Granularity-Driven Management for Reliable and Efficient Skyrmion Racetrack Memories

Skyrmion racetrack memory is a rising star of nonvolatile memories due to its outstanding access performance and storage density. However, the working principle of skyrmion racetrack memory introduces unique issues, such as the position error and data representation problem, which considerably impact the data reliability. Although brilliant error correction codes have been proposed to detect and correct bit errors, however, their time-consuming encoding and decoding processes cannot fully match the nanosecond-level access latency of skyrmion racetrack memory. Observing the dilemma among data reliability, access performance, and space utilization, we propose a granularity-driven management scheme for skyrmion racetrack memory. While eliminating the errors incurred due to the position error and data representation problem, the proposed management scheme aims to jointly optimize access performance and space utilization. To achieve this goal, the proposed scheme adaptively selects different combinations of data encoding, layout, and indexing schemes for the data of different granularities. Moreover, we investigate the port selection problem under our proposed data layouts to minimize shift overheads on data accesses. Through analytical and experimental studies, the proposed management scheme is evaluated, and the obtained results are quite encouraging.