Magnetic Architectures for Reservoir Computing Hardware (MARCH)

Lead Research Organisation: University of Sheffield
Department Name: Materials Science and Engineering


Classical digital computing is power hungry, fragile, and hard to interface to the analogue real world.

Unconventional computers such as in materio Reservoir Computers (RCs) can help overcome these issues, particularly by being able to perform embodied computation that can directly exploit the natural dynamics of their material composition, thereby dramatically reducing the power requirements. Such devices have been proven in principle, but systems need to be scaled up to provide sufficient and appropriate computational power for real world tasks. Furthermore, a range of device configurations are needed to support different computational tasks.

Our aim is to combine multiple material RCs:
1. with similar instance configurations, to provide scalability
2. with diverse instance configurations, to perform together tasks that no single RC configuration can readily perform alone.

Magnetic materials provide an excellent flexible testbed for developing a material RC design process. Such materials have the intrinsic memory and complex non-linear dynamics needed for RC operation,
and also have well-established methods of interfacing for data input/output, needed to build a practical device. We will exploit the properties of patterned 2D layouts of magnetic nanoring wires, which can be readily manufactured with existing technologies. This flexible design and experimental platform will allow us to develop generic techniques that will apply across a range of smart material RCs.

We will develop new multi-RC design tools, and will design, test and manufacture nanomagnetic RCs. This toolset and manufacture will provide a robust and reliable testbed on which we will develop and evaluate scalable RC architectures that can be configured for a range of practical computing tasks. We will demonstrate the resulting multi-reservoir architecture in an audio-controlled robot: a `turtle' following simple LOGO-like commands, controlled purely by in materio reservoirs, with no onboard digital computers.

The output will be a new design methodology and platform for multi-reservoir devices, that can be exploited to design low-power, robust, flexible, and efficient smart sensing and other `edge-computing' devices in a diverse range of materials.


10 25 50