An Opto-Electronic HfO _x -Based Transparent Memristive Synapse for Neuromorphic Computing System

Abstract

In this study, a transparent bilayer memristor showing both electrical and optical synapses along with good electrical properties after annealing is presented. In addition to 85% transparency, the device shows excellent electrical characteristics for 1000 cycles of stable LRS/HRS and more than 10 4 s retention at high temperatures. The annealed device also exhibits stable potentiation and depression cycles for more than 10 000 ac pulses with a low coefficient of nonlinearity. By applying consecutive ac pulses, synaptic properties of paired-pulse facilitation (PPF) and spike time-dependent plasticity (STDP) are calculated. The memristor is illuminated by a 405 nm light source in which different light intensities ranging from 20 to 40 mW/cm 2 are used for achieving multilevel cell (MLC) characteristics. Learning/Forgetting curve (PSC) and optical PPF are measured to mimic optical synaptic function. An image recognition comparison of optical and electrical synaptic properties with a normalized loss rate of <0.1 is obtained after just 100 epoch trainings. These excellent attributes of this transparent memristor make it a promising candidate for electrical/optical memory devices or for using it as an optically synaptic sensor device.

Tin Doping Induced High‐Performance Solution‐Processed Ga2O3 Photosensor toward Neuromorphic Visual System

“…To benchmark the performance of these Ga 2 O 3 :Sn devices, the recognition accuracy under weak light intensity stimulations was compared with various photonic synapses made of photoelectronic phototransistors, [ 7,27‐29,33,40‐42 ] photosensors, [ 6,30‐32,43 ] and optoelectronic memristors. [ 4,34‐39,44 ,45] As shown in Figure 5h, our device had comparable high recognition accuracy to the highest value of accuracy and ultra‐low light intensity response to the lowest value of intensity. This result demonstrates the outstanding performance of our devices toward high‐efficient NVS, which could broaden the application potential of the NVS from bright to dim scenes.…”

Linear and symmetric synaptic weight update characteristics by controlling filament geometry in oxide/suboxide HfOx bilayer memristive device for neuromorphic computing

“… 54 ITO/ZnO/HfO x /ITO + 1.8 V/− 2 V 10 3 10 4 ~ 10 2 3.19/2.4 Ref. 55 TiN/Ti/HfO x /TaO y /HfO x /Au − 5 V/+ 5 V 10 6 10 5 50 5.3/11.9 Ref. 56 Ti/HfO 2 /HfO 2−x /Pt + 1.1 V/− 0.84 V (DC measurement) 300 > 10 4 > 10 2 2.4/1.55 This work + 1 V/− 1.2 V (pulse measurement) 4 × 10 3 > 10 4 ~ 20 …”

LTP–LTD transformation of unipolar pulse voltage-driven zinc oxide memristors via TiO2 thin layer incorporation

“…This division hampers their ability to meet the demands of high-speed operations and low power consumption, which are crucial in the era of artificial intelligence development. 1,2 Inspired by the architecture of computing in memory in human brains, researchers have explored various memory devices for emulating artificial synaptic functions. 3,4 Among these, the memristor has garnered significant attention due to its capacity for high-speed processing with remarkable energy efficiency.…”