201115667 六、發明說明： 本申請案請求美國臨時專利序號第61/223,075號、申請 曰2009年7月6曰之優先權，並將該全文併入本文據以參考。201115667 VI. INSTRUCTIONS: This application claims priority to U.S. Provisional Patent No. 61/223,075, the entire disclosure of which is incorporated herein by reference.
【發明戶斤屬之技術領域:J 發明領域 本發明領域涉及用於自動秘方驗證及選擇之系統及方 法0 發明背景 晶圓缺損檢測是一種在半導體製造產業中的常見慣 例’其作為複雜產量分析與控制製程的一部分。雖然每一 個自動缺損檢測的工具供應商利用不同的檢測原則與方 法，但他們都有一個共通概念：檢測秘方（也稱為一檢測 工作）。該提到的檢測秘方是就某種代表晶圓（representing wafer)所建立，且被用於每一個相同來源（產品、製程等） 的晶圓。所有的晶圓檢測工具共享相同的「設定」+「執 行」情境。 多數自動檢測工具都面臨非常一致的挑戰，即對製程 變化要有残性。這意味著，_—個代表㈣所建立的 同個檢測秘方要被成功地用於相同種類的整批晶圓。 201115667 r發明内容】 發明概要 一種用以選擇一檢測秘方的方法’該方法根據本發明 的一實施例，可包括：獲取半導體裝置之一結構元件的— 影像；計算該結構元件的影像與藉由應用複數檢測秘方所 獲得之複數參考影像的每一者之間的多重類型距離；以及 基於該多重類型距離的數值自動地自該複數檢測秘方之中 選擇至少一個所選檢測秘方。 該方法可包括計算選自下列之多重類型距離、至少三 距離或是大多數類型的距離：（i)灰階直方圖間之一差距； (11)直線（L1)距離；（iii)歐幾里得（L2)長度；（iv) 卡方距離’（v)巴塔岭亞距離（Bhattacharyya distance); (vi)瓦塞斯;丁度量（Wasserstein metric) ; ( vii)司旺戶 量（Swainmetric);以及（viii)常態化關聯值（normalized correlation ) 〇 °玄方法可包括基於該結構元件影像與該參考影像之間 的多重類型距_數值，為每個參考影像產生—匹配值； 以及麟與—參考影像相關聯的每個檢測秘方，該參考影 像具有-在-可容許的匹配值範圍内的匹配值。 5玄方去可包括選擇多重所選檢測秘方。 得評藉由應用至少一個所選檢測秘方所 h變待計算義型距離。 201115667 牛了被塗佈-隨時間而改變光學性質的塗佈 材料’其中該複數心“…士 材料光學性質之數值。 匕們又凋整的塗佈 =供=檢測系統。根據本發明的_實施例，該系統 元:的一旦：像獲取模組’用以獲取半導體裝置之-結構 象’-距離計算器，用以計算該結構元件的影 像與藉由應轉歸_方㈣狀她㈣f彡像之每: 者間的多重類型距離；-選擇模組，用以基於多重類型距 離的數值自動地自該複數檢測秘方之中轉至少—個所選 檢測秘方·，以及-控制器，用以控制該影像擷取模組以應 用該至少一個所選檢測秘方之每一者。 該距離計算n可適詩計算選自下狀多重類型距 離、至少二距離或是大多數類型的距離：⑴灰階直方圖 間之一差距；（ii)直線（L1)距離；（iii)歐幾里得（L2) 長度；（iv)卡方距離；（v)巴塔恰亞距離（Bhauacharyya distance )’（ vi)瓦塞斯 >丁度量（wasserstein metric );( νϋ ) 司旺度量（Swain metric );以及（viii )常態化關聯值 (normalized correlation)。 該距離計算器可適用於基於該結構元件影像與該參考 影像之間的多重類型距離的數值，為每個參考影像產生一 匹配值；以及選擇與一參考影像相關聯的每個檢測秘方， 該參考影像具有一在一可容許的匹配值範圍之内的匹配 值。 201115667 該選擇模組可適用於選擇多重所選檢測秘方。 該距離計算器可適用於基於藉由應用至少一個所選檢 測秘方所獲取的評估結果，改變待計算的類型距離。 該結構元件可被塗佈一隨時間而改變光學性質的塗佈 材料，其中該複數檢測秘方彼此相差一它們受調整的塗佈 材料光學性質之數值。 提供一種電腦程式產品。根據本發明的一實施例，它 可包括一儲存指令的非暫態電腦可讀媒體，該等指令用 以：獲取半導體裝置之一結構元件的一影像；計算該結構 元件的影像與藉由應用複數檢測秘方所獲得之複數參考影 像的每一者之間的多重類型距離；以及基於該多重類型距 離的數值自動地自該複數檢測秘方之中選擇至少一個所選 檢測秘方。 該電腦程式產品可儲存指令，該等指令用以計算選自 下列之多重類型距離、大多數類型的距離或是至少三距 離：（0灰階直方圖間之一差距；（ii)直線（L1)距離；（iii) 歐幾里得（L2)長度；（iv)卡方距離；（v) 巴塔恰亞距 離；（vi ) 瓦塞斯汀度量；（vii )司旺度量；以及（viii ) 常態化關聯值。 該電腦程式產品可儲存指令，該等指令用以基於該結 構元件影像與該參考影像之間的多重類型距離的數值，為 每個參考影像產生一匹配值；以及選擇與一參考影像相關 聯的每個檢測秘方，該參考影像具有一在一可容許的匹配 值範圍之内的匹配值。 6 201115667 該電腦程式產品可儲存指令，該等指令用以選擇多重 所選檢測秘方。 該電腦程式產品可儲存指令，該等指令用以基於藉由 應用至少一個所選檢測秘方所獲取的評估結果，改變待計 算的該類型距離。 該結構元件可被塗佈一隨時間而改變光學性質的塗佈 材料，其中該複數檢測秘方彼此相差一它們受調整的塗佈 材料光學性質之數值。 圖式簡單說明 關於本發明之進一步細節、面相及實施例將僅就圖式 以實例的方式、參考至圖式來敘述。圖式中，同樣的參考 數字是用來識別同樣或功能性相似的元件。在該等圖中的 元件是為了簡明清晰而加以描繪，並未必依照比例繪圖。 第1圖說明本發明一實施例之方法； 第2圖說明本發明一實施例之方法； 第3圖說明本發明一實施例之方法；及 第4圖說明本發明一實施例之檢測系統。 7 201115667 I：實施♦式】 較佳實施例之詳細說明 視料發明 特別指明並清楚地加以主張 茶說明書的最後部分初 與操作方法兩者，造、不官如何，本發明關於組錯 逆叫具目的、特料β 7s 圖式一起閱讀時藉由夂 …，寸伋及優點，可在與隨阳 、下詳細5兒明最佳地加以理解。 在以下5平細說明中， 會提出許多特定細節。然而，孰f^明的—徹底理解， 發明可不料些較細節 1將會瞭解到，^ :熟知的方法、—=敘:=本: 以下方法、系統及電腦程式產品係針對至少一已知類 型的下列製&變化提供―健全與決定性解決方案：晶圓上 受檢測目標物的總體對比及/或紋理及/或反射率變化。所述 的製程變化種類通常會在晶圓塗佈物「老化」製程（―塗 佈液體從一批量中的第一月晶圓到該批量的最後一片晶圓 間改變其光學性質的現象）遇到。 晶圓塗佈製程能藉由將一晶圓浸泡在一塗佈液體中作 為一電鍍製程的一部分而加以實行。該塗佈液體能隨時門 改變其光學特性（反射率、透明度、吸收度）一導致在戶厅 得影像上有相當大的改變。因此，理想上同一結構元件之 所得影像基本上能互不相同。 8 201115667 。玄方去、系統與電腦程式產品基於預設標準，提供秘 方運轉時適應性’並就現前實施之晶圓容許健全與可靠的 秘方選擇/驗證。它們包括僅實行選自既存_化之檢測秘 方儲存庫中「良好」的檢測秘方。 該方法、系統與電腦程式產品可於一些預定義（諸如 丄m化)：之檢測秘方間選擇，每_者都為了該晶圓的特定 條件作最佳化。這絲件可从該晶圓的―總體反射率、 對比值、一塗佈液體的光學特性，及類似者。 母個檢測秘方可加以最佳化以備測與先前所提製程變 化狀心有關之aa|]典型影像之缺損。該檢測系統可劉覽整 個所提檢_方、使用各種統計方法相對於預存光學度量 的一組合而蚊、錢就最健配分料蚊實行與否。 所述比較及匹配技術（與該產業中其他常用者不同） 不是立基於其對f彡像舰或|彡像特定灰階的蚊。徤全性 是藉由-所提制料影縣析之專賴合魏成—即藉 由所存晶圓影像參考相對於運_晶^彡像來達成。 第1圖說明本發明一實施例之方法1〇〇。 方法觸起始於階段105或者接受或產生多重檢測秘 方0 階段105之後接續根據一檢浪u、, + > 懷，則秘方設定一檢測系統的 階段110。階段110可包括應用一你田土 使用者所定義的光學態： 包括但不限定於光照強度、光脬翻刑 〜頭型、收集建立，及倍率。 階段110可包括應用秘方的所存影像獲取離。 201115667 階段no之後接續至少獲取一影像的階段12〇。在階段 ⑽期間獲取的-影像能被稱為—所❹彡像。階段12〇可包 括獲取該晶圓之相同結構參考點（麵域）或是不同結構 元件（或多個區域）的一連串所獲影像。 階段120之後接續階段13〇，其評料至少—所獲影像 的每-者與至少-參考影像之__，該等參考影像與 在階段110期間用以設定該檢測系統的該檢測秘方有關。該 評估是響應於至少一預定義之匹配標準。 階段130之後可接續階段140，其係基於多重類型距離 的數值自動地自該複數檢測秘方中選擇至少一個所選檢測 秘方。該評估是響應於至少一預定義的匹配標準。 階段13 0可包括比較至少一所獲影像與至少一對應參 考影像。 階段130可包括計算一對所獲取影像與參考影像之間 的至少一類型距離。每對的參考影像與目前受評估的檢測 秘方有關：在階段110期間用以設定檢測系統的檢測秘方。 