Plasma Reactor Control via ML Latent State Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The processing of wafers for micro- and nano-scale devices is challenging due to high variability in plasma reactors, leading to inconsistent product quality, waste, and inefficiencies in existing control strategies that rely on time-consuming and costly metrology analyses.
Innovation Solution
A computer-implemented method using a trained machine learning model that receives real-time sensor data from multiple sensors to generate a latent representation of the plasma state, allowing for immediate adjustments to control parameters in the plasma reactor, thereby reducing process variability and improving product uniformity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If non-invasive measurement techniques are used to monitor plasma state, then plasma contamination is avoided, but specific information such as plasma density cannot be obtained
Solution Approach 1:
The patent introduces optical emission spectroscopy as an intermediary measurement technique that allows indirect observation of plasma properties without physical contact. By analyzing the spectral characteristics of light emitted by the plasma, the system can derive plasma density and other parameters while maintaining the non-invasive advantage of not contaminating the plasma with physical probes
Solution Approach 2:
The patent replaces mechanical/invasive probe-based measurement systems with optical measurement systems. Instead of using physical probes that interact directly with the plasma, the system uses optical emission spectroscopy to measure plasma properties remotely through light interaction, thereby eliminating contamination while preserving measurement capability
2Manufacturing precision
If detailed metrology analysis is performed on each wafer batch, then process quality is improved, but production time increases significantly
Solution Approach 1:
The patent performs preliminary real-time monitoring and analysis during the wafer fabrication process itself, rather than waiting until after batch completion. By continuously monitoring plasma parameters and wafer properties during processing, the system can identify quality issues early and make immediate adjustments, eliminating the need for time-consuming post-processing metrology analysis
Solution Approach 2:
The patent implements a feedback control system where real-time measurements of plasma state and wafer properties are continuously fed back to the control system. This enables dynamic adjustment of processing parameters during fabrication to maintain optimal quality, replacing the static post-batch analysis approach with an active continuous improvement loop that reduces both time and waste
3Manufacturing precision
If real-time control of plasma reactor parameters is implemented, then product consistency is improved, but system complexity increases
Solution Approach 1:
The patent employs a machine learning model that serves multiple functions simultaneously: it processes data from multiple sensors, identifies plasma state, predicts wafer properties, and generates control recommendations. This multi-functional approach consolidates what would otherwise require separate specialized systems into a single unified platform, managing complexity while maintaining comprehensive real-time control capability
Solution Approach 2:
The patent transforms complex multi-parameter plasma control into a simplified process by using the machine learning model to identify the critical subset of parameters that most influence product quality. The system dynamically adjusts these key parameters in real-time based on learned relationships, rather than attempting to control all possible parameters equally, thereby achieving effective control with managed complexity
Data Source
AI summary
Broadly speaking, the present techniques provide a method and system for controlling a wafer production process in real-time using a trained machine learning, ML, model. Advantageously, the ML model uses multiple sensed parameters to determine a state of a plasma used in the wafer production process, and this can be used to adjust at least one control parameter of a plasma reactor used in the wafer production process to reduce process variability.