講演情報

[16a-K306-8]Logical Stochastic Resonance in a Spintronics Stochastic Neuron with Colored Noise

〇廖 智強、Siyi Tang1、Hiroyasu Yamahara1、Munetoshi Seki1、Hitoshi Tabata1 (1.東大工)

キーワード:

Logical computation、Logical stochastic resonance、Colored noise

Superparamagnetic tunnel junctions (SMTJs) are nanoscale devices with thermal-induced spontaneous oscillation, enabled by their ultra-low energy barriers. When the stochastic behavior is controlled, SMTJs can perform neuromorphic computing tasks. Additionally, similar to biological synapses, SMTJs can undergo subthreshold synchronization in the presence of noise. This property indicates that SMTJs, like the human brain, possess ultra-low power consumption and efficient information processing capabilities. Among SMTJ-based applications, logical computing, which is a higher-order function of the human brain, is indispensable for neuromorphic systems. While the synchronization of SMTJs under noise injection has been previously reported, there has been no investigation into their logical operation performance under realistic noise conditions.
To address this gap, we studied the reconfigurable logic operations of the SMTJ under Gaussian colored noise. In real-world systems, noise is often colored, characterized by an uneven energy distribution across frequencies. In this study, we investigated four colored noises, ranging from low-frequency dominance (red and pink) to high-frequency dominance (blue and violet), with white noise serving as a broadband baseline. We identified logic stochastic resonance (LSR) behavior, which refers to enhancing logical operation performance through noise and nonlinear system interplay. Our results reveal that white, blue, and violet noise significantly enhance the robustness of SMTJ-based logic gates under subthreshold driving conditions while simultaneously reducing power consumption. These high-frequency-dominant noise types enable SMTJs to switch between two different LSR states by adjusting the amplitude and bias of logic inputs. Conversely, low-frequency-dominant noise raises the energy threshold for reliable logic operations and induces only one type of LSR.