Characterization of Resistance and Inductance of PIN Diode at mmWave Frequency Using 7-Layer Deep Neural Network

This paper presents a novel technique for extracting the resistance (R) and inductance (L) of an ultra-low capacitance PIN diode, a critical component in developing 5G mmWave reconfigurable circuits and antennas. In the proposed method, a PIN diode is mounted on a microstrip transmission line and biased by a DC biasing network and its S-parameters are measured. The measured S-parameters are calibrated by the thru-reflect-line calibration to reduce undesirable effects from the measurement fixture. Subsequently, the post-calibration transmission coefficient (
$S_{21}$
) is fed into a deep neural network (DNN) which has been trained with simulated
$S_{21}$
data obtained from a full-wave 3D electromagnetic simulation software. The output of the DNN provides frequency dependent R and L values at the frequency range from 27 GHz to 30 GHz. The results agree well the presumption that R decreases with the increase in bias current and frequency, while L increases as the frequency increases. This result was obtained with a MA4AGP907 p-i-n diode biased with three different forward currents i.e. 1 mA, 5 mA, and 7 mA.