System performance now depends as much on advanced semiconductor packaging as on transistor scaling, especially with chiplets, 2.5D interposers, and 3D stacks. The resulting electro-thermal-mechanical coupling makes conventional electromagnetic (EM), finite-element (FEM), and computational fluid dynamics (CFD) analysis costly for design-space exploration. This review synthesizes how artificial intelligence (AI) and machine learning (ML) accelerate and augment packaging workflows across four domains—signal/power integrity (SI/PI), thermal integrity, structural/assembly behavior (warpage and stress), and reliability—and how AI and ML enables multi-physics co-design. We organize prior work by design role (surrogate modeling, design tuning, time-series tracking and multi objective co- design) and by algorithm family for fast surrogates; deep models (convolutional neural networks and long short-term memory networks) for images and waveforms; emerging reinforcement learning for routing, stack-up, and parameter auto-tuning; and physics-informed networks for thermally constrained settings. Cross-domain analysis reveals which models generalize broadly and which are specialized for data types or physics constraints. Remaining challenges for industrial adoption include reliance on simulation-only datasets, limited generalizability across architectures, incomplete uncertainty quantification, and interpretability concerns. We emphasize ongoing opportunities in integrating physics knowledge, uncertainty-aware learning, digital twin frameworks, transfer- and semi-supervised learning strategies, standardized benchmarking, and embedding AI and ML into electronic design automation (EDA) flows. Overall, AI and ML are advancing semiconductor packaging from simulation-driven iteration toward learning-augmented co-design, enabling more efficient, robust, and holistic optimization. • Reviews AI/ML methods accelerating advanced semiconductor packaging design • Covers SI/PI, thermal integrity, warpage/stress, and reliability domains • Taxonomy by design roles: surrogates, tuning, tracking, multi-physics co-design • Highlights adoption gaps and opportunities and EDA integration
Rafiee et al. (Tue,) studied this question.