Hailo-8™: The Penny-sized Data Center on Wheels for ADAS
Interview with Orr Danon, CEO of Hailo
2021년 07월호 지면기사  / written by Han



Vehicles utilize the power of a full ‘data center on the wheels’ With the penny-sized Hailo AI chip. Hailo-8™ significantly improves cognitive sensor’s capabilities in ADAS with an innovative architecture that enables vehicles to run state-of-the-art deep learning applications that previously could not be run on the power envelope of automotive ECUs.
This is a short interview with Orr Danon, CEO of Hailo.


written by Han_han@autoelectronics.co.kr
한글버전






Q. Hailo was founded in 2017. Please tell me the major milestones. The announcement of the Hailo-8 chip and the ASIL-B and AEC-Q100 certifications sound like saying that Hailo has targeted and have capability in the automotive market from the start. right?

A. In 2019, Hailo released the Hailo-8™, the world’s top performing Edge AI processor. The Hailo-8™ processor, which features up to 26 tera operations per second(TOPS), significantly outperforms all other edge processors with area and power efficiency far superior to other leading solutions by a considerable order of magnitude - all at a size smaller than a penny, including the required memory. Hailo has achieved many milestones since its founding. In 2020, the company announced a $60 million Series B funding round, with key strategic investors participating including ABB Technology Ventures (ATV) - the strategic VC arm of ABB, a global leader in industrial automation and robotics; NEC Corporation - a leader in the integration of IT and network technologies; and London-based VC firm Latitude Ventures. This brings Hailo’s total funding to $88 million.

The automotive market has always been a key target market for Hailo, which designed the Hailo-8 to be an automotive grade AI accelerator. Hailo is currently working with automotive OEMs and Tier1s to leverage Hailo-8’s capabilities for ADAS and other AI applications.

Hailo’s AI processor is already enhancing state-of-the art products such as Foxconn's BOXiedge™ and Leopard Imaging’s EdgeTuring™. Powered by the Hailo-8™, these products offer next-generation AI processing solutions for video analytics that are cost-effective with high image quality and top energy efficiency for AI inference nodes. The solutions benefit applications including smart cities, smart medical, smart retail, and industrial IoT.

Hailo is the winner of many notable awards, including the 2021 Edge AI and Vision Product of the Year Awards for “Best Edge AI processor,” and was named a CES 2020 Innovation Awards Honoree, a Cool Vendor in the Cool Vendors in AI Semiconductors report by Gartner, and a winner of Red Herring’s Top 100 North America Award in the AI/Machine Learning Sector in 2019.



Hailo for ADAS and DMS



Q. Many people say that traditional automotive industry is staying between levels 2 and 3, and tech giants and startups preparing for level 4 or higher autonomous driving feel like they have lost momentum with Corona. What do you think of the current industry progress and future?

A. The automotive market is currently focusing on a range of effective driver assistance technologies that are making their way into most new models.

The market is currently somewhere between Level 1 to Level 2 and, driven by regulation and safety ratings, will likely head towards Level 2 and even Level 2+ in the coming years. Level 2+ refers to additional safety and comfort features that have been built on top of traditional ADAS applications to help the driver control the vehicle.

In the future, we expect more players to be focused on ADAS applications, which will help assist driving rather than operating the vehicle itself (i.e. automation). Automation features are continuously being developed (mostly premium Level 3 systems in the coming years) as well as more sophisticated and complex systems for very specific market niches due to their high cost. Additionally, expect to see higher performance of Level 1 and Level 2 ADAS in the future which will hopefully reduce accidents and casualties considerably.


Q. From Tesla's FSD computer, various car makers and Tier 1's ADAS ECUs, domain controllers and autonomous driving systems are competing in the market. What are the problems and inefficiencies of these systems?

A. As more and more advances are made in the automotive industry, more sensors will be required in vehicles with increased vehicular responsibility for driving. Efficiently processing environmental data for insights (such as accurately identifying pedestrians, vehicles and road conditions) remains a challenge, even for ADAS systems. In the perception phase, high resolution sensory data must be processed in real time, which is challenging with the area, power and cost of an embedded system. These are some of the inefficiencies that still must be overcome.






Penny-sized chip


Q. In terms of autonomous driving and ADAS, various AI accelerators are competing for Edge AI training, inference and prediction. What makes the Hailo processor so special?

A. Hailo’s AI processor, the Hailo-8™, is vastly improving the capabilities of perception-related sensors in autonomous vehicles and advanced driver-assistance systems (ADAS). The Hailo-8 is built with an innovative architecture that enables vehicles to run state-of-the-art deep learning applications that could not previously run in the power envelope of automotive ECUs, leading to improvements in collision avoidance, hazard detection, rear-collision warning, lane-change assist, and parking assist technologies. For example, the processor upgrades the performance of rear-view and front-facing cameras, ADAS ECUs, and surround view systems. By utilizing Hailo’s processor, vehicles on the road to autonomy can harness the power of a full “data center on wheels” via a chip the size of a penny.

