lawpalyer logo
in判決書

智慧財產及商業法院111年度行專訴字第54號

關鍵資訊

  • 裁判案由
    發明專利舉發
  • 案件類型
    智財
  • 審判法院
    智慧財產及商業法院
  • 裁判日期
    112 年 08 月 10 日

11154 112713 駿 11172711106305370 113 112313退315 (221223227229 ) (497) 386 113 I653605 (106145566N01)113 1 13使(195196 ) ()1061225 106145566 (13)I653605() ()108923221121213 1 13222111322(111)0426511120281190108 1213113 ()11172711106305370() ()1 3D31 232 23233 3 2 13 E33(Neu ral net-work)23 3 IBM 西2020817 2 3 10011023 ()23 3西200256 西2012123Geoffrey Everest HintonAlex KrizhevskyAdvances in Neural Information Processing Systems(NIPS) Imagenet classification with deep convolutional neural networks)23 2 3 2使() 3 2 3 3調 2 3 ()11323 123 231 2 13333 100110便231 2使 21調 3 ()使( ) 322 12331001101 2 231 123 2818913 2 3 133941(image from the reticle design database)232 ()11324 234 3 4調 4 23使1 124 281913 ()39101213235 25 23113 231132453 5調 5 2324使131910121323525 ()7236246 23113 23113 246 3 6 (30)6 2324使17236246 ()89101213237 247 23113 23113 247 3 7 7 2324使1819101213237247 ()11247 11324 737 724 使1111247 ()113 () 22323413131316100110130 1 23 3 2 23 3130調3 production reticles)3 調10275 2323102 75 IBM Cloud 2323 ()23 3(neural network) 3 (2DEFECT IMAGEREF. IMAGE) 2 使2 3 2 3 TRAIN ENGINE ()31 313131621 00(defect image)110(defect free reference image)155 test reticle)130(2AB100110 1 使[0035]3(test reticle)100110(23315325335310) 1 3100(defect image)110(defect free reference image)130 3(neural network) Training engine 130)memory 140) 102532使3 (Training engine 130) (memory 140)() ()211d 2 [0058]( [0084]256128128 32322 調2 11d 調 ()23123 使2 3 32使2 使6 7 3 ()233 image from the reticle design database) 23233 2TRAIN ENGINE 130)使 ()23121011112 13 () ()3neural network memory140 132729Training engine stores the results of the analysis of station 120 in a memory 140, preferably a neural network.(120 140 a neural networka memory 140使( ) RAMneural network memory140preferably) Once memory 140 is sufficiently trained,inspection of production reticles can commence.(140) memory140neural network3memory 140 a neural networkmemory 140is sufficiently trained(140) 調() 3neural networkmemory 140 training engine 3 ()( ) 14033 ()2323411b1d 211.2.3.2 21lb 2[0084] 256 12812832322 調211d調21lb( 1d)2()()2 21 3131316 100110130133739 100'110'31lb(3A pair of imagestwo images )(3 defect free reference image110reference image110')(3 defect image100image 100' of the defect)1d3 2 233 2 2312341 () () 713 1061225107121910611851() 21221 222 () 調使0 ( 000300055) 1 1081213 1081213 131 91113 1a 1b 1c 調1d 1e 1 1 調 1 1 5 5 5 使0 18 9 1 調調 18 1 8 () 27( 1061225) 2 1.2.3.4.(1) 3 使(1) 32155100110130( AB) 便(132821) 120100110115120(G)130(C)120 140 使/ Result=A+B+ 100(1332235) 140便 100'110'150使 140(F)使150(1343647) 2 4 [0103]使便() [0089]9 使5 [0110]使/ [0025]使()使7,570,7967,676,077 [0085] [0079]WB 6 使100(Duda西2012) 使0.01305(8718) 7 S1S2 S3 S4 S5 S6 S7(1) ()231101213 123 2[0039]使 142 1 1c1e1 1b調1d 313131621 55100 110130( AB) 1 1b3132729Training engine stores the results of the analysis of station 120 in a memory 140, preferably a neural network.120 140 140 Training engine331336Once memory140 is sufficiently trained, inspection of production reticlescan commence.140便 31403 313134532 2 (131012)3 2329100110115 120(G)130(C)1201403 3 (3)33 (3) 311b1 調1d 321 1b1d231 1231 20058 008400572526FC1 調1d2 21lb6()1(1)672( 1)6(4243) 122 ()122 2 2 1 2 lb2 4831調1d1b2 1 調1d2 2 31313162100(defect image) 110 130133739 31lb1d3 2 23neural networkmemory 1403 3 3 1 lb調1d 33148 61123/300 31090 320 330340350 315 325335345 33103203303403503153253353451 2 3 2 323 23233 2 3 31025323(9C783132729140(453103313272936 31403 3 281 918 109121318 1 231 232101213 ()234113 231 4 (0103)(0089)411b調1d2341 2131234 1234213 ()2352345 39101213 231 5 (0110)(0085)(0079)511b調1d235 1234123 451 391012131 235 23451 2352345 39101213()2362346 7 231 630 (81416)11b調1d236 1234123461 712362346 123623467 ()2372347 89101213 231 7ReLU (00250047)7 1 1b調1d237 123 4123471 891012131 237 23471 2372347 89101213 ()234711 23471 1112347 1234711 113222113 113 1 981         112    8     10            202020 24111 12 1.   2. 3. 1. 2. 3. 4.         112    8     21   7257 西2017728CN 106991368A 5650 (2) 西2006718US 7079235B2Reticle design inspection system 4943 (3) 西2016128US 2016/0358070A1Automatic tuning of artificial neural networks 4231 (4) 西2017127US 2017/0351952A1Systems and methods incorporating a neural network and a forwardphysical model for semiconductor applications 3019 (5) 西2017613 Classifying Radio Galaxies with the Convolutional Neural Network 1811 (6) 西2017613CN 106845549A 101 (7)