Description
When testing new RC values from segment-based correlation in asap7, cva6 showed some significant TNS discrepancy between GRT and DRT. That seems to be caused mainly by close long parallel wire segments such as the ones in M6 on the following image.
The linear regression model that we currently use can't capture such specific context. The accumulated coupling cap causes a certain path point to have a different delay, degrading subsequent skew ending up in a substantially larger slack at DRT.
Suggested Solution
Add a small cost at DRT to help spreading such segments apart from each other when there are available tracks.
Additional Context
No response
Description
When testing new RC values from segment-based correlation in asap7, cva6 showed some significant TNS discrepancy between GRT and DRT. That seems to be caused mainly by close long parallel wire segments such as the ones in M6 on the following image.
The linear regression model that we currently use can't capture such specific context. The accumulated coupling cap causes a certain path point to have a different delay, degrading subsequent skew ending up in a substantially larger slack at DRT.
Suggested Solution
Add a small cost at DRT to help spreading such segments apart from each other when there are available tracks.
Additional Context
No response