Production Forecasting in Semiconductor Manufacturing - By Ricardo Duarte
Dated: 16/11/2017

#ABSTRACT:

Semiconductor manufacturers are continuously ramping up production capacity and intensively searching for ways to optimize its usage. One important tool for production optimization and planning is the forecasting of future production performance. The current state-of-the-art for forecasting production is fab simulation, where different production scenarios are simulated in order to predict possible outcomes. This however consumes large amounts of time and computing resources and does not offer a continuously updated real-time representation of production forecasting. As the industry moves into new challenges like high diversification of products and services, production must be as flexible and agile as possible which requires the continuous creation and update of accurate forecasts.

In order to achieve this goal, production forecasting must replace complex and heavyweight simulations with simpler lightweight statistical forecast algorithms. While these methods are faster and more agile, they work in a different way than simulations. While a simulation can cover whole production scenarios, statistical forecast models have a more specific scope and typically focus on a single measure, e.g. a key performance indicator. As a consequence, multiple forecast models must be used in combination to achieve not only real-time capabilities but also coverage for whole production scenarios. The goal of this thesis is to analyze and select forecasting methods for production performance indicators and develop concepts for a multi-model forecasting system.