Orlando Moeira

GraiMatterLabs, The Netherlands

Neuronflow: a Neuromorphic Data Flow Architecture for Live AI at the Edge

Abstract

In this talk we will give an overview of GrAI Matter Labs’ Neuronflow many-core architecture, as implemented in GrAI One, our first chip. It is a neuromorphic data flow architecture that exploits brain-inspired concepts to deliver a scalable event-based processing engine for neuron networks in real-time applications at the edge. Its design is inspired by brain biology, but not necessarily biologically plausible. Neuronflow exploits structural, spatial and temporal sparsity to dramatically reduce latency and power consumption. It employs a data-flow model of execution for transport and synchronization, and threshold-based output activation to allow for partial input data processing. We will highlight some of the use cases running on GrAI One, some learnings, and discuss the future of neuromorphic processors.

Biography

Orlando Moreira is currently Fellow and Chief architect at GrAI Matter Labs, a neuromorphic chip design startup, where he is responsible for processor architecture and technology roadmap. From 2015 to August 2018 he lead the processor and compiler tools group for imaging and camera systems at Intel Corporation. He graduated in Electronics and Telecommunications Engineering from the University of Aveiro. He holds a PhD degree in Electrical Engineering from the Eindhoven University of Technology, with a thesis on real-time analysis and scheduling for streaming in embedded multiprocessors. He worked at Philips Research (2000-2007), NXP Semiconductors (2007-2008), ST-Ericsson/Ericsson (2008-2015). He has published research on reconfigurable computing, real-time multiprocessor scheduling and resource management, temporal analysis, neuromorphic computing, and software-defined radio.