Natural Ising Machines
Published:
Natural Ising Machines
We investigate natural Ising machines, where physical systems are used to solve computational problems. Specifically, we explore how magnetic frustration in artificial spin-ice lattices and optical interference patterns can map to the Ising Hamiltonian.
By fabricating nanoscale magnetic islands in specific geometries (e.g., kagome, square ice) and coupling them to free-space optical systems, we aim to utilize inherent parallelism for optimization tasks.
Engineering Frustrated Magnetism
Plasmonic Spin-Ice Metasurfaces ― We utilize nickel nano-island arrays to create tunable magnetic frustration with >5× magneto-optic enhancement. These systems allow for the observation of computational phase transitions and the exploration of energy landscapes.
Ultrafast Magnetic Imaging ― Using laser-illuminated photoemission electron microscopy (PEEM), we have captured femtosecond-resolution videos of magnetic vertices flipping in frustrated lattices, providing insight into dynamics at terahertz rates.
Optical Solver for Optimization ― We have developed a free-space machine that encodes spin couplings into spatial light patterns and reads out ground states through Fourier optics. This system addresses fully-connected Ising problems, such as Max-Cut, at room temperature using spatial light modulators.
Deep Learning and Magnetism ― We employ generative models to analyze spin configurations and discover physical relationships governing magnetic frustration from microscopy images.
Significance
This research demonstrates the potential of natural analogue computers to perform optimization tasks efficiently. By harnessing physical interactions, we aim to develop energy-efficient and scalable computing architectures.