應用粒子群演算法求解最適ramp-up區間及新舊製程之產能分配問題

Abstract

半導體產業必須持續提升技術以增加競爭力。經由技術不斷地改良與創新 可提升每片晶圓的利用率來降低產品成本。另一方面IC產品的生命週期太短，因此大部分公司都希望能夠快速地將產能轉換到新製程上生產，將所有產能投入到新製程前的這段期間，我們稱為新製程的ramp-up。然而在這段期間新製程的良率是非常不穩定且偏低，因此為了滿足客戶需求，新舊製程會同時在生產線上上線生產，新舊製程的產能分配成為公司面臨的重要議題。 本研究假設新製程的良率學習受到自我學習與實驗學習的影響，此兩因素分別受限於新製程的累積投料量與累積工程批量影響，並且假設已知新製程的生命週期，在瓶僅機台產能與滿足客戶需求下，以最大化利潤為目標，建構晶圓新、舊製程和工程的投料批量以及最適ramp-up區間的決策模式，並運用粒子群演算法求解。經由案例的敏感度分析，了解新製程的生命週期、良率學習模式參數，對新舊製程投料批量以及最適ramp-up區間之影響及其管理意涵。

Continuous improvement of process technology in a semiconductor firm is an important issue. A new production process is frequently introduced to the production line when its production process knowledge is incomplete for rapid product lifetime and expensive development cost. The period from the end of a new production process development to full capacity utilization with it to produce products is known as production ramp-up stage. During the beginning of this stage, the yield of the new production process is very unstable and low. Therefore, current production process is simultaneously existed with the new production process for meeting customer demands. The yield of a new production process is frequently improved by the accumulated production knowledge from the release lots in the production line. We call it as leaning. This creates a trade-off between the cost of production loss for low yield and the benefit of learning. In this study we formulate a mathematical model to determine the ramp-up time length , the capacity allocation for a new production process, current production process, and engineering lots during the ramp-up stage, for maximizing total profit during the life cycle of the new process. Then we propose a modified particle swarm approach to solve the problem.