國立高雄大學圖資館 |

語系: 繁體中文

說明(常見問題)

圖資館首頁

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

類神經網路與圖型分析於晶圓製程之良率預測 = Apply Artific...

國立高雄大學亞太工商管理學系碩士班

類神經網路與圖型分析於晶圓製程之良率預測 = Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes

紀錄類型:	書目-語言資料,印刷品 : 單行本
並列題名:	Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes
作者:	許光明,
其他團體作者:	國立高雄大學
出版地:	[高雄市]
出版者:	撰者;
出版年:	民99[2010]
面頁冊數:	63面圖，表 : 30公分;
標題:	良率模式
標題:	yield model
電子資源:	http://handle.ncl.edu.tw/11296/ndltd/46697382272572595975
摘要註:	在傳統良率模式中，必須事先假設缺陷密度函數才能進行良率的估計，但是在實際狀況下卻無法精準地以某一個分配完全描述缺陷密度，而且在出現群聚現象時可能會錯估良率。近代有學者提出以一些無須假設缺陷密度函數之數學模式取代傳統之良率模式，在這些模式中輸入參數的選擇就顯得相當重要，必須盡可能包含所有晶圓上之變異因素，其中群聚現象之因素通常以群聚指標加以描述，但某些特定的群聚圖型卻會使群聚指標產生誤判群聚程度的狀況而造成良率估計誤差變大。有鑑於此，本研究透過三道圖型特性分析將晶圓上之缺陷圖型分類成隨機分佈、區域集中分佈、直線分佈與環狀分佈等四種，接著將群聚圖型也納入與群聚指標共同描述群聚現象。最後比較只使用群聚指標描述群聚現象之模式與使用群聚指標及群聚圖型描述群聚現象之模式。研究結果顯示，在增加了群聚圖型參數的良率模式中所估計的良率確實比未加入群聚圖型參數的良率模式更接近實際良率。 In traditional wafer-yield models, it is assumed that the defect density on a wafer follows a pre-specified distribution beforehand. However, when the defects present the so-called "cluster phenomenon", it is impractical to use a single specific distribution to fit the defect density. Hence, previous work proposed some contemporary models as alternatives to the traditional ones. In con-temporary models, two important issues are: (1) the input parameters should include all of the process variations and (2) the similarity among parameters is minimized. Especially, the distribution of defect density is excluded from the input parameters. Furthermore, some cluster indices have been derived to present cluster phenomenon. However, cluster indices may incorrectly assess some special defect patterns. Hence, there is a need to devise a more adaptive procedure to predict the wafer yield. Towards this end, this research examines the effectiveness of two yield models: (1) cluster in-dices are incorporated into the contemporary model, and (2) both cluster indices and defect patterns are incorporated into the contemporary model. To present the special defect patterns, this research applies a pattern-analysis technique to classify the distribution of defects on a wafer into one of four patterns: random pattern, centered pattern, line pattern, and ring pattern. Experimental results show that the second model (combination of cluster indices and defect patterns) is much better than the first model (cluster indices).

類神經網路與圖型分析於晶圓製程之良率預測 = Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes
許, 光明

類神經網路與圖型分析於晶圓製程之良率預測 = Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes / 許光明撰 - [高雄市] : 撰者, 民99[2010]. - 63面 ; 圖，表 ; 30公分.
參考書目：面.
良率模式yield model

類神經網路與圖型分析於晶圓製程之良率預測 = Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes
LDR:03859nam0a2200277 450 001 272833
005 20170214101210.0
009 272833
010 0 $b 精裝
010 0 $b 平裝
100 $a 20170214y2010 k y0chiy05 e
101 1 $a chi $d chi $d eng
102 $a tw
105 $a ak am 000yy
200 1 $a 類神經網路與圖型分析於晶圓製程之良率預測 $d Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes $f 許光明撰
210 $a [高雄市] $c 撰者 $d 民99[2010]
215 0 $a 63面 $c 圖，表 $d 30公分
314 $a 指導教授：陳榮泰教授
320 $a 參考書目：面
328 $a 碩士論文--國立高雄大學亞太工商管理學系碩士班
330 $a 在傳統良率模式中，必須事先假設缺陷密度函數才能進行良率的估計，但是在實際狀況下卻無法精準地以某一個分配完全描述缺陷密度，而且在出現群聚現象時可能會錯估良率。近代有學者提出以一些無須假設缺陷密度函數之數學模式取代傳統之良率模式，在這些模式中輸入參數的選擇就顯得相當重要，必須盡可能包含所有晶圓上之變異因素，其中群聚現象之因素通常以群聚指標加以描述，但某些特定的群聚圖型卻會使群聚指標產生誤判群聚程度的狀況而造成良率估計誤差變大。有鑑於此，本研究透過三道圖型特性分析將晶圓上之缺陷圖型分類成隨機分佈、區域集中分佈、直線分佈與環狀分佈等四種，接著將群聚圖型也納入與群聚指標共同描述群聚現象。最後比較只使用群聚指標描述群聚現象之模式與使用群聚指標及群聚圖型描述群聚現象之模式。研究結果顯示，在增加了群聚圖型參數的良率模式中所估計的良率確實比未加入群聚圖型參數的良率模式更接近實際良率。 In traditional wafer-yield models, it is assumed that the defect density on a wafer follows a pre-specified distribution beforehand. However, when the defects present the so-called "cluster phenomenon", it is impractical to use a single specific distribution to fit the defect density. Hence, previous work proposed some contemporary models as alternatives to the traditional ones. In con-temporary models, two important issues are: (1) the input parameters should include all of the process variations and (2) the similarity among parameters is minimized. Especially, the distribution of defect density is excluded from the input parameters. Furthermore, some cluster indices have been derived to present cluster phenomenon. However, cluster indices may incorrectly assess some special defect patterns. Hence, there is a need to devise a more adaptive procedure to predict the wafer yield. Towards this end, this research examines the effectiveness of two yield models: (1) cluster in-dices are incorporated into the contemporary model, and (2) both cluster indices and defect patterns are incorporated into the contemporary model. To present the special defect patterns, this research applies a pattern-analysis technique to classify the distribution of defects on a wafer into one of four patterns: random pattern, centered pattern, line pattern, and ring pattern. Experimental results show that the second model (combination of cluster indices and defect patterns) is much better than the first model (cluster indices).
510 1 $a Apply Artificial Neural Network and Pattern Analysis to Yield Prediction in Wafer Manufacturing Processes
610 0 $a 良率模式 $a 半導體 $a 圖型分析 $a 群聚指標
610 1 $a yield model $a wafer $a pattern analysis $a cluster index
681 $a 008M/0019 $b 343425 0896 $v 2007年版
700 1 $a 許 $b 光明 $4 撰 $3 482999
712 0 2 $a 國立高雄大學 $b 亞太工商管理學系碩士班 $3 166023
801 0 $a tw $b 國立高雄大學 $c 20101224 $g CCR
856 7 $z 電子資源 $2 http $u http://handle.ncl.edu.tw/11296/ndltd/46697382272572595975