Presentation Information
[16a-D61-7]Reservoir Computing Utilizing Transient Dynamics of Spin-Hall Nano-Oscillators
〇Aakanksha Sud2,1, Akash Kumar3,6, Maha Khademi7, Johan Akerman1,6, Shunsuke Fukami1,3,4,5 (1.RIEC,Tohoku Univ., 2.FRIS,Tohoku Univ., 3.CSIS,Tohoku Univ., 4.CIES,Tohoku Univ., 5.WPI-AIMR, 6.Univ. of Gothenburg, 7.Chalmers Univ.)
Keywords:
Reservoir computing,spin hall nano-oscillators
Reservoir computing (RC) is an emerging computational paradigm that leverages the transient dynamics of physical systems to perform complex tasks such as pattern recognition and time-series prediction. In this work, we explore the use of Spin Hall nano-oscillators (SHNOs) as a novel medium for RC. Spin Hall nano-oscillators are nanoscale devices that exploit the Spin Hall Effect to generate and manipulate magnetization oscillations, which can be harnessed for computational purposes.
We present an experimental demonstration of a Physical Reservoir Computing scheme employing a SHNO. By utilizing the dynamic states of the SHNO, we tackle tasks such as waveform classification and prediction, achieving accuracies comparable to state-of-the-art neural networks.Our experiments involve a detailed analysis of the magnetization dynamics, and we pinpoint the specific regime that yields optimal performance. This regime is characterized by a delicate balance between non-linearity and memory capacity, essential for effective reservoir computing. We specify these regimes based on measurements conducted at different magnetic fields, highlighting the influence of magnetic field strength on SHNOs' computational capabilities. Metric calculations are employed to assess the computational effectiveness of our approach. The implications of our findings are significant: they lay the groundwork for developing swift, parallel, and energy-efficient computing systems based on oscillator networks.
We present an experimental demonstration of a Physical Reservoir Computing scheme employing a SHNO. By utilizing the dynamic states of the SHNO, we tackle tasks such as waveform classification and prediction, achieving accuracies comparable to state-of-the-art neural networks.Our experiments involve a detailed analysis of the magnetization dynamics, and we pinpoint the specific regime that yields optimal performance. This regime is characterized by a delicate balance between non-linearity and memory capacity, essential for effective reservoir computing. We specify these regimes based on measurements conducted at different magnetic fields, highlighting the influence of magnetic field strength on SHNOs' computational capabilities. Metric calculations are employed to assess the computational effectiveness of our approach. The implications of our findings are significant: they lay the groundwork for developing swift, parallel, and energy-efficient computing systems based on oscillator networks.
Comment
To browse or post comments, you must log in.Log in