About

Please note the venue has changed to the Hilton London Euston, 17-18 Upper Woburn Place, London WC1H 0HT. This venue is a 10 minute walk from Kings Cross train station.

The energy costs of AI & data processing are spiralling unsustainably, and are predicted to reach 20.9% of global energy production by 2030. Training Chat-GPT 3 generated an estimated carbon footprint of 550 tons of CO2, & the training cost of a cutting-edge machine learning model is doubling around every 3.4 months.

This exponentially growing demand for energy represents a real threat to a zero-carbon future, and is caused in large part by inefficiencies in Von Neumann architecture CMOS hardware.

The maths powering modern neural network software are derived from physical models invented to describe ‘spin glasses’, strongly-interacting magnetic arrays. From the inception of neural nets, there has been a desire to ‘cut out the middleman’ of CMOS-hosted simulation and implement neuromorphic computation directly in physical magnetic systems, with the heavy-lifting of computing the massively-parallel interactions and energy minimisation offloaded onto the intrinsic system physics at zero energy cost - netting massive energy savings & new functionality.

Finally, this vision has come of age & research groups internationally are working on fresh, diverse approaches to implement AI & neuromorphic computing directly in magnetic hardware, with the potential to reduce the future energy & carbon footprint of AI. Magnetic systems have much to offer to the young field of neuromorphic computing, from skyrmions, spintronic oscillators & spin waves to artificial spin ices.

With this meeting we hope to provide a forum for the exchange and discussion of fresh approaches on how to marry dynamic magnetic systems with the complex demands of tomorrow’s computing. This workshop will bring together groups across magnetism for tutorials, talks & a collaborative sharing of ideas.