Thermal Transient Modeling and Experimental Validation in the European Project PROFIT

Abstract

A major objective of the European project PROFIT is to generate boundary condition independent (BCI) dynamic compact thermal models (DCTM) of semiconductor products. Extending the methods for steady BCI-CTM developed in preceding projects DELPHI and SEED to the transient domain, a detailed numerical model of the component is needed, which is validated against four dual cold plate (DCP) experiments extracting heat along the main heat flow paths from a package. The validated detailed model is then used for numerical experiments in many environments represented by external BC. Results are used to optimize resistors and capacitors of a small network forming the DCTM. This work is focused on the first part of developing validated detailed dynamic models by comparison of modeling and measurements. Results of the European project PROFIT on thermal transient measurement and modeling of integrated circuit packages are presented. All together sixteen different packages from the three Semiconductor Manufacturers Infineon, Philips, and ST Microelectronics were measured in four DCP environments as defined in the preceding DELPHI and SEED projects. Solutions to measure TO-type and fine pitch packages in the DCP, especially for the critical DCP-4 boundary condition were demonstrated, as well as reduction of interface resistance and increased reproducibility by using Wood's alloy as an interface material. The measurements were simulated using the commercial software packages ANSYS, FLOTHERM, or MARC. The agreement between simulated and measured thermal impedance is 15% or better from steady state (t=1000 s) to transients with t>0.1, i.e., four orders of magnitude. In a few cases, this level of accuracy was kept even over seven orders of magnitude. Increasing relative inaccuracy with shorter transients corresponds to small absolute errors in temperature. So practical pulse temperature prediction will usually be correct within a few degrees. Extraction of geometrical and material parameters will need further improvement.