Machine learning impassioned assessment of trap states driven band bending analysis in organic Schottky formation

Abstract

Band bending at metal–semiconductor junction bears significant consequences providing analytical explanation of built-in-voltage (V_bi) in current–voltage (I-V) relationship. Knowing the band bending impact on charge transition, a theoretical model has been developed which relates the parameter with geometrical conductance ($g_m$) by introducing modification in empirical Mott-Gurney law. The theoretical relation is validated with the approximated outcome of organic semiconductor-based prototype simulated device structure ranges between 100 nm to 900 nm. The same has been examined on two different experimental semiconducting dye based organic diode. For each case, the result reveals a greater range of consistency satisfying a well trend of analogy with theoretically developed equation. Band bending (b) shows inverse linearity with $g_m$ at increasing voltage, whereas the following investigation explore its proportional nature with increasing device thickness (L). Since organic materials are trap prone and its influence on charge conduction is well known, therefore, to predict the impact of trap distributions on b, machine-learning based modeling approaches has been undertaken to leverage the prediction in a reliable way to quantify the trap energy (E_T). The comparison among different output set of computational modelling approach leads towards a non-linear approximation of b with variation of E_T and is best applicable at 700 nm device thickness. Furthermore, different linear and non-linear Machine Learning algorithm has been trained for validation employing the considered sample thickness in experimental dye-based devices and based on the assessment of performance metrics, the influence of E_T on variation of b has been detected.