Precharge free dynamic content addressable memory

Abstract

A precharge free dynamic content addressable memory (DCAM) is introduced for low-power and high-speed search applications. Elimination of precharge prior to search allows hardware engine to perform more number of searches within the stipulated time. The proposed DCAM cell not only removes precharge of matchline (ML) but also utilises decoupling of bitline and searchline so that unwanted capacitive couplings are minimised at charge storage nodes. A 512 bit of the proposed scheme is implemented using 45 nm CMOS technology and its efficacy is verified and proved through rigorous variations with 1000-point Monte-Carlo sampling of ML voltage as well as multi-search dissipation analysis.

The analogy of matchline sensing techniques for content addressable memory (CAM)

“…Segmented ML technique proved to be the best to handle all the performance metrics. Table 2 Performance comparison summary of ML-sensing techniques Precharge high [6,14,15] Current-race scheme [22,33] Precharge free [29][30][31][32] Low swing [11,[34][35][36] Segmented [37][38][39][40][41][42][43][44][45] Sele. precharge [46][47][48]…”

“…Most efforts have been made in reducing the ML switching power either by lowering the charging current or ML swing and few on enhancing the speed of ML evaluation [28]. Later designs have also introduced pre-charge free structures to invoke an increased number of searches in the same bandwidth [29][30][31][32]. Segmenting the MLs into multiple sections and performing/blocking of evaluation in the partitions separately with different ML structures can save ML power significantly.…”

“…(ii) Current race sensing [22,33]: a pre-charge low strategy on the ML using a current source. (iii) Pre-charge free ML sensing [29][30][31][32]: elimination of extra precharge phase to accommodate more number of searches. (iv) Low-swing scheme [11,[34][35][36]: low-swing ML approaches to reduce dynamic power.…”

“…Architectures without pre-charge logic are proposed recently [29][30][31][32] to increase the search rate. These schemes have used external [29] and self-control [30] voltage to avoid the ML precharging.…”

“…Segmenting the MLs into multiple sections and performing/blocking of evaluation in the partitions separately with different ML structures can save ML power significantly. All these techniques are categorised into six unique types as follows: (i) Pre‐charge high ML sensing technique [6, 14, 15]: a simple yet power hungry approach used by most CAM designers. (ii) Current race sensing [22, 33]: a pre‐charge low strategy on the ML using a current source. (iii) Pre‐charge free ML sensing [29–32]: elimination of extra pre‐charge phase to accommodate more number of searches. (iv) Low‐swing scheme [11, 34–36]: low‐swing ML approaches to reduce dynamic power. (v) Segmented ML evaluation [37–45]: a non‐uniform distribution of ML evaluation results by segmenting ML and designing sections with different lengths and structures to improve the power‐delay trade‐off. (vi) Selective pre‐charge sensing scheme [46–48]: another non‐uniform power distribution among the MLs based on the higher mismatch probability in initial sections. …”