Testing ReRAM-based TCAM for Computation-in-Memory Applications
Haneen G. Hezayyin, Mahta Mayahinia, Mehdi Baradaran Tahoori · 2024
Computation in memory (CiM) can potentially mitigate the high energy and latency associated with the back-and-forth data transfer between the processor core and memory sub-systems. This bottleneck is known as the memory wall problem and is further pressed by data-intensive applications. Measuring the similarity between two binary patterns is a data-intensive, highly frequent kernel in many modern applications that can be accelerated using the CiM concept. Content addressable memory (CAM) and its ternary version (TCAM) are the specific CiM blocks for the task of binary pattern similarity measurements. The performance and energy efficiency of the TCAM can be further improved by utilizing the emerging resistive non-volatile memory (NVM). Among different NVM technologies, redox-based RAM (ReRAM) is a promising candidate to be implemented within the NVM-based TCAM, mainly due to its highly distinct resistive levels. Besides the performance and energy benefits of the ReRAM-based TCAM (ReTCAM), its high functional quality is of utmost importance and needs to be carefully taken into account. More specifically, the co-integration of the ReRAM and transistor technologies results in new fault behaviors. To assure the high quality of the ReTCAM, in this paper, we aim to develop a March-like test algorithm tailored for ReTCAM array. The proposed March-like test algorithm is extendable and can achieve 100% fault coverage.