Flexible self-rectifying synapse array for energy-efficient edge multiplication in electrocardiogram diagnosis.

Area of Science:

• Neuromorphic Engineering
• Flexible Electronics
• Wearable Technology

Background:

• Edge computing enhances real-time data processing and responsiveness, especially in unstable networks.
• Wearable and skin-attached electronics require compact, energy-efficient, and flexible edge devices.
• Current digital approaches for edge AI in wearables face limitations in power consumption and form factor.

Purpose of the Study:

• To propose and demonstrate a flexible memristive dot product engine (f-MDPE) for efficient edge computing in wearable applications.
• To integrate the f-MDPE into a real-time electrocardiogram (ECG) monitoring system for edge diagnosis.
• To evaluate the performance, energy efficiency, and mechanical stability of the proposed f-MDPE.

Main Methods:

• Fabrication of a 32x32 crossbar array using low-temperature processed self-rectifying charge trap memristors on a flexible polyimide substrate.
• Development of a hardware-aware neural network training algorithm for real-time ECG classification.
• Testing of the f-MDPE's electrical and mechanical stability under bending conditions.

Main Results:

• The f-MDPE demonstrated high uniformity and robust stability under 5-mm bending.
• Real-time ECG diagnosis using the f-MDPE achieved 93.5% classification accuracy.
• The system consumed only 0.3% of the energy compared to traditional digital methods.

Conclusions:

• The proposed flexible memristive dot product engine is a viable solution for compact, low-power edge AI in wearables.
• This approach significantly reduces energy consumption for real-time data analysis in edge devices.
• The f-MDPE shows strong potential for emerging edge neuromorphic hardware in healthcare applications.