( ESNUG 464 Item 4 ) -------------------------------------------- [03/30/07]
Subject: ( ESNUG 458 #4 ) Conformal finds DC/PhysOpt was missing 40 DFFs!
> In my experience, there's a tight coupling between Design Compiler (which
> creates "dwsvf" files) and Formality. These extra "dwsvf" files makes
> Formality's job a lot easier. Without the files Formality sweats. In
> our last project we had to give up on Formality in our Magma flow because
> the tool just couldn't verify.
>
> - Jay Pragasam
> PLX Technology Sunnyvale, CA
From: [ Seek And Ye Shall Find ]
Hello John,
I would like to share with you bug I had with DC/PhysOpt at my previous
company. This DC/PhysOpt bug may not be detected if there is a side
information provided between DC and Formality.
We ran Verplex (Cadence Conformal) against a netlist that was generated from
Synopsys DC and optimized by PhysOpt, and found a non-matching number of end
points between our Golden RTL and our revised netlist.
We couldn't find why around 40 DFFs disappeared from our netlist?! They
were right there in our RTL. Using Verplex we proved that those 40 DFFs
shouldn't have been optimized out. Looking at the constants we traced the
issue to wrong/bad optimization by PhysOpt/DC.
By setting a switch in PhysOpt/DC that avoids constant propagation through
sequential cells, we were able to prove that the design is passing EC;
which meant that there was a bug in the synthesis tool.
It's our belief that it is best to avoid passing info directly from your
synthesis tool to EC as it may create assumptions that may prove wrong.
EC and synthesis should be unrelated tools, preferably from different
vendors. Also, EC should be a planned task and not assumed to be easy as
it's your gate keeper against faulty design.
Please keep me anonymous.
- [ Seek And Ye Shall Find ]
Index
Next->Item
|
|
