# US 2026/0094310 A1 · Generation of synthetic fault images of semiconductor specimens

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0094310 A1
- Document ID: US20260094310A1
- Status: Published application
- Kind: A1
- Application number: 18/902,793
- Filed: 2024-09-30
- Published: 2026-04-02
- Earliest claimed: 2024-09-30
- Assignees: Applied Materials Israel Ltd.
- Applicants: Applied Materials Israel Ltd.
- Inventors: Boris SHERMAN; Boris LEVANT; Ran BADANES; Ran YACOBY; Bar DUBOVSKI; Tomer YEMINY
- Web page: https://patentgrep.com/patent/US20260094310A1

## CPC

- G06T 11/00: Two-dimensional [2D] image generation
- G06T 7/0004: {Industrial image inspection}
- G06T 7/0008: {checking presence/absence}
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06T 2207/10061: from scanning electron microscope
- 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
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

The presently disclosed subject matter includes a computer system and a computer-implemented method of generating synthetic examination output images, including synthetic fault images and synthetic fault-free images. The synthetic fault images comprise artificially generated 3D defects. The proposed technique enables fast, accurate, and efficient generation of a large and diverse collection of synthetic fault images and fault-free images, which can be implemented in runtime, as part of the examination process of semiconductor specimens.

## Claims

### Claim 1

A computer system configured for automatic generation of synthetic images of a semiconductor examination tool, the system comprising a processing circuitry configured to:
obtain one or more images of a semiconductor specimen acquired by the semiconductor examination tool;
process the one or more images using a height map generation algorithm dedicated to generating, from each image, a respective height map of the semiconductor specimen, thereby obtaining one or more height maps;
modify at least some of the one or more height maps to include at least one artificially generated three-dimensional (3D) defect, thereby generating one or more respective synthetic fault height maps;
apply a first machine learning model to the one or more respective synthetic fault height maps, wherein the first machine learning model is trained to generate, from a height map of a semiconductor specimen, a corresponding synthetic image of the semiconductor specimen, thereby obtaining one or more respective synthetic fault images.

### Claim 2 (depends on 1)

The system of claim 1, wherein the one or more examination output images include one or more fault-free images; and wherein the processing circuitry is configured to:
process the one or more fault-free images using a height map generation algorithm dedicated for generating, from each image, a respective height map of the semiconductor specimen, thereby obtaining one or more height maps;
apply the first machine learning model to the one or more height maps, thereby obtaining one or more respective synthetic fault-free images.

### Claim 3 (depends on 1)

The system of claim 1, wherein generation of synthetic fault images is executed as part of a defects detection process, dedicated for detecting defects in examination output images of a semiconductor specimen; wherein the processing circuitry is configured to execute, as part of the defects detection process:
generate a training dataset from the synthetic fault images;
utilize the training dataset for training a second machine learning model dedicated for receiving examination output images of a semiconductor specimen and detecting defects therein.

### Claim 4 (depends on 2)

The system of claim 2, wherein generation of synthetic fault images and synthetic fault-free images is executed as part of a defects detection process, dedicated for detecting defects in examination output images of a semiconductor specimen; wherein the processing circuitry is configured to execute, as part of the defects detection process:
generate a training dataset from the synthetic fault images and fault-free images;
utilize the training dataset for training a second machine learning model dedicated for receiving examination output images of a semiconductor specimen and detecting defects therein.

### Claim 5 (depends on 4)

The system of claim 4 wherein the processing circuitry is configured, following training of the second machine learning model, to apply the second machine learning model to other examination output images of the semiconductor specimen acquired by the semiconductor examination tool, thereby detecting defects in the semiconductor specimens.

### Claim 6 (depends on 5)

The system of claim 5, wherein the defects detection process is executed as part of a semiconductor examination process for detecting defects in real-time during fabrication.

### Claim 7 (depends on 5)

The system according to claim 5, wherein the processing circuitry is configured to initiate execution of the defects detection process responsive to receiving data indicating that examination output images of a semiconductor specimen contain a type of defect that was not included in a training set used for training the second machine learning model, wherein the at least one artificially generated three-dimensional (3D) defect includes the type of defect, thereby updating the second machine learning model to detect the type of defect, on-the-fly.

### Claim 8 (depends on 1)

The system of claim 1, wherein the processing circuitry is configured for modifying each height map to include at least one artificially generated 3D defect to apply procedural generation methods for creating a 3D defect and implanting the 3D defect in the height maps.

