Time-Domain Content-Addressable Memory Based on Single Ambipolar Ferroelectric Memcapacitor for High-Density and Highly-Precise Distance Function Computation

초록

Single ambipolar ferroelectric memcapacitor-based time-domain (TD) content-addressable memory (CAM) is proposed and experimentally demonstrated. The proposed TD CAM design effectively resolves the critical challenges of limited integration density and computational reliability in conventional ferroelectric memcapacitor-based capacitive CAMs. The band-reject-filter-shaped and symmetric capacitance-voltage characteristics with high dynamic range of a gated p-i-n diode-structured ferroelectric memcapacitor are leveraged. This CAM performs dual-edge search operations, where the Hamming distance (HD) between entry and query vectors is computed based on the modulation of the variable capacitance of cells. The propagation delay of the TD CAM output signal is linearly correlated with the computed HD, enabling improved search accuracy and sensing margin. The error-free classification of previously unseen classes in a five-way one-shot learning task indicates the feasibility of the proposed TD CAM as an associative memory within memory-augmented neural networks toward real-world implementations. Moreover, modeling results confirm that the proposed operation scheme maintains robustness against process variations and interconnect parasitics in massive arrays of highly scaled devices. Overall, the proposed TD CAM array offers exceptional compactness, linearity, and in-memory search reliability, considerably outperforming the conventional ferroelectric CAMs for HD computation.

키워드