计算光刻的机器学习：实用技巧.pdf

1、ML for Computational Lithography:Practical RecipesYoungsoo ShinSchool of EE,KAISTChip Manufacturing Mask synthesis:layout to masks Lithography:pattern transfer from mask to wafer(through exposure to light)Wafer processing:etch,ion implantation,etc Packaging2Computational lithography:mathematical and

2、 algorithmic approaches to improve the resolution attainable with lithographyML for Lithography Has been popular since 2010 Why:(1)ML provides“higher”modeling capability,(2)many applications are“image recognition”or“image conversion”Some ML solutions are already being provided through vendor product

3、s(e.g.Synopsys,Mentor,Brion)3Motivations ML(for chip design and manufacturing)has its own limitations Lack of benchmark and common data set Difficult to analyze and debug Data belongs to users;Model provided by vendors This talk Which lithography applications are more promising with ML?When training

4、 samples are sparse,do we still use ML?4ML for Lithography Promising applications Test pattern classification Etch bias model OPC and ILT ML may not be an ideal solution in Optical model Hotspot Assist features5Lithography Test Patterns Parametric patterns Represented by a few geometrical parameters

5、(e.g.width,space)Easy to build and analyze Cannot cover complex patterns Actual patterns Extracted from sample layouts;can cover more complex shapes Should be well classified6Insignificant or covered by the parametric test patternsRedundant due to multiple capturing:1-10,7-12,13-15,23-24Test Pattern

6、 Classification Representation of sample patterns in parameter space Clustering Selection of representative patterns Representation parameters are important Hanan grid(or Squish pattern)Image parameter space(IPS):Imax,Imin,Islp ML features7Etch Bias Model Etch bias Amount of under-etch(positive bias