Machine Learning-Based Device Modeling and Performance Optimization for OTFT

In the huge growth of semiconductor industry, it is noticed that the device simulation is a very sluggish process. It is very promising to use Machine Learning (ML) techniques in device modeling as their combination will create great results in semiconductor industry and reduce the computational time. Organic Thin Film Transistor (OTFT) is a promising alternative to amorphous silicon devices due to its flexibility, low cost, and can be manufactured at reduced temperatures. In traditional TCAD simulation, at once only a single simulation of OTFT for fixed length, width and dielectric thickness can be done, for change in any of the input parameter again simulation has to be done. To avoid this ML is used to predict drain current for simultaneous changes in input parameters. This introduces a machine learning based structure to model OTFT integrated with ML algorithm named Random Forest Regressor (RFR). ML based device model for p-type OTFT takes length, width and thickness of dielectric layer as input parameters and drain current as output parameter. Experimental results has shown that our ML-based model can predict drain current accurately. <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{2}$</tex>-value is found be around 0.997253. ML based performance optimization is a promising alternative to traditional technology computer aided design (TCAD) tools. The highest <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{I}_{\text{ON}}/\mathbf{I}_{\text{OFF}}$</tex> ratio, very high ON current (I<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ON</inf>), very low OFF current (I<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">OFF</inf>) is achieved for OTFT. <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{I}_{\text{ON}}/\mathbf{I}_{\text{OFF}}$</tex> ratio is obtained to be 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11</sup>. The trained RFR models can accelerate the optimization in terms of performance and serves as promising alternative.