Semiconductor Tool Failure Prediction Using Feature-Quantity Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Semiconductor manufacturing apparatuses face abrupt failures due to motor and sensor degradation, leading to operational disruptions and substrate scrapage, as existing technologies lack effective predictive maintenance methods to prevent such failures.
Innovation Solution
A semiconductor manufacturing apparatus equipped with sensors that detect physical quantities, a calculation circuit to derive feature quantities, and a failure prediction circuit that compares these with model data to predict potential failures, allowing for timely intervention and preventing new substrate processing when a failure is imminent.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the motor and sensor are used continuously in the semiconductor manufacturing apparatus, then the productivity is maintained, but the reliability deteriorates due to aging degradation leading to abrupt failures
Solution Approach 1:
The system performs preliminary failure prediction by continuously monitoring physical quantities (vibration, temperature, current) and comparing them against pre-established model data representing normal and abnormal states. This allows the apparatus to detect degradation trends before actual failure occurs, enabling proactive maintenance scheduling that prevents abrupt failures while maintaining continuous operation
Solution Approach 2:
The system implements continuous feedback monitoring where sensors detect physical quantities during operation, the calculation circuit processes this data to determine device state, and the control circuit adjusts operations based on predicted failure risk. This closed-loop feedback enables real-time reliability management without interrupting productivity
2Device complexity
If the apparatus operates without failure prediction, then the device complexity is low, but the loss of time increases due to abrupt failures causing substrate scrapage and downtime
Solution Approach 1:
The system performs preliminary failure prediction by continuously monitoring physical quantities (vibration, temperature, current) and comparing them against pre-established model data representing normal and abnormal states. This allows the apparatus to detect degradation trends before actual failure occurs, enabling proactive maintenance scheduling that prevents abrupt failures while maintaining continuous operation
Solution Approach 2:
The patent replaces complex mechanical failure prediction methods with electronic sensing and computational analysis. Sensors detect physical quantities, which are then processed by calculation circuits using predetermined models to predict failure. This substitution maintains relative system simplicity while dramatically improving the ability to predict failures and minimize downtime
3Reliability
If the motor load factor is monitored continuously, then the reliability is improved through early detection, but the use of energy increases due to continuous sensing and processing
Solution Approach 1:
The system applies partial monitoring by focusing on specific critical physical quantities (vibration, temperature, current) rather than comprehensive continuous monitoring of all motor parameters. The calculation circuit processes data at intervals sufficient to detect degradation trends while avoiding excessive energy consumption from overly frequent sampling and processing
Data Source
AI summary
A semiconductor manufacturing apparatus including: a first device; a first calculation circuit that calculates one or more feature quantities of the first device from detected physical quantities; and a failure prediction circuit that determines a portion of model data with a minimum deviation between the measured feature quantities vector comprising the measured one or more feature quantities and a feature quantities vector comprising one or more feature quantities at each time in the plurality of pieces of model data, and calculates a predicted time until failure from a difference between the failure time point in the determined piece of model data and a point in time in the determined piece of model data at which the deviation between the measured feature quantities vector and the feature quantities vector at each time of the plurality of portions of model data is the minimum; and stops the receiving of a new substrate to prevent an introduction of defects on the new substrate.