NeuroSOFM-Classifier: Nanoscale FeFETs Based Low Power Neuromorphic Architecture for Classification Using Continuous Real-Time Unsupervised Clustering

Supervised machine learning techniques are becoming subject of significant interest in data analysis. However, the high memory bandwidth requirement of current implementations, scarcity of labeled data, and dynamic environments in many applications prevent implementation of supervised machine learning techniques. In this work, we propose a neuromorphic architecture implementing the self-organizing feature map algorithm using nanoscale ferroelectric field-effect transistors (FeFETs) and complementary metal-oxide-semiconductor (CMOS) technology to produce a semi-supervised NeuroSOFM-Classifier. A best matching input (BMI) identifier circuit allows for very few labeled samples to be used to provide supervised class labels for each hardware neuron in the NeuroSOFM-Classifier. The NeuroSOFM-Classifier architecture can then be used to inference or classify the new data in real-time. This NeuroSOFM-Classifier, trained on just 2% of the labeled data, is capable of classifying COVID-19 patient chest x-rays with an average accuracy of 83%.