A compact modeling approach using a genetic algorithm for accurate thermal simulation

Abstract

The rapid improvement in computer performance is intensifying the component thermal problem. It is becoming increasingly important for an optimal thermal design that thermal simulation is part of the design. Simplification of the thermal simulation model is inevitable as an enormous number of finite elements are required when the original CAD data set is adopted for modeling. However, the reduction of calculation time by model simplification and the maintenance of calculation accuracy are contradictory. Conventionally, model simplification is by empirical judgment, but a rational simplification technique using boundary conditions and material properties results in a more accurate and reliable calculation. Although simplification of the LSI component modeling method has been proposed by the Delphi project, it is difficult to apply other than to components, such as a keyboard. This paper proposes a new technique to generate the compact model of a keyboard with the required accuracy. First, some candidates for the simplified configurations are prepared. A genetic algorithm is proposed to identify the variables such as the boundary conditions and thermal conductivities that are most important in a high accuracy calculation. Finally, the optimum compact model which has the required accuracy is selected from the simplified models.