Left: schematic of biological neurons connected to a synapse. Right: schematic of an electroluminescent multilayer ITO/ZnO/Si NCs/CBP/MoO3/Au synaptic device. Reprinted with permission from Zhao et al., Nano Energy 54 (2018) 383–389.
Left: schematic of biological neurons connected to a synapse. Right: schematic of an electroluminescent multilayer ITO/ZnO/Si NCs/CBP/MoO3/Au synaptic device. Reprinted with permission from Zhao et al., Nano Energy 54 (2018) 383–389.

Conventional computing systems could be replaced in the future by massively parallel, low energy, more intelligent brain-like processors. Even the most sophisticated CMOS (complementary metal oxide semiconductor)-based technologies can only start to approach the human brain’s capacity for parallel processing and learning. Currently, ‘neuromorphic’ computing systems that mimic the brain rely on synaptic devices featuring electrical/optical inputs and electrical outputs. But now researchers at Zhejiang University have demonstrated a new type of synaptic device based on light-emitting silicon nanocrystals (Si NCs) [Zhao et al., Nano Energy 54 (2018) 383, https://doi.org/10.1016/j.nanoen.2018.10.018].

“The realization of neuromorphic computing critically depends on the development of synaptic devices,” explains Xiaodong Pi, who led the research with Deren Yang. “Up to now, synaptic devices could be categorized into two main types: memristors and transistors.”

Synaptic devices with optical outputs would be highly desirable for future neuromorphic computers because they could allow visualization of neural responses, ultimately perhaps enabling spatiotemporal monitoring of artificial neural systems. Si NCs could be the choice of material for this purpose as they can electrically stimulated to give out electroluminescence that decays on the timescale of tens of microseconds, which is the same as many important synaptic functionalities.

The researchers’ synaptic device comprises a multilayer structure of indium tin oxide (ITO), Si NCs, CBP (4,4-bis(N-carbazolyl)-1,1-biphenyl), zinc oxide (ZnO), molybdenum oxide (MoO3), and gold (Au) layers. The colloidal Si NCs act as the crucial light-emitters triggered by an electrical input.

“An electrical spike acts as an action potential, while electroluminescence is used as the medium for transmission of information from a presynaptic axon terminal to a postsynaptic dendrite terminal,” says Pi. “When two individual synaptic devices are intercoupled by sharing a common ITO electrode, basic logic functions can be realized.”

The team demonstrate that simple logic operations such as “AND” and “OR”, or “NAND” and “NOR” can be performed in this way.

“Our work has laid a good foundation for the realization of optogenetics-inspired artificial neural networks with optoelectronic integration,” says Pi.

Such artificial neuromorphic computing systems could be ideal for highly demanding operations such as image cognition or facial classification while maintaining low energy consumption. To realize such systems, electroluminescent synaptic devices will need to be much smaller, admit the researchers, ideally on the nanometer scale to enable a large number of devices to be packed together. Power efficiency could also be improved to realize even lower energy devices. Other electroluminescent devices with decay on the right timescale and high energy efficiency could also work just as well.

“These future efforts should significantly contribute to the development of artificial optoelectronic neural network for neuromorphic computing,” Pi told Materials Today.