A Novel Ferroelectric FET Based Universal Content Addressable Memory With Reconfigurability for Area- and Energy-Efficient In-Memory-Searching System

In this work, a novel ferroelectric FET (FeFET)-based universal content addressable memory (UCAM) is proposed and experimentally demonstrated for the first time. By utilizing the non-volatile multilevel ferroelectric polarization for entry storage and the contrary monotonicity of N-type and P-type FeFETs for query searching, the reconfigurable UCAM enables binary CAM, ternary CAM, multibit CAM and analog CAM modes without the need of extra circuit to generate complementary “0/1”, multilevel or analog search signals of twin branches in different CAM modes. Moreover, based on the proposed UCAM design, a novel interpretable and configurable tree-based hierarchical search for memory-augmented neural network (MANN) is further demonstrated with the higher accuracy (95%) of mixed pattern recognition tasks and the lower search energy consumption compared with conventional CAM-based MANN, providing a promising approach for area- and energy-efficient data-centric in-memory-searching system.