Combination-Encoding Content-Addressable Memory With High Content Density

Abstract

Recently, resistance switch-based content-addressable memory (RCAM) has been proposed as an alternative to the mainstream static random-access memory-based CAM because of its high integration potential and low static energy consumption. However, RCAM has a lower data density due to the use of a pair of resistance switches for a single bit of contents (0.5 bit/switch) than resistive random access memory (1 bit/switch). In this paper, we propose a new type of RCAM referred to as combination-encoding CAM (CECAM). In the N-CECAM, a single unit consists of N high and N low resistance state switches whose combination collectively represents binary contents, yielding a data density of approximately 0.85 bit/switch when N = 10, for instance. The key to the CECAM is the encoding of an n-bit search key as a 2N-digit key and its decoding. To this end, we propose a simple algorithm for encoding and decoding and its implementation in digital circuitry.

Experimental demonstration of combination-encoding content-addressable memory of 0.75 bits per switch utilizing Hf–Zr–O ferroelectric tunnel junctions

“…TCAM leverages its fast search capability for emerging processors like neuromorphic event processors (such as an artificial afferent nerve system, 7 artificial somatic reflex arc, [8][9][10] and an optical convolution processing unit 11 ) in which content-search speed for event routing dictates their key performance (synaptic operations). 12,13 Furthermore, machine learning (ML) techniques are often based on similarity measure operations; for instance, convolution operations in convolutional neural networks effectively measure the similarity between a given kernel and a feature map. Recommender systems based on graph neural networks use the measure of similarity between a given key and data.…”

“…The key requirement is an algorithm for encoding a w-bit word to a 2N-long binary array with N 0s and N 1s. To this end, we utilize the combination-encoding algorithm 12 that encodes w-bit words to a 2N-long binary array with N 0s and N 1s, satisfying…”

“…The key requirement is an algorithm for encoding a w -bit word to a 2 N -long binary array with N 0s and N 1s. To this end, we utilize the combination-encoding algorithm 12 that encodes w -bit words to a 2 N -long binary array with N 0s and N 1s, satisfying w = ⌊log 2 C (2 N , N )⌋,where ⌊·⌋ and C (2 N , N ) denote a floor function and 2 N choose N formula, respectively. Notably, eqn (1) is a supralinear function of N , and thus it underpins a larger content density than 0.5 bit per switch.…”

Memristor Crossbar Circuit for Ternary Content‐Addressable Memory with Fine‐Tuning Operation

“…Typically, memristorbased CAMs are configured as 2T2R or nT2R with selector devices or additional gate-connected transistors to improve read margin. [17][18][19][20][21][22][23][24] It is believed that a passive memristor crossbar circuit can also be utilized for CAM applications if the intrinsic nonideal problems of passive crossbar including program errors, line resistance, and sneak current can be sufficiently suppressed. However, few studies on CAM technology with passive crossbar array have been investigated although passive memristor crossbar circuit is considered advantageous for high-density integration compared with active array based on NOR flash array architecture.…”

“…Notably, memristor crossbar circuit offers the principal advantage of inherent vector-by-matrix multiplication (VMM) operation, which enables physical-level in-memory computing using Ohm's and Kirchhoff 's current laws and can be applied for various in-memory computing applications such as neuromorphic engineering, [1][2][3][4][5][6][7][8][9] physical unclonable function, [10][11][12][13][14][15][16] and content-addressable memory (CAM). [17][18][19][20][21][22][23][24][25][26][27] A CAM is a type of memory that receives an input vector, compares it with all stored vectors, and returns a related vector output. Generally, CAM is classified into two types: binary CAM (BCAM) having two states of "0" or "1", and ternary CAM (TCAM) having three states with the addition of "X" state (i.e., "do not care").…”