A Low-Power Analog Integrated Decision Tree Classifier for Laryngeal Voice Disorder Detection

This paper presents a low-power analog integrated decision tree classifier for real-time laryngeal voice disorder detection. Implemented using a TSMC 90 nm CMOS process, the classifier employs sub-threshold analog circuits to achieve ultralow power consumption of $527 n ~\mathrm{W}$ and a classification speed of 220 thousand inferences per second. Tested on a voice disorder dataset, it achieved 86.6% mean accuracy, demonstrating robustness across process, voltage, and temperature variations. The solution offers efficient, real-time edge-based classification for wearable medical devices, supporting early diagnosis and monitoring of voice disorders.