Low-Latency Visuotactile Neuron Using Self-Oscillating Memristor.

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Artificial Intelligence

Background:

• Low-latency perception is crucial for embodied intelligence and efficient human-machine interaction.
• Biological systems use time-to-first-spike (TTFS) and firing rate for real-time sensory processing.
• Hardware implementation of low-latency spike conversion remains a significant challenge.

Purpose of the Study:

• To demonstrate a self-oscillating neuron based on NbO_x memristors for simultaneous TTFS and rate encoding.
• To achieve ultra-low latency perception of visual and tactile stimuli.
• To evaluate the performance of combined TTFS and rate coding for neuromorphic systems.

Main Methods:

• Fabrication of a self-oscillating neuron utilizing NbO_x memristors and intrinsic parasitic capacitance.
• Implementation of simultaneous TTFS and rate encoding for visual and pressure stimuli.
• Evaluation using CIFAR-10 and Braille datasets for classification tasks.

Main Results:

• Achieved an ultra-low first spike latency of 260 ns for visual and pressure stimuli.
• Demonstrated superior accuracy, noise robustness, and reduced temporal latency compared to rate coding alone.
• Validated multisensory integration for Braille recognition, showing improved accuracy in visually constrained scenarios.

Conclusions:

• The NbO_x memristor-based neuron enables intrinsically low-latency multisensory perception.
• Combined TTFS and rate coding offers significant advantages for neuromorphic systems.
• This technology holds promise for advancing real-time human-machine interaction and embodied intelligence.