Thermal Profile Prediction for Ball Grid Array Solder Joints Using Physic-Informed Artificial Neural Network

According to comparable research studies, the reflow process has been identified as the most critical process to the Printed Circuit Board (PCB) solder joint quality in surface mounting technology (SMT) manufacturing. The ball grid array (BGA) are packages with one face partly covered with pins in a grid pattern connected to the pads on PCBs. Compared to passive components, the solder joints for BGAs are at the bottom of the packages. The joints for BGAs are harder to be heated because the convection heat from the air is blocked by the cover case of the BGA and relies more on the parasite conduction heat from the boards and the package cover. To ensure the quality of the solder joints, the solder paste manufacturers provide specifications, along with target temperature curves (i.e., thermal profiles), to secure the solder joint quality. One of the significant challenges for the BGA-related thermal study is to measure the temperature of the joints underneath the package and how to predict the temperature for the joints. In this study, (1) a physics-informed artificial neural network (PINN) was proposed for temperature prediction, and (2) the experiment was conducted to measure the solder joint temperature underneath the BGAs for validation. The prediction accuracy is higher than 96% in terms of the \({R}^{2}\) fitness to the actual thermal profile.