多物理
粒子群优化
计算机科学
人工神经网络
热的
工程类
有限元法
算法
人工智能
物理
结构工程
气象学
作者
Dongdong Chen,Xianglong Wang,Yintang Yang,Di Li,Gaoliang Li
标识
DOI:10.1109/ted.2023.3302825
摘要
The intelligent codesign strategy is proposed to balance the thermal management and cost control for 3-D integrated system with thermal through-silicon via (TTSV). The complex relationship between the thermal performance indexes and parameters of TTSV in the 3-D integrated system is investigated by the COMSOL Multiphysics software. According to the design logic, the row/column number of TTSV array and the distance of two TTSV unit are the global parameters, while the thickness of oxide layer, the radius of TTSV, and the height of oxide insulating layer are the local parameters. In order to improve the simulation efficiency, the modified neural network models are established based on the simulation data to replace the finite-element model. In the proposed codesign strategy, the thermal management and cost control are considered in the multiobjective evaluation function. According to the particle swarm optimization (PSO) algorithm, the multilevel PSO algorithm is developed to optimize the global and local parameters of TTSV. The effectiveness of the proposed codesign strategy is verified by six cases of two testbeds. The simulated temperature indexes are all lower than the desired targets (360, 320, and 20 K), and the number of TTSVs is decreased at the same time. Therefore, the proposed codesign strategy can not only realize the excellent thermal management but also achieve a compromise between thermal management and cost control.
科研通智能强力驱动
Strongly Powered by AbleSci AI