Analog Circuit Generator Using Semi-Supervised Neural Co-Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual circuit design is inefficient and time-consuming due to complex design parameters and process variations, while existing automated methods are either non-reusable or require extensive labeled training sets, leading to high computational costs and low accuracy.
Innovation Solution
A co-learning-based neural network approach that leverages datasets from simpler circuits to generate a fully-automated analog circuit generator (AnGeL), using semi-supervised learning to determine circuit behavior and reduce the need for labeled training samples, enabling fast and accurate design of complex circuits by dividing them into sub-circuits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If model-based approaches are used for automated circuit sizing, then reusability is improved, but accuracy deteriorates due to nonlinearity and large number of design parameters
Solution Approach 1:
The patent segments the circuit design problem into topology selection and element sizing sub-problems. The topology decision is made using a neural network that takes circuit specifications as input and outputs a selected topology, while element sizing is handled separately through optimization. This segmentation allows each sub-problem to be addressed with appropriate methods, improving both reusability and accuracy.
Solution Approach 2:
The patent introduces an intermediary neural network model that mediates between circuit specifications and topology selection. This neural network is trained on simulated data and serves as a fast, accurate intermediary that captures nonlinear relationships without requiring direct SPICE simulation for each design point, thus improving accuracy while maintaining reusability.
2Measurement precision
If simulation-based algorithms are used for automated circuit sizing, then accuracy is improved, but computational time increases and reusability deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-training a neural network model on circuit specifications and topologies using simulated data. This pre-computed model can then be reused for new circuit designs without requiring time-consuming SPICE simulations during the design process, significantly reducing computational time while maintaining accuracy through the pre-learned relationships.
Solution Approach 2:
The patent creates a computational copy of the circuit behavior through the neural network model. Instead of directly simulating circuits using SPICE for each design point, the neural network provides a fast copy that replicates circuit characteristics, enabling rapid evaluation and optimization without the time cost of repeated simulations.
3Measurement precision
If a large labeled training set is used to maintain high accuracy, then model accuracy is improved, but data collection time and computational cost increase
Solution Approach 1:
The patent applies self-service by using the neural network to generate its own training data through automated circuit simulation. The system simulates circuits, extracts features, and uses this self-generated data to train the neural network, eliminating the need for manual data collection and reducing the time and computational cost associated with assembling large labeled datasets.
Solution Approach 2:
The patent replaces the mechanical process of manual data collection and SPICE simulation with an automated neural network-based system. The neural network automatically learns circuit characteristics from simulated data, substituting the time-consuming manual processes with automated machine learning, thus reducing data collection time and computational cost while maintaining accuracy.
Data Source
AI summary
Machine Learning has shown promising results in predicting the behavior of analog circuits. However, in order to completely cover the design space for today's complicated circuits, supervised machine learning requires a large number of labeled samples which is time-consuming to provide. Furthermore, a separate dataset must be collected for each circuit topology making all other previously gathered datasets useless. In this disclosure, neural networks are used to determine the behavior of complicated topologies by combining simple ones. By generating a database with labeled and unlabeled data, the time for providing the training set is significantly reduced compared to the conventional approaches. Using this database, a fully-automated analog circuit generator framework is presented. The analog circuit generator performs all the schematic circuit design steps from deciding the circuit topology to determining the circuit parameters i.e. sizing.