Foxconn has combined its edge computing solution, BOXiedge, with the Japanese Socionext parallel processor SynQuacer” SC2A11, and the Hailo-8 deep learning processor developed by Tel Aviv based chipmaker Hailo. The new device aimed to provide powerful AI processing solution for video analytics at the edge. The new BOXiedge is capable of processing and analyzing over 20 streaming camera input feeds in real-time, all at the edge, including image classification, detection, pose estimation, and other AI-powered applications.
This is a milestone win for Hailo, a chipmaker startup who was established in 2017 and had completed a $60 million financing round in March 2020, with the participation ABB Technology Ventures (VC arm of ABB) and NEC Corporation. The funding will be used to enter mass production of the company’s Hailo-8 Deep Learning chip during 2020. Today the company employs approximately 80 employees.
The company’s Hailo-8 processor is a dedicated Neural Networks processor aimed to implement inference functions on edge devices in Automotive and Industrial applications. Hailo-8 processor reaches up to 26 Tera Operations Per Second (TOPS) and 3 TOPS per Watt. It is comprised of four main components: an Image Signal Processor to improve the image arriving from the sensor before its transfer for processing by the neural network core, an H.264 encoder that handles the video stream, an ARM-M4 processor to manages the chip, and a unique Neural Network core.
“Our vision is to pave the way for next generation AI solutions,” said Gene Liu, VP of Semiconductor Subgroup at Foxconn Technology Group. “We recognize the great potential in adopting AI solutions for a multitude of applications, such as tumor detection and robotic navigation. This platform will positively impact rapidly evolving sectors including smart cities, smart medical, smart retail, and industrial IoT.”
Foxconn has already deployed several in-house developed AI solutions on different production lines, leading to a reduction of at least one third of the operating costs for appearance defect inspection projects. “We are thrilled to collaborate with two of the global leaders in AI solutions,” said Orr Danon, CEO and Co-Founder of Hailo. “A new generation of chips means a new generation of capabilities at the edge.”
The Tel Aviv based chipmaker startup Hailo, has successfully completed a a $60 million financing round with the participation of key strategic investors ABB Technology Ventures (VC arm of ABB) and NEC Corporation. The funding will be used to enter mass production of the company’s processor Hailo-8 Deep Learning chip during 2020. Today the company employs approximately 80 employees.
Since its inception in February 2017, Hailo had raised $88 million. Following the last investment round, it begins to recruit 30-40 new employees for its research and development and support department, as well as offshore stuff for new offices in Europe, Japan and the US, to be opened in 2020. The company’s Hailo-8 processor is a dedicated Neural Networks processor aimed to implement inference functions on edge devices in Automotive and Industrial applications.
“We look forward to combining Hailo’s solution with our cutting-edge industrial technology as an important piece of the puzzle to drive the digital transformation of industries,” said Kurt Kaltenegger, Head of ABB Technology Ventures. Hiroto Sugahara, General Manager of Corporate Technology Division, NEC Corporation, said that Hailo’s technology will help NEC, “to dive deeper into the intelligent video analytics market. We look forward to incorporating Hailo’s technology into our next generation edge-based products.”
The Company’s co-founder and CEO, Orr Danon, told Techtime Hailo-8 chip presents a novel architecture to enable fast and power saving implementation of Neural Networks. “We identified that in during the processing of inference, there are differences in the behavior of the different layers in the neural network. Our solution provides the exact resources needed in each layer.
“In contrast, our competitors, who use solutions such as GPU processors, allocate to each and every layer the same level of resources. Our development software learns the specific problem of each application, characterizes it, and give the chip instructions on how to manage the resources of each layer in an optimal method.”
According to Hailo, its processor reaches up to 26 Tera Operations Per Second (TOPS) and 3 TOPS per Watt. It will meet the strict ISO 26262 ASIL-B as well as the AEC Q 100 Grade 2 standards. Hailo-8 is comprised of four main components: an Image Signal Processor to improves the image arriving from the sensor before its transfer for processing by the neural network core, an H.264 encoder that handles the video stream, an ARM-M4 processor to manages the chip, and the neural network core itself.
