Evaluating Performance Tradeoff in Defect-Tolerant Gate Programming Techniques for the Clock-Free Nanowire Crossbar Architecture

摘要

A novel asynchronous nanowire crossbar architecture has been recently proposed by authors' research group. The proposed clock-free architecture provides numerous significant benefits over its clocked counterparts which include better man-ufacturability, scalability, modularity and robustness. We also proposed various gate mapping and reconfiguration algorithms for defect-tolerant programming of PGMB (programmable gate macro blocks) - which is the primary building block of the proposed architecture. These algorithms were tested by simulations and a variety of parameter values were applied to show their performance characteristics. The most important performance metric of the proposed techniques is the programmability (i.e., the ratio of successfully programmed gates to the total number of gates). However, algorithms with higher programmability should come with higher time/space requirements. In this work, we will evaluate the tradeoff between programmability and time/space requirements and suggest a way to find the most suitable algorithm with acceptable combination of programmability and time/space requirements.