臺灣半導體供應鏈上游-晶圓廠與下游-測試廠營運關聯分析與評估

半導體製造始於1960年至今超過半個世紀的歷史，於全球經濟仍具有不可撼動的地位。無疑其產品應用多元廣泛，涵蓋工業、消費性電子、AI、雲端、交通、車用甚至軍用的相關功能皆需要透過半導體組件來執行。而動態隨機存取記憶體中的DRAM記憶體模組，更是半導體產業中不可或缺的一環，相對的，DRAM產業在整體半導體產業供應鏈中扮演關鍵的角色。然而影響記憶體的價格漲幅因素與市場經濟發展有著環環相扣的高度關聯。影響記憶體價格浮動因素會隨著時間而有所變動，準確預測影響因子具有相當的難度。倘若以單純歷史營運數據走勢做為未來績效的預測，進而使企業評估相對應的策略，以降低成本，達到營運目標。因此本研究欲利用公開資訊觀測的財報數據進行純益率計算，將特定上游晶圓廠與特定下游封裝測試廠近五年的財報數據，藉由上游廠季度間純益率之差走勢去預測下游廠商下一季度的走勢，做上下游之間數據分析與評估。分為兩個部分，一、分析數字變化的原因:將過去數據進行關聯度分析，透過數字變化去看經濟的歷史走向。二、準確度:依據數字去做二元邏輯回歸，評估數字間的關係是否可以進行未來走勢預測。以衡量評估上游晶圓廠經營績效與下游封測廠接單營運狀況的關聯因子分析。透過皮爾森相關係數可以得到2018年至2020年第一季度皆呈現為正相關;自2020第二季度到2022第三季度則呈現負相關。而藉由二元分類邏輯回歸可以看到在此條件下其數據表現的準確度為0.75。由於本研究時間僅四年，未來可將研究期間拉長以及增加相關廠商去觀察其變化的幅度是否會影響走勢的變化，再透過此模型或是相關預測模型驗證其準確度。進階透過成長正/負區間去安排人力以及向材料供應商討論價格調整幅度。

關鍵字

半導體； DRAM ；產業供應鏈；二元分類邏輯迴歸

並列摘要

Semiconductor manufacturing has a history of over half a century, starting in 1960s and still holding an unshakable position in the global economy. Undoubtedly, its products have a wide range of applications, covering industrial, consumer electronics, AI, Cloud computing, transportation, automotive, and even military-related functions that require semiconductor components to execute. Dynamic random-access memory (DRAM) memory modules are an essential part of the semiconductor industry, and the DRAM industry plays a crucial role in the overall semiconductor supply chain. However, the factors that affect the price fluctuations of memory are highly correlated with the development of the market economy.The factors that influence memory prices will change over time, and accurately predicting these factors is quite challenging. If future development is predicted based solely on historical operational data trends and corresponding strategies are evaluated by companies to reduce costs and achieve operational goals. Therefore, this study aims to calculate the net profit margin using publicly available financial statement data. The financial data of specific upstream wafer fabs and downstream packaging and testing plants for the past five years will be used to predict the trend of the next quarter for the downstream firms based on the difference in net profit margin between quarters for the upstream plants. This analysis aims to provide data analysis and assessment between upstream and downstream entities. The study consists of two parts. One is to analyze the reasons for numerical changes, past data will be subjected to correlation analysis to examine the historical trends in the economy through numerical variations. The other is accuracy assessment, use binary logistic regression will be used based on the numerical data to evaluate whether the relationship between the numbers can be used for future trend prediction. This analysis aims to measure the correlation factors between the operational performance of upstream wafer fabs and the order status of downstream packaging and testing plants. Using the Pearson correlation coefficient, it was found that from 2018 to the first quarter of 2020, there was a positive correlation. However, from the second quarter of 2020 to the first quarter of 2022, a negative correlation was observed. Through binary logistic regression, it was determined that the accuracy of the data performance under these conditions was 0.75. Since this study has a duration of only four years, it would be beneficial to extend the research period in the future and involve more relevant companies to observe the extent of their impact on the trends. This can be followed by validating the accuracy of the observations through this model or other related predictive models. Furthermore, advanced measures can be taken by allocating manpower based on positive/negative growth intervals and discussing price adjustment ranges with material suppliers.