Neural Network Prediction Model for Substrate Holder Maintenance

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The existing methods for maintaining substrate holders in plating apparatuses are inefficient, as they often require maintenance only after a failure occurs, leading to delayed preparation and prolonged downtime due to the inability to accurately predict when maintenance is necessary.

Innovation Solution

A prediction model using machine learning, specifically a neural network, is constructed to predict the number of substrates that can be plated before a substrate holder fails, and a selection model to identify the component causing the failure, allowing for timely maintenance and reducing downtime.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If maintenance is performed only after failure occurs, then device complexity is reduced, but productivity decreases due to prolonged downtime

Engineering Contradiction:

Improveplating productivityVSAvoidmaintenance system complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by collecting condition data from sensors during substrate holder usage and training prediction models in advance to forecast failures. This allows maintenance to be scheduled before actual failures occur, preventing unplanned downtime while maintaining operational simplicity

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The substrate holder performs self-service by incorporating built-in sensors that automatically monitor its own condition parameters (temperature, vibration, electrical characteristics). This self-monitoring capability enables the system to detect degradation without external intervention, improving productivity through early warning while adding minimal complexity

Inventive Principle:

Principle #25Self-service

2Measurement precision

If prediction model is constructed using machine learning, then measurement precision of maintenance timing is improved, but device complexity increases

Engineering Contradiction:

Improvemaintenance timing prediction accuracyVSAvoidprediction system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The prediction system is segmented into distinct functional modules: data collection from sensors, data preprocessing, model training, and prediction execution. This segmentation allows each component to be optimized independently and simplifies the overall system architecture while achieving high prediction accuracy through specialized processing in each segment

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

An intermediary computation device serves as a mediator between the substrate holder sensors and the control system. This intermediary collects raw condition data, performs preliminary processing, and feeds processed information to the prediction model, reducing the complexity burden on both the substrate holder and the central control system while maintaining high measurement precision

Inventive Principle:

Principle #24Intermediary (Mediator)

3Reliability

If condition data is collected from multiple components, then reliability of prediction is improved, but loss of information increases due to data management complexity

Engineering Contradiction:

Improveprediction reliabilityVSAvoiddata management overhead

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

Condition data from multiple components (temperature sensors, vibration sensors, electrical contact sensors) is merged into a unified data structure that captures the overall health state of the substrate holder. This consolidation maintains comprehensive information for reliable multi-component prediction while reducing data management overhead through integrated processing and standardized data formats

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11461647B2Method of constructing prediction model that predicts number of plateable substrates, method of constructing selection model for predicting component that causes failure, and method of predicting number of plateable substrates

Publication Date: 2022.10.04 EBARA CORP

AI summary

A method of the present disclosure includes: plating a plurality of substrates using a substrate holder; determining a total number of substrates that have been plated using the substrate holder until a failure occurs in the substrate holder; determining a first processable number and a second processable number; generating a first data set constituted by a combination of first condition data and the first processable number, the first condition data representing a state of a component of the substrate holder; generating a second data set constituted by a combination of second condition data and the second processable number, the second condition data representing a state of a component of the substrate holder; and optimizing a parameter of a prediction model constituted by a neural network using training data including the first data set and the second data set.