A 36-nW Electrocardiogram Anomaly Detector Based on a 1.5-bit Non-Feedback Delta Quantizer for Always-on Cardiac...

Area of Science:

• Biomedical Engineering
• Electrical Engineering
• Computer Science

Background:

• Continuous cardiac monitoring is crucial for detecting arrhythmias.
• Existing systems often face limitations in power consumption and complexity.
• There is a need for efficient, always-on ECG anomaly detection solutions.

Purpose of the Study:

• To propose and evaluate an ultra-low power (ULP) always-on electrocardiogram anomaly detector (EAD).
• To develop a novel feature extraction and classification engine for real-time cardiac monitoring.
• To demonstrate high detection accuracy with minimal power consumption.

Main Methods:

• Implemented a 1.5-bit non-feedback delta quantizer (DQ) with a computing-in-capacitor (CIC) subtractor for feature extraction.
• Utilized a multiplier-less convolutional neural network (CNN) engine for event-driven classification.
• Integrated DQ and CNN chips fabricated in 65-nm and 180-nm CMOS technologies, respectively.

Main Results:

• Achieved a detection accuracy of 90.6%–91.3% on the MIT-BIH arrhythmia database, identifying three ECG anomalies.
• Demonstrated an ultra-low average power consumption of 36 nW for the core circuits.
• Eliminated the need for traditional high-resolution analog-to-digital converters (ADCs).

Conclusions:

• The proposed ULP EAD offers a promising solution for continuous cardiac monitoring.
• The novel DQ and CNN architecture enables efficient and accurate anomaly detection.
• This technology represents a significant advancement in always-on wearable health devices.