A detailed IC package numerical model calibration methodology

Abstract

Experimentally derived structure functions can be used to provide insights into the thermal resistances and capacitances heat experiences as it travels from a die through and beyond an IC package.A 3D "detailed" numerical model of the package that purports to explicitly represent the internal construction of the package requires material properties and geometric sizes to be accurately specified. A structure function derived from simulating the detailed numerical model can itself be compared to the experimentally derived reference example. Deviations between experimental and numerical SFs indicate error sites within the detailed model and an indication of whether the thermal resistances or thermal capacitances of the numerical would need to be increased or decreased to match the experimentally observed values. Iterative modifications of the detailed model, based on successive structure function comparisons, will achieve a fully calibrated detailed numerical package model. Keywords NomenclatureSF -Structure Function R -Thermal Resistance (degC/W) C -Thermal Capacitance (J/DegC) Tj -Junction Temperature (degC)

Automatic calibration of detailed IC package models

“…For a package calibration task, these are typically physical dimensions or properties that are difficult or impossible to measure with accuracy. Thermal interface material (TIM) thickness and effective thermal conductivity, interfacial contact resistances, active area dimensions, and die size are examples from previous work [6,7,8].…”

“…The experimental data and the initial simulation model are the same as was previously used by Bornoff and Vass-Varnai [6] using the manual calibration methodology. The aspects of the model that were calibrated previously will become the calibration parameters in this work.…”

“…The aspects of the model that were calibrated previously will become the calibration parameters in this work. The T3Ster hardware experimental setup, and the details of the FloTHERM simulation model are found in the original paper [6].…”

“…The structure function comparison is qualitative in nature, using visual inspection of charts and experience to gauge goodness of fit. This methodology has been described in detail previously [6] and has been used successfully to produce simulation models that show qualitative agreement with experimentally derived structure functio ns for IGBT and MOSFET applications. [6,7,8] Structure function based model calibration is essentially an optimization problem.…”

“…This methodology has been described in detail previously [6] and has been used successfully to produce simulation models that show qualitative agreement with experimentally derived structure functio ns for IGBT and MOSFET applications. [6,7,8] Structure function based model calibration is essentially an optimization problem. This work will formalize the calibration process by describing an optimization cost function, and then minimizing this cost function within a prescribed design space.…”