ML-Based Circuit Pattern Encoding for Large Layout Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current electronic design automation techniques are impractical for processing large electronic circuits with billions of unique patterns due to memory and computational limitations, limiting the ability to perform effective pattern clustering and machine learning model training.
Innovation Solution
A distributed system using a machine learning-based encoder, such as a neural network autoencoder, for encoding circuit patterns to reduce data size, combined with hierarchical clustering, allows for parallel processing across multiple computer processors, enabling efficient classification and training of machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional pattern collection and clustering techniques are used on large electronic circuits, then pattern classification can be performed, but the process becomes impractical due to insufficient memory and computational power on single computing systems
Solution Approach 1:
The patent divides the large circuit layout into multiple smaller layout fragments that can be independently processed. Pattern extraction is performed on each fragment separately, and results are aggregated. This segmentation allows the system to handle billions of patterns by processing them in manageable chunks across multiple computing systems, resolving the contradiction between processing quantity and computational complexity.
Solution Approach 2:
The patent introduces a distributed computing dimension by deploying pattern extraction and clustering across multiple computing systems rather than relying on a single system. This dimensional shift from single-system to multi-system architecture enables the processing of large-scale circuit patterns by distributing the computational burden, thereby handling larger pattern quantities without proportionally increasing individual system complexity.
2Measurement precision
If all unique patterns from large circuits are collected and clustered, then comprehensive pattern classification is achieved, but the data size and processing time become prohibitive
Solution Approach 1:
The patent performs preliminary pattern extraction on layout fragments before full clustering is attempted. By pre-processing the layout to identify and extract patterns into a structured format, the system prepares data in advance for more efficient clustering operations. This preliminary action reduces the time required for the actual clustering process while maintaining comprehensive pattern classification coverage.
Solution Approach 2:
The patent extracts and processes patterns from layout fragments representing portions of the complete circuit rather than requiring the entire circuit to be loaded and processed simultaneously. This partial action approach allows the system to achieve representative pattern classification without the prohibitive time cost of processing all possible patterns in their entirety, balancing accuracy with processing efficiency.
3Reliability
If machine learning models are trained on complete circuit patterns, then model accuracy improves, but memory requirements exceed available resources on single computing systems
Solution Approach 1:
The patent segments the training data into patterns extracted from multiple layout fragments that can be loaded into memory incrementally. Instead of requiring all training patterns to reside in memory simultaneously, the system processes fragments in sequences or batches, enabling machine learning model training on large circuit datasets with limited memory capacity while maintaining model accuracy through comprehensive fragment coverage.
Data Source
AI summary
A system performs distributed or parallel pattern extraction and clustering for pattern classification of large layouts of electronic circuits. The system identifies circuit patterns with a layout representation. The system encodes the circuit patterns using a neural network based autoencoder to generate encoded circuit patterns that can be stored efficiently. The system clusters the encoded circuit patterns into an arbitrary number of clusters based upon a high degree of similarity. The clusters of circuit patterns may be used for training and evaluation of machine learning based models.