NeuroModeler, NeuroADS, and Examples

A neural network model developed by NeuroModeler can be used by Agilent ADS for circuit simulation and design through an ADS plug-in module called NeuroADS. All neural network structures supported by NeuroModeler (such as MLP, RBF, KBNN etc) are accepted by NeuroADS.

Steps to create Agilent ADS model

 A Passive Circuit Example : MMIC Bandpass Filter

This example illustrates ADS simulation of an MMIC bandpass filter using neural network models. A set of component neural models (bend, T-junction, cross-junction, microstrip line, open-stub) were developed by learning from EM data using NeuroModeler. These neural models are then plugged into ADS through NeuroADS plug. The overall filter was simulated in ADS. In conventional circuit design, there is always a trade-off between speed and accuracy: either sacrifice speed with EM simulation or sacrifice accuracy with equivalent circuit simulation. With NeuroADS, the neural network based filter simulation is as fast as equivalent circuit simulation but with accuracy near EM simulation.

Layout of the bandpass filter in ADS

Bandpass Filter Layout in Agilent ADS

Neural model for the cross-junction seen here in ADS

Neural model cross-junction

Schematic of the bandpass filter in ADS where each component is represented by an neural model

Schematic of a Bandpass Filter

Filter frequency response before and after optimization using neural network models in ADS

Filter frequency respone optimization

NeuroModeler/NeuroADS: Build and Then Use Neural Models in RF/Microwave Design

 An Active Circuit Example: 3-Stage Amplifier Design

This example shows statistical design of a 3-stage X band amplifier with physics-based FET models. FET neural models were developed using NeuroModeler by learning from physics-based FET data. The training data was generated using the Khatibzadeh and Trew model in OSA90 software. The trained FET neural model was then plugged into ADS through NeuroADS. The overall 3-stage amplifier circuit was simulated and optimized in ADS utilizing the physics-based FET neural models. This example implies that ADS users can now easily enjoy their proprietary (or new) semiconductor device in their ADS design since neural model can be trained to represent any device behavior. The example also implies that physics-based device models, very time-consuming for repetitive analysis, can now be efficiently used for statistical design and yield optimization through neural model representations.

Schematic of the 3-stage amplifier in ADS where each FET is represented by a neural model

3-Stage Amplifier

Physics-based FET neural model seen here in ADS. This is available through NeuroADS

FET Neural Model in Agilent ADS

DC curves of the FET neural model plotted in ADS

DC curves of the FET Neural model

Monte-Carlo analysis (gain vs frequency) of the amplifier in ADS. Simulation and optimizations were performed with neural FET models

Amplifier optimizations perform with neural FET models

NeuroModeler is a software by Professor Q.J. Zhang and his team, Department of Electronics, Carleton University
OSA90 was by OSA Inc., now Agilent EEsof Division
ADS is a software by Agilent EEsof Division
NeuroADS is a plug-in module to be plugged into ADS
NeuroADS is by Professor Q.J. Zhang and his team, Department of Electronics, Carleton University

For further information on NeuroModeler and NeuroADS, please contact: Professor Q.J. Zhang
Department of Electronics, Carleton University, Ottawa, Ontario, Canada K1S 5B6
email: Web: