# US 12,223,641 B2 · Defect detection of a semiconductor specimen > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,223,641 B2 - Document ID: US12223641B2 - Status: Granted - Kind: B2 - Application number: 17/748,996 - Filed: 2022-05-19 - Published: 2023-11-23 - Granted: 2025-02-11 - Earliest claimed: 2022-05-19 - Assignees: Applied Materials Israel Ltd. - Applicants: Applied Materials Israel Ltd. - Inventors: Boaz Dudovich; Assaf Ariel; Amir Bar; Lior Yehieli; Chen Itzikowitz; Shiran Ben Israel; Lior Katz; Eli Oren Joni; Eyal Rot - Web page: https://patentgrep.com/patent/US12223641B2 ## CPC - G06T 7/001: {using an image reference approach} - G06T 3/10: Selection of transformation methods according to the characteristics of the input images - H04N 7/01: Conversion of standards {, e.g. involving analogue television standards or digital television standards processed at pixel level} - G01N 21/9501: {Semiconductor wafers (manufacturing processes per se of semiconductor devices implementing a measuring step H10P74/20)} - G01N 2021/8883: {involving the calculation of gauges, generating models} - G01N 2201/1296: using neural networks - G06N 3/09: Supervised learning - G06T 2207/10061: from scanning electron microscope - G06T 2207/20021: Dividing image into blocks, subimages or windows - G06T 2207/20081: Training; Learning - G06T 2207/20084: Artificial neural networks [ANN] - G06T 2207/30148: Semiconductor; IC; Wafer - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION - G01N: INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES (measuring or testing apparatus or processes other than immunoassay, involving enzymes or microorganisms C12M, C12Q) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract There is provided a system and method of defect detection of a semiconductor specimen. The method includes obtaining a first image of the specimen acquired at a first bit depth, converting by a first processor the first image to a second image with a second bit depth lower than the first bit depth, transmitting the second image to a second processor configured to perform first defect detection on the second image using a first defect detection algorithm to obtain a first set of defect candidates, and sending locations of the first set of defect candidates to the first processor, extracting, from the first image, a set of image patches corresponding to the first set of defect candidates based on the locations, and performing second defect detection on the set of image patches using a second defect detection algorithm to obtain a second set of defect candidates. ## Claims ### Claim 1 A computerized system of defect detection of a semiconductor specimen, the system comprising a first processing and memory circuitry (PMC) configured to: obtain a first image representative of at least a part of the semiconductor specimen, the first image acquired by an inspection tool at a first bit depth; convert the first image to a second image with a second bit depth, wherein the second bit depth is a lower bit depth than the first bit depth; transmit the second image to a second PMC operatively connected to the first PMC, wherein the second PMC is configured to perform first defect detection on the second image using a first defect detection algorithm, giving rise to a first set of defect candidates, and send locations of the first set of defect candidates to the first PMC; upon receiving the locations from the second PMC, extract, from the first image, a set of image patches corresponding to the first set of defect candidates based on the locations; and perform second defect detection on the set of image patches using a second defect detection algorithm, giving rise to a second set of defect candidates. ### Claim 2 (depends on 1) The computerized system according to claim 1, wherein the first bit depth is one of: 32 bits per pixel, 24 bits per pixel, and 16 bits per pixel; and the second bit depth is one of: 16 bits per pixel and 8 bits per pixel. ### Claim 3 (depends on 1) The computerized system according to claim 1, wherein the first PMC comprises one or more graphics processing units (GPUs) optimized to process images with the first bit depth. ### Claim 4 (depends on 1) The computerized system according to claim 1, wherein the second PMC comprises one or more digital signal processors (DSPs) optimized to process images with the second bit depth. ### Claim 5 (depends on 1) The computerized system according to claim 1, wherein the first PMC is configured to convert the first image by selecting a transformation function based on pixel value distribution of the first image, the transformation function usable for converting a range of pixel values represented by the first bit depth to a range of pixel values represented by the second bit depth, and the pixel values of the first image are converted using the transformation function to obtain the second image. ### Claim 6 (depends on 5) The computerized system according to claim 5, wherein the transformation function is specifically selected and configured to improve accuracy of the second image with the second bit depth. ### Claim 7 (depends on 1) The computerized system according to claim 1, wherein the first defect detection algorithm is a die-to-reference detection algorithm using one or more reference images of the second image. ### Claim 8 (depends on 7) The computerized system according to claim 7, wherein the first defect detection algorithm is specifically configured for defect detection on one or more second images with the second bit depth. ### Claim 9 (depends on 1) The computerized system according to claim 1, wherein the second defect detection algorithm comprises a machine learning (ML) model trained using one or more training images with the first bit depth and respective labels associated therewith indicative of presence of defects of interest (DOIs) or nuisances thereof. ### Claim 10 (depends on 9) The computerized system according to claim 9, wherein the second set of defect candidates are reviewed by a review tool, giving rise to a set of DOIs, the ML model is re-trained using image patches corresponding to the set of DOIs and associated with a