Abstract
The demonstrated and expected advantages of neuromorphic over traditional von Neumann architectures is spurring a growing interest in using them for a variety of applications. While obviously well-suited for cognitive applications, there is a growing body of research highlighting the advantages of neuromorphic implementations of non-cognitive applications as well. Especially for applications where scalability is a concern, neuromorphic computing provides a promising solution. Here, we introduce a spiking neuromorphic algorithm that calculates the state-value function of a Markov reward process leveraging the parallel and event-driven nature of neuromorphic archi-tectures. We analyze the complexity of scaling the algorithm and provide some preliminary single-chip scaling results from implementations on the Loihi 2 and SpiNNaker 2 neuromorphic architectures.