Automating Data-Driven Decisions to Improve Key Financial and Operational Metrics in Semiconductor Manufacturing

Abstract

Semiconductor manufacturing is a complex, non-linear process. The processing order of wafer lots in a semiconductor fab are determined by thousands of decisions that must be made each day. Each decision impacts the cycle time of a lot which is compounded as it goes through up to 700 steps. Operators do not readily have access to the data they need to make optimal decisions.

This thesis focuses on automating data-driven decisions to empower operators to increase their productivity. By acquiring the right data and determining the key business decisions, lots can be prioritized more effectively to improve the fab’s KPIs.

We begin by performing a current state analysis to understand the fab’s performance to date. We then determine the decisions that drive outcomes in the fab. Data is then aggregated to properly inform those decisions. Next, we create a heuristic model that we hypothesize will improve the fab’s performance.

Although not completely optimal, the heuristic prioritization model was found to have significant process, performance, and visual management improvements. With the heuristic, lots are properly prioritized 50% more often, leading to cycle time being reduced 3.8 days for a single step in the process.

We conclude this thesis by discussing how to implement an optimized scheduler for the next iteration of improving lot prioritization.