tempo fast mask topography effect modeling with deep
play

TEMPO: Fast Mask Topography Effect Modeling with Deep Learning Wei - PowerPoint PPT Presentation

TEMPO: Fast Mask Topography Effect Modeling with Deep Learning Wei Ye 1 , Mohamed Baker Alawieh 1 , Yuki Watanabe 2 , Shigeki Nojima 2 , Yibo Lin 3 , David Z. Pan 1 1 ECE Department, University of Texas at Austin 2 Kioxia Corporation 3 CS


  1. TEMPO: Fast Mask Topography Effect Modeling with Deep Learning Wei Ye 1 , Mohamed Baker Alawieh 1 , Yuki Watanabe 2 , Shigeki Nojima 2 , Yibo Lin 3 , David Z. Pan 1 1 ECE Department, University of Texas at Austin 2 Kioxia Corporation 3 CS Department, Peking University

  2. Bottleneck in IC Manufacturing: Lithography ⧫ Moore’s law brings increasing manufacturing cost and challenges ⧫ Need to make sure design is manufacturable with high yield Light source Design target Wafer Lens Photomask Projection lens What you see (at design) ≠ what you get (at fab) Wafer 2

  3. Mask Topography Effects in Advanced Lithography Source Condenser Mask Pupils Near-field Near-field Lens Pupils Resist Aerial image Aerial image Substrate Thin mask approximation (Kirchhoff) Thick mask approximation 3

  4. Aerial Image Generation Resist Post Optical model processing model Mask Layout Aerial Image Slicing Threshold Resist Pattern h = 120 nm Intensity h = 110 nm . . . Optical model y (thin/thick mask) h = 70 nm x h = 60 nm 2D aerial image at . . . certain resist height h = 10 nm h = 0 nm 4

  5. Image-to-Image Translation Problems Semantic labeling [Long et al. 15’] Image colorization [Zhang et al. 16’] Boundary detection [Xie and Tu. 15’] Super-resolution [Johnson et al. 16’] Computer Graphics & Computer Vision & Computational Photography Machine Learning [“On Image-to-Image Translation”, Jun-Yan Zhu]

  6. Image-to-Image Translation In Lithography Fake/Real ILT Generator Engine Discriminator GAN-OPC [Yang+, DAC’18] Real Diff Mask pattern Aerial image Threshold Resist pattern Contour Encoder Decoder Optical Resist . processing model model Fake Generator . LithoGAN Input LithoGAN [Ye+, DAC’19] GAN-SRAF [Alawieh+, DAC’19] These applications are all single-domain transfer 6

  7. Cast as Multi-Domain Image-to-Image Translation ⧫ Facial image translation (facial attributes/expressions) › Bidirectional translation: original domain ⇔ target domain ⧫ Single mask pattern to multiple resist heights › Unidirectional translation: original domain ⇒ target domain Input h = 0 nm h = 10 nm h = 20 nm h = 90 nm h = 100 nm h = 110 nm h = 120 nm . . . 7

