Fig. 1. (Color online) (a) Grow the GaN nanowires (NWs) on Si substrate. (b) Ga₂O₃ nanowires formed after oxidation. (c) Transfer the Ga₂O₃ nanowires to a solution. (d) Transfer Ga₂O₃ nanowires to the electrodes. (e) Top-view SEM image of the synaptic device based on Ga₂O₃ nanowires. (f) Top-view SEM image of Ga₂O₃ nanowires. (g) Side-view AC-STEM image and high-resolution EDX mapping images of the Ga₂O₃ nanowires. Schematic diagrams of (h) two adjacent neurons and (i) a biological synapse.

Fig. 2. (Color online) (a) Current‒time (I‒t) curve of the device when subjected to a single light pulse stimulation (0.16 mW·cm^‒2) at 255 nm for 1 s. (b) I‒t curve of the device under two consecutive light pulses at 255 nm. (c) The decay time as a function of the number of light pulses. (d) The I‒t curve of the device under 5 continuous light pulses with an interval of 5 s at a constant bias voltage of 8 V.

Fig. 3. (Color online) (a) EPSC of the synaptic device at different frequencies triggered by two consecutive 255 nm light pulses of 0.16 mW·cm^‒2. (b) EPSC of a synaptic device stimulated by 10 consecutive 255 nm light pulses at different optical powers. (c) EPSC of synaptic device at various pulse numbers under an optical power density of 0.16 mW·cm^‒2. (d) EPSC of synaptic device stimulated by 5 and 2 Hz light pulses.

Fig. 4. (Color online) (a) EPSC of a synaptic device stimulated by two cycles of consecutive 5-light pulses at 5 s intervals (255 nm, 0.16 mW·cm^‒2). Synaptic weight results of the synaptic device stimulated by (b) different numbers and (c) different frequencies of light pulses. (d) Synaptic weight results of the artificial device stimulated by light pulses with various optical power densities.

Fig. 5. (Color online) (a) Equivalent circuit model of the synaptic nano-device. (b) I‒V curve of the synaptic nano-device. Schematic energy band diagrams of the Ga₂O₃ NWs (c) in dark, (d) under the first light stimulation, (e) without light stimulation, and (f) under the second light stimulation.

Fig. 6. (Color online) (a) Schematic diagram of an ANN simulation using 784 × 100 × 10 synaptic weights. (b) Schematic of a neuron node. (c) Experimental data of LTD/LTP characteristics triggered by optical pulses and their corresponding fitting curves. (d) Simulate the accuracy of different training sessions. (e) The results of 20 digital images randomly selected from the MNIST database for identification.

Table 1. Relevant parameters for the artificial neural network (ANN) training.