随机行走算法在IC芯片热分析中的应用

Applying Random Walk Algorithm to Chip Thermal Analysis

针对大规模IC芯片中局部高温热效应问题,提出一种应用随机行走算法对芯片进行热分析的技术.首先简述随机行走问题的基本概念,给出问题的理论解法;然后讨论基于三维网格划分的IC芯片热等效阻容网络模型,着重分析随机行走算法在热等效模型中的应用,并建立一个实际芯片的热等效模型,采用随机行走算法模拟计算模型的稳态温度分布和瞬态温度变化;最后针对随机行走算法应用中的不足,提出了伪并行行走和结果重用2种加速技术.实验结果表明,随机行走算法能够胜任大规模热等效网络的分析,计算精度满足热分析要求,所提出的加速技术能够显著地提高算法的执行效率.

A novel method for IC chip thermal analysis based on random walk algorithm is proposed to cope with the hotspot problem. Firstly, the random walk problem is introduced and the solution is schemed out theoretically. And then, the equivalent RC circuit for thermal analysis is discussed as the IC package is divided into many small blocks at the desired level of granularity. The techniques of analyzing the equivalent RC network by random walk algorithm are then discussed. At the same time, an equivalent model for thermal analysis is set up according to an IC chip. The proposed method is then applied to this model for both steady and transient analysis. Two speedup strategies, pseudo parallel walking and walk results reuse, are proposed sequentially to improve the algorithm performance. Experimental results indicated that the proposed method can afford large thermal equivalent RC network analysis with acceptable error, and the speedup strategies do effective work.