An Event-Driven Method for Fast Delay Distribution Estimation of Array Multipliers

This paper deals with the statistical timing problem of array multipliers which have exponentially increasing number of paths for timing. Obtaining the distribution of those longest paths is a hard problem due to the simulation cost to cover all test-vectors exhaustively. Finding the longest path timing statistics can enable the designer to tighten the clock and incorporate error correction circuitry when timing violation occurs. Traditional deterministic timing analysis methods such as static timing analysis (STA) are overly pessimistic to meet the clock tightening requirement and unable to provide test-vector-related timing information. We present in this paper a tree-search method and an event-driven method, then compare their performance in characterizing the delay distributions. Tests expose that the event-driven method is able to capture the delay distribution of array multipliers more accurately with much lower computational cost than SPICE. The proposed methods are able to produce timing distribution up to word-length 16-bit whereas SPICE fails to run exhaustive path timing for multipliers larger than 8-bit.