Computing with magnetic whirls

The rapid growth of AI exposes fundamental limitations in the power efficiency and scalability of conventional computing systems. In this talk, I will introduce physical reservoir computing (PRC) – a promising paradigm that harnesses the intrinsic dynamics of physical systems to transform complex computational tasks into simpler ones. Using simulations, we show that magnetic textures, such as skyrmion fabrics, provide a highly competitive, ultra-low-power physical implementation of the reservoir – the core component of PRC. We developed task-agnostic metrics to benchmark key reservoir properties: non-linearity, complexity, and fading memory. In spoken digit classification, our multi-input skyrmion reservoir achieves the highest reported performance among in-materio systems, rivaling digital counterparts. Experimental realizations further underscore the potential of skyrmion-based PRC for energy-efficient, high-performance computing. Future directions include combining complementary reservoirs or integrating different physical mechanisms to boost computational power and throughput.