Recent progress in compact thermal models

Abstract

The paper presents a number of recent developments in compact thermal modelling. A substantial part is devoted to a discussion on the accuracy of Boundary-Conditionlndependent (BCI) compact thermal models (CTMs) in practical situations. It bas been repeatedly shown that high accuracy can be obtained when comparing CTMs to detailed model results. However, these results are usually generated using certain assumptions. The validity of these assumptions is the main subject of this paper. An important conclusion is that the 'traditionally' generated CTMs perform very well in 'real-life' environments. Other recent developments that are shortly reviewed are: progress in dynamic compact modelling, the possibility of parameterisation, experimental calibration, and standardisation of CRvls. On the accuracy of compact thermal modelsSince the publication of the fust paper on BCI models by the DELPHI consortium [I], the concept has attracted wide attention because of its potentially high accuracy. These models enable the system level designer to predict tbe temperature of critical components with the accuracy required for reliability and performance prediction, without knowing all the details of the component in question. An overview of the principles of CTM generation can be found in Appendix A. The accuracy of Boundary-Condition-Independent (BCI) compact thermal models is still a matter of dispute because it is often not clear to what accuracy it should be compared.

Parameter independent model order reduction

“…The solid line is the error plot of the reduced system (38) whose dimension is q = 33. All the errors are below 0.01, which is already accurate enough for engineering applications [15,16]. If we want a more accurate small model, we can increase the dimension of the reduced system (38), the dotted line is the error plot of the reduced system (38), whose dimension is q = 49.…”

Parameterized physical compact thermal modeling

“…4) Our modeling approach is not as BCI as the DELPHI and other existing modeling approaches. This is because the surface area division method used by our model is not exactly the optimal one [23], although it is proved to be a reasonable one as shown in Section IV and [10]. It is also partly due to the fact that heat spreading in the package cannot be as well considered in our modeling approach as in the detailed models.…”

Boundary-Condition-Independent Compact Models

“…To address this topic in some detail, a simple trial and error approach was chosen by Lasance. 5 Using the dedicated compact model generation software (Lasance et al 4 ) it is very easy to select and deselect boundary conditions from the training set, run the optimization (usually within 20 seconds) and test the resulting CTM against the test set (in this case 67 boundary conditions). In this way, it is easy to conclude which boundary conditions are essential.…”