Hailo launches AI deep learning chip to challenge Intel, Nvidia

Hailo founders Photo: Eran Thierry

Israeli startup Hailo Technologies strong, cheap, versatile, and energy-saving AI processors empower autonomous cars, security cameras, drones, AR/VR platforms and other devices.

Israeli startup Hailo Technologies Ltd. has launched its first processor, designed for deep learning applications by end devices, such as autonomous cars, smart cameras, smartphones, drones, and augmented reality/virtual reality (AR/VR) platforms.

Artificial intelligence (AI) processors for end devices are a growing subsector of the semiconductor industry, the most important characteristics of which are a low-energy power source, small physical dimensions, a low price and maximum operational flexibility, in contrast to processors intended for data centers.

The first samples of the processor will soon be delivered to major manufacturers and suppliers in the automotive industry and camera companies for "performance envelope testing". To mark the launch, Hailo CEO Orr Danon told "Globes," that, "The company's objective is to develop the business partners’ different products for the autonomous car market around the Hailo processor, beginning with European and APAC automotive manufacturers."

Hailo, founded two years ago, has raised around $26 million in three financing rounds, including €2 million from Horizon 2020, the European Research and Innovation Program. Investors in the company include Hailo chairman Zohar Zisapel, the first investor in it, OurCrowd investment platform, venture capital funds Maniv Mobility and Next Gear Ventures, the lender for the Technion Drive Accelerator, Delek Motors Ltd., Gil Agmon, and China's Glory Ventures. Danon says that Hailo intends to shortly embark on another financing round of several tens of millions of dollars in order to expand its development and sales teams.

Danon jointly founded Hailo with CTO Avi Baum and VP business development Hadar Zeitlin, all of whom are graduates of the IDF’s Talpiot Project. Danon has won the Israel Defense Prize, Baum held a number of positions at Texas Instruments, and Zeitlin won the IDF Chief of Staff's Prize for Technological Excellence during her service. Before Hailo, Zeitlin served as the first product manager at Via Ltd. A fourth co-founder was Israel Defense Prize winner, the late Colonel (res.) Rami Feig, who died in a drowning accident a year ago. The Tel Aviv-based company has 55 employees.

Hailo stays opaque

One of the main markets that the new industry is targeting is the resource-intensive automotive market, but that is just one of the markets that will need strong, cheap, versatile, and energy-saving AI processors. The future autonomous car will include dozens of different sensors and will have to deal with vast quantities of data that the car will have to process in real time. The alternative is to process the data in the cloud, but because the autonomous car must make decisions immediately, it cannot rely on the cloud to process the data obtained from dozens of sensors simultaneously - decisions that can be a matter of life and death.

The technological challenge is actually even greater with the processors intended for industries of small and inexpensive end-devices such as security cameras that are widespread in large cities. The challenge is multiplied, because the processors that will be installed in the devices must be especially small and cheap to make widespread deployment possible. Hailo's new processor, which is the size of a US cent, consumes, according to the performance test that Hailo chose to present, 20 times less power than the corresponding processor by market leader Nvidia Corproation, the Jetson AGX Xavier, and is 15 times smaller in size. Hailo claims that its processor outperforms the TPU for Google end products, and Intel Corporation's Movidius Myriad X.

Indeed, a look at the Hailo board on which the processor is installed showed the absence of two prominent components: memory and a fan. According to Hailo, the processor’s low power renders fans redundant, while the memory is embedded within the processor itself. The result is that Hailo's processor is smaller than the corresponding Nvidia processor , and the latter comes surrounded by a row of external memory processors that increase the total space that the product occupies multifold.

Other companies developing artificial intelligence processors are Britain's Graphcore Ltd., which raised $310 million at a company value of $1.7 billion, and Avigdor Willenz's Habana Labs Ltd., which has raised a total of $120 million. Another Israeli startup in Hailo's sector is NeuroBlade Ltd., which plans to unveil its first prototype in the second half of the year. An investor in the company, StageOne Ventures managing partner Tal Slobodkin once told "Globes", "One NeuroBlade processor should consume a tenth of the voltage of Nvidia, while offering 50 times greater performance."

