Statistical STA (SSTA) with delay distributions

[robtaylor]

Summary

Model gate delays as statistical distributions rather than single deterministic values, enabling probabilistic timing analysis.

Current State

GEM uses deterministic min/typ/max delays from SDF. This requires running multiple corners to explore the delay space, and still doesn't capture the full statistical picture.

Proposed Approach

• Model delays as Gaussian or other distributions parameterized by mean and sigma
• Propagate distributions through the circuit (sum of Gaussians for series paths, max of Gaussians for convergent paths)
• Report timing yield: probability that a path meets timing
• Can use Monte Carlo sampling on GPU for complex distributions

Impact

Low — academic interest, not widely adopted even in commercial flows.

Effort

Very High — fundamentally different analysis framework.

Notes

This is a long-term research direction. Monte Carlo on GPU could be a unique advantage of the GEM platform.