Speaker
Description
The field of magnonics, which utilises magnons for energy-efficient data processing, has made significant advances through inverse design. AI-based optimisation is emerging as a powerful tool in such efforts. Prior approaches have optimised the magnetic material itself, through geometry or saturation magnetisation landscapes, as well as reconfigurable scattering media, realising functionalities including RF filtering, logic operations, and neural-network tasks. Inverse design has also been applied to spin-wave excitation waveforms for targeted pulse generation. Here we demonstrate a fundamentally new approach: inverse-designing the coplanar waveguide transducer geometry. We apply this framework to form directional spin-wave beams, channeling energy without patterned waveguides, thereby reducing fabrication complexity.
The conductor profile is parameterised by piecewise cubic splines through a small set of normalised control points. Fabrication constraints, minimum feature sizes and gap tolerances, are encoded as bounds on these control points, so every candidate geometry explored by the optimiser is intrinsically compatible with lithographic limits. A gradient-based algorithm maximises excitation efficiency at a target wavenumber, sculpting the emission into a defined beam window in the yttrium iron garnet film.
The optimised transducer geometry enables spin-wave beam optics. Beams can be focused, defocused, or steered simply by varying the conductor profile, without additional waveguide patterning. Focused beams concentrate spin-wave intensity, directly lowering the power threshold for nonlinear interactions that drive wave-based computation. Furthermore, a single transducer optimised simultaneously at two frequencies produces two spatially separate beams, realising frequency-demultiplexing directly within the excitation structure.