Virtual Production Control of Order Suggestion Optimization by Using Genetic Algorithm

[[abstract]]在專業分工的供應鏈體系下，Fabless (如IC設計廠) 專注於其設計本業，而設計後續之製造生產則委由晶圓與封測廠所負責。對Fabless而言，當晶圓製造段供給時程不確定性偏高時，為降低風險而提高緩衝時間，或增加庫存常為普遍因應之道，從而造成Fabless庫存成本居高不下。需求規劃系統若缺乏包含如供應產出與需求計畫等以彙整物料需求時間，將導致下單時機與數量不佳，而造成物料發生短溢風險。 本研究基於基因演算法，並結合虛擬生產管制系統(Virtual Production Control System VPCS)的產出推估，以提供Fabless 最佳晶圓訂單建議。在缺料分析上，利用庫存，已開單數量，良率，產品結構，需求計畫，與預期交貨計畫，可獲得缺料計畫量。在下單規劃上，目標為最大化達交下之最小累積庫存，應用穩態型基因演算法，採用固定族群數，並確保歷代基因可滿足月產能限制，在合理前推時間下，求得多種限制條件下的最佳訂單建議計畫。此外，本研究以VPCS推估可能交貨時程，以評估預期交貨計畫的有效性。 在研究成果上，與實際某Fabless之數期下單建議比較，經校正人為提前下單後，本研究與實際之下單量相同度為100%；就達交率而言，除因加工時間不足不予考慮外，可在產能限制下，規劃週期 (一般為四個月)內的98%達交，達交率提升了4 8%。此外，由於所提基因演算法的改善，產品數40可於2分鐘內獲得可行之近最佳解，使得本研究具實務使用價值。[[abstract]]To specialize in a supply chain a fabless company i e IC design house focuses its IC design and releases the following manufacturing activities to a foundry and packaging/testing companies A fabless company often increases buffering time or inventory for preventing the risks which are induced from high uncertainty of wafer supplying such that it suffers high inventory costs If the demand planning system of a fabless company fails to integrate supplying and ordering plans as a material demand it could induce poor timings and quantities of a placing demand; consequently material shortage or overflow risks occur to the company This study based on a genetic algorithm provides a fabless company with the optimal demand suggestion by integrating production estimation of a VPCS (virtual production control system) In material shortage analysis a shortage demand can be derived from quantities of placed demands yields product structures demand plans and projected delivery plans In planning of placing demand the goal of the optimal demand suggestion is to maximize order fulfillment and minimize accumulated inventory under reasonable lead times and various constraints by using a steady-state genetic algorithm with fixed number of chromosomes and monthly capacity limits Besides this study estimates a possible delivery schedule by using VPCS in order to assess effectiveness of a projected delivery plan Results of this work indicate that the similarity of placing demands is 100% of actual demands of a fabless company for several periods after correction of artificial shifted demands For order fulfillment this work shows mean order fill rate 98% in the planning periods the order fill rate is improved 4 8% e g four months under capacity limits and excluding the demands with insufficient lead times In addition the work presents its practice value while improving a genetic algorithm to derive a near-optimal suggestion of 40 products in 2 minute