This neural network is comprised of a flexible matrix of software-configurable processing, controls, computational resources and memory units. “Hailo’s Deep Learning chip is a real game changer in industries such as automotive, industry 4.0, robotics, smart cities, and more,” said Hailo Chairman Zohar Zisapel, “A new age of AI chips means a new age of computing capabilities at the edge.”
Intel Corporation announced that it has acquired Habana Labs, an Israel-based developer of programmable deep learning accelerators for approximately $2 billion. Habana will continue to be based in Israel where Intel also has a significant presence and long history of investment. Prior to this transaction, Intel Capital was an investor in Habana. Habana chairman Avigdor Willenz will serve as a senior adviser to Intel.
“Habana turbo-charges our AI offerings for the data center,” said Navin Shenoy, general manager of the Data Platforms Group at Intel. “This acquisition advances our AI strategy, which is to provide solutions to fit every performance need – from the intelligent edge to the data center. Our combined IP and expertise will deliver unmatched computing performance and efficiency for AI workloads in the data center.”
Intel expects that the fast-growing AI silicon market be greater than $25 billion by 2024, and within that, AI silicon in the data center is expected to be greater than $10 billion. In 2019, Intel expects to generate over $3.5 billion in AI-driven revenue, up more than 20 percent year-over-year.
Based in Caesarea, Israel, Habana labs was established in 2016 with Willenz as its first investor, and has developed dedicated chips for Deep Learning Training and Inference. Its Goya AI Inference Processor, which is commercially available, has demonstrated excellent inference performance including throughput and real-time latency in a highly competitive power envelope.
The Gaudi AI Training Processor is currently sampling with select hyperscale customers. Large-node training systems based on Gaudi are expected to deliver up to a 4x increase in throughput versus systems built with the equivalent number of GPUs. It is produced in TSMC’s 16 manometer process.
The acquisition gives Habana access to Intel AI capabilities, including deep expertise in AI software, algorithms and research that will help Habana scale and accelerate. In November 2018, Intel Capital led a $75 million investment round in Habana Labs. “We have been fortunate to get to know and collaborate with Intel given its investment in Habana, and we’re thrilled to be officially joining the team,” said David Dahan, CEO of Habana.
Last week, Intel and Microsoft brought together nearly 100 security and Artificial Intelligence (AI) experts to discuss new standards for Homomorphic Encryption (HE), which is emerging as a leading method to protect privacy in machine learning and cloud computing. The HE standards workshop took place on Intel’s Santa Clara, California campus. Following the first meeting in October, 2018, Intel and Microsoft initiated the founding of the HomomorphicEncryption.org group.
As more data is collected and used to power AI systems, concerns about privacy are on the rise. Casimir Wierzynski from the office of the CTO of AI Products Group at Intel, said that Intel is collaborating with Microsoft Research and Duality Technologies on standardizing HE, “to unlock the power of AI while still protecting data privacy.”
Fully homomorphic encryption, or simply homomorphic encryption, refers to a class of encryption methods envisioned by Rivest, Adleman, and Dertouzos already in 1978, and first constructed by Craig Gentry in 2009. Homomorphic encryption differs from typical encryption methods in that it allows computation to be performed directly on encrypted data without requiring access to a secret key. The result of such a computation remains in encrypted form, and can at a later point be revealed by the owner of the secret key.
It allows AI computation on encrypted data, thus enabling data scientists and researchers to gain valuable insights without decrypting or exposing the underlying data or models. This is particularly useful in instances where data may be sensitive – such as with medical or financial data. Homomorphic encryption also enables training models directly on encrypted data, without exposing its content. Such encryption would enable researchers to operate on data in a secure and private way, while still delivering insightful results.
