The W7019E PathWave IC-CAP NeuroFET Extraction license includes a dedicated UI Toolkit to drive measurements and run the extraction procedure for the Keysight NeuroFET model for GaAs and MOSFET devices.
The W7019E PathWave IC-CAP NeuroFET Extraction Package includes:
- One-stop package includes advanced measurement control, neural network training, and verification
- Improved DC and RF convergence and S-parameters fitting versus the Root model
- Improved distortion simulation at low amplitude versus the Root model
- Export and use the NeuroFET model in PathWave Advanced Design System (ADS)
NeuroFET is an evolution of the Root model, with the same topology but with tables replaced by Artificial Neural Networks (ANN). The Root model has some limitations due to the table-based representation of device currents and charges.
ANNs are a computational paradigm based on how the brain works - they can smoothly approximate any nonlinear function using a network of highly interconnected nonlinear processing functions called neurons. A sophisticated machine learning (AI) training algorithm identifies the weights between nodes to optimally and smoothly approximate the model nonlinear current and charge functions. Smooth derivatives are critical for accurately representing distortion behavior, particularly at low signal levels, where high-order derivatives must be continuous to achieve the best accuracy.
The NeuroFET model also adds a sophisticated extrapolation methodology beyond the boundaries of the measured data, resulting in robust DC and large-signal simulator convergence behavior. The resultant model is technology-independent and works well for both HEMT and FET devices in silicon and III-V materials. NeuroFET does not cover passive mixer or switching applications.
Advantages of the NeuroFET over table-based Root models:
- Improved DC and RF convergence
- Improved distortion simulation at low amplitude
- More accurate S-parameters versus bias
The IC-CAP NeuroFET Extraction package controls DC and S-parameters measurements necessary to extract the model. The software then performs ANN training through a specific error optimization procedure. The output model file can then be simulated via ADS within the IC-CAP environment or exported to ADS for use with the built-in NeuroFET circuit simulation component.