Memristive Logic-in-Memory Implementation with Area Efficiency and Parallelism

In-memory computing that performs logic operations inside the memory unit has emerged as an alternative to the traditional von Neumann architecture. In particular, logic-in-memory (LiM) architecture with memristor crossbars shows promising results in terms of power and performance, which operates the Boolean functions within the crossbar in parallel by utilizing memristor-aided logic (MAGIC) operations. In this work, we propose an efficient LUT mapping framework of Boolean logic functions into the memristor crossbar, which leverages the parallelism of logic computation to maximize performance and enhances area efficiency by optimizing crossbar utility. In detail, we map the LUTs of NoN (NOR-of-NOR) into the crossbar in a topological order in a way that maximizes the horizontal and vertical NOR operations. Moreover, we add an algorithmic rearranging stage for the intermediate results to more utilize the crossbar space during the next LUT placement. Our experiments prove that the proposed LUT mapping framework incorporates parallel operations with higher area efficiency, thereby achieving a 13.81 % speedup on average with a smaller crossbar dimension compared to the previous approach. We also successfully map the benchmark circuits with a large number of LUTs into the small-sized crossbar, in which the previous method has failed.