NEURO-MAPP: Neuromorphic shortest path runs faster, uses less energy than CPU

Efficient shortest path computation in weighted graphs is fundamental to routing, network analysis, and AI. Neuromorphic hardware promises massive parallelism and low power, but traditional algorithms like Dijkstra don't exploit its architecture. Researchers from several institutions now present NEURO-MAPP (Neuromorphic-based Min-Add Parallel Propagation), a distributed algorithm designed for neuromorphic systems that uses local processing and spike-based communication. They implement it on the SpiNNaker 2 platform, a digital neuromorphic chip with 2 million neurons.

Testing on synthetic and real-world graphs, NEURO-MAPP on SpiNNaker 2 scales favorably in runtime for many graph structures compared to Dijkstra on a modern CPU, and consumes less energy per shortest-path query in nearly all cases. This demonstrates that neuromorphic hardware with sparse, event-driven communication can efficiently solve classical graph problems, opening the door to low-power graph processors for applications in logistics, social networks, and biological neural simulations.