Inspired by biology, one of the important ways for information processing, especially at the edge of limited resources, is to develop neuromorphic devices with similar biological neural network functions.

In a study published in Nature Communications, Prof. SHANG Dashan’s group at the Institute of Microelectronics of the Chinese Academy of Sciences (IMECAS) developed a vertical dual-gate electrolyte-gated transistor, named neurotransistor, with a 30 nm channel length, short-term memory characteristic, and the stackability for 3D integration. The read power and energy reach ~3.16 fW and ~30 fJ, close to the biological level.

The electrolyte-gated transistor uses electrolyte materials with mobile ions (such as H⁺, Li⁺, O^2-) as gate dielectric. The migration of ions driven by the gate voltage towards the channel produces multi-level short-term memory effects of the channel conductance, which is very similar to the dendritic behavior of neurons.

The researchers used the short-time memory characteristics of the neurotransistors to realize the dendrite computing function in biological neurons, such as dendrite integration and coincidence detection. The dendrite computing ability is extended to the recognition of sound azimuth and distance by integrating neurotransistors into a bionic sound localization neural network.

This work not only demonstrates the potential of neurotransistors as building blocks to emulate the advanced functions of biological neural networks, but also provides a novel approach for the development of edge-oriented, high-density, low-energy neuromorphic computing hardware systems from the device level.

This study was done in collaboration with Fudan University, Tsinghua University, and the University of Hong Kong.