該計算可包括進行每對所獲取影像與參考影像之像素值的 至少一統計分析。 階段130可包括使用以下演算法：過簡灰階直方圖均值 (Simplistic gray level histogram mean)、L1 或L2分布、卡 方分布、巴塔恰亞距離、移土者距離（Earth-mover’s distance ’又名瓦塞斯汀度量）、司旺度量，以及常態化關 聯值。 10 201115667 階段130之後可接續階段14〇，其係處理在階段130期間 所計算的該至少一類型距離，以提供一匹配值。該處理可 包括應用統計演算法，諸如均值、變異數、多數權重總和， 及類似者。 階段140之後可接續階段150 ’其係基於該匹配值，決 定該檢測秘方是否應該被執行。該數值分數能與一匹配值 極小值、一匹配值極大值，或是一與該目前受評估檢測秘 方有關的匹配值範圍比較。 階段150之後可接續：（i)階段160,其係確認其他檢 測秘方是否應該加以評估、（ii)階段170,其係選擇一剩餘 檢測秘方（假如存在）並跳到階段11〇、及（iii)階段18〇， 其係在驗證所有檢測秘方之後執行每個應該被執行之檢測 秘方。 或者，假如階段150決定了檢測秘方應該要被執行，則 階段150之後可接續階段19〇執行該檢測秘方。階段D0之後 可接續一連串階段，其中可包括階段16〇及階段170。 方法100之各種階段的重複可導致：（丨）計算介於晶圓 之一結構元件的一所獲取影像與藉由應用複數檢測秘方所 獲得之多數參考影像的每—者之間的多重類型距離、及（ii) 基於該等多重類型距離的數值，自動地自該等複數檢測秘 方之中選擇至少一所選檢測秘方。 第2圖說明本發明一實施例之方法2〇〇。該方法2〇〇包 括.（i·)階段210’其係獲取半導體裝置結構元件之一影像。 201115667 該結構元素可以是一凸塊、一溝渠，及類似者；（ii.)階段 2 20，其係計算介於該結構元件的所得影像與藉由應用複數 檢測秘方所獲得之複數參考影像之每一者之間的多重類型 距離；（iii.)階段230，其係基於在結構元件之所得影像與 參考影像之間的該等多重類型距離的數值，為每個參考影 像產生一匹配值。受成像的結構元件可被塗佈一隨時間而 改變光學性質的塗佈材料，且該等複數檢測秘方可彼此相 差一它們受調整的塗佈材料光學性質之數值；（iv.)階段 240,其係基於該等多重類型距離的數值自動地自複數檢測 秘方之中選擇至少一所選檢測秘方。階段240可響應於每對 參考影像與受檢影像的匹配值，以及（V.)階段250，其係 執行每個所選檢測秘方。 階段220可包括計算選自下列之（a)多重類型距離、 (b)大多數多重類型距離，或（c)至少三類型距離： (i) 灰階直方圖間之一差距； (ii) 直線（L1)距離； (iii) 歐幾里得（L2)長度； (iv) 卡方距離； (v) 巴塔恰亞距離； (vi) 瓦塞斯汀度量； (vii) 司旺度量；以及 (viii) 常態化關聯值。 12 201115667 使用多重類型距離可協助提供一檢測秘方的一健全 選擇。該健全性可促成不同類型距離對不同類型改變的不 同響應，該等不同類型改變諸如製裎變異、塗佈材2之反 射率的改變，及類似者。要注意的是其他類型距離可加以 計算。 階段240可包括選擇與一參考影像有關的每個檢測秘 方，該參考影像具有一在一可容許匹配值範圍之内的匹配 值。該可容許匹配值範圍係事先決定且可藉由一操作子（或 其他被授權人）被決定。 階段240可包括選擇多重所選檢測秘方。這些所選檢測 秘方的每一者應在階段250期間被執行。 在執行階段210-260的一次或更多重覆後，方法2〇〇可 藉由階段270更新至少一檢測秘方而進行。 階段270可包括基於藉由應用該至少一所選檢測秘方 所獲取的評估結果，改變待計算類型距離。階段27〇可包括 省略一先前所選類型距離，假如那個類型距離沒有促成該 所選類型距離的選擇的話（假如，例如在那個類型距離數 值與決定是否選擇一相關檢測秘方之間的關聯性非常 低）。階段270可包括在方法200的下一次重覆期間增加一新 的待計算類型距離。階段270之後可接續階段21〇。 方法100及200的各個方法可包括一從檢測系統的一操 作子（或從任何其他被授權人）接受指令的階段。這些指 令可決定每個檢測秘方的脈絡、可決定要計算哪些類型距 13 201115667 離如何D十算匹配分數及如何基於匹配分數選擇-檢測秘 方。一指令能限制所選檢測秘方的數目且能在參考影像及 所獲取影像之產生㈣選擇要成像哪些結構元件。 第3圖說明本發明實施例之階段105。 階段105起始於階段310，其係產生至少一檢測秘方。 階段310可包括定義一用於影像取得之設定（一狀 態）’例如選擇影像取得之光學特性、一晶圓上的位置，或 其他所證物體等等。 階段310之後接續階段32〇，其係在決定是否使用該檢 冽秘方時定義至少一待使用的匹配標準◦該匹配標準可包 括在一所得影像及一參考影像之間待計算的類型距離、其 中—或更多類型距離的數值要被處理以提供一匹配值的方 式、以及將導致選擇該檢測秘方或是駁回它之匹配值的數 值。 階段330之後接續階段33〇，其係抓取一參考影像—使 用所選設定。 階段330之後接續階段34〇，其係儲存至少一匹配標準 及參考影像。 階段310之後可接續階段35〇，其係決定是否有定義另 一個檢測秘方的需求。假如答案是正面的，則階段35〇後接 續階段310。 201115667 假設例如只有晶圓（或塗佈材料）光學特丨生上 」’則可定 改變的一部份被先前所定義之檢測秘方所「 可今。午 復蓋I， 義另一個檢測秘方。 階段350可包括決定定義另一個檢判 秘方以順應已知 的製造改變製程（例如光學液體反射率下降等等） 該檢測系 第4圖說明本發明一實施例之檢測系統 統400包括： 用以獲取半導體艘置之 (0 一影像獲取模組410, 一結構元件的一影像。 (U) -距離計算器42〇,用以計算在該結構元件的 影像及藉由應用複數檢測秘方所獲得之複數參考影像的每 一者之間的多重類型距離。 類型距離的數 個所選檢測秘 (iii) 一選擇模組430，用以基於多重 值自動地自該複數檢測秘方之中選擇至少一 方；以及 以應 方法 圖所 及該 (W) 一控制器440，用以控制該影像獲取模組 用該至少一個所選檢測秘方之每一者。 裘 該檢測系統4〇〇可執行方法1〇〇及2〇〇。為了執〜 105，檢測系統4〇〇它應該包括秘方產生器45〇〜如第4 示。已知檢測系統400可從其他裝置或系統接受秘方， 秘方產生器450能與檢測系統分離。 15 201115667 距離計算器420及選擇模組430可為處理器444的一部 份。 距離計算器420可適於計算選自下列之多重類型距 離、大多數類型距離或是三個或以上的類型距離：（0灰 階直方圖間之一差距；（ii)直線（L1)距離；（iii)歐幾里 得（L2)長度；（iv)卡方距離；（v)巴塔恰亞距離；（vi) 瓦塞斯汀度量；（vii)司旺度量；以及（viii)常態化關聯 值。 距離計算器420可適於基於該結構元件影像與該參考 影像之間的多重類型距離的數值，為每個參考影像產生一 匹配值；以及選擇與一參考影像相關聯的每個檢測秘方， 該參考影像具有一在一可容許的匹配值範圍之内的匹配 值0 選擇模組430可適於選擇多重所選檢測秘方。 距離計算器420可適於基於藉由應用至少一所選檢測 秘方所獲取的評估結果來改變待計算的類型距離。 影像獲取模組410可包括一或更多相機、光學儀器及一 或更多感應器、影像抓取器（未顯示）及一影像處理器。 影像獲取模組410能響應於儲存在不同檢測秘方的資訊而 設定成不同狀態。不同狀態可在其等之光學參數（例如亮 度、焦距）、物理參數（例如與晶圓的距離、曝光時間、所 使用的化學材料）等等上有所不同。 16 201115667 方法100、200及300或其任一者能被一執行指令之處理 器（諸如處理器444)所執行。該指令能被儲存在非暫態電 腦可讀媒體中，諸如磁碟、磁帶、磁片，及類似者。 例如，電腦程式產品可包括一儲存指令的非暫態電腦 可讀媒體，該等指令用以：獲取半導體裝置之一結構元件 的一影像；計算在該結構元件的影像與藉由應用複數檢測 秘方所獲得之複數參考影像的每一者之間的多重類型距 離；以及基於該多重類型距離的數值自動地自該複數檢測 秘方之中選擇至少一個所選檢測秘方。 雖然本發明的特定特徵在此已加以描繪及說明，熟習 此藝者由此將可想到許多修改、替代、變化及等效物。因 此要瞭解的是，所附申請專利範圍是意圖涵蓋所有落入本 發明真實精神之中的此等修改及變化。 17 201115667 I：圖式簡單説明3 第1圖說明本發明一實施例之方法； 第2圖說明本發明一實施例之方法； 第3圖說明本發明一實施例之方法；及 第4圖說明本發明一實施例之檢測系統。 【主要元件符號說明】 100···方法 100 250…階段250 105…階段105 270…階段270 110…階段110 300…方法300 120…階段120 310···階段 310 130···階段 130 320···階段 320 140···階段 140 330···階段 330 142···階段 142 340…階段340 144…階段144 350…階段350 160···階段 160 400…檢測系統 170…階段170 410…影像獲取模組 180…階段180 420…距離計算器 190…階段190 430…選擇模組 200···方法200 440…控制器 210···階段 210 444.