the rails
presumably it's the sustaining engineer's job to right it.
Just don't make the poor guy any more miserable.

 

[rikie_rizza]

The organization itself is a matured top 10 company, but we live with "let it run now, think about the stability and improvement later" principle. This is okay for the guys in business unit, but people at production site will suffer from maintaining the output and yield without sure improvement future.
(Not to mention the expensive Tera...oops...tester are not as robust as its suppose to. Flex...isble it is, but most of the time we spent maintaining the tester more than the product itself.)

Maybe youre right, Im just an engineer not a fortune teller...statistic will not get me what the future of testing will be for a particular device. I dont need stochastic or things, just make it as my regular ordinary work.