label of DOI, and image patches corresponding to non-DOI defect candidates from the second set of defect candidates and associated with a label of nuisance, and the re-trained ML model is used to re-perform the second defect detection. ### Claim 11 (depends on 1) The computerized system according to claim 1, wherein the first PMC is configured to obtain a plurality of first images sequentially acquired by the inspection tool during runtime inspection and representative of different parts of the semiconductor specimen, and perform the converting, transmitting, extracting and performing for each given first image of the plurality of first images, and wherein each given first image is converted adaptively by selecting a specific transformation function based on pixel value distribution of the given first image. ### Claim 12 (depends on 1) The computerized system according to claim 1, further comprising the second PMC, wherein the first PMC is optimized for first bit depth processing and is thereby capable of improving detection sensitivity, the second PMC is optimized for second bit depth processing and is thereby capable of improving computation efficiency and lowering computation cost, and wherein the first PMC and the second PMC are leveraged in a specific way for performing respective defect detections on images with different bit depths, thereby enabling improved throughput without sacrificing detection sensitivity. ### Claim 13 A computerized method of defect detection of a semiconductor specimen, the method performed by a first processing and memory circuitry (PMC) and comprising: obtaining a first image representative of at least a part of the semiconductor specimen, the first image acquired by an inspection tool at a first bit depth; converting the first image to a second image with a second bit depth, wherein the second bit depth is a lower bit depth than the first bit depth; transmitting the second image to a second PMC operatively connected to the first PMC, wherein the second PMC is configured to perform first defect detection on the second image using a first defect detection algorithm, giving rise to a first set of defect candidates, and send locations of the first set of defect candidates to the first PMC; upon receiving the locations from the second PMC, extracting, from the first image, a set of image patches corresponding to the first set of defect candidates based on the locations; and performing second defect detection on the set of image patches using a second defect detection algorithm, giving rise to a second set of defect candidates. ### Claim 14 (depends on 13) The computerized method according to claim 13, wherein the first PMC comprises one or more graphic processing units (GPUs) optimized to process images with the first bit depth. ### Claim 15 (depends on 13) The computerized method according to claim 13, wherein the second PMC comprises one or more digital signal processors (DSPs) optimized to process images with the second bit depth. ### Claim 16 (depends on 13) The computerized method according to claim 13, wherein the converting the first image comprises selecting a transformation function based on pixel value distribution of the first image, the transformation function usable for converting a range of pixel values represented by the first bit depth to a range of pixel values represented by the second bit depth, and the pixel values of the first image are converted using the transformation function to obtain the second image. ### Claim 17 (depends on 13) The computerized method according to claim 13, wherein the second defect detection algorithm comprises a machine learning (ML) model trained using one or more training images with the first bit depth and respective labels associated therewith indicative of presence of defects of interest (DOIs) or nuisances thereof. ### Claim 18 (depends on 17) The computerized method according to claim 17, wherein the second set of defect candidates are reviewed by a review tool, giving rise to a set of DOIs, the ML model is re-trained using image patches corresponding to the set of DOIs and associated with a label of DOI, and image patches corresponding to non-DOI defect candidates from the second set of defect candidates and associated with a label of nuisance, and the re-trained ML model is used to re-perform the second defect detection. ### Claim 19 (depends on 13) The computerized method according to claim 13, wherein the first PMC is optimized for first bit depth processing and is thereby capable of improving detection sensitivity, the second PMC is optimized for second bit depth processing and is thereby capable of improving computation efficiency and lowering computation cost, and wherein the first PMC and the second PMC are leveraged in a specific way for performing respective defect detections on images with different bit depths, thereby enabling improved throughput without sacrificing detection sensitivity. ### Claim 20 A non-transitory computer readable storage medium tangibly embodying a program of instructions that, when executed by a computer, cause the computer to perform a method of defect detection of a semiconductor specimen, the method comprising: obtaining a first image representative of at least a part of the semiconductor specimen, the first image acquired by an inspection tool at a first bit depth; converting the first image to a second image with a second bit depth, wherein the second bit depth is a lower bit depth than the first bit depth; transmitting the second image to a second PMC operatively connected to the first PMC, wherein the second PMC is configured to perform first defect detection on the second image using a first defect detection algorithm, giving rise to a first set of defect candidates, and send locations of the first set of defect candidates to the first PMC; upon receiving the locations from the second PMC, extracting, from the first image, a set of image patches corresponding to the first set of defect candidates based on the locations; and performing second defect detection on the set of image patches using a second defect detection algorithm, giving rise to a second set of defect candidates.