Deep learning transcends the edges of our imagination
The explosive growth in mobile connectivity – from cell phones and tablets to the expanding internet of things (IoT) and Industry 4.0 – is driving demand for increased speed, decreased latency and power consumption, and enhanced functionality of connected devices. To meet this demand, embedded sensors and edge computing (processing close to the edge of the network rather than centrally) are playing an increasingly important role relative to centralised data processing and cloud-based services. The EU-funded Hailo-8 project is preparing a step change in edge computing. Its pioneering embedded Hailo-8™ artificial intelligence (AI) processor brings deep learning to AI-based edge devices with a focus on processing of vision sensor data.
Deep learning, broad reach
Deep learning leverages a brain-inspired, multilayered, artificial neural network architecture that, like the brain, can learn without a priori rules, hardwired instructions or human supervision. Further, it can deduce structure in raw data. Hailo-8™’s AI hardware architecture unleashes the power of deep learning to support edge devices in advanced applications. The processor delivers unparalleled AI performance with low power consumption and compact size. Extremely high processing resolution harnesses the full potential of advanced sensors. The processor’s fully programmable AI accelerator chip and its comprehensive software development kit support numerous neural network types, enhancing the flexibility of programming. The software seamlessly fuses with existing machine language development frameworks to streamline integration in products. Avi Baum, chief technology officer at Israeli SME Hailo Technologies and project coordinator, explains the potential: “As an embedded AI processor, the Hailo-8™ is a perfect fit for a wide variety of embedded computing platforms ranging from automotive and heavy industries to consumer electronics. Given its automotive and industrial grade qualifications and functional safety attributes designed into the product to address automotive needs, it is uniquely positioned for industrial applications in an IoT-4.0 context as well as smart mobility, public safety and other IoT applications.”
Heading to the edge
The Hailo-8 project focused on two aspects. On the engineering side, the team built the capacity for mass production and design for testability and manufacturability. On the business side, the company targeted product promotion and customer engagement. The project launched an early access evaluation programme to gain market traction. It attracted more than a dozen potential customers seeking early access to the technology, including several automotive suppliers and original equipment manufacturers. In fact, Baum reflects, “the overwhelming interest from new markets that we were not directly targeting was quite exciting and led to many more business opportunities than we originally anticipated. With the Hailo-8™ AI processor now in volume production and ready for deployment, we already have over 30 potential customers in the sales pipeline.” Hailo-8™’s comprehensive and flexible software will facilitate ease of use and rapid deployment. Baum concludes: “Not only have we lowered the barrier for introducing deep learning to embedded platforms, but we have also delivered the most efficient AI processor for edge devices available today. Its unparalleled processing capabilities will unleash the awesome potential of edge computing.”
Keywords
Hailo-8, embedded, AI processor, deep learning, automotive, edge computing, edge devices, IoT, neural network