GPU Substrate Routing for Faster Throughput Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches for substrate routing and throughput modeling in semiconductor processing are time-consuming, heuristic, and fail to accurately account for events and failures in cluster tools, leading to inefficiencies and inaccuracies in processing sequences.
Innovation Solution
The use of one or more Graphics Processing Units (GPUs) to generate processing models for semiconductor substrates, allowing for concurrent processing of parallel inputs to produce parallel outputs that optimize substrate routing and predict throughput, enabling real-time adjustments to processing sequences in response to failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional heuristic approaches are used for substrate routing and throughput modeling, then processing sequences can be generated, but the processing duration is excessively long and accuracy is insufficient
Solution Approach 1:
The patent replaces conventional CPU-based sequential processing with GPU-based parallel processing architecture. The GPU's massively parallel compute units execute routing and throughput modeling calculations simultaneously, achieving both higher accuracy through refined algorithms and reduced processing time through parallel computation of multiple substrate processing scenarios
Solution Approach 2:
The patent segments the substrate batch into individual substrate processing tasks that can be processed in parallel. Each substrate's routing and throughput modeling is calculated independently on separate GPU compute units, allowing simultaneous computation of multiple processing sequences and enabling accurate comparison of different routing options without sequential bottlenecks
2Productivity
If conventional sequential processing is used for substrate routing calculations, then routing sequences can be determined, but productivity is reduced due to long wait times
Solution Approach 1:
The patent substitutes traditional CPU sequential execution with GPU parallel architecture, where thousands of cores simultaneously compute routing paths and throughput metrics for multiple substrates. This parallelization eliminates sequential wait times and dramatically increases substrate processing throughput by calculating all routing options concurrently rather than one at a time
Solution Approach 2:
The patent performs preliminary parallel computation of all possible routing paths and throughput scenarios before actual substrate processing begins. By pre-calculating optimal routes and predicting throughput for all substrates simultaneously using GPU acceleration, the system eliminates idle wait times during production and enables immediate execution of optimized sequences
3Measurement precision
If detailed processing models are generated for each substrate, then routing accuracy improves, but device complexity increases
Solution Approach 1:
The patent uses GPU parallel architecture to manage the complexity of detailed processing models for each substrate. The GPU's massive parallel compute capacity handles the computational burden of calculating precise routing paths, throughput metrics, and constraint satisfaction for multiple substrates simultaneously, allowing high accuracy without proportionally increasing system complexity
Solution Approach 2:
The patent implements a universal GPU-based processing framework that handles all substrate routing and throughput modeling tasks through a single parallel architecture. This multi-functional system can process different substrate types, routing scenarios, and constraint conditions using the same hardware platform, reducing overall device complexity compared to dedicated systems for each function
Data Source
AI summary
A method includes receiving a matrix including start times associated with substrate operations in a substrate processing system. The method further includes generating, by a first graphics processing unit (GPU) of one or more GPUs, a plurality of matrices based on the matrix. The method further includes concurrently processing, by a plurality of cores of the one or more GPUs, the plurality of matrices to generate parallel outputs. A schedule for processing substrates in the substrate processing system is to be generated based on the parallel outputs.