Drafting and Multi-Input Switching in Digital Dynamic Timing Simulation for Multi-Input Gates

TL;DR

This paper extends a digital timing simulation tool to support multi-input gates with novel delay formulas, improving prediction accuracy for integrated circuit timing and power analysis without sacrificing performance.

Contribution

Introduces a multi-input gate extension with analytic delay formulas based on a hybrid model, enhancing simulation accuracy and simplifying gate characterization.

Findings

Improved delay prediction accuracy over previous models

Supports multi-input gates like NOR and NAND with new formulas

Maintains performance while enhancing simulation fidelity

Abstract

We present a prototype multi-input gate extension of the publicly available Involution Tool for accurate digital timing simulation and power analysis of integrated circuits introduced by Oehlinger et al. (Integration, 2021). Relying on discrete event simulation, the Involution Tool allows fast timing simulation of circuits made up of an arbitrary composition of supported gates, provides automatic random input stimulus generation, and supports parameter sweeping. It also enables a detailed comparison of the delay predictions obtained by different models, including pure and inertial delays as well as digitized SPICE-generated reference traces. Our extension added support for 2-input gates like NOR and NAND, by implementing novel analytic delay formulas obtained via a refined analysis of a recently proposed thresholded first-order hybrid model of such gates. The resulting formulas faithfully cover not only multi-input switching effects (also known as Charlie effects), but also the decay of short pulses (aka Drafting effects). Besides the fact that our analytic models not only allow the derivation of closed-form delay formulas for arbitrary compositions of such gates, they are also key for a strikingly simple procedure for model parametrization, i.e., for gate characterization, which only needs three characteristic delay values. Using the extended Involution Tool, we compare the delay and power predictions for some benchmarking circuits stimulated by randomly generated input traces. Overall, our results reveal considerably improved prediction accuracy compared to the original Involution Tool, without a noticeable performance penalty.