··處理器 220···階段 220 450…秘方產生器 230…階段230 240…階段240 18FIELD OF THE INVENTION The field of the invention relates to systems and methods for automatic recipe verification and selection. BACKGROUND OF THE INVENTION Wafer defect detection is a common practice in the semiconductor manufacturing industry as a complex yield analysis. And part of the control process. While each toolkit for automatic defect detection uses different detection principles and methods, they all share a common concept: the detection recipe (also known as a test job). The detection recipe mentioned is a wafer built on a representative wafer and used for every same source (product, process, etc.). All wafer inspection tools share the same “setup” + “execution” scenario. Most automated inspection tools face a very consistent challenge of being remnant of process changes. This means that the same test recipe established by _-(4) is successfully used for the same batch of wafers. SUMMARY OF THE INVENTION A method for selecting a detection recipe is provided. The method according to an embodiment of the present invention may include: acquiring an image of a structural component of a semiconductor device; calculating an image of the structural component and Applying a multiple type distance between each of the plurality of reference images obtained by the complex detection recipe; and automatically selecting at least one selected detection recipe from the complex detection recipe based on the value of the multiple type distance. The method can include calculating a plurality of types of distances selected from the group consisting of: at least three distances or most types of distances: (i) one of the grayscale histograms; (11) a straight line (L1) distance; (iii) a few Length of Led (L2); (iv) Chi-square distance '(v) Bhattacharyya distance; (vi) Vasese; Wasserstein metric; (vii) Swan household (Swainmetric And (viii) normalized correlation 〇° 方法 method may include generating a matching value for each reference image based on a multiple type of distance_value between the structural element image and the reference image; Each of the detection recipes associated with the reference image has a matching value within the range of -to-tolerable matching values. 5 Xuanfang can include multiple selection of detection recipes. The evaluation determines the prototype distance by applying at least one selected detection recipe. 201115667 The cow is coated - the coating material that changes the optical properties over time 'where the complex heart "...the value of the optical properties of the material. We are also coated = supply = detection system. According to the invention _ In an embodiment, the system element: once the image acquisition module is used to acquire a structure image of the semiconductor device - a distance calculator for calculating an image of the structural component and by means of a _ square (four) shape (four) f Each of the images: multiple types of distances between the two; - a selection module for automatically translating at least one selected detection secret from the complex detection secret based on the value of the multiple type distance, and - the controller Controlling the image capture module to apply each of the at least one selected detection recipe. The distance calculation n can be selected from the following multiple type distances, at least two distances, or most types of distances: (1) gray One of the differences between the order histograms; (ii) the straight line (L1) distance; (iii) the Euclidean (L2) length; (iv) the chi-square distance; (v) the Bhauacharyya distance' Vi) Vases > Ding metrics (wass Erstein metric ); ( νϋ ) Swain metric; and (viii ) normalized correlation. The distance calculator can be applied to multiple types of distances based on the image of the structural component and the reference image. a value that produces a match value for each reference image; and selects each detection secret associated with a reference image that has a match value within a tolerable range of matching values. 