Related Experiment Video
Updated: Jun 25, 2025
07:13
3D Modeling of Dendritic Spines with Synaptic Plasticity
Published on: May 18, 2020
6.8K
Graphene-Based Artificial Dendrites for Bio-Inspired Learning in Spiking Neuromorphic Systems
Samuel Liu1,2, Deji Akinwande1,2, Dmitry Kireev1,2,3
1Chandra Family Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, Texas 78712, United States.
Nano Letters
|May 31, 2024
Summary
Researchers developed biocompatible graphene-based artificial dendrites (GrADs) for brain-inspired computing. These GrADs enable efficient dendritic processing in artificial neural networks, reducing energy consumption without compromising accuracy.
Area of Science:
- Neuromorphic Engineering
- Materials Science
- Neuroscience
Background:
- Analog neuromorphic computing offers enhanced expressivity and energy efficiency over digital systems by mimicking brain parallelism.
- While artificial neurons and synapses are advancing, artificial dendrites for dendritic processing remain underdeveloped.
- Biocompatible materials are crucial for developing effective bio-interfaced computing systems.
Purpose of the Study:
- To introduce and characterize novel biocompatible graphene-based artificial dendrites (GrADs).
- To demonstrate the capability of GrADs in implementing dendritic processing and higher-order neuronal responses.
- To evaluate the performance of GrADs in energy-efficient spiking neural network simulations.
Main Methods:
- Fabrication of trilayer graphene devices with a dual side-gate configuration and Nafion membrane.
- Characterization of GrADs to exhibit spatiotemporal responses mimicking dendritic potentials (leaky recurrent, alpha, Gaussian).
- Integration of GrADs into data-driven spiking neural network simulations to assess their impact on network activity and accuracy.
Main Results:
- GrADs successfully demonstrated controllable conductance and spatiotemporal dendritic potential responses.
- Variable connectivity of GrADs enabled higher-order neuronal processing.
- Spiking neural network simulations showed a reduction in spiking activity by up to 15% without accuracy loss.
- Low-frequency operation of the networks was stabilized by the inclusion of GrADs.
Conclusions:
- Graphene-based artificial dendrites (GrADs) are a viable technology for implementing dendritic processing.
- GrADs contribute to energy-efficient bio-interfaced spiking neural networks.
- The developed GrADs show promise for advancing neuromorphic computing applications.
