Toward autonomous decision support in semiconductor manufacturing: a review of planning, scheduling, dispatching, and execution control

Abstract

Abstract Semiconductor manufacturing is a high-stakes cyber-physical production domain in which planning targets, scheduling feasibility, dispatching actions, process-control feedback, and human oversight must remain continuously aligned. This paper presents a structured state-of-the-art review of semiconductor smart manufacturing from the perspective of autonomous intelligent systems (AIS). The review does not treat planning, scheduling, dispatching, real-time dispatching (RTD), advanced process control (APC), manufacturing execution systems (MES), advanced planning and scheduling (APS), and enterprise resource planning (ERP) as separate production-management topics. Instead, it synthesizes them as bounded components of a human-supervised closed-loop decision architecture. The central artifact is an AIS reference model that maps fab layers to state inputs, decision-support functions, human authority, and audit or safeguard mechanisms. The review also develops a deployment-maturity matrix for classical optimization, simulation, rule-based dispatching, predictive learning, reinforcement learning, digital twins, LLM-enabled orchestration, and industrial software infrastructure. The synthesis shows that classical methods remain central because they provide interpretable constraints and executable interfaces. Learning-based and large-language-model-enabled methods can strengthen state estimation, policy adaptation, semantic mediation, and human-system interaction, but only under validated simulation or offline evaluation, bounded authority, safe fallback, access control, and auditable records. Semiconductor fabs should therefore not be described as fully autonomous systems. A more defensible AIS framing is trustworthy, human-supervised autonomous decision support: systems that sense changing fab states, evaluate alternatives, recommend or execute approved actions, and feed outcomes back into planning, scheduling, dispatching, and process-control decisions.

The authors' abstract, as published at the source. Autonomous Intelligent Systems, 2026 · DOI ↗