Chiplet Multi-Objective Optimization Algorithm Based on Communication Consumption and Temperature

Abstract

A chiplet multi-objective optimization algorithm for 2.5-D integrated circuit (IC) based on a passive interposer is discussed in this article. Inspired by the network-on-chip mapping problem, we propose a novel algorithm, called chiplet multi-objective optimization, which minimizes the average temperature and the communication consumption between chiplets at the same time. The algorithm considers the specificities of 2.5-D IC chiplets, such as the spacing and different sizes of chiplets. In addition to the weight factor, α is also introduced to make a balance between temperature and the communication consumption. The designer can change the weight factor according to their own requirement. The multi-window display system is used as an example in this article to demonstrate the algorithm’s efficiency and the accuracy. According to our algorithm, the system temperature of the most ideal solution can be reduced by 8.34 K and the communication consumption reduced by 232.13 μJ.