TFT-LCD Mura & Y.H. Tseng 2006.12.4 Outline Mura Mura Mura 類 度 Mura Mura JND Mura Convolution filter (Filter design) Statistical method (ANOVA,EWMA) Backgroup estimation (LSD) 2
What is Mura- Mura Mura 不 不易 不 度 Otsu s Mura 離 3 What is Mura-Mura 類 Blob Mura 亮 Spot, particle Line Mura 亮 Large area Mura Rubbing Mura Gap Mura Gravity Mura Rubbing mura 4
What is Mura- Mura types -Blob Mura images 5 What is Mura- Mura types -Line Mura images 6
What is Mura- Mura types - Large area Mura images 7 What is Mura - 度 Mura 8
Mura -JND 率 where C is the local contrast of a candidate area-mura and S is the size of a candidate area-mura. 9 Mura Convolution filter (Filter design) Statistical method (ANOVA,EWMA) Backgroup estimation (LSD) 10
Convolution filter (filter design) 立 濾 Mura ( 濾 ) 路 不 11 立 濾 Mura 利 濾 (1) 濾 (2) 利 數 臨 濾 値 値 異 12
立 濾 Mura 濾 令 f ( x, ( x, 値 H ( x, m n 濾 f ( x, H ( x, Q( x, = m 1 n 1 i= 0 j= 0 f ( x + i, y + j) H ( i, j) 臨 若 µ Q Kσ Q < Q( x, < µ Q + Kσ ( x, Q Q ( x, : 濾 値 ( x, y (1) ) 13 立 濾 Mura ICA 濾 ICA 濾 Y = W LCD 不 立 來 利 立 來 不 X x11 x12 X = M x1k x x x 21 22 M 2k L L L xh1 x h2 M xhk The training data matrix W = H ( i, j) 14
立 濾 Mura ICA 濾 - ICA ICA Max Non-Gaussianity St. µ Kσ < Q( x, < µ + Kσ Q Q Q Q 濾 更 15 立 濾 Mura ICA 濾 - * * * * ICA 練 濾 w = ( w1, w2,..., w k ) 濾 便 行 * H ( i 1, j 1) = w( i 1) n+ j i = 1, 2,..., m; j =1, 2,..., n Q( x, m 1 n 1 = i= 1 j= 1 f ( x + i, y + j) H ( i, j) H ( i, j) 濾 m n, i = 1, 2,..., m; j = 1, 2,..., n f ( x, Q( x, 値 16
異 立 濾 Mura - 17 ICA 濾 Mura (1/2) Mura 度不 度 異 不易 識 易 MuraMura 18
ICA 濾 Mura (2/2) 行 Mura Mura Mura 19 ICA 濾 Mura (Detect defective line) 行 異 (Detect defect location) 異 行 率 20
ICA 濾 Mura ICA 練 ICA 來 行 練 濾 Y = WX Y W: X : X by row 濾 行 Mura 異 () 異 Mura 21 ( ) ICA 濾 Mura ICA SPC 行 異 () ICA ( ) 異 Mura 異 22
ICA 濾 Mura ICA SPC 不 SPC 異 行 若 ICA 異 行 23 ICA 濾 Mura (1/2) 利 行 convolution( ) 滑 異 Mura 輪廓 不 利 來 異 24
ICA 濾 Mura (2/2) 異 異 異 異 SPC 來 25 ICA 濾 Mura 異 行 行 SPC 異 識 26
ICA 濾 Mura 了 行聯 Mura 27 Mura Mura Convolution filter (Filter design) Statistical method (ANOVA, EWMA) Back-group estimation (LSD) 28
異數 (ANOVA) 數 數 量 ( 異數 )- (1) Ho µ 1=µ 2=µ 3=µ 4=µ 5=µ6=..=µ n (2) Ha 不 29 ANOVA Mura 立 異數 240 241 242 243 244 245 數 異不 Ho µ 1=µ 2,,=µ n MURA 1 5 9 14 20 26 32 38 44 50 56 62 68 74 80 86 92 98 105 112 119 126 133 140 245 250 255 數 異 Ho µ 1=µ 2,,=µ n 1 5 9 14 20 26 32 38 44 50 56 62 68 74 80 86 92 98 105 112 119 126 133 140 MURA 30
