# US 12,181,793 B1 · Mask synthesis using tensor-based computing platforms > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,181,793 B1 - Document ID: US12181793B1 - Status: Granted - Kind: B1 - Application number: 17/216,056 - Filed: 2021-03-29 - Granted: 2024-12-31 - Earliest claimed: 2020-04-01 - Assignees: Synopsys, Inc. - Applicants: Synopsys, Inc. - Inventors: Peng Liu - Web page: https://patentgrep.com/patent/US12181793B1 ## CPC - G03F 1/70: Adapting basic layout or design of masks to lithographic process requirements, e.g., second iteration correction of mask patterns for imaging - G03F 1/36: Masks having proximity correction features; Preparation thereof, e.g. optical proximity correction [OPC] design processes - G06N 3/045: Combinations of networks - G06N 3/08: Learning methods - G03F: PHOTOMECHANICAL PRODUCTION OF TEXTURED OR PATTERNED SURFACES, e.g. FOR PRINTING, FOR PROCESSING OF SEMICONDUCTOR DEVICES; MATERIALS THEREFOR; ORIGINALS THEREFOR; APPARATUS SPECIALLY ADAPTED THEREFOR; (phototypographic composing devices B41B; photosensitive materials or processes for photographic purposes G03C; electrophotography, sensitive layers or processes therefor G03G) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A tensor-based computing platform performs mask synthesis. A method includes accessing a layout of a lithographic mask and estimating a printed pattern resulting from use of the lithographic mask in a lithographic process. The lithographic process is modeled by a sequence of at least two forward models. A first of the forward models uses the layout of the lithographic mask as input and a last of the forward models produces the estimated printed pattern as output. The method further includes modifying the layout of the lithographic mask based on differences between the estimated printed pattern and a target printed pattern. All of the forward models are implemented on the tensor-based computing platform. ## Claims ### Claim 1 A method comprising: accessing a layout of a lithographic mask; estimating a printed pattern resulting from use of the lithographic mask in a lithographic process, wherein the lithographic process is modeled by a sequence of at least two forward models, a first of the forward models using the layout of the lithographic mask as input and a last of the forward models producing the estimated printed pattern as output; and applying a mask synthesis model to modify the layout of the lithographic mask based on differences between the estimated printed pattern and a target printed pattern; wherein the mask synthesis model includes a deep neural network followed by a shallow neural network, the deep neural network has more hidden layers than the shallow neural network, and the shallow neural network is pattern specific and has not more than two hidden layers; wherein the sequence of at least two forward models and the mask synthesis model are implemented on a tensor-based computing platform. ### Claim 2 (depends on 1) The method of claim 1, wherein the sequence of forward models comprises at least one of: a mask model that estimates a near field resulting from an illuminating field applied to the lithographic mask; an imaging model that estimates an aerial image resulting from imaging of the near field onto a wafer; and a resist model that estimates the printed pattern resulting from the aerial image exposing a resist process. ### Claim 3 (depends on 1) The method of claim 1, wherein the forward models in the sequence are separable and are replaceable without requiring changes to the other forward models in the sequence. ### Claim 4 (depends on 1) The method of claim 1, wherein at least one of the forward models includes constraints based on a physical process modeled by the forward model. ### Claim 5 (depends on 1) The method of claim 1, wherein at least one of the forward models is based on a physical process modeled by the forward model. ### Claim 6 (depends on 1) The method of claim 1, wherein at least one of the forward models is a machine learning model trained by supervised learning. ### Claim 7 (depends on 1) The method of claim 1, wherein at least one of the forward models is not a machine learning model trained by supervised learning. ### Claim 8 (depends on 1) The method of claim 1 wherein the sequence of forward models consists of: a mask model that estimates a near field resulting from an illuminating field applied to the lithographic mask, wherein the mask model is a neural network model trained by supervised learning; an Abbe or Hopkins imaging model that estimates an aerial image resulting from imaging of the near field onto a wafer; and a resist model that estimates the printed pattern resulting from the aerial image exposing a resist process, wherein the resist model is also a neural network model trained by supervised learning. ### Claim 9 (depends on 1) The method of claim 1, wherein the mask synthesis model is based on a gradient of a loss function between the estimated and target printed patterns. ### Claim 10 A system comprising: a memory storing instructions; and a tensor-based computing platform, coupled with the memory and to execute the instructions, the instructions when executed cause the tensor-based computing platform to: execute a sequence of at least two forward models to estimate a printed pattern resulting from use of a lithographic mask in a lithographic process, wherein a first of the forward models uses a layout of the lithographic mask as input and a last of the forward models produces the estimated printed pattern as output; and execute a mask synthesis model to modify the layout of the lithographic mask based on differences between the estimated printed pattern and a target printed pattern; wherein the mask synthesis model includes a deep neural network followed by a shallow neural network, the deep neural network has more hidden layers than the shallow neural network, and the shallow neural network is pattern specific and has not more than two hidden layers. ### Claim 11 (depends on 10) The system of claim 10, wherein the tensor-based computing platform comprises a software framework comprising tensor multiplications and tensor operations. ### Claim 12 (depends on 10) The system of claim 10, wherein the tensor-based computing platform comprises a hardware architecture configured for tensor multiplications and tensor operations. ### Claim 13 (depends on 10) The system of claim 10, wherein the sequence of forward models comprises at least one of: a mask model that estimates a near field resulting from an illuminating field applied to the lithographic mask; an imaging model that estimates an aerial image resulting from imaging of the near field onto a wafer; and a resist model that estimates the printed pattern resulting from the aerial image exposing a resist process. ### Claim 14 (depends on 10) The system of claim 10, wherein: at least one of the forward models is a machine learning model trained by supervised learning; and at least another one of the forward models is not a machine learning model trained by supervised learning. ### Claim 15 A non-transitory computer readable medium comprising stored instructions, which when executed by a processor, cause the processor to: access a layout of a lithographic mask; estimate a printed pattern resulting from use of the lithographic mask in a lithographic process, wherein the lithographic process is modeled by a sequence of at least two forward models, a first of the forward models using the layout of the lithographic mask as input and a last of the forward models producing the estimated printed pattern as output; and execute a mask synthesis model to modify the layout of the lithographic mask based on differences between the estimated printed pattern and a target printed pattern, wherein the mask synthesis model includes a deep neural network followed by a shallow neural network, the deep neural network has more hidden layers than the shallow neural network, and the shallow neural network is pattern specific and has not more than two hidden layers; wherein all of the forward models and the mask synthesis model use tensor multiplications and tensor operations. ### Claim 16 (depends on 15) The computer readable medium of claim 15, wherein the sequence of forward models comprises at least one of: a mask model that estimates a near field resulting from an illuminating field applied to the lithographic mask; an imaging model that estimates an aerial image resulting from imaging of the near field onto a wafer; and a resist model that estimates the printed pattern resulting from the aerial image exposing a resist process. ### Claim 17 (depends on 15) The computer readable medium of claim 15, wherein the forward models in the sequence are separable and are replaceable without requiring changes to the other forward models in the sequence. ### Claim 18 (depends on 15) The computer readable medium of claim 15, wherein at least one of the forward models is a machine learning model trained by supervised learning.