ML-Based Substrate State Prediction in CMP

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

It is challenging to accurately monitor and predict the state of a substrate during chemical mechanical polishing due to the complex interactions between the substrate and the operating states of the polishing apparatus, making it difficult to maintain production quality and yield.

Innovation Solution

An information processing apparatus that acquires polishing conditions, including top-ring, polishing-table, and polishing-fluid-supply-nozzle state information, and uses a machine-learning model to predict substrate state information, thereby enabling accurate prediction of substrate states during or after polishing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If sensors are directly attached to each substrate to detect substrate state, then measurement precision is improved, but device complexity and cost increase significantly

Engineering Contradiction:

Improvesubstrate state detection accuracyVSAvoidsensor attachment complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary system consisting of sensors on the polishing table and a machine learning model that acts as a mediator between the polishing process and substrate state measurement. Instead of directly measuring substrate stress, the system measures polishing table states and uses the learning model to infer substrate states, thereby avoiding direct sensor attachment to substrates while maintaining measurement capability

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the mechanical approach of directly attaching sensors to substrates with an information processing approach using machine learning. The system substitutes physical measurement on the substrate with computational inference based on polishing process parameters, eliminating the need for complex sensor attachment mechanisms

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If multiple sensors are attached to monitor complex interactions between top ring, polishing table, and polishing fluid supply nozzle, then measurement precision is improved, but device complexity increases

Engineering Contradiction:

Improvepolishing process monitoring accuracyVSAvoidsensor system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential polishing process parameters (top ring state, polishing table state, polishing fluid supply nozzle state) that have the most significant impact on substrate stress, rather than attempting to measure all possible interactions. This selective extraction reduces the number of sensors needed while maintaining monitoring effectiveness

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The machine learning model serves as a universal processing unit that handles the complex analysis of multiple polishing parameters simultaneously. Instead of requiring separate measurement systems for each parameter interaction, the single learning model processes all inputs to predict substrate state, reducing overall system complexity

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Manufacturing precision

If comprehensive polishing conditions are collected and analyzed using machine learning, then manufacturing precision is improved, but loss of time for data processing increases

Engineering Contradiction:

Improvesubstrate state prediction accuracyVSAvoiddata processing time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by training the machine learning model in advance using historical polishing data. Once trained, the model can quickly predict substrate states during actual polishing operations without requiring time-consuming real-time analysis of all polishing parameters. The heavy computational work is done beforehand, enabling fast online predictions

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250050461A1Information processing apparatus, inference apparatus, machine-learning apparatus, information processing method, inference method, and machine-learning method

Publication Date: 2025.02.13 EBARA CORP

AI summary

An information processing apparatus (5) includes: an information acquisition section (500) configured to acquire polishing conditions including top-ring state information, polishing-table state information, and polishing-fluid-supply-nozzle state information in the chemical mechanical polishing in chemical mechanical polishing of a substrate performed by a substrate processing apparatus including a polishing table configured to rotatably support a polishing pad, a top ring configured to press the substrate against the polishing pad, and a polishing-fluid supply nozzle configured to supply a polishing fluid onto the polishing pad; and a state prediction section (501) configured to predict substrate state information for the substrate on which the chemical mechanical polishing is performed under the polishing conditions by inputting the polishing conditions acquired by the information acquisition section (500) to a learning model (10A) having been generated by machine learning that causes the learning model (10A) to learn a correlation between the polishing conditions and the substrate state information indicating a state of the substrate on which the chemical mechanical polishing is performed under the polishing conditions.