Reconfiguring Global Electronics and Semiconductor Supply Chains under Escalating Tariff Risks: A Multi-Echelon Robust-Stochastic Optimization and Hybrid Machine Learning Approach

Tariff volatility has emerged as a central source of disruption in global semiconductor supply chains, where sudden policy interventions and shifting geopolitical priorities can rapidly alter sourcing incentives, production footprints and cross-border logistics flows. Traditional deterministic planning is often unable to accommodate these uncertainties or guide long-term network redesign. This study develops an integrated framework that combines data-driven tariff forecasting with a multi-echelon optimization model to support strategic reconfiguration of semiconductor supply networks under tariff uncertainty. A hybrid machine learning approach is employed to generate realistic tariff scenarios, where long short-term memory networks capture temporal patterns, Bayesian neural networks quantify predictive uncertainty and random forests detect structural regime shifts linked to policy changes. These scenarios are embedded in a multi-stage robust–stochastic optimization formulation that determines global sourcing, production allocation, capacity investment and inventory decisions while explicitly accounting for tariff ambiguity and operational recourse. Theoretical analysis establishes conditions for tractability and demonstrates that the hybrid robust–stochastic structure provides superior resilience compared to deterministic, purely robust or purely stochastic models. A customized Benders decomposition method is developed to solve the resulting large-scale formulation efficiently. Using a representative semiconductor supply network with suppliers in East Asia, fabrication facilities in Taiwan and the United States and assembly operations in South and Southeast Asia, the study illustrates how the proposed framework enhances cost efficiency, mitigates tariff-induced risks and supports more stable long-term planning. The findings highlight the practical value of integrating machine learning with advanced optimization for strengthening resilience in globally distributed, policy-sensitive supply chains.