Tunable Stochasticity in an Artificial Spin Network

Dédalo Sanz-Hernández (1) , Maryam Massouras (2) , Nicolas Reyren (1) , Nicolas Rougemaille (3) , Vojtěch Schánilec (3) , Karim Bouzehouane (1) , Michel Hehn (2) , Benjamin Canals (4) , Damien Querlioz (5) , Julie Grollier (1) , François Montaigne (2) , Daniel Lacour (2)

1 Laboratoire Albert Fert (ex-UMPhy Unité mixte de physique CNRS/Thales)
2 IJL - Institut Jean Lamour
3 NEEL - MNM - Micro et NanoMagnétisme
4 NEEL - TMC - Théorie de la Matière Condensée
5 C2N - Centre de Nanosciences et de Nanotechnologies

Magnetic domain-walls travelling through a magnetic circuit 1 perform naturally and simultaneously logic and memory operations, eliminating the von Neumann information bottleneck 2. The motion of magnetic domain-walls along nanoscale tracks is thus promising to achieve high-speed, low-power and non-volatile information processing, and an extensive range of domain-wall-based logic architectures is being explored 3-6. Traditional domain-wall devices suppress intrinsic stochastic processes to enhance accuracy 7,8. Still, domain-wall stochasticity could be turned into an asset by using stochastic computing frameworks, such as Bayesian sensing 9 or random neural networks 10. These approaches however require controlling and tuning stochasticity. An iconic device used to illustrate the emergence of order from controlled randomness is the Galton board 11. In this device, multiple balls fall into an array of pegs to generate a bell-shaped curve that can be modified

Mots clés: Galton board; stochastic; domain-wall; nanomagnetic; neuromorphic; computing

Domaines: Physique [physics]; Matière Condensée [cond-mat]

Dédalo Sanz-Hernández, Maryam Massouras, Nicolas Reyren, Nicolas Rougemaille, Vojtěch Schánilec, et al.. Tunable Stochasticity in an Artificial Spin Network. Advanced Materials, 2021, 33 (17), pp.2008135. ⟨10.1002/adma.202008135⟩. ⟨hal-03001203⟩