New-Tech Europe Magazine | H2 2023

H2 2023

16 A simpler method for learning to control a robot 18 Hailo Expands Hailo-8 AI Accelerator Portfolio, Delivering Unprecedented Performance to a Diverse Range of Edge AI Applications 20 High-performance boost for modular systems 24 How to Greatly Improve Battery Power Efficiency for IoT Devices

2024

2024

EXPO Tel Aviv | 28-29.5.2024

SAVE THE DATE 28-29.05.2024

Sponsor by:

For further details: www.new-techevents.com www.new-techevents.com Pavilion 1, EXPO Tel Aviv | 28-29.5.2024

Widest selection of electronic components ™ In stock and ready to ship

mouser.co.il

HIGH-PERFORMANCE INTERCONNECT SOLUTIONS

High-Speed Board-to-Board • High-Speed Cable • Precision RF • Optics • Rugged / Power Systems

Samtec Israel 21 Bar-Kochva St. • Concord Tower • B’nei Brak, Israel 51260

Tel: +972 3 7526600 • Fax: +972 3 7526690 www.samtec.com • e-mail: israel@samtec.com

King of Coax Connection.

© eiCan

#KINGOFCOAX

Coaxial Connectors The SMA connector is designed for a frequency range of DC to 18 GHz at 50 Ω impedance. With its threaded-type coupling mechanism, it is perfect for securing your connection in intensive vibration environments.

n Standard interface in accordance with MIL-STD-348 n Precise CNC machining with up to 0.003 mm tolerance n Center contact captivation n Resistant to severe conditions for up to 48 hours

For further information, please visit: www.we-online.com/coax

High Reliability Interconnection Powell Electronics in Europe Connectors, cables, value-added assembly

Quell Glenair Harwin Amphenol Aerospace AB Connectors Lemo Conesys and more…

WHAT IF TECHNOLOGY COULD HELP US CONNECT ON A HUMAN LEVEL?

High-performance, portable technology isn’t limited to activity monitoring or location sharing. With advanced sensing and contextually-aware technologies, ADI is powering multidimensional solutions that know how we feel and deepen our human connection.

Analog Devices. Where what if becomes what is. See What If: analog.com/WhatIf

H2 2023

About New-Tech Magazines Group Read To Lead ‘New-Tech Magazines’ A world leader in publishing high-tech and electronics, producing top quality publications read by tens of thousands professionals from all over the world especially from Europe, innovative electronics, IoT, microwave, homeland security, aerospace, automotive and technological industries. Our specialized target audiences prefer New-Tech Europe because they know that our publications are a reliable source of the latest information in their respective fields. Our multidimensional editorials, news items, interviews and feature articles provide them with a full, well-rounded picture of the markets in which they operate - an essential asset for every technological leader striving to stay ahead, make the right decisions, and generate the next global innovation. Moreover, as an attractive platform for advertisers from around the world, New-Tech Europe has become a hub for bustling international commercial activity. Here, through ads and other promotional materials, Israeli readers obtain crucial information about developers and manufacturers worldwide, finding the tools, instruments, systems and components they need to facilitate their innovative endeavors. Targeting the needs of both the global and european industries and global advertisers, New-Tech Magazines Group constantly expands and upgrades its services. Over the years, the company has been able to formulate a remarkably effective, multi-medium mix of offerings, combining magazine publications with useful online activities, newsletters and special events and exhibitions.

Editor: Tomer Gur-Arie

COO & CFO: Liat Gur-Arie

Journalist:

Amir Bar-Shalom

Technical journalist:

Arik Weinstein

Graphic Design: Hadas Vidmayer

Concept design: Maya Cohen

mayaco@gmail.com

Technical counselor:

Arik Weinstein

Sales and advertising:

sales@new-techmagazine.com

Account Manager: Sivan Bekerman

Project Manager/Events: Shiri Abdi

Project Manager/Events: Yaara shahaf

Data system: Liat Tsarfati

Editorial coordinator: Danit Segev

Operation Manager: Sivan Bekerman

Mail: Office:

info@new-techmagazine.com

Publisher :

NEW-TECH MAGAZINE GROUP LTD

www.new-techeurope.com

8 l New-Tech Magazine Europe

Contents

10 LATEST NEWS 16 A simpler method for learning to control a robot 18 Hailo Expands Hailo-8 AI Accelerator Portfolio, Delivering Unprecedented Performance to a Diverse Range of Edge AI Applications 20 High-performance boost for modular systems 24 How to Greatly Improve Battery Power Efficiency for IoT Devices 28 The future of standardized defense platforms using MOSA, SOSA and VPX open architectures 32 Towards a process flow for monolithic CFET transistor architectures 36 How to avoid delays & accelerate development cycles with optimized design processes

