A Novel Instruction Driven 1-D CNN Processor for ECG Classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Diagnostics

Background:

• Electrocardiography (ECG) is crucial for diagnosing cardiovascular diseases.
• Wearable devices with on-device AI processors are transforming ECG data analysis.
• Existing AI processors face challenges in flexibility, portability, power efficiency, and latency.

Purpose of the Study:

• To propose an instruction-driven convolutional neural network (CNN) processor for versatile ECG analysis on wearable devices.
• To optimize AI processor performance for portable, low-power, and low-latency applications.

Main Methods:

• Development of an instruction-driven CNN processor with a Processing Element (PE) array designed for parallelism and data reuse.
• Integration of a CORDIC-based activation unit supporting Tanh and Sigmoid computations.
• Implementation using 110 nm CMOS technology.

Main Results:

• The processor achieved a die area of 1.35 mm² and a power consumption of 12.94 µW.
• Demonstrated high accuracy in two typical ECG AI applications: 97.95% for two-class (normal/abnormal) classification and 97.9% for five-class classification.
• The 1-D CNN algorithm showed excellent performance for both classification tasks.

Conclusions:

• The proposed instruction-driven CNN processor offers a flexible, portable, and power-efficient solution for on-device ECG analysis.
• This advancement enables enhanced capabilities for wearable health monitoring systems in cardiovascular diagnostics.
• The design successfully addresses key challenges in implementing AI for real-time ECG interpretation.