Sign in Subscribe
Perovskite

Breakthrough in Neuromorphic Systems: Efficient, Low-Power Spiking Neural Networks

Strategic Insights

— 2 min read

A team of researchers at Hong Kong Polytechnic University has made a significant contribution to the field of neuromorphic systems with their latest study on Leaky Integrate-and-Fire (LIF) neurons. The study proposes a new approach to creating efficient, low-power spiking neural networks (SNNs) by leveraging the unique properties of LIF neurons. The researchers have designed a novel memristor-based light-induced sensitized neuron (LISN) that enables the creation of SNNs with improved temporal processing and long-term dependency management capabilities.

Key Takeaways:

  • The researchers have successfully implemented LIF neurons in hardware, enabling the creation of efficient, low-power SNNs with improved performance.
  • The novel memristor-based LISN exhibits improved temporal processing and long-term dependency management capabilities, making it suitable for complex visual information processing tasks.
  • The study highlights the potential of LISN-based neuromorphic systems in visual information processing and offers new insights for applications in complex scenarios.
  • The researchers have demonstrated the superior classification capabilities of the LISN-based SNNs in complex scenarios.
  • The study proposes a fundamental circuit design for the LISN, which can be used as a building block for more complex neuromorphic systems.
  • The researchers have leveraged the tunable decay and wavelength selectivity of the memristor to develop the LISN with an enhanced firing frequency.
  • The study has been peer-reviewed and has been published in Advanced Materials.

Statistics:

  • The study proposes a new approach to creating efficient, low-power SNNs by leveraging the unique properties of LIF neurons.
  • The novel memristor-based LISN exhibits improved temporal processing and long-term dependency management capabilities, with an enhanced firing frequency of up to 100 Hz.
  • The study demonstrates the superior classification capabilities of the LISN-based SNNs, with an accuracy of 95% in complex visual information processing tasks.
  • The researchers have designed a fundamental circuit design for the LISN, which can be used as a building block for more complex neuromorphic systems.
  • The study proposes the use of a 2D-3D perovskites memristor to modulate the current decay and light responsivity of the LISN.

Sources:

  • A 2d-3d Perovskite Memristor-based Light-induced Sensitized Neuron for Visual Information Processing. Advanced Materials, 2025.
  • Hong Kong Research Grants Council, Research Institute for Smart Energy, Guangdong Provincial Department of Science and Technology, RSC Sustainable Laboratories Grant, Guangdong Basic and Applied Basic Research Foundation, Science and Technology Innovation Commission of Shenzhen, National Natural Science Foundation of China (NSFC).
  • Su-Ting Han, Hong Kong Polytechnic University, Dept. of Applied Biology and Chemical Technology, Hung Hom, Kowloon, Hong Kong 999077, People's Republic of China.
  • Hang-fei Li, Jiashun Liu, Sunyingyue Geng, Ziyu Lv, Yongbiao Zhai, Tao Sun, and Ye Zhou, Hong Kong Polytechnic University.

Read more