201115667 The selection mode The group may be adapted to select multiple selected detection recipes. The distance calculator may be adapted to change the type distance to be calculated based on the evaluation results obtained by applying at least one selected detection recipe. The structural element may be coated with A coating material that changes optical properties over time, wherein the complex detection recipes differ from each other by the value of the optical properties of the coated material being adjusted. A computer program product is provided. According to an embodiment of the invention, it may include a storage instruction Non-transitory computer readable medium for obtaining structural components of a semiconductor device An image; calculating a multiple type distance between the image of the structural element and each of the plurality of reference images obtained by applying the complex detection recipe; and automatically selecting a value based on the multiple type distance from the complex detection recipe At least one selected detection recipe. The computer program product can store instructions for calculating multiple types of distances selected from the following, most types of distances, or at least three distances: (0 gap between grayscale histograms) (ii) straight line (L1) distance; (iii) Euclidean (L2) length; (iv) chi-square distance; (v) Batačia distance; (vi) Vassestin metric; (vii) Siwang metrics; and (viii) normalized correlation values. The computer program product can store instructions for generating a matching value for each reference image based on a value of a multiple type distance between the structural component image and the reference image; and selecting a reference image associated with a reference image For each detection recipe, the reference image has a matching value within a tolerable range of matching values. 6 201115667 This computer program product can store instructions for selecting multiple selected detection recipes. The computer program product can store instructions for changing the type of distance to be calculated based on the evaluation results obtained by applying at least one selected detection recipe. The structural element can be coated with a coating material that changes optical properties over time, wherein the complex detection recipes differ from each other by the value of their optical properties of the coated material. BRIEF DESCRIPTION OF THE DRAWINGS Further details, aspects, and embodiments of the present invention will be described by way of example only, with reference to the drawings. In the drawings, the same reference numerals are used to identify the same or functionally similar elements. The elements in the figures are for clarity and clarity and are not necessarily drawn to scale. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 illustrates a method of an embodiment of the present invention; Fig. 2 illustrates a method of an embodiment of the present invention; Fig. 3 illustrates a method of an embodiment of the present invention; and Fig. 4 illustrates a detection system of an embodiment of the present invention. 7 201115667 I: Implementation ♦ Equations Detailed Description of the Preferred Embodiments The invention is specifically pointed out and clearly claims that both the initial part and the operating method of the tea specification are both inaccurate and inconsistent. The purpose and special content of the β 7s pattern can be best understood by reading 随..., inch 汲 and advantages when reading together. In the following 5 flat descriptions, many specific details are presented. However, 孰f^明—completely understand, the invention can be understood in more detail. 