The conceivable accuracy of experimental and numerical thermal analyses of electronic systems

The past decade has seen significantly increased use of computational codes to calculate the thermal behaviour of electronic systems. The benefits of these virtual prototyping tools are undisputed when performing parametric studies in early design phases. However, when the objective of the calculation is accuracy, as required for reliability and performance assessments, discussion about the level of accuracy to expect in practice becomes far from trivial. A natural question is how accurate numerical simulations are when compared to well-designed experiments in prototypes or final products. Many studies show amazing agreement, the conclusion often being that 'validation of the numerical model' has been proven. These conclusions are subject to serious doubt. This paper discusses 'every' topic associated with a comparison between numerical and experimental results that is based on first principles, not on fitting parameters until the two results match. The final conclusion is inevitable: the situation when all computations at system level can be used for accurate temperature prediction is still a long way off. There are two reasons for this. The first is the lack of sufficiently accurate input parameters and boundary conditions. The second is the fact that several complex physical phenomena that rule the heat transfer behaviour of electronic systems concur with complex geometries. From a practical point of view, detailed analysis of reality cannot and will not be performed for a long time to come. Several suggestions of how to solve these problems are presented.