Egg shell membrane based electrolyte gated oxide neuromorphic transistor
LI Yan,CHEN Xin-Li,WANG Wei-Sheng,SHI Zhi-Wen,ZHU Li-Qiang,
DOI: https://doi.org/10.7498/aps.72.20230411
IF: 0.906
2023-01-01
Acta Physica Sinica
Abstract:In recent years, the studies of neuromorphic devices has received extensive attention. It is becoming an important branch of the developments of artificial intelligence technology. At the same time, natural biomaterials have several priorities, such as biodegradability, good biocompatibility, non-toxicity, etc., and have important value in novel portable intelligent systems. The egg shell membrane (ESM) is a fiber scaffold composed of highly crosslinked collagen, glycoprotein and cysteine-rich eggshell membrane proteins. It has porous morphology, thermal stability, mechanical strength, etc. Therefore, these protein-based fiber membranes have several potential applications, including nanocatalysts, microbial fuel cells, adsorption of toxic dyes, etc. This study adopts ESM as electrolyte, exhibiting extremely high proton conductivity of ~6.4×10 -3 S/cm and extremely high electric-double-layer (EDL) capacitance of ~2.8μF/cm 2 at room temperature. Thus, it has extremely strong interfacial EDL electrostatic modulation ability. Then, indium tin oxide EDL transistor has been fabricated using a single step masking processing with magnetron sputtering. The device demonstrates typical n-type output curves and transfer curves, at low operation voltage. In addition, transfer curves have been scanned for two times. It is observed that the curves approach with each other quite good, indicating the good stabilities. The device exhibits excellent electrical performances thanks to the extremely strong proton gating effects, including ON/OFF ratio of ~2.5×10 6 , mobility of ~3.2 cm 2 /(vs), sub-threshold swing of ~213 mV/dec. With the unique interfacial EDL modulation activities of ESM, the transistor can mimic some important synaptic plasticity behaviors, including excitatory postsynaptic current, paired pulse facilitation. With the increased pre-synaptic spike amplitude, the EPSC value increases correspondingly. And with the increased pre-synaptic spike frequency, the EPSC grain increases correspondingly, indicating the potentials in high-pass synaptic filtering. By loading 64 potentiation spikes and 64 depression spikes, multi-level synaptic weight can be updated, demonstrating potentiation and depression activities. Again, with the same potentiation spikes and depression spikes, synaptic weight value curves approach with each quite well, indicating the good stabilities for the present ESM gated oxide neuromorphic transistor. Then, an artificial neural network is adopted to perform supervised learning with Modified National Institute of Standards and Technology (MNIST) database. For simulation, a two-layer multilayer perceptron (MLP) neural network with 400 input neurons, 100 hidden neurons and 10 output neurons was adopted. The best recognition accuracy is as high as 92.59%. The proposed ESM gated oxide neuromorphic transistors have certain potentials in low-cost biodegradable neuromorphic system.
physics, multidisciplinary