Photo above: A development team at Hailo. The Tel Aviv based company employs around 65 workers today
Tel Aviv-based Hailo is preparing for the mass production of its AI chip that meets the ASIL-B standard of the automotive industry, and start full-scale production during 2020. Company co-founder and CEO Orr Danon told Techtime that the new chip is named Hailo-8 and was developed as part of the cooperation between Hailo and auto manufacturers. The chip enables the meeting of demands for critical life-saving systems, including the meeting of working under conditions of up to 105 degrees Celsius.
According to company data, the Hailo-8 processor reaches up to 26 tera operations per second (TOPS) and 3 TOPS per Watt power efficiency. It will meet the strict ISO 26262 ASIL-B as well as the AEC Q 100 Grade 2 standards. Hailo-8 is comprised of four main components: an Image Signal Processor that improves the image from the sensor before its transfer for processing by the neural network core, an H.264 encoder that handles the video stream, an ARM-M4 processor that manages the chip, and the neural network core itself, that is comprised of a flexible matrix of software-configurable processing, controls, computational resources and memory units.
Renew the Old Idea of DFP Processors
Hailo was established in February 2017 by CEO Orr Danon, CTO Avi Baum and the Business Development Manager Hadar Zeitlin. The first investor in the company was Zohar Zisapel, who serves today as Chairman of the board. To date Hailo has raised around $24 million, of which $21 million was in the latest round of financing which was completed in January 2019. The company developed a new architecture for AI chips for edge devices that carry out the execution phase – i.e. the application of the inference of a neural network in edge devices at a rapid speed and with a huge saving in energy. According to Danon, the architecture, which is protected by dozens of pending patents, “belongs to a forgotten family of processors from the Data Flow Processors type.”
In DFP processors, the processing activity takes place only when data is fed into the processor, which conducts a fixed series of operations on top of this information, and then transfers the processed information. “In recent years, it turned out that the reliance on neural networks is an efficient and reliable means of solving many problems, and therefore most of the AI systems that we see in the market are based on neural networks. Here the challenge is structural, since the chip needs to implement a structure of a neural network. In a neural network one infuses experience into the description of a structure, and therefore it is a very efficient solution in solving problems which are based on recognizing examples.”
How is your chip designed? What are the main principles of the architecture?
Danon: “Our architecture describes the structure of the neural network and allocates resources to every layer in the network. We identified that in during the processing of inference, there were differences in the behavior of the various layers of the neural network, and therefore there was a need to provide them with different resources. This runs counter to our competitors, who use solutions such as GPU processors that allocate to each and every layer the same level of resources. Our development software learns the specific problem, characterizes it, and knows how to transfer to the chip instructions on how to manage the resources of each layer in an optimal method.”
What are the components of the chip?
“The idea is to use the memory units that are located very close to the processing units. We allocate memory and processing units in accordance with every task, and in this way achieve very fast processing, and extensive savings in the chip’s power consumption. This allows us to meet the extremely strict standards of the automotive industry, since the chip does not heat up and is capable of operating in the environmental temperatures that the industry demands.”
You claim that your chip is more efficient that other solutions in the market. However, there is no universally accepted means of measuring AI chips.
“We measure the performance of our chips by checking how many operations per Watt we execute a specific neural network. Nowadays there is the MLPerf consortium that is attempting to define a benchmark which will serve as a basis for comparing different deep learning processors. Regarding edge devices, the industry is apparently going in the direction of measuring the number of operations per Watt (TOPS/W) that the neural network carries out for a specific task, like an image.”
Today new methods are being developed for deceiving neural networks. How are you dealing with this problem?
“It is possible to relate to AI deception as a weakness, just as one relates to security vulnerabilities. At the outset, the weaknesses of the software systems surprised the industry, but gradually they found solutions. In the AI field, first of all this is a conceptual problem that lacks a solution on the silicon level. However, if the network was trained in the wrong way, and the attacker is aware of how the network was trained, then he is capable planning an attack. We also deal with this problem, and in principle it shows the advantage of installing AI systems at the edges of the network, since in this way there are less vulnerabilities along the route of transferring information.“
Hailo is growing quickly, and currently employs around 65 workers. The company is in the process of hiring additional manpower. Hailo is focusing its efforts on two key markets: the automotive and IoT. These two markets are expected to be huge and are also very demanding as in both there is a need for a product which is very reliable, low-cost, with very low power consumption. Danon: “In many respects the camera in a vehicle is no different than the IoT camera in a smart city. These are two areas that will be very dominant, and they share many common requirements.”
