Event-Driven Neuromorphic Gaze Decoding via e-Skin Electrooculography

Area of Science:

• Bioelectronics
• Neuromorphic Computing
• Wearable Technology

Background:

• Traditional wearable eye-tracking systems face limitations in bulkiness, power consumption, and reliance on external computation.
• There is a need for unobtrusive, energy-efficient, and private eye-tracking solutions for advanced human-computer interfaces.

Purpose of the Study:

• To develop a hardware-software codesigned electrooculography (EOG) interface for real-time gaze decoding.
• To integrate ultrathin conformal e-skin sensors with resistive random-access memory (RRAM) for efficient neuromorphic processing.
• To create a glassless, low-power, and private wearable eye-tracking system.

Main Methods:

• Utilized ultrathin conformal e-skin sensors for stable acquisition of vertical and horizontal oculomotor signals.
• Implemented a neuromorphic processing pipeline using RRAM crossbars for synaptic vector-matrix multiplication.
• Developed a lightweight spiking neural network (SNN) for classifying attention-guided spike features from EOG signals.
• Employed noise-aware training for quantized weights on the RRAM array to ensure robust inference.

Main Results:

• Achieved real-time gaze decoding with a hardware-software codesigned EOG interface.
• Demonstrated substantial reduction in latency and energy demand compared to conventional methods.
• Ensured robust inference through noise-aware training and quantized weights on RRAM.
• Enabled edge-level computation for enhanced user privacy and elimination of cloud dependency.

Conclusions:

• The proposed glassless, energy-efficient wearable interface advances immersive computing, assistive technologies, and mobile health monitoring.
• Flexible bioelectronics integrated with neuromorphic processors offer a promising pathway for next-generation wearable interfaces.
• The EOG-based system provides a private and unobtrusive solution for real-time gaze tracking.