From autonomous mobile robots to self-driving cars and industrial automation, the need for rapid and accurate video analytics at the network edge has risen. Foxconn Technology Group today said it is working with Socionext and Hailo to offer a next-generation artificial intelligence processing system.
Hailo Technologies Ltd. recently announced the Hailo-8 deep learning processor that will be part of the joint offering. The Tel Aviv, Israel-based company raised $60 million in Series B funding in March 2020.
Taipei, Taiwan-based Foxconn, formally known as Hon Hai Precision Industry Co., is a global leader in smart manufacturing. It is combining its high-density, fan-less BOXiedge edge-computing system with processors from its partners. Yokahama, Japan-based Socionext Inc., which provides system-on-a-chip (SoC) solutions for video and imaging, is contributing the SynQuacer SC2A11 high-efficiency parallel processor.
The companies said their new combination could benefit applications including smart cities, smart medical, smart retail, and the industrial Internet of Things (IIoT). The global market for AI will experience a compound annual growth rate of 28.5% between 2018 and 2023, approaching $98.4 billion in revenue, predicted research firm IDC.
Foxconn, Hailo, and Socionext said their joint offering addresses the need for cost-effective multiprocessing for video analytics, image classification, and object segmentation. The robust product can process and analyze more than 20 streaming camera input feeds in real time, all at the edge, they said. The high-density, low-power local video management system (VMS) server is designed for video analytics, including detection, pose estimation, and other AI-powered applications.
“Our vision at Foxconn is to pave the way for next generation AI solutions,” stated Gene Liu, vice president of the Semiconductor Subgroup at Foxconn. “We are confident that this strategic collaboration with our long-standing partner, Socionext, alongside Hailo, will do more than that. We recognize the great potential in adopting AI solutions for a multitude of applications, such as tumor detection and robotic navigation. This is why we are proud to say that our edge-computing solution, combined with Hailo’s deep-learning processor, will create even better energy efficiency for standalone AI inference nodes.”
Foxconn said it has already deployed several in-house developed AI solutions on different electronics production lines, leading to an improvement in reporting accuracy from 95% to 99% and a reduction of at least one-third of the operating costs for defect-inspection projects.
“We are very pleased with this joint effort by the companies and to officially announce our strategic partnership with Hailo,” said Noriaki Kubo, executive vice president at Socionext. “This collaboration will lead to more innovative solutions that specifically address the growing demand from our AI customers in multiple sectors. We are confident that this product will enable endpoint devices to operate with better performance, lower power, more flexibility, and minimal latency.”
Hailo’s specialized Hailo-8 deep learning processor delivers up to 26 Tera Operations Per Second (TOPS). The chip’s architecture is designed to enable edge devices to run sophisticated deep learning applications that could previously only run on the cloud, said the company. This translates into higher performance, lower power, and minimal latency, enabling enhanced privacy and better reliability for edge devices, the company said.
“We are thrilled to announce our collaboration with two of the global leaders in AI solutions,” said Orr Danon, co-founder and CEO of Hailo. “Our deep learning processor significantly upgrades the capabilities of smart devices operating at the edge, and this collaboration will impact a wide range of industries increasingly driven by edge technology.”
“It took two years from inception to a fully functional processor in silicon,” said Avi Baum, chief technology officer at Hailo. “We have more than 10 patents pending in structure-defined data-flow architecture, and 80 employees with extensive experience.”
“General-purpose architectures have evolved from low-power CPUs and MCUs to server-class GPUs and CPUs, but they’re very costly,” he told The Robot Report. “Deep-learning applications such as pedestrian detection, collision avoidance, and quality inspection require more domain-specific processors for neural network inference.”
“We didn’t want to go down the path of adapting general-purpose processors,” said Baum. “Instead, we wanted to rebuild computing ingredients back from the basics into one architecture, with a tight understanding of memory, computing, controls, and interconnections among them.”
“The workload that neural networks represent are very different than what traditional computing devices assume, which is highly controlled,” he explained. “Rather than deciding cycle by cycle what to do next, there is no need for a big chunk of centralized memory, since the control fabric is very thin, and most of it is predetermined.”
Foxconn said the next generation of its BOXiedge, along with its partners’ AI processors, is intended for a wide range of applications relying on low latency, a high data rate, high reliability, and quick processing at the edge.
For example, smart retailers and smart cities require hundreds of cameras — either in-store or in traffic monitoring — to generate video streams that need to be processed locally, quickly, and efficiently with minimal latency, said the company. Similarly, for IIoT, acquiring, processing, inferencing, and presenting data on the production floor rather than in the cloud can result in significant cost savings, as well as more efficient processing for tasks such as inspection and quality assurance, said Foxconn.
“Lots of players were providing enough compute capacity within a reasonable performance envelope, but capable processors were very costly and not capable of being mounted in mobile devices,” Baum noted. “For example, a delivery robot requires safe autonomous navigation similar to that of an autonomous vehicle, but you don’t want to sacrifice accuracy to be within the cost and power envelope.”
Hailo is more focused on building an enabling technology than on a single application, and it is working with ABB, among other robotics vendors, said Baum. However, because Hailo-8 is self-contained and can provide high availability and functional safety in harsh environments, it conforms with automotive industry standards, he added.
“In a 10-by-20-cm or 4-by-6-in. card, we can cram everything that runs on a server-class device,” he added. “We see a lot of traction in ADAS [advanced driver-assistance systems], which have already deployed AI processors. Safety regulations are requiring OEMs to make sure that cars can detect pedestrians from greater distances and at higher speeds, so they need higher-resolution image processing at lower cost.”
“Response from the market has been very positive, and we’re moving from prototypes and samples to mass production,” Baum said. “We’re definitely hiring and are looking for strategic partners and customers worldwide.”