Recent Articles
Due to the pandemic a couple of my projects ended up being a little later than planned so in the last few months we have had a couple more articles published on our neuromorphic and reservoir computing work. The first paper was published in Advanced Functional Materials on the emergent dynamics in a set of magnetic rings that we found could be used for reservoir computing. The experimental work was performed by researchers in the Department of Materials Science who manufactured and measured the properties of a grid of interconnected rings. The domain walls in these rings become pinned at the interfaces and can interact causing the emergent dynamics. As a reservoir it has a range of important properties (fading memory, non-linearity, etc) and we found excellent performance on a speech recognition task. The paper is available here.
The next paper was an invited paper in Applied Physics Letters for a special topic on Mesoscopic Magnetic Systems in which we introduced a novel reservoir based on thermally activated magnetic dots. While it might appear that the stochastic behaviour of these dots do not have much to offer, we found by applying a voltage controlled anisotropy we can tune the relaxation rates from one orientiation to another. Importantly this system is low energy so could find use in edge-computing systems where energy is limited. You can find the full article here.
Most recently we have puclished our work on a non-linear domain wall oscillator as a physical reservoir. This was again in collaboration with Materials Science and we built a 1D model of the domain wall motion when it is pinned between two anti-notches. The pinning potential exhibits chaotic behaviour for fields upto 1.2 KA/m and a non-linear response beyond this. From this we could train it using the reservoir computing approach and found good performance on a set of tasks. You can read more here.