Wafer Topography Prediction for Early Post-Polishing Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wafer manufacturing processes face inefficiencies in detecting and addressing nanotopography degradation due to high noise in post-polishing measurements, leading to material losses and reduced production efficiency, with feedback occurring only at late stages and varying device-specific characteristics.
Innovation Solution
A computer device with a processor and memory system analyzes wafer scans using neural network models to predict post-processing conditions, comparing them to thresholds, and adjusts manufacturing devices in real-time to maintain quality control and reduce losses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional post-polishing measurement tools are used, then nanotopography can be measured, but the measurements contain high noise and are unreliable
Solution Approach 1:
The patent introduces an intermediary computational model that maps early-stage shape map data to predicted post-polishing nanotopography. This model acts as a mediator between the grinding process measurements and the final polishing outcome, avoiding direct measurement of high-noise post-polishing surfaces while still providing reliable nanotopography predictions through the established mapping relationship.
2Loss of information
If post-polishing maps are collected at late stages, then complete production data is available, but feedback efficiency is reduced and material losses increase
Solution Approach 1:
The patent performs preliminary prediction of post-polishing nanotopography using shape map data collected during early grinding stages. By establishing a computational mapping model beforehand that can predict final nanotopography from intermediate shape data, the system enables early detection of potential defects before polishing occurs, allowing timely process adjustments and preventing material losses.
3Adaptability or versatility
If individual production lines are monitored separately, then device-specific characteristics are captured, but system complexity increases
Solution Approach 1:
The patent implements local quality adaptation by training separate computational models for each production line and grinder, capturing device-specific characteristics. Each model learns the unique mapping relationship between shape maps and nanotopography for its specific equipment, ensuring high accuracy for that particular line while maintaining a standardized overall system architecture that manages complexity through modular, equipment-specific model instances.
Data Source
AI summary
A computer device is programmed to store a model for converting shape maps to simulate a portion of an assembly line, receive scan data of a first inspection of a product being assembled, generate a shape map from the scan data of the first inspection, execute the model using the shape map as an input to generate a final shape map of the product, compare the final shape map to one or more thresholds, determine if the final shape map exceeds at least one of the one or more thresholds, and if the determination is that the final shape map exceeds at least one of the one or more thresholds, cause the first device to be adjusted.