  8. <latexit sha1_base64="/vU8h/lX1rESZaAHqitk6dwHbs=">AB63icbVBNS8NAEJ34WetX1aOXxSJ4KokU9Fj04rFCv6ANZbOdtEt3N2F3I5TQv+DFgyJe/UPe/DcmbQ7a+mDg8d4M/OCWHBjXfb2djc2t7ZLe2V9w8Oj4rJ6cdEyWaYZtFItK9gBoUXGHbciuwF2ukMhDYDab3ud9Qm14pFp2FqMv6VjxkDNqc6k1lOVhperW3AXIOvEKUoUCzWHlazCKWCJRWSaoMX3Pja2fUm05EzgvDxKDMWVTOsZ+RhWVaPx0ceucXGbKiISRzkpZslB/T6RUGjOTQdYpqZ2YVS8X/P6iQ1v/ZSrOLGo2HJRmAhiI5I/TkZcI7NilhHKNM9uJWxCNWU2iycPwVt9eZ10rmtevVZ/rFcbd0UcJTiHC7gCD26gAQ/QhDYwmMAzvMKbI50X5935WLZuOMXMGfyB8/kDZz6N0w=</latexit> <latexit sha1_base64="rdI1VPpmf7sfroM+JomejvJLTug=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkoMeiF48V+wVtKJvtpl262YTdiVBCf4IXD4p49Rd589+4bXPQ1gcDj/dmJkXJFIYdN1vp7CxubW9U9wt7e0fHB6Vj0/aJk414y0Wy1h3A2q4FIq3UKDk3URzGgWSd4LJ3dzvPHFtRKyaOE24H9GREqFgFK302Bx4g3LFrboLkHXi5aQCORqD8ld/GLM04gqZpMb0PDdBP6MaBZN8VuqnhieUTeiI9yxVNOLGzxanzsiFVYkjLUthWSh/p7IaGTMNApsZ0RxbFa9ufif10sxvPEzoZIUuWLRWEqCcZk/jcZCs0ZyqklGlhbyVsTDVlaNMp2RC81ZfXSfuq6tWqtYdapX6bx1GEMziHS/DgGupwDw1oAYMRPMrvDnSeXHenY9la8HJZ07hD5zPH9cPjYM=</latexit> <latexit sha1_base64="RGY3D1jZgdJQcGWGlGkDWHCORWM=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkoMeiF48V+wVtKJvtpF262YTdjVBCf4IXD4p49Rd589+4bXPQ1gcDj/dmJkXJIJr47rfTmFjc2t7p7hb2ts/ODwqH5+0dZwqhi0Wi1h1A6pRcIktw43AbqKQRoHATjC5m/udJ1Sax7Jpgn6ER1JHnJGjZUemwM+KFfcqrsAWSdeTiqQozEof/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ6mkEWo/W5w6IxdWGZIwVrakIQv190RGI62nUWA7I2rGetWbi/95vdSEN37GZIalGy5KEwFMTGZ/02GXCEzYmoJZYrbWwkbU0WZsemUbAje6svrpH1V9WrV2kOtUr/N4yjCGZzDJXhwDXW4hwa0gMEInuEV3hzhvDjvzseyteDkM6fwB87nDyv+jbs=</latexit> <latexit sha1_base64="rdI1VPpmf7sfroM+JomejvJLTug=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkoMeiF48V+wVtKJvtpl262YTdiVBCf4IXD4p49Rd589+4bXPQ1gcDj/dmJkXJFIYdN1vp7CxubW9U9wt7e0fHB6Vj0/aJk414y0Wy1h3A2q4FIq3UKDk3URzGgWSd4LJ3dzvPHFtRKyaOE24H9GREqFgFK302Bx4g3LFrboLkHXi5aQCORqD8ld/GLM04gqZpMb0PDdBP6MaBZN8VuqnhieUTeiI9yxVNOLGzxanzsiFVYkjLUthWSh/p7IaGTMNApsZ0RxbFa9ufif10sxvPEzoZIUuWLRWEqCcZk/jcZCs0ZyqklGlhbyVsTDVlaNMp2RC81ZfXSfuq6tWqtYdapX6bx1GEMziHS/DgGupwDw1oAYMRPMrvDnSeXHenY9la8HJZ07hD5zPH9cPjYM=</latexit> <latexit