Spiking Transformers for Neuromorphic AI
This project seeks to leverage neuromorphic computing and spiking neural networks to analyze scientific data at unprecedented rates and scales with lower costs and higher speeds than conventional AI while maintaining equivalent accuracy. Current AI models carry high operational costs, are computationally expensive and consume vast amounts of power. Neuromorphic processors, which mimic the operation of neurons in the human brain, consume from a thousand to ten thousand times less energy than existing CPUs or GPUs yet remain capable of performing the same parallel computations. This neuromorphic advantage, when used effectively, holds the potential to achieve AI-driven autonomous laboratories that dramatically advance U.S. scientific leadership and economic competitiveness at greatly reduced expense. These spiking neural networks will be integrated with the Genesis Mission and run on the American Science Cloud, with results shared for community use.
Partners
- Duke University
- University of Delaware
