Abstract
The authors find two types of input-driven bifurcation in the spin-torque oscillator, and they reveal the information processing capacity as a reservoir computing changes according to these bifurcations.
Highlights
Physical reservoir computing (PRC) is a method that originated from schemes of recurrent neural network training [1,2], and it exploits physical dynamics as computational resources [3]
We reveal that the memory capacity (MC) and information processing capacity (IPC) properties change completely depending on bifurcations, and there are singular phenomena of MC at certain input intervals that are multiples of the original oscillation period of the spin-torque oscillators (STOs) dynamics
We revealed two input-driven bifurcations in the STO and confirmed that the information processing capacity property changed drastically through these bifurcations
Summary
Physical reservoir computing (PRC) is a method that originated from schemes of recurrent neural network training [1,2], and it exploits physical dynamics as computational resources [3]. We will show that the information processing capability of the spintronics reservoir changes drastically according to these bifurcations. We analyze these bifurcations numerically using the Lyapunov exponent and echo state property (ESP) [2,34]. The dynamics possess (linear) memory capacity (MC) [35] and information processing capacity (IPC) [36], which provide comprehensive criteria for analyzing linear and nonlinear MCs. As a result, we reveal that the MC and IPC properties change completely depending on bifurcations, and there are singular phenomena of MC at certain input intervals that are multiples of the original oscillation period of the STO dynamics.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have