Ultralow-Power Single-Sensor-Based E-Nose System Powered by Duty Cycling and Deep Learning for Real-Time Gas...

Area of Science:

• Materials Science
• Sensor Technology
• Artificial Intelligence

Background:

• Conventional electronic nose (e-nose) systems often rely on sensor arrays, leading to increased power consumption and cost.
• Metal oxide semiconductor (MOS) sensors are widely used for gas detection but can be power-intensive.

Purpose of the Study:

• To develop a novel, ultralow-power single-sensor e-nose system for real-time gas identification.
• To reduce the power demands of e-nose technology for practical IoT applications.

Main Methods:

• Utilized a single MOS sensor on a suspended 1D nanoheater driven by duty cycling (pulsed power).
• Leveraged the sensor's ultrafast thermal response to decouple temperature and surface charge effects.
• Employed a convolutional neural network for gas identification and concentration regression.

Main Results:

• Achieved early-stage identification of five gas types within 30 seconds with 93.9% classification accuracy.
• Demonstrated a significant power reduction of up to 90%, achieving 160 μW.
• Obtained a concentration regression error of 19.8%.

Conclusions:

• The single-sensor e-nose system effectively identifies gases with high accuracy and minimal power consumption.
• The technology is manufacturable via wafer-level batch processes, enabling cost-effective, battery-driven IoT monitoring.