16

18

40 OUT OF THE BOX 42 NEW PRODUCTS 52 INDEX

20

24

www.new-techeurope.com

New-Tech Magazine Europe l 9

Latest News

NeoLogic Unveils Novel Processor Design Technology at 16nm, Promising Significant Power, Cost, and Area Reductions

NeoLogic Unveils Novel Processor Design Technology at 16nm, Promising Significant Power, Cost, and Area Reductions. Novel standard cells for 5nm and 3nm technology nodes are under development. The company expects to tapeout a 16nm ARM processor for demonstration this coming December.

the number of transistors. This breakthrough benefits the logic synthesis as well as the physical design. Dr. Avi Messica, Co-founder and CEO of NeoLogic, stated: “Utilizing Quasi-CMOS for

processor development delivers a technological leap in performance. Our design technology enables us to design a 16nm processor that delivers performance equivalent to more

Credit: NeoLogic

advanced – sub 16nm – technology nodes, while saving development (NRE) and manufacturing (OPEX) costs. Reducing the processor’s power consumption in data centers leads to significant cost savings (cooling, electricity, infrastructure).” NeoLogic, which recently secured an 8-million-dollar seed funding, was founded in 2021 by Dr. Avi Messica (CEO) and Ziv Leshem (CTO), both of whom have decades of experience in R&D and management of microprocessors design and fabrication. Dr. Avi Messica (Ph.D. Weizmann Institute of Science) is an expert in solid-state physics and quantum devices and in ultrafast transistors in particular with 26 years of managerial experience in a variety of hi-tech companies. Messica previously served as a device group manager at Tower Semiconductors and has hands on experience in the design and fabrication of CMOS devices. He also served as VP of Engineering at Shellcase and founded and served as the CEO of three semiconductor companies in the fields of image sensors, MEMS-based optical switches, and photonic chips. Ziv Leshem has 25 years of experience in processor design. He worked for some of the world’s leading semiconductor companies, such as National Semiconductors, DSPG, and Synopsys, and managed complex processor design projects. He was one of the founders of LogixL, a company that developed a hardware-based HDL simulator and also served as a manager at NewSight Imaging, a developer of LiDAR and iTOF sensors. Before founding NeoLogic he was the manager of the physical design group at Inomize where he managed a group of engineers and developed processors for customers in various industrial sectors in CMOS technologies ranging from 40nm to 7nm.

Israeli processor technology startup NeoLogic is launching a groundbreaking processor design technology that is poised to revolutionize chip design. The company expects to tapeout an ARM processor at 16 nanometers for demonstration this December. The technology and the processor will be available for evaluation to key selected customers. NeoLogic’s Quasi-CMOS technology serves as a platform for processor design. It delivers high computing power in tandem with reduced power consumption and cost. The company has completed the development of new, non existing, standard cells for the 16nm technology node, on top of the existing CMOS standard cells library. NeoLogic’s standard cells are single-stage high fan-in (8 to 16 inputs), among others, leading to up to 50% reduction in power consumption compared to the most advanced equivalent CMOS cells while saving up to 40% of the area. The technology was conceived to address the increasing workloads in data centers and the need to reduce the high costs associated with developing processors using advanced technology nodes. Designing processors with Quasi-CMOS delivers superior computing power per watt per millimeter square, catering to the escalating workloads of artificial intelligence, machine learning, data analysis, video streaming, and more in data centers. CMOS technology, which has been the “workhorse” of processor design and fabrication for the past 40 years, is nearing its limits and is challenging to improve. Quasi-CMOS breaks through these limitations by significantly increasing the maximum number of inputs of standard cells and by changing their topology to reduce

10 l New-Tech Magazine Europe

Latest News

Advantech and Hailo Partner to Expand Energy-Efficient Edge AI Portfolio

Advantech, a leading provider of AIoT platforms and services, is excited to announce a new strategic partnership with Hailo, a pioneering AI-focused chipmaker, to expand its energy-efficient edge AI portfolio. Through this collaboration, Advantech will leverage Hailo’s AI accelerators to develop exceptional performance-and price-per-watt edge AI systems

acceleration modules, including EPC-R3720, ARK-1250, ARK 3533, and EI-53 during September and October.The collaboration also encompasses software services, providing comprehensive end-to-end

