Automatic tool for real-time estimation of QFN-related heat transfer in multi-layer PCB by using SPICE simulations

Abstract

Power hungry electronic components such as CPU, GPUs as well as voltage regulators heat up during operation. Several sensing applications require ambient air temperature and humidity measurements. Nevertheless, ambient air temperature and humidity measures with a surface mounting technology devices is challenging due to the thermal influence of nearby components. In this context, thermal design of Printed Circuit Boards (PCB) becomes a critical step to ensure the reliability of electronic systems and to preserve those components that are vulnerable to heat-accelerated failure mechanisms. Although PCB thermal analysis increasingly relies on software embedding complex but accurate fluid dynamics or finite elements solvers, simulation times during the design phase result very long and are not suitable for rapid prototyping processes. To bridge this gap, this paper proposes an automatic tool for the rapid simulation of heat transfer pathways inside a PCB when quad-flat no-lead (QFN) packages are employed. The proposed tool exploits a resistive networks-based model able to adapt to the metal/dielectric/soldermask composition of the analyzed area. It is made possible by integrating image processing algorithms to identify thermal connections between the analyzed elements, allowing a multi-layer reconstruction even with irregularly shaped metal areas. The proposed tool has been tested on a PCB and the results compared with the ones from professional software for FEM thermal analysis. The proposed modeling system can ensure optimal accuracy on the chip area (error compared to $\mathrm{F E M}\lt 1^{\circ} \mathrm{C}$), and within areas of 9 $\mathrm{cm}^{2}$, resulting $\sim 91$ times faster than the equivalent FEM in estimating heating trend on the same board.