How to extend a thermal-RC-network model (derived from experimental data) to respond to an arbitrarily fast input

For a growing number of customers, the transient thermal response of packaged semiconductor devices is a critical issue. It is not enough to predict "time averaged" junction temperatures based on average power dissipation, because the actual duty cycle and associated transient response of a device may lead to peak junction temperatures vastly higher than such steady-state predictions. Experimental techniques exist for accurately measuring the thermal transient response of a physical device in the 100 /spl mu/s range, and thermal RC-networks can readily be generated to match these measurements. Such networks can then be exercised with modeling tools such as SPICE to obtain predictions for time-varying power inputs in this time range. Faster experimental measurements are often complicated by uncontrollable electronic interactions, yet in real-life applications, power cycling of a device may necessitate accurate predictions of the system thermal response in the 1 /spl mu/s time frame (or faster). It is shown that a thermal RC-network model based on experimental data can be readily modified by following straightforward rules, such that an arbitrarily fast response can be accommodated. This concept can also be used to optimize the meshing of 3D finite element models such that accurate transient response is obtained in the desired time domain, yet model size is minimized.