Research Background
image
Sequence/superimposition multimodal modulation in synaptic devices is one of the foundations for the realization of complex neurobehavior and activities, and this is still a major challenge faced by traditional artificial synapses. Coupling triboelectric potential modulation and photon sensitization in artificial synapses may provide an active and direct way to take advantage of synaptic plasticity and achieve multimodal neuromorphic calculations.



image
Innovation
image
Sun Qijun and Wang Zhonglin’s research team from the Institute of Nano Energy and Systems, Chinese Academy of Sciences reported a mechanical-photonic artificial synapse based on graphene/molybdenum disulfide (Gr/MoS2) heterostructures, which has synergistic mechanical and optical plasticity (ie, mechanical displacement). Auxiliary photo-synaptic plasticity). Synaptic devices include phototransistors based on Gr/MoS2 heterostructures and integrated TENG in contact separation mode. The triboelectric potential generated by the TENG displacement can effectively drive the synaptic transistor. By controlling the charge transfer/exchange between Gr and MoS2, the triboelectric potential can also easily adjust the photo-synaptic behavior. Under the synergistic effect of mechanical displacement (as a state parameter) and light pulses containing temporal and spatial information (such as intensity and light duration), photonic synaptic plasticity including long-term memory and neural facilitation has been studied. It is also easy to implement photon programming and mechanical erasing processes in mechanical photonic artificial synapses. In addition to device-level simulation of synaptic function, an artificial neural network (ANN) was further constructed to prove the feasibility of improving image recognition with the help of mechanical plasticization.



image
Article analysis

Picture Figure 1. Biotactile/optical nerve and mechanical-photonic artificial synapse: (A) Schematic diagram of biotactile/visual perception system; (B) Schematic diagram of device.

Picture Figure 2. Mechanical-phototransistor based on Gr/MoS2 heterostructure and its working mechanism: (A) the relationship between TENG output voltage and displacement; (B) transmission curve; (C) working mechanism

Picture Figure 3. Mechanical-photonic artificial synapse based on Gr/MoS2 heterostructure: (A) real-time post-synaptic current in the dark; (B) photo-activated post-synaptic current; (C) under light and different displacements Changes of postsynaptic current under synergistic effect; (D) Density of states and carrier distribution of heterostructures

Picture Figure 4. Mechanical-photon artificial synapse mechanical signal and visual signal synergy: (A) schematic diagram; (BD) artificial synapse response to light signal; (E, F) light under different mechanical displacements Pulse-enhanced plasticity

Figure 5. Artificial neural network based on mechanical-photon artificial synapse for image recognition
image
After reading
image
The author successfully demonstrated a mechanical-photonic artificial synapse with cooperative multimodal plasticity based on the heterostructure of Gr/MoS2. Such synaptic devices have great prospects for the application of mixed-modal neuromorphic chips and unconventional convolutional neural networks in the fields of interactive optical-mechanical interfaces, artificial retinas, and intelligent robots. Programmable permanent photoconductivity and mechanical behavior-based modulation/erasing/plasticizing have great application prospects in memory neuromorphic calculation of various sensory data (such as optics, photons, touch, pressure, displacement, etc.). However, the response of this synaptic device to light signals is only inhibitory postsynaptic current (Δ<0), and excitatory postsynaptic current is also important for building an artificial visual system, and the device can be further optimized in the future.



【references】

https://advances.sciencemag.org/content/7/12/eabd9117