Circuit Performance Variation Prediction via Adjoint Sensitivity Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting circuit performance variations due to device mismatch in integrated circuits, such as Monte Carlo and worst case analyses, require a large number of repetitive circuit simulations, making them inefficient.
Innovation Solution
The approach involves performing circuit simulation using a harmonic balance method to generate sensitivity information and compute current/charge deviations, allowing for the determination of steady-state mismatch effects without the need for repeated simulations by employing adjoint sensitivity analysis and statistical models for circuit element parameter variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo analysis is used to estimate statistical distribution of circuit performance, then prediction accuracy is improved, but computational time and simulation quantity increase significantly
Solution Approach 1:
The patent introduces sensitivity information as an intermediary that bridges device parameter variations and circuit performance. By computing sensitivity information through adjoint sensitivity analysis, the method avoids direct repetitive simulations while capturing the relationship between device mismatch and performance variations, thus achieving accurate predictions with reduced computational cost
Solution Approach 2:
The patent replaces the mechanical simulation-based approach (Monte Carlo and worst case analyses requiring hundreds to thousands of simulations) with a mathematical analytical approach using adjoint sensitivity analysis and statistical models. This substitution eliminates the need for repetitive circuit simulations while maintaining prediction accuracy
2Measurement precision
If worst case analysis is used to estimate circuit performance variations, then prediction accuracy is improved, but the number of simulations required increases largely dependent on the number of device parameters
Solution Approach 1:
The patent uses sensitivity information as an intermediary that encapsulates the impact of device parameter variations without requiring exhaustive simulation. The adjoint sensitivity analysis computes this intermediary efficiently, allowing the method to handle complex circuits with many device parameters without increasing the simulation burden
Solution Approach 2:
The patent transforms the problem from simulating multiple device parameter combinations to computing sensitivity information once and then using statistical models to predict performance variations. This parameter transformation reduces the simulation requirement from hundreds to a single simulation, significantly reducing device complexity
Data Source
AI summary
Aspects of the invention relate to techniques for predicting circuit performance variations due to device mismatch. Circuit simulation is performed to generate circuit simulation results based on a circuit description and information of circuit element parameters. Based on the simulation results, sensitivity information for the circuit design and current/charge deviations caused by individual circuit element parameter variations may be computed. Based on the sensitivity information and the current/charge deviations, steady-state mismatch effect information is determined. The determination may comprise first computing output parameter deviations caused by the individual variations of the circuit element parameters and then computing a total output parameter deviation based on the output parameter deviations.