Deep Learning to Automate Fitting and Parameter Extraction of 2D Transistors

We present a deep learning approach to extract physical parameters (e.g., mobility, Schottky barrier height) of 2D field-effect transistors from <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$I_{\mathbf{d}}-V_{\text{gs}}$</tex> measurements, enabling automated parameter extraction and technology computer-aided design (TCAD) fitting. Our neural network is trained on TCAD-generated <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{I}_{\mathbf{d}}-\boldsymbol{V}_{\text{gs}}$</tex> data and achieves excellent fits with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&gt; 5\times$</tex> less training data than previ-ous efforts. We validate our approach experimen-tally with 51 measured monolayer WS<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> transistors, achieving a median <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}=0.986$</tex> fit. We also show that this approach generalizes to other transistor types, even when fitting 35 parameters simultaneously.