Hailo, an embedded-AI chipmaker, is utilizing non-Von Neumann data flow architecture for deep learning operations on the edge to enable efficient, cost-effective capabilities. The company’s Hailo-15 chip is suitable for embedding into cameras for the purpose of cost-efficiently offloading the more expensive work of cloud vision analytics, while also conserving power.
Recently, the company CEO, Orr Danon spoke to VentureBeat, discussing the different approach they are taking when compared to the preeminent player in the AI space, NVIDIA. Danon tell VentureBeat that their target is the embedded space and to optimize for power, rather than performance as NVIDIA looks towards targeting bigger, data center deployments.
The Hailo-15 chip handles incoming video signal robustly, factoring in image processing, encoding, and a heterogenous compute stack based on an ARM-licensed CPU. The ‘heavy lifting’ is accomplished through a neural net core and unlike NVIDIA, data is processed without trying to mimic a hard-coded neural network. In this way, problems and applications can more simply and efficiently map from the software level to the hardware, literally streaming the insights and not the video.
Hailo is dedicated to performing neural network efficiency and with deep learning operations taking center stage in the internet of things (IoT), cost-effective processing is more important than ever. AI chip technology is developing rapidly and Hailo is leading the charge in the embedded market.
The company was founded in 2017 and is comprised of a young and tech-savvy team that is passionate about developing cutting-edge technologies for the purpose of providing high-performance yet energy-efficient deep learning solutions. The Hailo-15 chip is a fantastic example of their innovative capabilities. They’ll be joining top executives in San Francisco on July 11-12 to discuss how leaders are inserting and optimizing AI investments for success.
