A Digitally Enhanced Multi-Fab Discrete-Event Simulator for Semiconductor Manufacturing

Front-end semiconductor manufacturing is among the most complex job-shop environments. With a high mix of products and hundreds of processing steps on shared equipment, production flows result to be dense, intersecting, and re-entrant, making production system management particularly challenging. Modern fabs collect huge volumes of shop-floor data, enabling the precise determination of the system state at a chosen time. Given this context, Discrete-Event Simulation (DES) allows to capture the system dynamics and its stochastic behavior, serving as a key decision-support tool for performance assessment and what-if analysis. This paper presents an industrial case study of a large semiconductor fab modeled using a data-driven simulator. First, a data cleaning and integration framework is defined to combine data from sources such as Computer-Aided Design (CAD) layouts, Manufacturing Execution Systems (MES), and Radio-Frequency Identification (RFID) tracking systems into a consistent dataset. An input-generation module produces simulator input files and initializes the model to an exact snapshot at the selected time, with the goal of ensuring an accurate level of alignment between the simulated system and the real fab conditions. Finally, the approach operates on a large-scale DES (350+ tools, 200+ products) that incorporates a multi-fab mechanism: selected services are routed to a neighboring fab via a statistical interface that uses probability distributions derived from historical data to represent the transfer behaviour. The model is validated by directly comparing simulation outputs with real fab data. Work-in-Progress (WIP) profiles, lead time distributions, throughput, and tool utilization from the simulator are quantitatively measured and matched against the corresponding values extracted from the MES over the same time horizon. Leveraging the validated simulator, queue times, lead times, throughput, and tool utilization can be quantified under alternative management policies and configuration alternatives, enabling systematic testing and optimization of dispatching and release rules, and other production control levers.