1 : Well-known methods, —= Syria: = This: The following methods, systems and computer programs are for at least one known The following types of & variations provide a sound and decisive solution: the overall contrast and/or texture and/or reflectance changes of the object under test on the wafer. The process variation type is usually encountered in the wafer coating "aging" process (the phenomenon that the coating liquid changes its optical properties from the first month wafer in a batch to the last wafer in the batch). To. The wafer coating process can be carried out by immersing a wafer in a coating liquid as part of an electroplating process. The coating liquid can change its optical properties (reflectance, transparency, absorbance) at any time, resulting in considerable changes in the image of the household. Therefore, ideally, the resulting images of the same structural element can be substantially different from each other. 8 201115667. Xuanfang went, system and computer program products based on pre-set standards, providing the flexibility of the secret operating time' and allowing the sound and reliability of the current implementation of the wafer selection/verification. They include detection recipes that are only "good" selected from the existing detection recipes. The method, system, and computer program product can be selected between a number of predefined (such as 丄m) detection recipes, each of which is optimized for the particular conditions of the wafer. This filament can be derived from the "total reflectivity, contrast value, optical properties of a coating liquid, and the like." The parent detection recipe can be optimized to measure the defect of the typical image associated with the previously proposed process change heart. The detection system can view the entire inspection, using a combination of various statistical methods and pre-existing optical metrics, and the mosquitoes and money are the most suitable for the distribution of mosquitoes. The comparison and matching technique (unlike other commonly used in the industry) is not based on the mosquitoes of the specific gray scale of the ship or the image.徤 徤 是 是 藉 藉 藉 藉 藉 藉 藉 藉 藉 藉 藉 藉 藉 县 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏Figure 1 illustrates a method 1 of an embodiment of the present invention. The method starts at stage 105 or accepts or generates a multi-detection secret 0 stage 105 followed by a check wave u,, + >, then the secret set a stage 110 of the detection system. Stage 110 may include applying an optical state defined by your field user: including but not limited to light intensity, light retire to head type, collection establishment, and magnification. Stage 110 may include the application of the stored secret image acquisition. 201115667 After the stage no, at least the stage 12 of acquiring an image is obtained. The image acquired during phase (10) can be referred to as the image. Stage 12 can include obtaining a series of acquired images of the same structural reference point (area) of the wafer or different structural elements (or regions). Stage 120 is followed by stage 13A, which evaluates at least - each of the obtained images and at least - the reference image, which is associated with the detection recipe used to set the detection system during stage 110. The evaluation is in response to at least one predefined matching criterion. Stage 130 may be followed by stage 140, which automatically selects at least one selected detection recipe from the complex detection recipe based on the value of the multiple type distance. The evaluation is in response to at least one predefined matching criterion. Stage 130 may include comparing at least one acquired image with at least one corresponding reference image. Stage 130 can include calculating at least one type of distance between a pair of acquired images and a reference image. The reference image for each pair is related to the currently detected detection recipe: during phase 110, the detection recipe for the detection system is set. The calculating can include performing at least one statistical analysis of pixel values of each of the acquired image and the reference image. Stage 130 may include the use of the following algorithms: Simplistic gray level histogram mean, L1 or L2 distribution, chi-square distribution, Batachya distance, and earth-mover's distance' Named Wassestin metrics, Swan metrics, and normalized correlation values. 