2D Magnetic Material for Sustainable AI

About

The team led by Professors Deblina Sarkar (MIT) and Ralf Herbrich (HPI) is developing a new class of energy-efficient hardware for probabilistic computing, inspired by the stochastic nature of 2D magnetic materials. Addressing the growing energy demands and architectural limitations of conventional AI systems, the project explores how spontaneous switching behavior in magnetic tunnel junctions can serve as a physical source of randomness, enabling low-power, hardware-native approaches to sampling and inference.

By combining these magnetic devices with digital logic, the team has designed hybrid architectures tailored for uncertainty-aware AI tasks such as Bayesian learning and Monte Carlo simulations. Building on their prior theoretical and FPGA-based work, the next phase focuses on fabricating nanostructured devices, expanding to more complex distributions, and applying the technology to real-world inference problems. Ultimately, the project aims to demonstrate a new paradigm of sustainable, stochastic hardware aligned with the future of AI and climate-conscious computing.