Biological brains efficiently process slow input signals in their living environments. It is believed that by mimicking these characteristics (slow operation and ultra-low power consumption) with artificial devices, we can pave the way for ultra-low power information processing. However, it has been challenging to achieve both characteristics at the same time. This time, with ingenious control of the charged ions present in a solid-state material, we have developed a new-concept transistor that converts input signals into slowly changing output signals over time. Based on the control of ions, this MOS transistor, using strontium titanate for a channel, operates more than a million times slower than traditional MOS transistors using silicon. It was demonstrated that it operates with an extremely low power consumption of 500 pW. The operation verification of this transistor, which can mimic the operation of biological neurons with a very long time constant (the time scale in which output current changes in response to input voltage), contributes to the realization of edge devices capable of complex learning and inference with ultra-low power consumption, similar to living organisms.
Papers
Journal: Advanced Materials
Title: Taming Prolonged Ionic Drift-Diffusion Dynamics for Brain-Inspired Computation
Authors: Hisashi Inoue, Hiroto Tamura, Ai Kitoh, Xiangyu Chen, Zolboo Byambadorj, Takeaki Yajima, Yasushi Hotta, Tetsuya Iizuka, Gouhei Tanaka, Isao H. Inoue