Hailo prioritizes power efficiency, size and cost, making it an attractive solution for edge devices that would benefit from independent processing - such as automotive vehicles. According to research done by Hailo, when compared to available solutions in automotive such as Nvidia’s AGX Xavier, which runs neural network benchmarks including ResNet-50, the Hailo-8 is 10 times better on area efficiency and consumes almost 20 times less power while performing the same tasks. Moreover, the inherent scalability of the Hailo-8 architecture allows Hailo to tailor its offering to various market requirements.







Q. How much positive impact will the Hailo processor have in the entire autonomous driving and ADAS environment, from sensors and domain controls in cars to the cloud and autonomous driving IT infra?

A. Hailo is working with leading OEMs and tier-1 automotive companies on advanced driver-assistance systems (ADAS) to enable them to process sensory information more efficiently and at greater speeds - in order to pave the way towards full autonomy. Hailo’s processor vastly improves the performance of ADAS ECUs: There have been several improvement vectors for mainstream Level 2/2+ ADAS systems currently under development.

One example is VRU (vulnerable road user) detection. Many accidents involve VRUs, which leads to grim statistics, so the motivation to improve ADAS capabilities in detecting VRUs is quite high. In addition to the advances in regulation requirements in this domain, sensors are being developed with improved resolution, accuracy and coverage, which will enable much better detection and avoidance of VRUs.

Another example are the advances made in driver monitoring. Driver monitoring, in conjunction with ADAS features such as adaptive cruise control and lane keeping assistance, has become crucial to avoiding situations in which drivers take their attention off the road. Regulatory requirements are promoting better integration of driver monitoring and in-cabin monitoring with ADAS (and in the future, automated driving) in order to better assist drivers to maintain attention and control of the vehicle.
 

Q. Should AI solutions be different depending on applications and algorithms such as lane change and overtaking assistance, pedestrian recognition, and driver and occupant condition monitoring? In addition, while autonomous driving AI is mainly for images now, the future of automobiles will be related to all areas and industries such as mileage extension, predictive maintenance, voice assistant, entertainment, POI, finance, etc., and I think Hailo's target market will be really diverse.

A. Hailo’s AI acceleration capabilities help enable a variety of AI applications in the vehicular environment via low power consumption and high efficiency. Since vehicle sensors such as cameras, LiDAR’s and radars generate very large amounts of data which need to be processed in real time, the Hailo-8 capabilities enable ADAS applications to operate with high efficiency. The Hailo-8 can facilitate a multitude of AI applications for the mobility industry, especially ones requiring high AI performance in a low area and power envelope.
 

Q. Processor supply isn't all Hailo can do, right? What are the products and supports for customers in the automotive industry?

A. Hailo’s main offering is its edge AI chip, the Hailo-8™. Hailo also offers its standard form factor M.2 and Mini PCIe high-performance AI acceleration modules for empowering edge devices. The modules enable customers to integrate high performance AI capabilities into edge devices, providing a more flexible and optimized solution for accelerating a wide range of Deep Learning-based applications with high efficiency - optimizing time to market with a standard form factor. Read more about the M.2 and Mini PCIe high-performance AI acceleration modules here.


Q. The automotive market is very conservative in applying new technologies because of safety. There are start-ups that are performing proof-of-concept in partnership with renowned chip makers and software suppliers targeting car makers, and in some cases, they are already supplying them. Please tell me if there are any references you can reveal. Or, let me know when good news will start to be released.

A. We are unable to disclose more details at this time.


Q. Hailo is already a world famous startup. And Hailo has been targeting Asian markets such as Korea and Japan. I am curious about the current status of collaboration with major companies in this region. What is Hailo's next technology and business plan for the global market, including Korea?

A. Hailo is continuing to bolster its relationships with existing customers in Japan. In 2020, in response to growing demand from Japanese customers developing products requiring AI technology at the edge, Hailo opened a new subsidiary in Tokyo to better serve current customers and to continue to grow in market segments such as automotive, smart cities, smart retail, smart homes, industry 4.0, and beyond. Hailo also has a presence in Taiwan and is expecting to expand into additional Asia-Pacific markets in the near future. 

Japanese Corporation NEC - a leader in the integration of IT and network technologies - joined Hailo’s $60 million Series B funding round as a strategic investor. Hailo also has established key partnerships with Socionext, a major provider of advanced SoC solutions for video and imaging systems headquartered in Japan, and Foxconn, a global leader in smart manufacturing. A significant portion of Hailo’s customers are based in Asia, which is a strategic target market for Hailo for the automotive and other industries.  [AEM]



<저작권자 © AEM. 무단전재 및 재배포, AI학습 이용 금지>


  • 100자평 쓰기
  • 로그인


  • 미분류

TOP