solutions. Advantech’s Edge AI software, a toolset that assists customers in evaluating, developing, and deploying diverse edge AI applications, will integrate Hailo’s AI SDK TAPPAS, which has been verified on Advantech hardware models. “We are excited to team up with Hailo to enable scalable high-efficiency AI capabilities at the edge.” said Aaron Su, Vice President of Embedded IoT Group, Advantech. “The collaboration between Hailo and Advantech signifies a mutual dedication to accelerate edge AI inference application development with a diverse range of technologies, solutions, and services.” “We are happy to expand our cooperation with Advantech to bring efficiency and high performance to edge AI applications” said Gary Huang, General Manager Greater China at Hailo. “Hailo is proud to be working with a leading partner like Advantech to leverage AI to empower a wide variety of market segments like industrial automation, security and smart cities”

credit: Advantech

and AI acceleration modules, catering to compact and robust edge AI applications such as factory AOI defect inspection, warehouse AMR object detection, and public parking lot management. Edge AI use cases are highly diverse, driving the need for diversity in AI computing capabilities, power consumption, and industrial designs. Power sensitivity is a crucial consideration when integrating edge devices into robotics or autonomous devices, where efficient power consumption preserves battery life and prolongs operational duration. Hailo’s product portfolio includes the Hailo-8™ AI accelerator, which seamlessly integrates with edge platforms, empowering real-time deep learning inference tasks at the edge with 26 TOPS at a low power consumption of 2.5W, small size, and low cost. Advantech will launch a series of edge AI systems bundled with Hailo-8 M2 AI

Hyundai Motor Company, Kia Corporation and Infineon sign multi-year supply agreement for power semiconductors

Infineon Technologies AG (FSE: IFX / OTCQX: IFNNY) and Hyundai Motor Company and Kia Corporation have signed a multi-year supply agreement for silicon carbide (SiC) and silicon (Si) power semiconductors. Infineon will build and reserve manufacturing capacity to supply SiC as well as Si power modules and chips to Hyundai/Kia until 2030. Hyundai/ Kia will support the capacity build-up and capacity reservation with financial contributions.

“Infineon stands as a valued strategic partner, boasting steadfast production capabilities and distinct technological prowess within the power semiconductor market,” said Heung Soo Kim, Executive Vice President and Head of Global Strategy Office (GSO) at Hyundai Motor Group. “This partnership not only empowers Hyundai Motor and Kia to stabilize its semiconductor supply but also positions us to solidify our leadership in the global EV market, underpinned by

New-Tech Magazine Europe l 11

Latest News our competitive product lineups.”

lead to strong market growth for power semiconductors, especially those based on wide bandgap materials like SiC. With the significant expansion of its Kulim fab, Infineon will build the world’s largest 200-millimeter SiC power fab and further strengthen its market-leading role as a high-quality, high volume supplier to the automotive industry. In line with Infineon’s multi-site strategy, the Kulim facility will complement Infineon’s current manufacturing capacity in Villach, Austria, and further capacity expansions in Dresden, Germany.

“The future car will be clean, safe and smart and semiconductors are at the heart of this transformation. As a trusted partner, we are proud to advance our long-term partnership with Hyundai/Kia,” said Peter Schiefer, President of Infineon’s Automotive Division. “We contribute premium products of high quality, our system knowledge and application understanding combined with continued investments in manufacturing capacity to address the increasing demand for automotive power electronics.”

From left to right: Mr. Jason Chae (Vice President of the Semicon Strategy Group at Hyundai Motor Company), Mr. Heung Soo Kim (Executive Vice President and Head of Global Strategy Office at Hyundai Motor Group), Mr. Peter Schiefer (President of Infineon’s Automotive Division), Mr. Peter Schaefer

Infineon’s power semiconductors are key enablers for the transition to electromobility. This transition will Samtec Opens New Design, Manufacturing Center In Taiwan (Executive Vice President, Sales, Marketing and Distribution of Infineon’s Automotive Division) Credit: Hyundai Motor Company

Samtec, Inc., has announced the grand opening of its new design and manufacturing center in Taiwan that will design, develop, and build leading edge connector products for multiple industries, including data com, telecom, artificial intelligence, machine learning, test and measurement, industrial, and medical.

Vice President of Engineering. “The Taiwan Design Center enables Samtec to realize its aggressive new product development roadmap, while leveraging Taiwan’s wealth of talented associates to achieve our engineering and manufacturing goals.”

