Thermal placement optimization of multichip modules using a sequential metamodeling-based optimization approach

• Published in: Applied thermal engineering Vol. 30; no. 17; pp. 2632 - 2642
• Main Authors: Cheng, Hsien-Chie, Tsai, Yang-Howa, Chen, Kun-Nan, Fang, Jiunn
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.12.2010; Elsevier
• Subjects: Applied sciences; Chips (electronics); Design engineering; Energy; Energy. Thermal use of fuels; Exact sciences and technology; Face-centered central composite design; Finite element analysis; Finite element method; Halton sequence; Heat transfer; Mathematical analysis; Multichip module; Optimization; Placement; Response surface methodology; Strategy; Subspaces; Theoretical studies. Data and constants. Metering; Thermal placement optimization; Halton sequence; Response surface methodology; Finite element analysis; Multichip module; Thermal placement optimization; Face-centered central composite design; Polynomial approximation; Iterative process; Natural convection; Iteration; Composite material; Optimization; Case study; Geometry; Finite element method; Accuracy; Feasibility; Sampling; Performance
• ISSN: 1359-4311
• DOI: 10.1016/j.applthermaleng.2010.07.004

The study attempts to seek the optimal thermal design of planar multichip module (MCMs) under natural convection through optimal chip placement design. To attain the goal, a sequential metamodeling-based optimization approach is introduced. This approach incorporates a response surface methodology (RSM)-based design of experiment (DOE), three-dimensional (3D) thermal finite element analysis (FEA) and an updating scheme. Essentially, the RSM is used to construct, via quadratic polynomial approximation, the global RS of the chip junction temperature in terms of design variables. For speeding up the DOE and the solution of the optimization, several dynamic experimental design strategies using move limits and different proposed sampling techniques are introduced. The feasibility of the strategies is demonstrated, and their solution accuracy and efficiency are also compared with each other. By the explicit RS-based performance function together with geometry constraints, a constrained thermal optimization subproblem is formed. The optimum of the subspace optimization is sought, which is considered as the nominal starting point of next iteration. The iterative process continues with a new defined design subspace and factorial design plan until convergence is attained. The applicability of the proposed design optimization technique is demonstrated through several design case studies involving various planar MCMs.