A recent report made by BIS Research, Elbit Systems from Israel is mentioned as one of the 10 key players in emerging global Cognitive Electronic Warfare systems market. BIS estimates that the market will witness high growth during 2023-2028, as cognitive electronic warfare systems have the capability to help overcome the problem of reprogramming the electronic warfare models.
“Artificial Intelligence and Machine Learning with computational hardware are likely to revolutionize the electronic warfare (EW) systems.” According to BIS Research analysis, the global cognitive electronic warfare system market is expected to report a revenue of $385.7 million in 2023 and is estimated to cross $928.4 million by 2028, at a CAGR of 19.20% during 2023-2028.
The analysts indicate that since cognitive EW systems make use of AI and machine learning techniques, they are expected to overcome most of the drawbacks of current electronic warfare system, such as incapability of handling excessive load and being non-responsive to unknown threats. This market is expected to witness a high growth rate owing to countries such as the U.S., Russia, and China, who plan to implement artificial intelligence (AI) in the military.
A European Surprise in the Market
AI is required for handling the data and information of huge sensors and communication networks, as the quantity of information is increasing. AI and machine learning offer potential benefits to military organizations by providing fast and high-quality data, which can support in deep analysis of complex and strategic data. Europe, including some of the major companies such as the U.K, Brazil, Russia, Germany, and France is the most prominent region for the cognitive electronic warfare systems market.
North America is the second most prominent regions, which is continuously progressing in the field of cognitive electronic warfare systems market. According to Swati Chaturvedi, Senior Research Analyst at BIS Research, Europe is one of the most prominent regions for the growth of the global cognitive EW market and will dominate it in 2023. “Europe is keenly focusing and investing in the research and development of AI for military.
“However, the geographical analysis of this market unveils an immense potential for its growth in the region of North America. Similarly, the Asia-Pacific market is also likely to witness numerous growth opportunities during the forecast period 2023-2028.” According to BIS Research, the 10 key players in the global cognitive electronic warfare system market are BAE Systems, Elbit Systems, General Dynamics, Harris Corporation, Leonardo S.p.A., Lockheed Martin, Raytheon Company, Saab AB, Textron Inc., and Thales Group.
Hailo from Tel Aviv announced the the first samples of its Hailo-8 Deep Learning Processor were delivered to selected partners, mainly from the automotive industry. To allow running deep learning applications on the edge devices, Hailo re-designed the main pillars of computer architecture: memory, control, and compute and added a comprehensive Software Development Kit (SDK) co-developed with the hardware.
The result is a 16 nano-meter chip produced by TSMC capable of delivering up to 26 Tera Operations Per Second (TOPS). The company said that during ResNet-50 benchmark tests, Hailo-8 outperformed Nvidia’s Xavier AGX and consumed almost 20 times less power while performing the same tasks. Hailo was established in 2017 by former members of the Israel Defense Forces’ intelligence unit.
AI Beyond the Cloud
Its deep learning processor is designed to run complex algorithms on autonomous vehicles, smart cameras, smartphones, drones, AR/VR platforms, and wearable devices. The company said it is now working in cooperation with tier-1 automotive companies on advanced driver-assistance systems (ADAS), and with major players in smart city and smart home markets, to empower powerful IoT devices.
“In recent years, we’ve witnessed an ever-growing list of applications unlocked by deep learning, which were made possible thanks to server-class GPUs,” said Orr Danon, CEO of Hailo. “However, there is a crucial need for an analogous architecture that replaces processors of the past, enabling deep learning to run devices at the edge. Hailo’s chip was designed from the ground up to do just that.”