Comparative analysis of parameter extraction techniques for AlGaN/GaN HEMT on silicon/sapphire substrate

• Published in: Microelectronics and reliability Vol. 78; pp. 389 - 395
• Main Authors: Majumdar, Shubhankar, Bag, Ankush, Biswas, Dhrubes
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-11-2017
• Subjects: Artificial neural network (ANN) model; GaN HEMT on silicon/sapphire; RF characterization; Small signal model (SSM); Small signal model (SSM); Artificial neural network (ANN) model; RF characterization; GaN HEMT on silicon/sapphire
• ISSN: 0026-2714; 1872-941X
• DOI: 10.1016/j.microrel.2017.08.016

We report a comparative study of artificial neural network (ANN) model and small signal model (SSM) based on extracted parameters. ANN model training is done using Levenberg-Marquardt back propagation algorithm, whereas SSM is formed by extracting circuit parameters from measured S-parameters of GaN HEMT on Silicon and Sapphire. It has been found that, for the GaN HEMT parameter extraction, it takes 85 hidden layer neurons to produce the output with higher accuracy. The optimized test and training error/performance are found to be 1.12×10−8/0.97 and 1×10−8/0.99, respectively.
[Display omitted]
•Fabrication and RF characterization of the GaN HEMT on silicon (Si) and sapphire (Al2O3).•Implementation of ANN model for determining the small signal model parameter.•ANN model utilizes variable weights & same activation function for extraction of SSM parameters of GaN HEMT.