10 201115667 Stage 130 may be followed by stage 14〇, which processes the at least one type of distance calculated during stage 130 to provide a match value. This processing may include applying statistical algorithms such as mean, variance, majority weight sum, and the like. Stage 140 may be followed by stage 150' based on the match value to determine if the detection recipe should be executed. The value score can be compared to a match value minimum value, a match value maximum value, or a range of match values associated with the currently evaluated detection secret. Stage 150 can be followed by: (i) stage 160, which confirms whether other detection recipes should be evaluated, and (ii) stage 170, which selects a residual detection recipe (if present) and jumps to stage 11〇, and (iii) Phase 18, which performs each detection recipe that should be performed after verifying all detection recipes. Alternatively, if stage 150 determines that the detection recipe should be executed, then phase 150 can be followed by stage 19 to execute the detection recipe. Stage D0 can be followed by a series of stages, which can include stage 16 and stage 170. The repetition of the various stages of the method 100 can result in: (丨) calculating the multiple type of distance between each acquired image of one of the structural elements of the wafer and each of the majority of the reference images obtained by applying the complex detection recipe. And (ii) automatically selecting at least one selected detection recipe from among the plurality of detection secrets based on the values of the multiple types of distances. Figure 2 illustrates a method 2 of an embodiment of the invention. The method 2 includes an (i.) stage 210' of acquiring an image of a structural element of the semiconductor device. 201115667 The structural element can be a bump, a trench, and the like; (ii.) Stage 2 20, which calculates the resulting image between the structural component and the complex reference image obtained by applying the complex detection recipe. Multiple type distances between each; (iii.) Stage 230, which generates a match value for each reference image based on the values of the multiple types of distances between the resulting image of the structural element and the reference image. The imaged structural element can be coated with a coating material that changes optical properties over time, and the complex detection recipes can differ from each other by the value of their adjusted coating material optical properties; (iv.) stage 240, It automatically selects at least one selected detection recipe from among the complex detection recipes based on the values of the multiple types of distances. Stage 240 may be responsive to a match between each pair of reference images and the image being examined, and (V.) stage 250, which performs each selected detection recipe. Stage 220 can include calculating (a) multiple type distances, (b) most multiple type distances, or (c) at least three types of distances selected from: (i) one of the grayscale histograms; (ii) a straight line (L1) distance; (iii) Euclidean (L2) length; (iv) Chi-square distance; (v) Batačaya distance; (vi) Vassestin metric; (vii) Siwang metric; (viii) Normalize the associated value. 