The Taiwan Design Center is the latest addition to Samtec’s global design, manufacturing and customer service footprint in support of Samtec Sudden Service®. In 2023,

The state-of-the-art 70,000 square foot facility (6,503 square meter) consists of three floors of manufacturing, new

Credit: Samtec

product development and engineering, machine and equipment design, and connector and cable assembly. The manufacturing capabilities primarily focus on Samtec’s precision RF product line, which are used in the wireless infrastructure, aerospace and defense, computer, and test and measurement industries. “The new Taiwan Design Center allows Samtec to produce increased quantities of next-generation interconnect product. In turn, this empowers our customers and partners to drive innovation and remain at the forefront of their markets,” said Brian Vicich, Samtec’s

Samtec opened a Singapore Distribution Hub, built and fitted a new cable manufacturing facility in Pennsylvania (USA), and began expanding the Vietnam and Costa Rica manufacturing facilities. Samtec serves customers in over 125 countries. With 22 sales offices, 13 design centers, and 14 operations/inventory locations across the globe, Samtec’s Sudden Service® supports a customer base of 50,000+, spanning well known global tech giants to universities and small start-ups. Samtec sells directly to customers as well as through highly respected distribution partners.

12 l New-Tech Magazine Europe

Latest News

Autotalks and Hyundai Mobis collaborate to implement cutting-edge V2X technology for connected cars

Autotalks, a world leader in V2X (Vehicle-to-Everything) communication solutions, and Hyundai Mobis, a leading global automotive supplier, are set to unveil their latest joint effort: Hyundai Mobis’ MTCU (Multi-functional Telematics Control Unit) connectivity modules. MTCU Gen1, Hyundai Mobis’ most advanced connectivity module to date, supports Day

IAA Mobility exhibition, which will take place in Munich, Germany, between September 5th and September 10th, 2023. Hyundai Mobis will display its latest technical advancements and provide visitors with a glimpse into the future of connected-car user experiences at booth B2, Hall D20. Ram Shallom, VP Business Development & Marketing

credit: Hyundai Mobis

for APAC (Asia-Pacific) at Autotalks, emphasized the collaborative success between Autotalks and Hyundai Mobis in advancing V2X technology, stating, “Over half a decade of partnership with Hyundai Mobis has enabled us to take a central part in the MTCU development process and fuel the cutting-edge MTCU product lineup, setting new standards for safety and mobility in connected vehicle environments. By utilizing Autotalks’ advanced chipsets and Hyundai Mobis’ automotive safety expertise, we’re poised to help automakers realize the full potential of V2X technology. The product of this partnership will enable car manufacturers an effortless integration of V2X as a safety critical sensor, advancing ADAS reliability to its ultimate level.” Dr. Joseph Kim, VP Product Planning Group at Hyundai Mobis says, “Partnership with the latest technology of Autotalks will bring a great advantage to the automotive industry. Our customers, future V2X drivers, pedestrians, and whomever on the roads will experience high performing autonomous driving features, safety and convenience while connected.”

1 V2X services in all geographies. Powered by Autotalks’ V2X chipset, the MTCU Gen1 sets the standard for high performance, globally compatible automotive V2X. Day-1 applications alert drivers in dangerous, non-line-of-sight situations, for example to avert a possible collision with a vehicle or bicycle at an intersection. Additional applications include alerts about road hazards. As the demand for V2X continues to grow, primarily driven by the global NCAP roadmap, Hyundai Mobis selected Autotalks’ 3rd generation chipset to power its next generation connectivity module solution, MTCU Gen2. The module, expected to be available in 2024, will enable car manufacturers to offer advanced safety applications based on V2X communication. MTCU Gen2 enables a range of use-cases from basic Day 1 to advanced Day-2 V2X applications. This is achieved through its inherent support for the latest global V2X standards, dual-channel operation, and the integration of Functional Safety-grade implementation, enabling car actuation based on V2X data. The MTCU Gen1 module will be showcased at the upcoming

onsemi and Magna Sign Strategic Agreements to Invest in Silicon Carbide for Growing Electric Vehicle Market

Magna to integrate onsemi EliteSiC into its traction inverter solutions to improve range and efficiency of electric vehicles onsemi (Nasdaq: ON), a leader in intelligent power and

sensing technologies, and Magna (NYSE: MGA; TSX: MG), a mobility technology company and one of the world’s largest automotive suppliers, announced a long-term supply agreement (LTSA) for Magna to integrate onsemi’s