Hailo has stayed opaque since it was founded, not only about its customers, but also the processor's specs. The company has not yet published the full specifications, making it impossible to fully compare it with its competitors. The opacity is understandable, given Danon's remarks, when he explained a year ago that the secrecy was due to the goal to beat the corporate giants that are seeking a similar achievement. "Everyone understands that artificial intelligence is the main interest today, but current processors are not suitable for dealing with the sector's problems. We believe that the solution is at the level of architecture: change how the computer is built - no set of rules, but intuitive thinking based on a lot of data, which requires a different level of component design."

Danon says, "The basic structure of current embedded processor architecture was developed 70 years ago and suffers from significant disadvantages. Hailo's holistic solution deals with these problems by rethinking the basic elements in computer architecture - the memory, control, and processing. The solution also includes a comprehensive software development kit (SDK) that was jointly developed with the hardware.” Put simply: Hailo's processor carries out actions that are usually divided among several different components on the circuit board (the card on which the processors are embedded). Instead of the processor referring to the memory via the circuit board's communications components, the control and retrieval from the memory and the processing are wholly undertaken within the processor. This avoids bottlenecks and contributes to accelerated processor performance resulting in faster performance by the artificial-intelligence based insights fulfillment system and decision-making, which is also comprised on end-processors similar to that of Hailo.

On the giants' radar

The sector of edge devices processors is considered one of the hottest in the semiconductor industry. This week, Lip-Bu Tan, the founding partner of WRV Capital, which specializes in processors, predicted that the AI inference subsector in edge devices (the subsector in which Hailo operates) will grow to $4-4.5 billion in 2025, compared with less than $100 million in 2017. Tan has invested in AI processor company Habana Labs, among others.

Tan, who spoke at the Samsung conference in Tel Aviv, also serves as the president and CEO of Cadence Inc., one of the top two multinationals in software for processor design and planning (alongside Synopsis Inc.). Cadence has a development and sales center in Israel. Tan also predicted that the AI processor sector as a whole will be the main growth engine of the semiconductor industry. According to the data he presented, the sector will have three times faster revenue growth than the semiconductor industry as a whole through 2027.

Although Danon admits that he would be happy for Hailo to grow independently, the semiconductor giants are presumably already casting their eyes at it. During a visit to Israel in late March, Nvidia CEO Jensen Huang told "Globes" that the company might acquire one of the Israeli companies, including Habana Labs and Hailo. "I haven’t met any of them, but I'd obviously be happy to do so. I've been in the high-tech industry for 26 years. I've seen a lot of big companies and they seem to be very promising companies," he said.

It cannot be ruled out that Hailo is also on the radar of Intel, which has already bet on the automotive industry end-processor market with its acquisition of Mobileye. Technology like Hailo's could definitely interest Intel, which is in distress and is still looking for future growth engines. The interest is partly because the technology in Mobileye's processors is based on machine vision, while Hailo's processor is designed to be versatile: it is suited for multiple uses, including for other sensors that will be included in future products.

An industry source, who examined for "Globes" the details that Hailo disclosed, explained that the processor is apparently one of the most flexible processors in the growing market of artificial intelligence edge processors. The processor is also designed to perform tasks of the core processor for sensors and tasks of the artificial intelligence edge processors. As a result, it makes possible the use of advanced neural network technologies, such as the inputting and processing of data on the processor to send binary data to the cloud in a fraction of the volume of the original data - a feature that turn it into an especially efficient processor for the Internet of Things (IoT) industry installations.

Founded: 2017

Business: Development of an AI processor

Founders: Orr Danon, Avi Baum and Hadar Zeitline

No. of employees: 55

Capital raised: about $25 million

Prominent investors: Zohar Zisapel, OurCrowd, and Maniv Mobility

Published by Globes, Israel business news - en.globes.co.il - on May 16, 2019

© Copyright of Globes Publisher Itonut (1983) Ltd. 2019

5 Comments
View comments in rows
Update by email about comments talkback
POST
Comments
Hailo founders Photo: Eran Thierry
Hailo founders Photo: Eran Thierry
Twitter Facebook Linkedin RSS Newsletters גלובס Israel Business Conference 2018