An Efficient and Accurate Approach for Characterizing Non-Linear Capacitance in MESFET/HEMT Devices

This paper presents an accurate and efficient approach of characterizing the non-linear capacitance in MESFET/HEMT devices. This approach utilizes a feed-forward neural network program using the back-propagation method with the Levenberg-Marquardt (LM) algorithm implemented. Using the LM algorithm provides accurate modeling of the non-linear capacitance with minimal training time. This approach is truly independent of device process and technology, and can therefore be applicable to FET devices from SiC, GaN, MOSFETs to GaAs and InP HEMTs. Excellent agreement is also observed between the novel approach and measured results.