12 201115667 Using multiple types of distances can help provide a robust selection of detection recipes. This robustness can contribute to different responses of different types of distances to different types of changes, such as variations in the sputum, changes in the reflectivity of the coated material 2, and the like. It should be noted that other types of distances can be calculated. Stage 240 can include selecting each detection recipe associated with a reference image having a match value within a range of allowable match values. The allowable range of matching values is determined in advance and can be determined by an operator (or other authorized person). Stage 240 can include selecting multiple selected detection recipes. Each of these selected detection recipes should be executed during phase 250. After performing one or more repetitions of stages 210-260, method 2 can be performed by stage 270 updating at least one detection recipe. Stage 270 can include changing the type distance to be calculated based on the evaluation results obtained by applying the at least one selected detection recipe. Stage 27〇 may include omitting a previously selected type of distance, if that type of distance does not result in a selection of the selected type of distance (if, for example, the relationship between the type of distance value and the decision to choose whether to select a related detection recipe is very low). Stage 270 can include adding a new type of distance to be calculated during the next iteration of method 200. Phase 270 can be followed by phase 21〇. The various methods of methods 100 and 200 can include a stage of accepting instructions from an operator of the detection system (or from any other authorized person). These instructions determine the context of each detection recipe and determine which type of distance to calculate. How to calculate the match score based on how to match the match score and how to select the match based on the match score. An instruction can limit the number of selected detection recipes and can select which structural components to image in the reference image and the generated image (4). Figure 3 illustrates stage 105 of an embodiment of the invention. Stage 105 begins at stage 310, which produces at least one detection recipe. Stage 310 can include defining a setting (a state) for image acquisition, e.g., selecting an optical characteristic for image acquisition, a position on a wafer, or other evidenced object, and the like. Stage 310 is followed by stage 32〇, which defines at least one matching criterion to be used when deciding whether to use the check secret. The matching criterion may include a type distance to be calculated between a obtained image and a reference image, wherein The value of the value of the more or more type of distance to be processed to provide a match value, and the value that will result in the selection of the detection recipe or the rejection of its matching value. Stage 330 is followed by stage 33, which captures a reference image - using the selected settings. Stage 330 is followed by stage 34, which stores at least one matching standard and reference image. Phase 310 can be followed by a phase 35, which determines if there is a need to define another detection recipe. If the answer is positive, then stage 35 continues with stage 310. 201115667 Assume, for example, that only wafers (or coating materials) are optically "on", then a portion of the change can be determined by the previously defined detection recipe. "This is a cover for I. Another detection recipe." Stage 350 can include determining to define another test recipe to conform to a known manufacturing change process (eg, optical liquid reflectance reduction, etc.). FIG. 4 illustrates a detection system 400 of an embodiment of the present invention including: Obtaining a semiconductor device (0 image acquisition module 410, an image of a structural component. (U) - distance calculator 42A, for calculating an image of the structural component and obtaining the secret by applying a complex detection recipe Multiple types of distances between each of the plurality of reference images. A plurality of selected detection secrets of the type distance (iii) a selection module 430 for automatically selecting at least one of the complex detection recipes based on the multiple values; And (W) a controller 440 for controlling the image acquisition module to use each of the at least one selected detection recipe. 