Simulating Organic Thin Film Transistors Using Multilayer Perceptron Regression Models to Enable Circuit Design

There is increasing interest in using specialized circuits based on emerging technologies to develop a new generation of smart devices. The process and device variability exhibited by such materials, however, can present substantial challenges for designing circuits. Three models are considered here: a physical compact model, an empirical look-up table, and an empirical surrogate model based on a multilayer perceptron (MLP) regression. Each one is fit to measurements of discrete organic thin film transistors in the low voltage regime. It is shown that the models provide consistent results when designing artificial neuron circuits, but that the MLP regression provides the highest accuracy and is much simpler to fit compared to the compact model. The targeted technology exhibits non-ideal behavior such as variable threshold voltage and hysteresis. Using the MLP regression model, the effect of such variability on the performance of an artificial neuron circuit is compared. It is found that these effects alter the neuron firing rate and change the time spent in the on/off states but do not change the basic operation.