AI-enabled Automatic Molding Compound Selection for A Power Device with High Solder Joint Reliability

Abstract

Solder joint fatigue crack is one of the key failure modes in IC packaging, particularly for power packages under a thermal cycling test (TCT). Design optimization and material selection were widely reported to achieve a high solder joint reliability performance. Epoxy molding compound (EMC), as an encapsulant, takes an important role in package reliability. Its major parameters include modulus, coefficient of thermal expansion (CTE), and glass transition temperature (Tg). For EMC selection, the traditional DOE approach based on reliability test and/or finite element (FE) modeling is constrained by time and cost, which in turn limits the possibility to get the best combinations. With the development of artificial intelligence (AI) technology, it enables package design and material selection automatically, which can significantly enhance the efficiency and robustness of devices. In this work, an AI-enabled framework for automatic EMC selection for power packages was developed to achieve high solder joint reliability performance under TCT. The model consists of a FE model, a multi-fidelity Bayesian optimization (MFBO), and an artificial neural network (ANN). A FE model was first constructed to characterize the accumulated creep strain in the solder joint under TCT. In the MFBO model, the properties of EMC including Prony series, WLF shift function, and CTE were employed as input, and accumulated creep strain per cycle in the solder joint obtained from the FE simulation was employed as output. The optimized Prony series and WLF shift function corresponding to the lowest creep strain in solder joint was then forwarded to the trained ANN model and translated into modulus at different temperatures and Tg.The EMC selected by this framework was validated by experimental results. Besides solder joint fatigue life enhancement, this framework can be further extended to other applications with automatic materials selection, such as thermal management, WLP warpage, etc.