Spintronics for AI at the Intelligent Edge: From Non-Volatile Compute-in-Memory to Probabilistic Computing

The rapid growth of artificial intelligence is increasingly constrained not by arithmetic capability, but by the energy and latency required to move data between memory and processing units. This challenge is particularly acute at the intelligent edge, where battery capacity, thermal limits, connectivity and response time impose strict constraints. The next architectural shift must therefore bring computation closer to where data is stored and generated.

This presentation explores how spintronic compute-in-memory can help overcome the memory wall by combining non-volatile storage and computation within the same physical platform. Keeping weights in place and exploiting the intrinsic parallelism of magnetic devices could enable a new generation of compact, instant-on and energy-efficient intelligence, starting with autonomous drones, robots, industrial systems and connected mobility, and scaling towards increasingly complex and large-scale AI models.

Looking further ahead, the same spintronic platform may extend beyond deterministic AI inference towards probabilistic computing, optimisation, neuromorphic dynamics and quantum-adjacent technologies. Spintronics should therefore be viewed not simply as a new memory technology, but as a broader computational foundation for future autonomous and intelligent machines.