sha1_base64="RGY3D1jZgdJQcGWGlGkDWHCORWM=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkoMeiF48V+wVtKJvtpF262YTdjVBCf4IXD4p49Rd589+4bXPQ1gcDj/dmJkXJIJr47rfTmFjc2t7p7hb2ts/ODwqH5+0dZwqhi0Wi1h1A6pRcIktw43AbqKQRoHATjC5m/udJ1Sax7Jpgn6ER1JHnJGjZUemwM+KFfcqrsAWSdeTiqQozEof/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ6mkEWo/W5w6IxdWGZIwVrakIQv190RGI62nUWA7I2rGetWbi/95vdSEN37GZIalGy5KEwFMTGZ/02GXCEzYmoJZYrbWwkbU0WZsemUbAje6svrpH1V9WrV2kOtUr/N4yjCGZzDJXhwDXW4hwa0gMEInuEV3hzhvDjvzseyteDkM6fwB87nDyv+jbs=</latexit> <latexit sha1_base64="JGwn3HCr1rEwiyxU5H7FGr7jg=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0lKQY9FLx4rtrXQhrLZbtqlm03YnQgl9Cd48aCIV3+RN/+N2zYHbX0w8Hhvhpl5QSKFQdf9dgobm1vbO8Xd0t7+weFR+fikY+JUM95msYx1N6CGS6F4GwVK3k0p1Eg+WMwuZ37j09cGxGrFk4T7kd0pEQoGEUrPbQGtUG54lbdBcg68XJSgRzNQfmrP4xZGnGFTFJjep6boJ9RjYJPiv1U8MTyiZ0xHuWKhpx42eLU2fkwipDEsbalkKyUH9PZDQyZhoFtjOiODar3lz8z+ulGF7mVBJilyx5aIwlQRjMv+bDIXmDOXUEsq0sLcSNqaMrTplGwI3urL6RTq3r1av2+Xmnc5HEU4QzO4RI8uIG3ET2sBgBM/wCm+OdF6cd+dj2Vpw8plT+APn8wfYk42E</latexit> <latexit sha1_base64="/vU8h/lX1rESZaAHqitk6dwHbs=">AB63icbVBNS8NAEJ34WetX1aOXxSJ4KokU9Fj04rFCv6ANZbOdtEt3N2F3I5TQv+DFgyJe/UPe/DcmbQ7a+mDg8d4M/OCWHBjXfb2djc2t7ZLe2V9w8Oj4rJ6cdEyWaYZtFItK9gBoUXGHbciuwF2ukMhDYDab3ud9Qm14pFp2FqMv6VjxkDNqc6k1lOVhperW3AXIOvEKUoUCzWHlazCKWCJRWSaoMX3Pja2fUm05EzgvDxKDMWVTOsZ+RhWVaPx0ceucXGbKiISRzkpZslB/T6RUGjOTQdYpqZ2YVS8X/P6iQ1v/ZSrOLGo2HJRmAhiI5I/TkZcI7NilhHKNM9uJWxCNWU2iycPwVt9eZ10rmtevVZ/rFcbd0UcJTiHC7gCD26gAQ/QhDYwmMAzvMKbI50X5935WLZuOMXMGfyB8/kDZz6N0w=</latexit> <latexit sha1_base64="JGwn3HCr1rEwiyxU5H7FGr7jg=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0lKQY9FLx4rtrXQhrLZbtqlm03YnQgl9Cd48aCIV3+RN/+N2zYHbX0w8Hhvhpl5QSKFQdf9dgobm1vbO8Xd0t7+weFR+fikY+JUM95msYx1N6CGS6F4GwVK3k0p1Eg+WMwuZ37j09cGxGrFk4T7kd0pEQoGEUrPbQGtUG54lbdBcg68XJSgRzNQfmrP4xZGnGFTFJjep6boJ9RjYJPiv1U8MTyiZ0xHuWKhpx42eLU2fkwipDEsbalkKyUH9PZDQyZhoFtjOiODar3lz8z+ulGF7mVBJilyx5aIwlQRjMv+bDIXmDOXUEsq0sLcSNqaMrTplGwI3urL6RTq3r1av2+Xmnc5HEU4QzO4RI8uIG3ET2sBgBM/wCm+OdF6cd+dj2Vpw8plT+APn8wfYk42E</latexit> Multi-Domain Image-to-Image Translation Simple model Ideal model Encoders Decoders Target domain 1 Target domain 1 T 1 T 1 Target domain 2 Target domain 2 Target domain label i Source Source T 2 T 2 domain domain Target domain i Target domain i T i T i Encoder Decoder Target domain m Target domain m T m T m ⧫ #Models scales up with #domains ⧫ Exploit the high correlation between different domains ⧫ Models are independent 8

Recommend


More recommend