Fast solid-state neuromorphic proton transistor trained to convert triangles into sinewaves

Author: EIS Release Date: Aug 14, 2023

Japan’s National Institute for Materials Science (NIMS) and the Tokyo University of Science have created a solid-state electric double layer neuromorphic transistor, claiming it to be the fastest yet, and proving its computational ability by training it to convert waveforms.

 

 

 

NIMS UTokyo neuromorphic transistor

“The team measured the speed at which the transistor operates by applying pulsed voltage to it, and found that it operates 8.5 times faster than existing electric double layer transistors,” according to the university. “This transistor may be used to develop AI devices with applications including future event prediction, and facial, voice and odour recognition.”

 

27μs was the switching speed achieved, which the researchers attribute to developing a solid electrolyte with high proton conductivity: yttria-stabilised porous zirconia ceramic thin film, whose nano-scale pores can absorb water to provide a path for hydrogen ions, explained the team.

 

 

In the transistor (see diagram), this is paired with a hydrogenated diamond thin film channel – an electric double layer can be formed at ceramic-diamond interface, rapidly charged and discharged through the conductive electrolyte.

 

 

 

 

 

 

The device works with the source at 0V, and both drain and gate at more negative potentials.

 

Current flow though the channel is minimal with no bias on the gate, and increases as the gate is pulled negative (-1V is the maximum useful bias voltage). On/off ratio is ~ five current decades.

 

Takashi Tsuchiya National Institute for Materials Science (NIMS) Waveform_transform_IGRIn a proof-of-concept, the transistor was used to in a ‘reservoir’ style neuromorphic computer, which was trained to implement a non-linear waveform transform, converting a triangle voltage wave into sine wave, square wave, π/2-shifted triangle wave or frequency-doubled triangle wave.

 

It was successful – creating all of them with over 90% fidelity, except for the square wave, at which it still scored over 70%.

 

And how many transistors were needed?

 

“Interestingly, we need only one transistor to perform this waveform transformation task,” NIMS scientist Takashi Tsuchiya told Electronics Weekly. “It is due to the unique transient electrical characteristic of the transistor, originated from charging and discharging dynamics of electric double layer which is quite sensitive to the past input history. We can harness the unique, inherent and fast dynamics of solid/solid electrolyte interface with about 1nm thickness for high performance neuromorphic computing. That is the main attractiveness of the work.”