Spectrally Tuned Floating-Gate Synapse Based on Blue- and Red-Absorbing Organic Molecules for Wavelength-Selective Neural Networks and Fashion Image Classifications

Abstract

A neuromorphic optical synapse transistor based on a spectrally tuned floating-gate synapse (STFGS), designed to achieve optoelectronic synaptic behavior, is presented. The device incorporates a heterojunction structure composed of a dinaphtho[2,3-b:2',3'-f]selenopheno[3,2-b]selenophene (DNSS) upper channel and an E)-2-(2-((6-(di-p-tolylamino)-4,4-dimethyl-4H-indeno[1,2-b]thiophen-2-yl)methylene)-3-oxo-2,3-dihydro-1H-inden-1-ylidene)malononitrile (Dta-Inth-IC) floating-gate layer. A parylene dielectric layer strategically positioned between the DNSS and Dta-Inth-IC layers functions as a barrier, enabling selective charge storage within the floating-gate architecture. Synaptic plasticity is analyzed by varying stimulation conditions, such as the on-time, off-time, and pulse number of optical pulses. Long-term potentiation (LTP) is observed with efficient charge trapping in the floating-gate under 660 nm light stimulation. Energy band alignment analysis confirms charge accumulation in Dta-Inth-IC under 660 nm light, while 455 nm light stimulation induced rapid recombination in DNSS. The applicability of artificial neural networks (ANN) based on the potentiation curves obtained from STFGS is evaluated. For this purpose, a convolutional neural network (CNN)-based ANN is designed and performs classification tasks using the Fashion Modified National Institute of Standards and Technology (Fashion MNIST) dataset. Through repeated training, a maximum recognition rate of 91.37% for 660 nm light stimulation is achieved, demonstrating that the STFGS successfully mimics synaptic behavior.