Bio-plausible reconfigurable spiking neuron for neuromorphic computing

Area of Science:

• Neuroscience
• Materials Science
• Computer Engineering

Background:

• Biological neurons exhibit complex spiking patterns crucial for neural computation.
• Current neuromorphic systems often use simplified models, limiting their bio-plausibility and functionality.
• Emulating intricate biological spike patterns in hardware is computationally expensive.

Purpose of the Study:

• To develop a compact, reconfigurable neuron design that emulates bio-plausible spiking dynamics.
• To leverage the intrinsic properties of NbO2-based spiking units and electrochemical memory (ECRAM) for neuron modeling.
• To demonstrate flexible reconfiguration of firing modes and adaptive behaviors in a neuromorphic context.

Main Methods:

• Proposed a novel neuron design integrating a NbO2-based spiking unit with an ECRAM.
• Utilized the ECRAM's tunable resistance to control the membrane potential's temporal dynamics.
• Implemented various bio-plausible firing modes, including phasic and burst spiking.
• Built spiking neural networks (SNNs) incorporating the developed bio-plausible neuron model.

Main Results:

• Successfully emulated fast-slow dynamics characteristic of biological neurons.
• Achieved flexible reconfiguration of firing modes (phasic, burst) and adaptive spiking.
• Demonstrated improved classification accuracies in SNNs using bursting neurons compared to simplified models.
• Showcased the potential for more bio-plausible neuromorphic computing systems.

Conclusions:

• The proposed ECRAM-based neuron design offers a cost-effective solution for emulating complex biological neuron dynamics.
• This approach enables flexible and adaptive spiking behaviors, enhancing neuromorphic system capabilities.
• The developed model shows significant promise for advancing bio-plausible neuromorphic computing and artificial intelligence.