半導體後段一元化代工服務之協同式生產規劃及策略聯盟廠商遴選

Abstract

半導體生產一元化代工服務係由晶圓製造廠整合產業之生產資源，同時提供 晶圓針測(Probing Testing)、封裝(Assembly )與最終測試(Final Testing)的一元化半 導體生產加工服務，讓客戶省下繁複的訂單與跟催作業，並降低半成品在各階段 廠商與客戶間運輸之前置時間，此舉大大地提高的客戶下單的意願，進而創造出 晶圓代工廠的競爭優勢。然而，對於提供一元化代工服務的晶圓製造廠而言，則 必須肩起後段加工的訂單分派作業，因此，唯有建構一套半導體生產一元化代工 服務之協同式生產規劃，方能有效整合後段生產資源，與前段晶圓製造銜接，達 成準時交貨之任務。再者，為了控管前置時間及成本，必須將運輸時間與成本納 入考量。因此本計劃擬建構一套整數線性規劃模式，以最小化總成本為目標，求 解涵蓋前後段加工之多產品、多階段與多廠區之協同式生產暨運輸規劃。 半導體生產一元化代工服務同時面臨生產規劃與運輸規劃之問題，其中單是 運輸路線規劃問題，已被證明為NP-Complete。基於此，本計劃於第一年度進而 提出一套網路轉換方法，將原整數規劃問題轉換成TPP(Traveling Purchaser Problem, TPP)問題。要言之，吾人以TPP 能同時考慮多樣產品的特性，來對應一 元化代工服務多階段與多種產品的特性，再以TPP 中的市場遴選機制來對應一元 化代工服務之多廠區遴選問題。一旦完成一對一對應之後，吾人將可藉由TPP 中 發展已臻成熟之啟髮式演算法求解，以期達到兼具效率與品質之結果。 對於擁有多座晶圓製造廠之公司而言，提供一元化代工服務時必須考量所有 製造基地的地理位置，因此，在第二年度本計劃將原先屬於單一來料的半導體生 產協同式生產規劃問題，延伸為一個多方來料的協同式生產規劃問題，以期能夠 在考量多晶圓廠區的來料資訊下，完成最佳化的生產規劃。我們將原TPP 網路模 型修正為多方來料之網路模型，並進一步提出適用於MDSPTS 模型之啟髮式演 算法，作為多晶圓製造廠之一元化服務生產規劃問題之網路轉換使用。 在一元化代工服務的趨勢之下，晶圓製造廠可藉由與外包廠商簽訂策略聯 盟，以取得優先預約產能的權利，來確保長期穩定供貨的承諾。因此，第三年度 計劃擬採用資料包絡法(Data Envelopment Analysis, DEA)，從各規劃週期的生產 規劃結果中，進行各外包廠商的績效評比。吾人將從可提供產能、製程能力、加 工成本、最小下單數量...等績效指標的表現，挑選出最優先合作的協同夥伴，作為簽訂策略聯盟的外包廠商。接著再依據策略聯盟外包廠商之產能與製程能力資 訊，制定長期產能規劃，並訂出對各產品最低可承諾之產能。一旦規劃完成，則 此案將可作為提供半導體生產一元化代工服務之業者在接單時之依據，或作為晶 圓製造廠擴增產能之參考。

The semiconductor production turnkey service integrates the resources of outsourcing companies by wafer fabrication factory and offers the service of wafer probe testing, assembly, and final testing for customers. The customer can save the heavy and complicated ordering and tracing process, and also save the time spend on the comings and goings of the semi-manufactured goods between manufacturers and the customer. Therefore, the turnkey service is favoring by customers, and then creates the competition advantage of the wafer fabrication factory. Moreover, in order to control the lead time and cost, the transportation time and cost must be considered in the production planning and the cost control. Thus, collaborative production planning and transportation problem contains the job assignment and route selection among multi-products, multi-stages, and multi-factories and covers the front-end and back-end semiconductor processes. The first year plan regards minimizing the total cost as the objective function and constructs an integer linear programming (ILP) model to solve the problem. However, once the scale of the problem becomes large, the time for solving the ILP model will elongate excessively. This project then proposes a network transformation method to mate all components of original ILP problem with essential components of Traveling Purchaser Problem (TPP). The TPP is adopted because it can handle multi-products purchasing at the same time that is corresponding to the existing of multi-stage and multi-product for semiconductor turnkey service. Then the market selecting mechanism in the TPP can be used to deal with the multi-factory selecting problem for turnkey service. Once finishing one-to-one correspondence, the TPP can be solved by using heuristic algorithms already developed for TPP in the hope of reaching the result with efficiency and quality. If the fabrication company owns several factories, the semiconductor production turnkey service has to consider the geographical position of all factories further. Therefore, the collaborative production planning and transportation problem must extend the type of single material supplier into the type of multiple material suppliers. To achieve the optimization of production and transportation planning for multiple wafer fabs, in the second year plan, we will build an ILP model and revise the network model of original TPP as the MDSPTS network model. Then, we will propose MDSPTS heuristic algorithms and test their effectiveness for solving such a production planning problem. Under the trend of semiconductor production turnkey service, the wafer fab can make the strategic alliance with outsourcing companies to guarantee the commitment of long-term capacity supply. In the third year, this plan adopts the data envelopment analysis (DEA) to implement the performance evaluation for each outsourcing company by reviewing planning results in past planning periods. The performance indicators are the available capacity, the process capability, the manufacturing and transportation cost, and the minimum quantity of order, etc. Then, the collaborative partners will be selected for strategic alliance. With the available capacity and process capability in each allied factory company, we can make long-term capacity planning so as to confirm the minimum capacity provided for each product type. Such a result can be the reference of order selection for semiconductor turnkey service, or can be deemed as the reference when the wafer fab considers capacity expansion.