Neural Network Circuit Performance Distribution Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional Monte Carlo simulations face scalability issues due to increasing process variables in semiconductor manufacturing, leading to inefficient statistical fluctuation analysis of circuits under process variations.
Innovation Solution
A method using a neural network to learn the probability distribution of circuit performance metrics, trained with random noise and performance metrics data, to generate synthetic samples that follow the distribution of performance metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Monte Carlo simulations are used to analyze circuit performance under process variations, then statistical fluctuation analysis can be performed, but the computational complexity and time required increase exponentially as the number of process variables increases
Solution Approach 1:
The patent uses a neural network to learn and copy the complex mapping relationship between process variables and circuit performance metrics from a limited set of Monte Carlo simulation data. Once trained, the neural network generates synthetic performance data that replicates the statistical characteristics of full Monte Carlo simulations, enabling rapid statistical fluctuation analysis without re-running expensive simulations.
Solution Approach 2:
The patent performs preliminary Monte Carlo simulations to generate training data before the actual statistical analysis is needed. The neural network is trained in advance on this data to capture the relationship between process variations and performance metrics. When statistical fluctuation analysis is required, the pre-trained network can rapidly generate results without needing to perform new full-scale Monte Carlo simulations.
2Manufacturing precision
If the number of process variables increases to account for manufacturing complexity and tighter tolerances, then manufacturing precision can be maintained, but the scalability of traditional simulation methods deteriorates
Solution Approach 1:
The patent replaces the mechanical computation process of traditional Monte Carlo simulations with a neural network-based system. Instead of repeatedly solving circuit equations for each simulation sample, the neural network learns the underlying patterns from a limited number of simulations and can rapidly predict performance metrics for any combination of process variables, scaling efficiently to handle increased numbers of process variables.
Solution Approach 2:
The patent changes the approach from direct simulation parameter variation to neural network parameter optimization. The neural network is trained by adjusting its internal parameters (weights and biases) to match the input-output relationships observed in Monte Carlo simulations. Once trained, the network can evaluate any process variable combination without additional simulation overhead, enabling efficient analysis of manufacturing precision across multiple process variables.
Data Source
AI summary
Disclosed is a probability distribution learning method for efficiently performing a statistical fluctuation analysis of a circuit according to process variations. The data generation method includes generating data representing performance metrics of the circuit, training a neural network by applying random noise as training input data and the performance metrics as training output data to the neural network, and generating synthetic samples that follow a distribution of the performance metrics by applying the random noise to the trained neural network.