### Claim 9 (depends on 8)

The system of claim 8, wherein the procedural generation methods include at least one of Perlin noise and Voronoi noise.

### Claim 10 (depends on 1)

The system of claim 1 wherein the semiconductor examination tool is a Scanning Electron Microscope (SEM), and the examination output images are SEM images of the semiconductor specimen.

### Claim 11

A computer-implemented method of generating synthetic images of a semiconductor examination tool, comprising:
obtaining one or more examination output images of a semiconductor specimen acquired by the semiconductor examination tool;
processing the one or more examination output images using a height map generation algorithm dedicated for generating, from each image, a respective height map of the semiconductor specimen, thereby obtaining one or more height maps;
modifying at least some of the one or more height maps to include at least one artificially generated three-dimensional (3D) defect, thereby generating one or more respective synthetic fault height maps;
applying a first machine learning model to the one or more respective synthetic fault height maps, wherein the first machine learning model is trained to generate, from height maps of a semiconductor specimen, corresponding synthetic images of the semiconductor specimen, thereby obtaining one or more respective synthetic fault images.

### Claim 12 (depends on 11)

The computer-implemented method of claim 11, wherein the one or more examination output images include one or more fault-free images; the method comprising:
processing the one or more fault-free images using a height map generation algorithm dedicated for generating, from each image, a respective height map of the semiconductor specimen, thereby obtaining one or more height maps;
applying the first machine learning model to the one or more height maps, thereby obtaining one or more respective synthetic fault-free images.

### Claim 13 (depends on 11)

The computer-implemented method of claim 11, wherein generation of synthetic fault images is executed as part of a defects detection process, dedicated for detecting defects in examination output images of a semiconductor specimen; the method comprising executing, as part of the defects detection process:
generating a training dataset from the synthetic fault images and the fault-free images;
utilizing the training dataset for training a second machine learning model dedicated for receiving examination output images of a semiconductor specimen and detecting defects therein.

### Claim 14 (depends on 12)

The computer-implemented method of claim 12, wherein generation of synthetic fault images and synthetic fault-free images is executed as part of a defects detection process, dedicated for detecting defects in examination output images of a semiconductor specimen; the method comprising, executing as part of the defects detection process, operations comprising:
generating a training dataset from the synthetic fault images and synthetic fault-free images;
utilizing the training dataset for training a second machine learning model dedicated for receiving examination output images of a semiconductor specimen and detecting defects therein.

### Claim 15 (depends on 14)

The computer-implemented method of claim 14 comprising: following training of the second machine learning model, applying a second machine learning model to examination output images of the semiconductor specimen acquired by the semiconductor examination tool, thereby detecting defects in the semiconductor specimens.

### Claim 16 (depends on 15)

The computer-implemented method of claim 15, wherein the defects detection process is executed as part of a semiconductor fabrication process for detecting defects in real-time.

### Claim 17 (depends on 15)

The computer-implemented method according to claim 15 comprising:
responsive to receiving data indicating that examination output images of a semiconductor specimen contain a type of defect that was not included in a training set used for training the second machine learning model, executing the defects detection process where the at least one artificially generated three-dimensional (3D) defect includes the type of defect, thereby updating the second machine learning model to detect the type of defect, on-the-fly.

### Claim 18 (depends on 11)

The computer-implemented method of claim 11, wherein modifying each height map to include at least one artificially generated 3D defect comprises applying procedural generation methods for creating a 3D defect, and implanting the 3D defect in the height maps.

### Claim 19 (depends on 11)

The computer-implemented method of claim 11 wherein the examination tool is a Scanning Electron Microscope (SEM), and the examination output images are SEM images of the semiconductor specimen.

### 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 generating synthetic images of a semiconductor examination tool, obtaining one or more examination output images of a semiconductor specimen acquired by the semiconductor examination tool; the method comprising:
processing the one or more examination output images using a height map generation algorithm dedicated for generating, from each image, a respective height map of the semiconductor specimen, thereby obtaining one or more height maps;
modifying at least some of the one or more height maps to include at least one artificially generated three-dimensional (3D) defect, thereby generating one or more respective synthetic fault height maps;
applying a first machine learning model to the one or more respective synthetic fault height maps, wherein the first machine learning model is trained to generate, from height maps of a semiconductor specimen, corresponding synthetic images of the semiconductor specimen, thereby obtaining one or more respective synthetic fault images.
