fall-rise uncertainty vs. simple

[shaharkl]

Hi,

For synthesis I use 5% guard on the clock period (setting 95% of nominal period) and also 5% setup uncertainty to cover for clock waveform impurities (jitter etc.).
My design has some fall-rise timing paths which, for these, I feel this 5% uncertainty is a little overhead.
I am considering to relax the fall-rise uncertainty to 2.5%.
So I would like to ask you all for a second opinion.

Thanks
Shahar

 

[ThisIsNotSam]

uncertainty is a function of your system. it's very hard for us to judge whether 5% is high or low.

 

[dick_freebird]

If you have a 5% supply tolerance then you will
have more than 5% drive strength (hence timing)
variation.

If you are on the hook for MIL or full industrial
temp range then your timing span will be even
wider (leaving out P, V, even). I've seen 2X delay
going from -55C to +125C.

Process variation of 10% is entirely reasonable
and some elements of this can alter the relation
between rise and fall delays / edge rates.

 

[danadakk]

If you have a 5% supply tolerance then you will
have more than 5% drive strength (hence timing)
variation.

If you are on the hook for MIL or full industrial
temp range then your timing span will be even
wider (leaving out P, V, even). I've seen 2X delay
going from -55C to +125C.

Process variation of 10% is entirely reasonable
and some elements of this can alter the relation
between rise and fall delays / edge rates.

Dick my background was 70's manufacturing, and we would see 50% or more speed
variations on 20 - 50 u processes. I am curious what does 70 nm look like
in todays world ?

Regards, Dana.

 

[ThisIsNotSam]

I think we are discussing too entirely different things, variation and uncertainty are not connected. process variation is modelled by corners and OCV, it already is captured by STA. what the OP seems to be asking about is clock uncertainty, which is a function of the system. where is the clock source coming from? how good is the source?

 

[shaharkl]

Thanks fellas, but you took it to the wrong direction.
I was really asking about fall-2-rise uncertainty vs. rise-2-rise uncertainty. I want to define the fall-2-rise uncertainty requirement to be half of the rise-2-rise uncertainty requirement.
By default the timing tool will use the same uncertainty requirement for full-cycle path and for half-cycle path. I think its over-constraint for half-cycle (fall-2-rise).
What do you think ?

Thanks