Novel Thermal Model Development for System Level Thermal Performance

摘要

In general, the system level-thermal performance analysis, IC packages are simplified to a compact thermal model (CTM) from the detailed model for computer calculated time reduction. The CTMs include 2R, Delphi compact thermal model (DCTM) and compact conduction model (CCM) are widely used. However, these simply models and methods are difficulty to achieve accurate temperature result in complex environments and the prediction of temperature interacted with other components are also challenge. In this work, the novel boundary condition (BC) training scheme through numerical calculation methods was developed, which will create a more accurate and efficient DCTM and CCM. The BC set considers the impacts of the IC power, radiation emissivity, ambient temperature, other heat sources, PCB size, heat sink size and air velocity; the levels of these factors cover the range of common environments, which makes the training result could close to the divergence system environments. These models can use to provide more realistic CTM in system level thermal performance evaluation.The novel DCTM include multiple layers, each layer has 9 nodes, and 55 thermal resistances. Based on the partial result of 86 BCs training, the temperature is the similar result for outer nodes, and each layer has been decreased to 3 nodes and only 11 thermal resistances form the optimal resistance set by response surface optimization (RSO) and genetic algorithm-artificial neural network (GA-ANN) mthods. In the other hand, novel CCM has several cuboids based on different package structure and replaces the thermal resistance to thermal conductivity. Finally, the optimal thermal conductivity and size set is determined by RSO and GA-ANNThe results show that the new BC set can be used to train for the novel DCTM and CCM by numerical method. Compared with these current modeling methods, the novel CCM can meet the requirements of time performance, and it is also superior to other modeling in the accuracy of the model. In addition, it can also correctly reflect the impact of different environment and the effect on the temperature of other components.