Microfabrication Tool Fault Detection Using Neural Network Monitoring

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Semiconductor processing tools often fail, leading to improperly processed wafers that need to be scrapped, increasing production costs and causing unscheduled processing stoppages.

Innovation Solution

Implement an AI-based neural network system that learns from machine data to predict tool faults, allowing for proactive intervention to prevent failures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If semiconductor processing tools operate continuously without intervention, then productivity increases, but tool faults occur leading to wafer scrap and production cost increases

Engineering Contradiction:

Improveprocessing throughputVSAvoidtool fault rate

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The neural network system performs preliminary fault detection by analyzing machine data patterns before actual tool failures occur. The system trains on historical data to recognize precursors to faults, enabling proactive intervention that prevents wafer scrap while maintaining continuous operation

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements continuous feedback by monitoring machine data from processing tools and using the neural network to predict faults. This closed-loop approach allows real-time adjustments and alerts, enabling operators to address issues before they cause production failures

Inventive Principle:

Principle #23Feedback

2Reliability

If processing tools are monitored and maintained more frequently, then reliability improves, but production time increases due to stoppages

Engineering Contradiction:

Improvetool operational stabilityVSAvoidprocessing stoppage time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary assessments of tool health status using neural network predictions, allowing scheduled maintenance to be performed only when faults are anticipated. This prevents unnecessary stoppages while ensuring maintenance occurs before actual failures

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The monitoring system operates autonomously, continuously analyzing machine data without requiring manual intervention or frequent operator checks. This automated self-monitoring reduces the need for manual inspections and associated production stoppages

Inventive Principle:

Principle #25Self-service

3Difficulty of detecting and measuring

If manual monitoring and inspection of processing tools is performed, then fault detection capability improves, but labor costs and processing time increase

Engineering Contradiction:

Improvefault detection accuracyVSAvoidmonitoring system complexity

Core Design Contradiction:

Difficulty of detecting and measuringVSDevice complexity

Solution Approach 1:

The system replaces manual monitoring and inspection with an automated neural network-based detection system. The neural network automatically analyzes machine data patterns to detect faults, eliminating the need for manual intervention while improving detection accuracy and consistency

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS12593646B2Automated fault detection in microfabrication

Publication Date: 2026.03.31 美国泰尔制造与工程公司

AI summary

A method including: collecting first processing tool machine data from a first processing tool while treating semiconductor substrates, the first processing tool machine data including process data and operational codes associated with one or more discrete intervals of time during the treatments, training a first neural network with the first processing tool machine data from the first processing tool, and generating a first output indicative of a fault of the first processing tool from the first neural network, based, at least in part, on applying subsequent machine data from at least one processing tool.