Skyrmionics-aware and Compiler-centric Racetrack Architecture

• Applicants: Professor Dr.-Ing. Jeronimo Castrillon-Mazo; Professor Dr. Jian-Jia Chen
• Subject Area: Computer Architecture, Embedded and Massively Parallel Systems
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 565882679

Beyond the wide application of conventional volatile memory technologies, emerging non-volatile memories receive a bright focus in research and in practical applications already. Up to now, they promise higher densities, lower power consumption, and smaller access latency, beyond the capabilities of conventional memory. Specifically, technologies like ferroelectric field-effect transistors (FeFETs) and racetrack memories (RTMs) feature great potential to significantly improve system-level metrics. Compared to other NVMs, skyrmion-based racetrack memories (SK-RTMs) promise at least an order of magnitude better density with practically unlimited write-endurance. Moreover, like other NVMs, they have demonstrated great potential for implementing various logic and arithmetic operations in-place. On the down side, SK-RTMs are more prone to errors and, as other RTMs, require shift operations to access data sequentially. Shifts lead to latency and energy overheads and errors due to misalignment. Longer sequences of zero bits, represented by non-skyrmions, are an additional source of errors in SK-RTMs. While isolated hardware and software optimizations exists for SK-RTMs, there is no research on co-design and cross-layer optimization. In SKYWALKER, we will investigate bleeding edge SK-RTM-based systems and devise cross-layer optimizations. We will evaluate the impact of access patterns and data representation on SK-RTM-based systems. This collaborative project involves two German universities and four Taiwanese partners, each contributing expertise in orthogonal and complementary domains such as RTM devices and system designs, compiler optimizations, and runtime and system support. The research will focus on identifying and developing optimization strategies at different layers and exploring cross-layer optimizations to improve the SK-RTMs reliability, performance and energy consumption. For evaluation, our open-source simulation tools (gem5 and RTSim) will serve as foundational work and will be extended to support SK-RTMs.

• DFG Programme: Research Grants
• International Connection: Taiwan
• Partner Organisation: National Science and Technology Council (NSTC)
• Cooperation Partners: Professor Dr. Yuan-Hao Chang; Professor Dr. Meng-Fan Chang; Professor Shuo-Han Chen; Professor Po-Chun Huang