AI-Driven Discovery and Design of Spintronic Hardware for Probabilistic and Neuromorphic Computing

Probabilistic (p-bit) and oscillator-based computing are leading routes to energy-efficient AI and optimisation hardware, both built from nanomagnetic devices whose figures of merit emerge only from finite-temperature micromagnetic dynamics. This project designs such hardware by large-scale GPU simulation driven by an autonomous AI research methodology: AI agents orchestrate hypothesis-free regime scans across the device design space, automatically flag and falsify anomalies, and adversarially verify every design-relevant claim, with fabrication and measurement in the loop at MPI Halle. Two device families are targeted - stochastic magnetic tunnel junctions (p-bits) and domain-wall / spintronic oscillators (oscillator-Ising and neuromorphic primitives) - across multiple stack compositions, device-to-device variability ensembles and temperature dependence, delivering quantitative variability-aware design maps and candidate stacks plus an open, reproducible GPU simulation engine. The team requests 20,000 node-hours on the LUMI-G partition (AMD MI250X) over 6 months.