Summary
Re-analyze critical paths with path-specific slew propagation to recover pessimism from graph-based analysis (GBA).
Current State
GEM's STA (compute_timing()) uses graph-based analysis — worst-case delay at each pin regardless of which path is being analyzed. This is pessimistic because a pin's worst-case delay may come from a different path than the one being analyzed.
Proposed Approach
- After GBA identifies top N critical paths, re-analyze each with path-specific input slew values
- Use Liberty NLDM (Non-Linear Delay Model) tables to compute delay as f(input_slew, output_load)
- Typically recovers 5-10% margin on critical paths
- Only needed for paths close to violation threshold
Impact
Medium — reduces false-positive timing violations and gives more realistic slack values.
Effort
Medium-High — requires NLDM table lookup and slew propagation along specific paths.
Summary
Re-analyze critical paths with path-specific slew propagation to recover pessimism from graph-based analysis (GBA).
Current State
GEM's STA (
compute_timing()) uses graph-based analysis — worst-case delay at each pin regardless of which path is being analyzed. This is pessimistic because a pin's worst-case delay may come from a different path than the one being analyzed.Proposed Approach
Impact
Medium — reduces false-positive timing violations and gives more realistic slack values.
Effort
Medium-High — requires NLDM table lookup and slew propagation along specific paths.