(Invited) A Neural Compact Model Based on Transfer Learning for Organic FETs

Abstract

Organic field-effect transistors (organic FETs) exhibit highly non-linear electrical characteristics due to their distinctive transport mechanism manifested by thermally-assisted hopping via localized states randomly distributed in both energy and space domains. This feature impedes the establishment of an analytical compact model. In this context, we present an approach to adopt artificial neural networks (ANN) for the development of a compact model, namely a neural compact model, by exploiting the powerful ability of the ANN to describe a non-linear function accurately. In addition, we evaluate the knowledge transfer method to reduce learning time and improve accuracy even when the data is scarce and costly. By technology computer-aided design simulation, we constructed a dataset of electrical characteristics of organic FETs with Gaussian disorder which is calibrated to the experimental measurement. Subsequently, we constructed and compared neural compact models based on conventional deep learning and transfer learning. We showed that the neural compact model with transfer learning provides an equivalent accuracy at a shorter modeling time. A complex dependence on the gate voltage, drain voltage, and temperature is successfully modeled over a wide range of operation regimes and temperatures.