Enhancement of Thermal Transient Prediction Accuracy for Mobile AP Package

As the demand for high performance of the mobile application processor (AP) grows gradually, the power consumption increases so that the precise estimation of AP temperature becomes more important. Accurate prediction of Dynamic Thermal Management Starting Time (DTMST), which is the starting time of frequency throttling to lower the operating temperature is a key for designing AP. In this paper, the accurate prediction methods for modeling thermal transient behavior of AP are presented and demonstrated by using Thermal Test Vehicle (TTV). First the machine-learning is used to calculate an effective thermal conductivity (ETC) to account for a complex metal pattern in substrates. Second the construction and comparison of RC network is used for the further enhancement in accuracy. With the methodologies, the prediction accuracy of DTMST in simulation has been improved to 95.6% which can estimate performance of mobile AP successfully. The transient thermal modeling predicts the superior performance of FO-PLP compared to I-PoP due to the thermally enhanced package specifications.