A Novel Ferroelectric Tunnel FET-based Time-Domain Content Addressable Memory with High Distance-Metric Linearity and Energy Efficiency for Edge Machine Learning

In this work, a novel ferroelectric tunnel FET (FeTFET) based time-domain content addressable memory (TD-CAM) is proposed and demonstrated with high distance-metric linearity and energy efficiency. By utilizing the modulated ambipolar feature of FeTFET, the XNOR-like comparison operation in time domain can be realized with only 3T, along with the dual-edge operation for further energy saving. Moreover, by utilizing the cascaded TD-CAM chain to accumulate propagation delay time, the proposed TD-CAM architecture can perform distance metric with ideally high linearity. Furthermore, benefiting from the full-TFET TD-CAM design, the search energy consumption can be further reduced with VDD scaling, providing a promising approach for energy-efficient machine learning.