Transistor Statistical Model Using Neural Network Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transistor modeling methods face challenges in efficiently constructing models for emerging devices due to complex physical mechanisms and long development cycles, with physics-based models being slow and unreliable, especially in capturing the impact of process parameter fluctuations on device performance.
Innovation Solution
A method is introduced to establish a transistor statistical model using an artificial neural network system, which generates a nominal model based on data sets including gate-source voltage, drain-source voltage, and drain-source current data, screens neurons to obtain final variational neurons, and calculates the distribution of weights and threshold voltages to create a reliable and efficient model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physics-based models are used for transistor modeling, then model reliability is improved, but simulation speed deteriorates
Solution Approach 1:
The patent segments the transistor model into two parts: a physics-based nominal model for accurate baseline characterization and data-driven variational components for capturing process fluctuations. This segmentation allows the physics-based part to ensure reliability while the data-driven part accelerates simulation by pre-computing statistical variations.
Solution Approach 2:
The patent introduces an intermediary neural network model that bridges physics-based models and statistical variations. The neural network is trained on physics-based model data and serves as a faster surrogate, capturing both nominal behavior and process variations without directly invoking slow physics-based simulations during circuit analysis.
2Measurement precision
If physics-based models are used for transistor modeling, then model accuracy is improved, but development cycle length deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing process variation effects during the neural network training phase using physics-based models. Once trained, the neural network rapidly predicts statistical variations without requiring repeated physics-based simulations, significantly reducing development cycles for new circuit designs.
Solution Approach 2:
The patent creates a copy of the physics-based model's behavior through the neural network surrogate model. The neural network learns to replicate the complex physics-based model outputs including process variations, providing a fast copy that maintains accuracy while eliminating the computational burden of the original physics-based model.
3Reliability
If process parameter fluctuations are considered in device models, then manufacturing reliability is improved, but model complexity deteriorates
Solution Approach 1:
The patent changes the approach from modeling individual process parameters to directly modeling their combined effect on device performance. Instead of tracking multiple process parameters separately, the neural network learns the net effect of all process variations on current-voltage characteristics, simplifying the model while maintaining reliability.
Solution Approach 2:
The patent extracts the essential statistical variation information from complex process parameter fluctuations. By training the neural network on physics-based model data that includes process variations, it extracts and stores only the relevant statistical effects, eliminating the need to explicitly model each individual process parameter.
4Productivity
If neural network models are used for transistor modeling, then simulation speed is improved, but model reliability deteriorates
Solution Approach 1:
The patent applies local quality by using physics-based models for the nominal (typical) case where high accuracy is critical, and neural networks for capturing statistical variations where speed is more important. This localized approach ensures reliability is maintained where needed while achieving speed improvements in statistical analysis.
Solution Approach 2:
The patent incorporates feedback by training the neural network using physics-based model data as ground truth. The physics-based model provides feedback to validate and correct the neural network's predictions, ensuring that the fast data-driven model maintains reliability by anchoring it to accurate physics-based simulations during the training phase.
Data Source
AI summary
A method of establishing a transistor statistical model based on an artificial neural network system comprising receiving a first data set and generating a nominal model of a baseline transistor by the artificial neural network system based on the first data set; screening neurons in the artificial neural network system based on the first data set and the nominal model to obtain final variational neurons; obtaining distribution of weights of the final variational neurons and distribution of threshold voltages based on variation of the nominal model with respect to weights of the final variational neurons, variation of the nominal model with respect to the threshold voltages, distribution of the drain-source current and distribution of the gate-source voltage in the first data set; and establishing the transistor statistical model based on the nominal model, the distribution of weights of the final variational neurons and the distribution of the threshold voltages.