1. Finalized the two submissions to ISPD.

   Including formating and latex source submission.
 
2. Discussion with Joanna on Multi-level routing

   The discussion is mainly on the terminology of ML routing and
   the general concept of global routing and detailed routing.

   I went through the code with her a bit, explained the
   format of the benchmark, and the framework of ML routing.

3. Read papers in ASP-DAC
   
   I read one paper from D.F. Martin on statistical shortest path
   algorithm, which was the one they used in their buffer insertion
   paper published in ICCAD 2005.

   I had high expectation on their paper, but it turns out that what
   they exclaimed is not what I expected. The shortest path problem
   is solved with a lot of simplifed assumptions (indepence).

   There is another paper from Georgia Tech which claimed to have 
   solved  a similar problem (statistical Bellman-Ford Shortest 
   paths problem). I will read more into this paper to see if there is
   any new contribution.

=== Carry-over from last week ==== 
4. Detailed discussion with Yan on 
   (1) how to model spatial correlation in SSTA for FPGA
       e.g., how to get the first-order canonical form, and
             how to obtain the coefficient in the canonical form?
                 There are generally two ways, one is from SPICE simulation
                 and the other is from analytical derivation, assuming
                 we know what the underlying relation is, either from
                 ckt physics or from ckt theorem).
                 
                 In fact, I also discussed this with Hao as he may 
                 be interested in extending his work on this subject. 
                  

   (2) how to speed-up SSTA computation considering spatial correlation
       e.g., how to improve accuracy without loosing correlation,
             how to speed-up by pruning

             I showed Yan that why the accuracy will be an issue. It
             is mainly because of the loose of correlation, both 
             due to multi-way maxing, and due to path convergence.
             This problem can be solved in two ways (1) keep all
             correlation terms without lumping them into a random 
             un-correlated term; (2) introduce an extra correlated term
             to match the variance.

             I also showed Yan the SSTA run time can be reduced by
             effectively pruning out some small correlated terms
             in the cannoical form. Yan will implement this
             technique as far as I understand. 

5. Discussion with Lerong on
   (1) how to improve the spatial correlation model for DAC submission
       e.g., how to consider systematic spatial variation
      
       This is done by extending the linear model with an extra term
       to capture the systematic variation (or true die-to-die
       variation).

   (2) experiments to do to better present the exp data
 
       Details, please see Lerong's report


-Jinjun