A Neuron Imitation Device Based on Ovonic Threshold Switch
The Korea Institute of Science and Technology (KIST) announced on March 6 that its research team led by Dr. Lee Su-yeon has developed a low-power consumption and high-efficiency artificial sensory neuron device, which is expected to facilitate the commercialization of in-sensor computing imitating the human nervous system.
A neuron refines external sensory stimuli in a spike form. With the spike signal, the human brain can carry out cognition, learning, inference, prediction, and so on with a small amount of energy.
The research team has developed a neuron imitation device using an ovonic threshold switch (OTS) device. The OTS device is a two-terminal switching device, maintains a high resistance at or below its switching voltage, and shows a rapid decrease in resistance above the switching voltage.
Previously, the research team developed an artificial neuron device imitating a neuron generating a spike signal at a certain signal intensity. In its recent research, the team aimed to imitate the neuron quickly finding a pattern from massive sensory data and abstracting it. To this end, the team developed a three-terminal OTS device capable of switching voltage control, connected a stimulus-voltage conversion sensor to the third electrode of the device, and completed a neuron device whose spike signal form varies with external stimuli.
In addition, the three-terminal OTS device was connected to a photoelectric conversion sensor to realize an artificial visual neuron device imitating human information processing. Also, the artificial visual neuron device was connected to an artificial neural network, lung X-ray image learning was conducted, and viral pneumonia and COVID-19 were distinguished with an accuracy of 86.5 percent.
“The artificial sensory neuron device developed by the team can be applied to sensory organs by connection to existing sensors and this platform technology is of great importance in the field of in-sensor computing,” the institute said, adding, “The newly developed artificial intelligence system requires just 1,000 neurons and can be directly connected to medical diagnostic sensors and, as a result, can help make medical devices providing diagnosis simultaneously with sensor signal pattern generation.”
According to the team, the applications of the technology include acute cardiac disorder prediction based on time-series blood pressure or pulse pattern analysis, disaster prevention based on vibration detection outside the audible frequency range, and many more. Details of the research have been published in the latest edition of Nano Letters.