運用倒傳遞類神經網路於CMOS影像感測器生產機台異常分析

Using Back-Propagation Neural Network Analysis for the Failure Production of CMOS Image Sensor Machines

陳哲豪(Che-Hao Chen)

指導教授 ： 陳建良

國立清華大學/工學院/工業工程與工程管理學系碩士在職專班/碩士(2014年)

摘要

摘 要
CMOS影像感測器(Complementary Metal-Oxide Semiconductor Image Sensor, CIS)產業乃是一種技術、資本高度密集之產業，因此如何使生產機台利用率最大化，以其於在短時間內能將設備成本攤平，降低產品製造成本，創造公司競爭優勢，成為業界不斷追尋之目標。
就CMOS影像感測器生產測試而言，Chroma測試機與電源供應機為CMOS影像感測器生產測試中，最為關鍵之設備。為能維護此機台正常運作與確保產出量，機台之異常需能即使診斷出發生原因為何。目前於生產機台之維修，依然仰賴資深維修人員之經驗，但諸多因素，造成這些資深維修人員流動率偏高，以至於知識、技術與經驗傳承不易，企業難以獲得有效蓄積。
透過學者成功運用類神經網路於半導體機台異常分析之經驗並結合產線機台之維修人員經驗、相關測試資料與類神經網路（Artificial Neural Network, ANN）技術進行異常分類與解析，建構一套CMOS影像感測器生產測試機台異常診斷模型。倒傳遞類神經網路(Back Propagation Neural Network, BPN)用以找出異常現象與異常原因之關係，模型之效能則以網路績效加以衡量。
研究結果顯示，本研究所提出之模式對於CMOS影像感測器生產機台之異常原因診斷，可獲得良好之效果，因此本模式應可應用於相關領域之生產測試機台。

關鍵字：類神經網路、製造成本、網路績效

關鍵字

類神經網路 ； 製造成本 ； 網路績效

並列摘要

Abstract
CMOS image sensor (Complementary Metal-Oxide Semiconductor Image Sensor, CIS) is technology driven and highly capital -intensive industry. Hence the way to maximize the utilization of production equipment becomes the stepping stone for business to minimize its cost of equipment and manufacturing, also to create competitive advantage.

For production testing of CMOS image sensors, Chroma testing machine with the power supply unit is the most critical tool among all. In order to maintain normal operations and ensure the output volume of the machine, currently the production line is still dependent on the experience of senior maintenance staff. However due to many reasons, high turnover rate of senior maintenance staff results in difficulty of obtaining effective accumulation the knowledge, skills and experience.

Through scholars used of neural networks in the success experience of semiconductor machine failure analysis and combined with the experience machine production line maintenance personnel, the relevant test data and neural network (Artificial Neural Network, ANN) technology for fault classification and analysis, constructing a CMOS image sensor production test machine fault diagnosis mode. Back Propagation Neural Network (BPN) is used to identify the relationship between symptom and cause of the fault. Model efficiency is measured by the network performance.

The results show that the proposed model for the diagnostics of CMOS image sensor test tool malfunction obtained good results, so this model should be applied to related fields of test equipment.

Key word: Neural Network、Cost Of Manufacturing、Network Performance

並列關鍵字

Neural Network ； Cost Of Manufacturing ； Network Performance

參考文獻

Circuits, IEEE Journal, Volume 34, Issue 12, Pages 1821-

Circuits, IEEE Journal, Volume 35, Issue 8, Pages 1146-

A single-chip image sensor for mobile applications, Solid-

State Circuits, IEEE Journal, Volume 37, Issue 12, Pages

4.L. Chua and L. Yang(1988), Cellular Neural Networks: