Department of Electrical and Computer Engineering University of Wisconsin - Madison Computer Engineering Seminar (Spring 2007-2008)

        Speaker:    Lin Xie and Professor Azadeh Davoodi
        Time:       12:00 pm 
        Date:       February 15, 2008 
        Location:   Room 4610, Engineering Hall 

        Subject:    New Challenges in Statistical Static Timing Analysis 


As usual soft drinks will be available for those who show up in
time for the seminar. 

*************************************************************************

            New Challenges in Statistical Static Timing Analysis

                                  by

                       Lin Xie and Professor Azadeh Davoodi 
            Department of Electrical and Computer Engineering
                           UW - Madison



Abstract:

With further scaling into sub-45nm technology, we face new challenges to account
for the effects of process variations on the circuit timing. So far, the 
assumption has been that the exact distributional information of the process 
variations are known. However in practice we might only know the average and 
range of the process variations as provided by the foundries, or we might only 
have an approximate estimate of within-die variations, as modeled using the 
positions of the logic gates obtained from a coarse placement tool. Under 
these statistical uncertainties, our goal is to find a robust estimate of 
probabilistic measures of timing. One such measure is the "timing-yield" which 
is the probability of a circuits timing to be less than a target timing constraint. 
A robust estimate should be made all the scenarios that match with the 
partially available statistical information.

The talk is presented in two sections. The first part discusses a technique for 
robust estimation of timing-yield of a circuit under incomplete statistical 
information on process variations. In the second part, we model the distribution 
of the process variations as a Skew-Normal distribution, based on the available 
variability information. This is because the Skew-Normal distribution has a 
lot of flexibility and captures a variety of shapes. We then show a very fast 
and highly accurate Statistical Static Timing Analysis technique which becomes 
possible as a result of the special properties of the Skew-Normal distribution 
that we take advantage from. 

Biography: 

Azadeh Davoodi is an Assistant Professor at the ECE department since Fall 2006. 
Lin Xie is a PhD student working with Azadeh since Spring 2007. Their research 
interests include addressing the recent challenges of process variations including 
uncertainty in their distributions and the very high dimension of variation sources. 
Other research interests include physically-aware logic synthesis for incorporation 
of interconnects and within-die variations, and on parallel optimization techniques 
for IC-CAD problems in different computing environments such as multi-core and 
computing grids.