Author and Affiliation: Orlando Moreira, Neural Processing Core Chief Architect, Snap Inc.

Bio: Orlando Moreira is the Neural Processing Core Chief Architect at Snap Inc. His expertise encompasses computer architecture, edge AI, embedded systems, real-time systems, and data flow methodologies. Before Snap Inc., he held the position of Chief Architect at GrAI Matter Labs, where he was responsible for the computer architecture definition and specification, and for the roadmaps for computer architecture and software development kit (SDK).

Before, he worked at Philips Research, ST-Ericsson, Ericsson, and Intel, where he was group leader for imaging programming and core tools - compiler, simulator, debugger and hw generation.

Throughout his career, he has contributed peer-reviewed work to the fields of embedded computer architecture and techniques for formal modeling and temporal analysis of multiprocessor real-time systems. He published a book with Springer titled "Scheduling real-time streaming applications onto an embedded multiprocessor".

He holds a PhD in Electrical Engineering from the University of Eindhoven.

Title: Neuronflow, an event-driven architecture for Augmented Reality AI
Abstract: We provide an overview of the GrAIcore Neural Processing Unit (NPU) developed at Snap Inc. for Augmented Reality applications, emphasizing the role of sparsity in achieving significant power and performance improvements compared with traditional architectures. We propose the adoption of an event-driven execution model to exploit the benefits of sparsity. Additionally, the talk will explore advanced training optimization techniques that enhance neuron activation sparsity, enabling the potential of the event-driven multicore architecture to be realized, resulting in dramatic gains in power and latency in the inference of vision networks. If time allows for, we will also demonstrate Snap's See-through AR Spectacles.

(image: Jazi Araújo / Pixbay)