Variation-Tolerant Ferroelectric FET-based Ternary Content-Addressable Memories (TCAM) Cell for Meta-learning Application

Ferroelectric FET (FeFET)-based ternary content-addressable memories (TCAM) have been proposed for memory augmented neural networks (MANNs). By sensing the discharging rate of the matchline (ML), FeFET-based TCAMs can get the Hamming distances between a search word and an array of stored words parallelly. In this work, we demonstrate that the accuracy of calculating the Hamming distance corrupts when considering the FeFET device variations. To solve this problem, we propose a 2FeFET-1R cell to reduce the variability of discharging currents of the TCAM cells. Adopting 2FeFET-1R cell, we demonstrate an error-free sensing can be achieved in an 8-bit TCAM array.