Charge-domain content addressable memory based on ferroelectric capacitive memory for reliable and energy-efficient one-shot learning.

Abstract

Non-volatile content addressable memories (NV-CAMs) accelerate memory augmented neural networks (MANNs) for brain-like efficient learning from a few examples or even one example. However, most existing NV-CAMs operate in current domain, posing challenges in reliable, low-power, and sensing-friendly Hamming distance (HD) computation. To address these challenges, this work proposes transferring the computation to charge domain using ferroelectric capacitive memory (FCM). For the first time, a charge-domain 2FCM CAM based on the inversion-type FCM is reported. By storing data as device capacitance, this CAM structure directly outputs HD as linear multi-level voltages, enabling simplified sensing processes and reduced peripheral costs. Its differential nature further exhibits immunity to device variation, ensuring accuracy in the computation of long data vectors. Parallel 16-bit HD computation using a fabricated 16 × 16 2FCM CAM array is experimentally demonstrated with record performance at array level, evidencing the superiority of charge-domain computation and showcasing tremendous potential for in-memory-search applications.