In a recent groundbreaking discovery, researchers from CEA-LETI Université Grenoble Alpes, University of Zurich, and ETH Zurich have unveiled a revolutionary approach to neural network circuit architectures. Their technical paper, titled “Mosaic: in-memory computing and routing for small-world spike-based neuromorphic systems,” presents an innovative systolic architecture called Mosaic that leverages distributed memristors for efficient information routing and computation within Spiking Neural Networks (SNNs).
Drawing inspiration from the connectivity of the human brain, the researchers embrace the concept of small-world graphs, characterized by local density and global sparsity. This principle, evident in various species’ evolutionary processes, holds the key to robust and efficient information routing. However, current artificial neural network structures fail to fully incorporate the benefits of small-world neural network models.
Mosaic bridges this gap by introducing a novel non-von Neumann architecture that combines in-memory computing and in-memory routing using distributed memristors. By implementing small-world graph topologies, Mosaic takes advantage of the power of SNNs to achieve unparalleled routing efficiency. The team’s design, fabrication, and experimental demonstrations of Mosaic’s components exhibit remarkable results, utilizing integrated memristors with cutting-edge 130 nm CMOS technology.
One of the most significant findings from their research is the outstanding routing efficiency achieved by Mosaic compared to existing SNN hardware platforms. The enforced locality in connectivity ensures that Mosaic surpasses others by at least an order of magnitude in terms of routing efficiency. Remarkably, while excelling in routing, Mosaic also delivers competitive accuracy across a range of edge benchmarks.
Furthermore, Mosaic offers a scalable approach for edge systems, harnessing the power of distributed spike-based computing and in-memory routing. With its potential to revolutionize information routing in neuromorphic systems, Mosaic paves the way for more efficient and powerful AI applications.
As the field of AI rapidly progresses, the research conducted by these visionary scientists points towards a new era of neural network circuit architectures. By embracing the principles of small-world neural networks, Mosaic sets the stage for improved efficiency and performance in information routing, pushing the boundaries of AI and machine learning capabilities.
The source of the article is from the blog publicsectortravel.org.uk