裘 The detection system 4 Line method 1〇〇 and 2〇〇. In order to execute ~105, the detection system 4〇〇 should include the secret generator 45〇~ as shown in Fig. 4. The known detection system 400 can accept the secret recipe from other devices or systems, and the secret recipe is generated. The device 450 can be separated from the detection system. 15 201115667 The distance calculator 420 and the selection module 430 can be part of the processor 444. The distance calculator 420 can be adapted to calculate multiple types of distances, most types of distances, or Is the distance of three or more types: (0 gap between grayscale histograms; (ii) straight line (L1) distance; (iii) Euclidean (L2) length; (iv) chi-square distance; (v a Batacia distance; (vi) a Vassestin metric; (vii) a Swan metric; and (viii) a normalized correlation value. The distance calculator 420 can be adapted to be based on the structural component image and the reference image a value of the multiple type distance, generating a matching value for each reference image; and selecting each detection secret associated with a reference image having a matching value within a tolerable matching value range of 0 Selection module 430 can be adapted The multiple selected detection recipes are selected. The distance calculator 420 can be adapted to change the type distance to be calculated based on the evaluation results obtained by applying the at least one selected detection recipe. The image acquisition module 410 can include one or more cameras, An optical instrument and one or more sensors, an image grabber (not shown), and an image processor. The image acquisition module 410 can be set to different states in response to information stored in different detection recipes. Different states can be Such optical parameters (such as brightness, focal length), physical parameters (such as distance from the wafer, exposure time, chemical materials used), etc. are different. 16 201115667 Methods 100, 200 and 300 or any of them It can be executed by a processor (such as processor 444) that executes instructions. The instructions can be stored in a non-transitory computer readable medium such as a magnetic disk, a magnetic tape, a magnetic disk, and the like. For example, the computer program product can include a non-transitory computer readable medium storing instructions for acquiring an image of a structural component of the semiconductor device; calculating an image of the structural component and applying a complex detection recipe Multiple type distances between each of the obtained plurality of reference images; and automatically selecting at least one selected detection recipe from the complex detection recipe based on the value of the multiple type distance. Many modifications, substitutions, changes and equivalents will be apparent to those skilled in the art. It is therefore to be understood that the appended claims are intended to cover all such modifications and 17 201115667 I: BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 illustrates a method of an embodiment of the present invention; FIG. 2 illustrates a method of an embodiment of the present invention; FIG. 3 illustrates a method of an embodiment of the present invention; and FIG. A detection system in accordance with an embodiment of the present invention. [Description of main component symbols] 100···Method 100 250... Stage 250 105... Stage 105 270... Stage 270 110... Stage 110 300... Method 300 120... Stage 120 310··· Stage 310 130··· Stage 130 320· · Stage 320 140··· Stage 140 330··· Stage 330 142··· Stage 142 340... Stage 340 144... Stage 144 350... Stage 350 160··· Stage 160 400...Detection System 170... Stage 170 410... Image acquisition module 180...stage 180 420...distance calculator 190...stage 190 430...selection module 200···method 200 440...controller 210···stage 210 444.·processor 220···stage 220 450... secret recipe generator 230... stage 230 240... stage 240 18