Posted by Godwin Maben on February 6, 2009
Wow, it’s been almost 2 months since I blogged. My apologies for the same. Lately I have been very busy with many UPF engagements and UPF based tape-outs.
I would like to mention one incident related to the ongoing structural error topic, wherein the Silicon failed. Detailed debugging showed how Power Intent being “incomplete” can lead to this catastrophic event.
When we look at this from 10000 feet, it is a simple case of missing special cells . Some special cells did not get inserted because Power Intent was not complete and most of the LP checkers said that everything is ok and design is electrically safe. I would not blame this on checkers though, as they verified the design as per the Power Intent provided.
Later detailed review of the intent, clearly pointed towards incomplete Power Intent. On fixing this, pretty much every issue was caught and fixed peacefully. Now this could be avoided if
(a) Full blown MV aware Functional Simulation was in place to verify the design at RTL level
(b) Use of Power Intent Critique feature within LP checkers to validate your intent at a very early stage.
Initially we might feel that the design is very simple and has only 2-3 power domains, but just having less number of power domains does not reduce the minimize the complexity of the issue.
lets look into more such cases in next few posts.
- Godwin Maben