Spintronics-based Neuromorphic Computing

Reservoir Computing (RC) harnesses the transient dynamics of nonlinear systems for temporal information processing. Physical Reservoir Computing (PRC) extends this concept to real physical media, exploiting intrinsic material nonlinearities for efficient unconventional computation. This thesis advances PRC through magnetic systems, connecting material physics, spectral dynamics, and hardware implementation to establish magnetism-driven computation as a scalable and energy-efficient paradigm. The study begins with a chiral magnet, Cu2OSeO3, where the spin-wave spectra evolve across helical, conical, and skyrmion phases. By applying field-cycling protocols, metastable skyrmion states are created at T ∼ 4 K, exhibiting historydependent and reconfigurable responses suitable for adaptive computation. To realise operation under ambient conditions, Yttrium Iron Garnet (YIG) is then employed as a room-temperature ferrimagnet with flexible spin-wave dynamics, achieving real-time signal classification accuracy comparable to algorithmic machine learning. Building on these findings, the work systematically compares the reservoir performance of several magnetic materials, including Cu2OSeO3, FeGe, YIG, and Co8Zn8Mn8.5, across a wide temperature range. Quantitative metrics such as nonlinearity, memory capacity, and signal-to-noise ratio are introduced to link spectral dynamics with computational performance, forming a practical framework for selecting efficient magnetic reservoirs. To overcome the limitations of field-driven systems, a field-free PRC platform based on synthetic antiferromagnets (SyAF) is developed. Using second-harmonic detection and time-multiplexed processing, the system operates with only milliampere-level currents and an energy cost of about 155 µJ per operation, demonstrating compact and low-power functionality. Finally, leveraging the electrical accessibility of the SyAF reservoir, the study explores its integration with microcontrollers for autonomous, real-time computation. Together, these developments trace a complete pathway from material discovery to embedded hardware, paving the way for scalable and adaptive physical intelligence.