Application of Stochastic Deconvolution Methods to improve the Identification of Complex BCI Multi-port Thermal RC Networks

The thermal modeling of electronic components is more and more crucial to prevent ageing phenomena when the component temperatures exceed their operating limits. At board level, the analysis of their mutual thermal interactions is done using multi-port RC networks as thermal models. Those compact models are usually extracted from a full physical representation of the component and its thermal behavior for a set of boundary conditions. Unfortunately, that fine description of the device requires a set of information about the package that is often not available.To complete the missing thermal properties, transient measurements of the junction-to-case behavior proved to be very useful. Using step function responses, a set of network identification by deconvolution of RC thermal model was conducted using iterative methods such as Bayesian ones. As a main result, a practical procedure is proposed that allows a direct extraction from discrete time constant spectrum of a very low-stage RC thermal model in the form of Foster ladder. The derived RC model demonstrates a high agreement with experimental data. The comparison is done on a set of devices mounted on a test vehicle. Further, the network identification procedure is conducted on the detailed numerical model of each device for calibration prospect. That calibration procedure of unidirectional responses permits to fix model discrepancies with the aim of creating relevant Boundary-Condition-Independent multipath RC networks.