Real-Time Edge Neuromorphic Tasting From Chemical Microsensor Arrays

Area of Science:

• Neuromorphic Engineering
• Artificial Intelligence
• Sensor Technology
• Chemical Analysis

Background:

• Continuous liquid analysis is crucial for quality control in food, beverage, and chemical industries.
• Existing automated monitoring systems face challenges in miniaturization, energy autonomy, and real-time operation.
• The need for on-site, real-time chemical sensing necessitates advanced edge computing solutions.

Purpose of the Study:

• To present the first implementation of an artificial taste system on neuromorphic hardware for edge monitoring.
• To develop an energy-efficient and low-latency solution for continuous, automated liquid analysis.
• To evaluate the system's performance in a realistic beverage classification task.

Main Methods:

• Utilized a solid-state electrochemical microsensor array for multivariate, time-varying chemical data acquisition.
• Employed temporal filtering to enhance sensor readout dynamics.
• Deployed a rate-based, deep convolutional spiking neural network for efficient electrochemical sensor data fusion on neuromorphic hardware.

Main Results:

• Achieved 97% accuracy in beverage classification using the artificial taste system.
• Demonstrated 15x greater energy efficiency compared to similar convolutional architectures on commercial edge AI devices.
• Realized over 178x lower latencies than the sensor readout sampling period.

Conclusions:

• The neuromorphic artificial taste system offers a highly energy-efficient and low-latency solution for edge-based liquid analysis.
• This technology enables continuous, automated quality monitoring directly at the point of interest.
• The developed system, including the MicroBeTa dataset, shows significant promise for industrial applications requiring real-time chemical sensing.