Inspired by the nonlinear dynamics of neural networks, new unconventional computing hardware has emerged under the name of physical reservoir computing. In this paradigm, an input-driven dynamical system (the reservoir) is exploited and trained to perform computational tasks. Recent spintronic thin-film reservoirs show state-of-the-art performances despite simplicity in their design. Here, we explore film geometry and show that simple changes to film shape and input location can lead to greater memory and improved performance across various time-series tasks.
@inproceedings(Dale++:2021-UCNC, author = "Matthew Dale and Simon O’Keefe and Angelika Sebald and Susan Stepney and Martin A. Trefzer", title = "Computing with Magnetic Thin Films: Using Film Geometry to Improve Dynamics", pages = "19-34", doi = "10.1007/978-3-030-87993-8_2", crossref = "UCNC-2021" ) @proceedings(UCNC-2021, title = "UCNC 2021 Espoo, Finland, October 2021", booktitle = "UCNC 2021 Espoo, Finland, October 2021", series = "LNCS", volume = 12984, publisher = "Springer", year = 2021 )