Unified Neural Network for AMS Circuit Characterization Across PVT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network implementations for analog circuit modeling require circuit-specific architectures, leading to increased design time, resource consumption, and limited reusability, especially when dealing with multiple circuits or evolving design requirements, and lack generalization across different circuit topologies.
Innovation Solution
A Unified Deep Learning Neural Network (U-DNN) with a standardized architecture that models diverse analog and mixed-signal circuits across various PVT conditions without circuit-specific structural modifications, utilizing Bayesian optimization for hyperparameter tuning and leveraging a fixed architecture with five hidden layers to achieve high accuracy and generalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If circuit-specific neural network architectures are used for each analog circuit, then modeling accuracy for that specific circuit is improved, but design time and computational resources increase linearly with the number of circuits
Solution Approach 1:
The patent implements a unified neural network architecture that can model multiple different analog circuit topologies (operational amplifiers, voltage-controlled oscillators, charge pumps, phase-locked loops) using the same structure. This universal architecture eliminates the need to design separate neural networks for each circuit type, thereby reducing design time while maintaining modeling accuracy through the network's ability to adapt to different circuit configurations.
2Measurement precision
If circuit-specific neural network architectures are developed for each circuit, then precise characterization of that circuit is achieved, but resource consumption and maintenance effort increase
Solution Approach 1:
The patent merges the functionality of multiple circuit-specific neural networks into a single unified architecture. By combining the modeling capabilities for different circuit topologies into one network, the system reduces computational overhead, memory requirements, and maintenance effort while preserving the ability to accurately characterize each circuit type through the unified structure.
3Reliability
If traditional parameter variability analysis methods are used, then design expertise can be applied, but the difficulty and time required increase exponentially with circuit complexity
Solution Approach 1:
The patent replaces traditional manual parameter variability analysis methods with a neural network-based automated system. The unified neural network automatically handles complex PVT variability analysis for different circuit topologies, substituting the need for manual expert analysis. This automation maintains reliability analysis capability while reducing the exponential increase in complexity that would otherwise occur with more complex circuits.
Data Source
AI summary
The present invention discloses a Unified Deep-Learning Neural Network (U-DNN) Architecture capable of modeling a wide range of AMS circuits effectively and shows the ways it can be used for on demand circuit specification, design optimization and self-adaptation. Applying the U-DNN architecture for analog circuit characterization demonstrated consistent performance across all the DUTs, with an R2Score exceeding 0.95 with {μE,σE}<1% for the test split data, validating its accuracy. The U-DNN architecture exhibited an average MaPE of less than 1% over unseen test cases, showcasing its strong generalization capabilities. Remarkably, the knowledge encapsulated in the U-DNN architecture, gained from modeling diverse AMS circuits in CMOS 180 nm and 65 nm technologies, translated into accurate modeling of AMS circuits in CMOS 28 nm. The U-DNN architecture serves as a versatile platform for modeling various AMS circuits for different applications, reducing the extensive design exploration time needed to select appropriate NN structures.