利用雙層模型架構達成半導體製程參數設置最佳化;The Study of Dual Discriminated Generative Adversarial Network

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/95519

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95519

Title:	利用雙層模型架構達成半導體製程參數設置最佳化;The Study of Dual Discriminated Generative Adversarial Network
Authors:	楊泓益;YANG, HONG-YI
Contributors:	資訊管理學系
Keywords:	工業4.0;生成對抗網路;強化式學習;深度學習;參數最佳化;Industry 4.0;Generative Adversarial Networks;Reinforcement Learning;Deep Learning;Parameter Optimization
Date:	2024-07-18
Issue Date:	2024-10-09 16:55:19 (UTC+8)
Publisher:	國立中央大學
Abstract:	近幾年來，工業4.0已經大幅改變了人們的生活，整合物聯網、大數據、人工智慧、雲端運算，利用大量的資訊進行分析、預測以及應用達成更高效的生產技術。半導體在其中扮演著十分關鍵的角色，半導體作為電子產品的核心，其製造流程對電子產品影響很大。在半導體製造過程中，參數設定將極大地影響生產良率和能源效率。目前的方法無法有效處理大量參數推薦以及參數之間的相互關係。在本文中，我們提出一個能夠有效處理這些問題的模型-DDGAN。DDGAN使用雙層架構模型來推薦序列參數，內層模型訓練目標為生成類似真實的資料，而外層模型的訓練目標則為生成良率為正常的資料，藉由這種模型架構能夠有效的推薦出符合需求的參數序列。能夠使機台的參數設置降低試錯成本並找出新的最佳參數組合。;In recent years, Industry 4.0 has significantly transformed people′s lives by integrating the Internet of Things (IoT), big data, artificial intelligence (AI), and cloud computing. These technologies leverage vast amounts of information to conduct analysis, make predictions, and enhance the efficiency of production techniques. Semiconductors play a crucial role in this context, serving as the core of electronic products, with their manufacturing processes greatly influencing the performance of these products. During semiconductor manufacturing, parameter settings profoundly impact production yield and energy efficiency. Current methods are insufficient in effectively handling large-scale parameter recommendations and the interrelationships among parameters. In this paper, we propose a model named DDGAN to address these challenges effectively. DDGAN employs a dual-layer architecture model for sequential parameter recommendation. The inner model is trained to generate data similar to real-world data, while the Exterior model aims to generate data with normal yield rates. This model architecture allows for the efficient recommendation of parameter sequences that meet specific requirements, thereby reducing trial-and-error costs and identifying new optimal parameter combinations for machine settings.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	93	View/Open

社群 sharing

Loading...