Thermal imaging and power mapping at IBM

By careful control of chip cooling conditions and accurate modeling, the images produced by a typical IR camera - which show radiant intensity, and thus temperature, as a function of position - can be utilized to derive considerably more information than simple hotspot location. The focus of this work is to review in detail a practical method by which a quantitative map of power density can be produced by de-convolving the thermal map with the local heating point spread function. The resulting power density maps can then be used to predict the chip temperature under a range of conditions, for example, minimum and maximum ambient air temperature, air flow, thickness variations in the heat sink or thermal interface material, etc. In this way, the worst case chip temperature can be estimated under the prevailing constraints. A primary objective of the early work described here was to produce estimates of on-chip temperatures in the system, for a range of expected manufacturing variations in packaging. In addition, we comment on some tradeoffs involved when applying these techniques to other situations.