Related Experiment Video
Updated: May 15, 2026
05:18
A Rapid and Chemical-free Hemoglobin Assay with Photothermal Angular Light Scattering
Published on: December 7, 2016
Calibration-free and drift-robust AlGaN/GaN HEMT sensor arrays for intelligent pH detection
Jiang Zhu1, Heqiu Zhang1, Dawei Guo1
1School of Integrated Circuits, Dalian University of Technology, Dalian, 116620, PR China; Dalian Key Laboratory of Wide Bandgap Semiconductor Devices Integration and System, Dalian, 116024, PR China.
Analytica Chimica Acta
|May 13, 2026
Summary
This study presents a co-optimized biosensing platform using AlGaN/GaN HEMTs and artificial neural networks (ANNs) to overcome variability in biosensor arrays. The system achieves high accuracy and robustness without recalibration, even with sensor degradation over time.
Area of Science:
- Semiconductor device physics
- Biosensor technology
- Machine learning applications
Background:
- AlGaN/GaN HEMTs offer high electron mobility and chemical stability, ideal for liquid-phase biosensing.
- Array-level biosensor implementations face challenges with sensor-to-sensor variation, chip inconsistency, and measurement reproducibility.
Purpose of the Study:
- To develop a hardware-algorithm co-optimized sensing platform to address variability in AlGaN/GaN HEMT biosensor arrays.
- To improve the robustness and reproducibility of biosensing measurements across different sensors, chips, and time.
Main Methods:
- Integration of a monolithically embedded on-chip reference electrode and multilayer metallization for electrical isolation and matrix-addressed signal acquisition.
- Application of a data-driven artificial neural network (ANN) to manage array-level variability and improve performance.
- Utilizing X-ray photoelectron spectroscopy and Kelvin probe measurements to analyze surface-state evolution.
- Employing an attention-enhanced ANN to mitigate performance degradation due to long-term storage.
Main Results:
- The platform demonstrated robust recognition performance across diverse chips and measurement conditions, achieving ~95% classification accuracy.
- Stable within-chip recognition and effective cross-chip transfer were achieved without explicit recalibration.
- Surface-state evolution was observed on the GaN cap layer over a 30-day storage period.
- The attention-enhanced ANN successfully alleviated classification accuracy degradation on aged datasets, improving prediction accuracy.
Conclusions:
- A co-optimized hardware-algorithm approach effectively addresses variability in AlGaN/GaN HEMT biosensor arrays.
- Data-driven ANNs provide a robust solution for consistent biosensing performance without frequent recalibration.
- The developed platform demonstrates potential for reliable, long-term biosensing applications, even with material degradation.
