Distributed Prediction Network for Confidential Collaborative Diagnosis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The complexity of semiconductor production processes makes it difficult to accurately diagnose issues and perform root-cause analysis, especially when multiple parties with confidential information are involved.

Innovation Solution

A distributed prediction network using peer-to-peer probabilistic models allows multiple parties to cooperate while maintaining privacy, by sharing intermediate inference results that do not reveal confidential data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple parties share their expertise and data to improve prediction accuracy, then the diagnostic accuracy improves, but the confidentiality of each party's data and expertise is compromised

Engineering Contradiction:
Improvediagnostic accuracyVSAvoidconfidentiality
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent introduces an intermediary mechanism where parties share intermediate inference results (partial CPTs) rather than raw data or complete models. These intermediates act as mediators that convey diagnostic information without revealing confidential underlying data, allowing collaborative diagnosis while preserving privacy boundaries.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the diagnostic information into separate components: each party maintains their own probabilistic model and data separately, sharing only specific intermediate results (partial conditional probability tables) that are necessary for the overall diagnosis. This segmentation allows collaborative accuracy improvement without full data exposure.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If parties share complete probabilistic models and data, then the root-cause analysis accuracy improves, but the complexity of managing and securing multiple confidential datasets increases

Engineering Contradiction:
Improveroot-cause analysis accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the necessary intermediate inference results (partial CPTs) from each party's complete probabilistic model for sharing. Instead of managing and securing entire datasets and models, parties exchange only the specific intermediate components needed for collaborative root-cause analysis, reducing system complexity while maintaining diagnostic accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

3Reliability

If parties maintain separate confidential models, then data security is improved, but the ability to perform accurate collaborative diagnosis is reduced

Engineering Contradiction:
Improvedata securityVSAvoidcollaborative diagnosis accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent uses intermediate inference results as mediators that enable collaborative diagnosis between parties with separate confidential models. These intermediates (partial CPTs) carry the necessary diagnostic information for accurate collaborative analysis while preserving the security boundaries of each party's complete model and data.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4538938A1Confidentiality-preserving collaborative model for deriving information on a production system
Publication Date: 2025.04.16 ASML NETHERLANDS BV
  • EP4538938A1 patent drawingFigure 1
  • EP4538938A1 patent drawingFigure 2
  • EP4538938A1 patent drawingFigure 3

AI summary

A method is presented for using a distributed network to derive information characterising a production system described by variables. The network comprises two probabilistic models implementing a graph comprising nodes associated with a corresponding one of the variables, and directed edges connecting respective pairs of the nodes. Each graph comprises an interface node for which the corresponding associated variable is common and normal nodes to which edge(s) are directed that are associated with a respective conditional probability table (CPT) specifying a probability of the variable being in a set of states based on the variables associated with the corresponding one or more nodes from which the one or more edges are directed. The interface node is associated with a partial CPT of the common variable. The method comprises generating, using the at least two models, conditioned data specifying a probability distribution of the state of the common variable.