Artificial Neural Network Modelling of Sub-Millimeter-Wave Silicon IMPATT Sources

This study investigates artificial neural networks (ANNs) for predicting the millimeter-wave (mm-wave) and sub-millimeter-wave (sub-mm-wave) characteristics of Double-Drift Region (DDR) Si IMPATT diodes. The ANN models predict key parameters, including DC, large-signal (L-S) performance, and avalanche noise across 94–500 GHz. Training data from self-consistent quantum drift-diffusion (SCQDD) simulations enable the models to capture structural, doping, and biasing effects accurately. The ANN approach significantly reduces computational time, predicting device characteristics in just 4.4% to 15% of the time required for SCQDD simulations while maintaining high accuracy. The mean square error (MSE) for breakdown voltage and power output remains in the order of 10⁻<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> Unit<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, demonstrating strong predictive performance. Validation against experimental data confirms excellent agreement in power output, efficiency, and noise characteristics. These results highlight the potential of machine learning to replace conventional time-intensive simulations, offering a powerful tool for the rapid design and optimization of high-frequency semiconductor devices.