Neural Network Super-Resolution for Defect Review Imaging

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current Automated Defect Review (ADR) processes in sample analysis systems are hindered by long processing times, defocus issues due to irregular wafer flatness, and the risk of physical contact, which significantly impact throughput and image quality.

Innovation Solution

Employing a neural network to generate high-resolution defect images from lower-resolution scan images, trained during recipe setup, allowing for faster ADR without the need for high-resolution image collection, thus reducing processing time and eliminating defocus issues while preventing wafer damage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If high-resolution images are collected at each defect location during ADR, then image quality is improved, but processing time increases significantly

Engineering Contradiction:

Improveimage qualityVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary scanning at lower resolution to identify defect locations before the ADR process. This preliminary action allows the subsequent high-resolution imaging to be targeted only at relevant locations, reducing overall processing time while maintaining image quality where needed

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system applies different imaging resolutions to different regions: lower resolution for routine scanning and higher resolution only at identified defect locations. This local differentiation optimizes the balance between image quality and processing time by concentrating computational resources only where necessary

Inventive Principle:

Principle #3Local quality

2Measurement precision

If review objective moves to each defect location for imaging, then detailed inspection is improved, but risk of physical contact with wafer increases

Engineering Contradiction:

Improveinspection detailVSAvoidwafer damage risk

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The system creates optical copies or images of the wafer surface at defect locations without requiring physical contact. The review objective captures images through optical fields, eliminating the need for mechanical contact while still providing detailed inspection capability

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system replaces mechanical positioning and contact-based inspection with optical field-based imaging. The review objective uses optical fields to probe and image defect locations, substituting mechanical interaction with electromagnetic field interaction to eliminate wafer damage risk

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Ease of operation

If ADR is performed on wafers with irregular flatness profile, then defect review capability is maintained, but defocus issues occur

Engineering Contradiction:

Improvedefect review capabilityVSAvoidimage focus

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The system dynamically adjusts the review objective position or wafer stage height in real-time during ADR based on measured wafer flatness variations. This dynamic compensation maintains focus across irregular surfaces, enabling defect review capability while preventing defocus issues

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system uses feedback from wafer flatness measurements to adjust imaging parameters during ADR. The flatness profile data feeds back to control the review objective positioning, automatically compensating for irregularities and maintaining optimal focus throughout the defect review process

Inventive Principle:

Principle #23Feedback

Data Source

PatentEP3853592B1Super-resolution defect review image generation through generative adversarial networks

Publication Date: 2023.09.06 KLA CORP

Figure 1
Figure 2A
Figure 2B

AI summary

A system for analyzing a sample includes an inspection sub-system and at least one controller. The inspection sub-system is configured to scan a sample to collect a first plurality of sample images having a first image resolution. The controller is configured to generate a defect list based on the first plurality of sample images. The controller is further configured to input images corresponding to the defect list into a neural network that is trained with source data including sample images having the first image resolution and sample images having a second image resolution higher than the first image resolution. The controller is further configured to generate a second plurality of sample images with the neural network based on the images corresponding to the defect list, where the second plurality of sample images have the second image resolution and correspond to the defect list.