Mura ANOVA - 31 EWMA Mura 利 EWMA Exponentially Weighted Moving Average 量 力 UCL = x + 3σ λ ( 2 λ)n LCL = x 3σ λ ( 2 λ)n 滑 數 0 1 EWMA Chart for NO MURA EWMA Chart for MURA 244 243 3.0SL=243.9 252 EWMA 242 X=242.1 EWMA 247 3.0SL=246.5 241 X=244.6 240-3.0SL=240.4 0 50 100 150 Sample Number (1) Mura EWMA -3.0SL=242.7 242 0 50 100 150 Sample Number (2) Mura EWMA 32
Mura Convolution filter (filter design) Statistical method (ANOVA) Backgroup estimation (LS) 33 不 e.g. 異 ( 利 LS ) e.g. : where 利 Mura 34
參數 αregion-mura 了 α % window size p 料 (Least-Squares regression method) 料 行 (l) f p 1 J ( p) = WH 1 Ψ P ( l) xy f p ( x, 數 行 料 z J. Y. LEE and S. I. Yoo, Automatic Detection of Region-Mura Defect in TFT-LCD, 200435 來 行 r * xy = z xy f ( h) B ( x, * r xy z xy ( ) f h B ( x, 36
37 Thanks!
立 濾 立 (Independent Component Analysis, ICA) 來 濾 立 來 數 40
立 (ICA) Mixing matrix A s 1 Sources s 2 x 1 Observations x 2 x x 1 2 = a = a 11 21 s 1 s 1 + a 12 + a 22 s 2 s 2 41 立 (ICA) ICA(X) 立 數 (S) (A) X = A S x = A i s i X = [ xi ], S = [ si ] 數 立 s s j 立 i i j 42
立 (ICA) 了 立 數 (Demixing matrix) W 數 X 立 (Independent components) 立 Y 來 數 S Y = W X 1 W = A Y S 立 立 Y = [ y ], i y i s i 43 立 (ICA) s 1 x 1 x 2 X = A ( ) s 2 S y1 x1 y 2 Y = W ( ) x 2 X 44
立 (ICA) 立 兩 利 立 量 數 數 ICA method=objective function of independence + Optimization algorithm for de-mixing matrix W 45 ICA 濾 ICA 濾 粒 (Particle Swarm Optimization, PSO ) 列 量 濾 W 濾 更 46
ICA 濾 TFT-LCD 不 不 立 來 利 立 來 不 利 ICA 練濾 濾 47 Independent Component Analysis (ICA) The Independence of ICs are measured by non-gaussianity. The non-gaussianity of ICs can be measured by kurtosis. Kurtosis kurt 4 2 ( = E( y ) 3( E( y )) 2 For a Gaussian variable y, E ( y Zero for a Gaussian variable, and greater than zero for non- Gaussian random variables. Maximize 4 2 2 ) = 3( E ( y )) kurt ( = { kurt( } 2 0 48
Independent Component Analysis (ICA) X (observed signals) S(original signals) W (de - mixing matrix) = IC1 IC2 IC3 X (observed signals) Y (IC matrix) 49 ANOVA
(Flat Panel Display Measurement System-510 FPM-510) 51 量 度 (BM-5A) 量 1000(pixels) BM-5ALCD 離 量 離 (WD) BM-5A R 量 :D 量 = 1000pixels = R 2 52
