Using ARIMA Analysis to Forecast the Wafer-Out Demand of a Foundry - A Case Study of A Company

• Main Authors: Hu, Ya-Chih, 胡雅智
• Other Authors: Tong, Lee-Ing
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/2hx585

碩士 === 國立交通大學 === 管理學院工業工程與管理學程 === 107 === Taiwanese foundry industry has the largest share in the global market. In recent years, the American Globalfoundries and South Korea's Samsung Electronics are actively challenging Taiwanese foundry, and the Chinese government also actively supports its semiconductors. The competition of Taiwanese foundry industry in the future becomes quite fierce. Wafer foundry is a highly capital-intensive industry and its production model is make-to-order, therefore, forecasting wafer demand is a very important issue in foundry industry. This study utilizes a Taiwanese foundry as a case study and proposes a time-series model to forecast its wafer output demand. The proposed forecasting model can provide information of the wafer-out demand of a foundry in addition to its traditional prediction method (i.e. Bottom-up method) to avoid wrong strategy of wafer production which may result in unmeet wafer demand or wasting capital expenditure. In this study, the ARIMA analysis was used to establish a wafer demand forecasting model. The results show that the multivariate ARIMA model including the customer’s behavior (e.g. new tape-out) predictor has the highest prediction accuracy among other ARIMA models.