異數 (ANOVA) - 1024x768 X16 Y9 144 度 量 數 行 異數 53 異數 (ANOVA)- 144 --- (1) Ho µ 1=µ 2=µ 3=µ 4=µ 5=µ6=..=µ 144 (2) Ha 不 54
ANOVA - Mura Mura 數 ---, 異數 立 行 Residual Frequency 4 3 2 1 0-1 -2-3 -4 45 40 35 30 25 20 15 10 5 0 Normal Plot of Residuals -3-2 -1 0 1 2 3 Normal Score Histogram of Residuals -4-3 -2-1 0 1 2 3 4 Residual Residual Analysis Residual Residual 5 4 3 2 1-1 0-2 -3-4 -5 7 7 7 7 I Chart of Residuals 6 0 100 200 300 400 500 600 700 Observation Number Residuals v s. Fits 4 3 2 1 0-1 -2-3 -4 241.2 242.2 243.2 Fit UCL=4.374 Mean=-5.9E-15 LCL=-4.374 (1) 不 ( F-test Robust,,, ANOVA ) (2) 立 (3) 異數 55 ANOVA - Mura Mura 數 異數 --- 144 數 異 240 241 242 243 244 245 1 5 9 14 20 26 32 38 44 50 56 62 68 74 80 86 92 98 105 112 119 126 133 140 56
ANOVA -Gap Mura Mura 數 ---, 異數 立 行 Residual Frequency 5 4 3 2-1 01-2 -3-4 -5 100 90 80 70 60 50 40 30 20 10 0 Normal Plot of Residuals -3-2 -1 0 1 2 3 Normal Score Histogram of Residuals -5 0 5 Residual Residual Analysis Residual Residual 5 0-5 5 4 3 2 1 0-1 -2-3 -4-5 7 7 7 7 7 7 7 I Chart of Residuals 1 6 6 5 1 0 100 200 300 400 500 600 700 Observation Number Residuals v s. Fits 244 249 254 Fit 7 7 7 UCL=4.510 Mean=-3.3E-15 LCL=-4.510 (1) 不 ( F-test Robust,,, ANOVA ) (2) 立 (3) 異數 57 ANOVA -Gap Mura Mura 數 異數 --- 144 數 異 245 250 255 異 1 5 9 14 20 26 32 38 44 50 56 62 68 74 80 86 92 98 105 112 119 126 133 140 58
Mura ANOVA - 59 EWMA 數 (Exponentially Weighted Moving Average, EWMA) EWMA 量 (ex:) 更 率 利 EWMA 來 LCD 不 60
EWMA 利 EWMA 量 力 UCL = x + 3σ λ ( 2 λ)n LCL = x 3σ λ ( 2 λ)n 滑 數 0 1 61 EWMA - =0.8 EWMA EWMA Chart for NO MURA EWMA Chart for MURA 244 243 3.0SL=243.9 252 EWMA 242 X=242.1 EWMA 247 3.0SL=246.5 241 X=244.6 240 0 50 100 150 Sample Number -3.0SL=240.4 242 0 50 100 150 Sample Number -3.0SL=242.7 (1) Mura EWMA (2) Mura EWMA 62
EWMA- - =0.2~1.0 率 0.2 0.4 0.6 0.8 1.0 Mura 率 144 /144 =100% 144 /144 =100% 144 /144 =100% 144 /144 =100% 144 /144 =100% Mura 率 132 /144 =91.7% 140 /144 =97.2% 142 /144 =98.6% 144 /144 =100% 143 /144 =99.3% (1) =0.2,0.4,0.6 1.0 (2) =0.8 (3) EWMA 來 Mura,=0.8 63 異數 (ANOVA) - 流 BM-5A 量 BM-5A 量 離 不良 略 量 數 利 BM-5A 行量 度 ANOVA, =95% 異 EWMA, =0.8 YES NO YES NO Mura Mura Mura Mura 64
(2/4) ( ) J ( p) 來 行 1( ) 0( ) median-filter 濾 濾 LS 行 66
(4/4) ( ) r µ 2 * xy 異數 σ Z ( x, Z 1, x, = { 0, r * xy ( * rxy µ / σ > T µ / σ T T 來 行 * r xy 67 類
69