New-Tech Magazine Europe l 13

Latest News

EliteSiC intelligent power solutions into its eDrive systems. By integrating onsemi’s industry-leading EliteSiC MOSFET technology, Magna eDrive systems can offer better cooling performance and faster acceleration and charging rates, improving efficiency and increasing the range of electric vehicles (EVs). Additionally, onsemi’s end-to-end silicon carbide (SiC) manufacturing capability, combined with its ability to ramp production quickly, improves Magna’s vertical integration and simplifies its supply chain to meet the growing demand for its SiC-based products for EVs. “With range anxiety still a top deterrent to EV adoption, our technology enables Magna to go further, easing the transition to an electrified future,” said Asif Jakwani, senior vice president and general manager, Advanced Power Division, onsemi. “Our latest EliteSiC MOSFET technology enables increased power density and higher efficiency in traction inverters, resulting in improved gas-equivalent miles per gallon without compromising driving dynamics and safety.” Simultaneous with the signing of the LTSA, the companies

entered a separate agreement for Magna to also invest approximately $40 million for the procurement of new SiC equipment at onsemi’s New Hampshire and Czech Republic facilities to ensure access to future supply. “We believe that a secure supply of silicon carbide chips will be critical to our ability to continue delivering innovative and efficient eDrive systems for our customers,” said Diba Ilunga, president Magna Powertrain. “Accordingly, we are both investing to grow SiC production capacity, and establishing the commercial basis for long-term supply of SiC-based chips to advance our electrification strategy and outpace the competition.” Silicon carbide is a wide bandgap semiconductor substrate that is ideal for high-temperature, high-power applications such as electric vehicles, but it is incredibly difficult to produce. With a limited number of manufacturers and significant demand for SiC-based designs, OEMs and automotive suppliers are increasingly looking to secure long term, reliable supply.

Analog Devices Invests More Than $1 Billion in Semiconductor Facility Expansion in Oregon

Analog Devices, Inc. (Nasdaq: ADI), a global semiconductor leader, celebrated its more than $1 billion investment to expand its semiconductor wafer fab in Beaverton, Oregon. The Beaverton site, built in 1978, is ADI’s largest wafer fabrication facility by volume and serves customers in critical industries, such as industrial, automotive, communications, consumer, and healthcare.

nearly doubling production output, the facility is targeting to reduce absolute greenhouse gas emissions by approximately 75% and water usage per production unit by approximately 50%.

“By expanding ADI’s Beaverton facility, we are increasing our production capacity in critical industries, boosting domestic manufacturing in line with the vision of the CHIPS Act, and enhancing the global resiliency of ADI’s

Credit: Analog Devices

hybrid manufacturing model,” said Vincent Roche, ADI’s CEO and Chair. “While our investment in Beaverton will facilitate these goals, they will be achieved through the incredible dedication and talent of ADI’s existing workforce and further tapping into Oregon’s strong talent pool.” Commenting on the announcement, U.S. Senator Ron Wyden of Oregon said, “Today’s good news from Analog Devices marks a significant step to ensure Oregon is taking full advantage of the federal CHIPS Act I worked to pass

The facility investment expands cleanroom space to about 118,000 sq-ft and nearly doubles internal manufacturing of products running on the 180-nanometer technology node and above. It is also expected to create hundreds of new, long term employment opportunities, a significant increase to ADI’s current roughly 950 employees in Oregon. More than 10% of the total investment is for new, state-of-the art fab tools intended to increase overall efficiency and utilize chemistries that are more environmentally friendly. Despite

14 l New-Tech Magazine Europe

Latest News

into law. This announcement strengthens a signature industry in Oregon by solidifying semiconductors’ place in the Silicon Forest and beyond, creating statewide impact with new good-paying jobs in Beaverton and rippling out to benefit the entire state economy.” In addition, the Beaverton facility will host the Semiconductor Advanced Manufacturing University (SAMU), a workforce development training center that will offer eight-week courses to train groups of around 25 students each on semiconductor equipment maintenance. A key driver of ADI’s efforts with the training center is providing training opportunities to many diverse groups within the local community, including U.S. military veterans, people re-entering the workforce, and existing ADI factory operators, to learn semiconductor

manufacturing fundamentals and other curriculum to advance their professional development. ADI’s hybrid manufacturing model is a broad network of internal factories and external partners. This model enables effective management of ADI operations through economic cycles, enhances the resiliency of its global supply chain, and better serves its customer needs. To improve ADI-owned operations, the company has increased capital investment to a high single-digit percent of revenue in fiscal year 2022 and to date in fiscal year 2023 from its historical precedent of around four percent. These manufacturing investments span the globe, including in the states of Washington and Massachusetts, as well as in Ireland, Thailand, Malaysia, and the Philippines.

Voia Raises $3 Million to Bring Hollywood Production Quality to Everyone

The founders include serial entrepreneurs who sold companies to Twitter, VMWare and IBM, and past executives from Sony Pictures and Warner Bros Voia, a startup developing AI based technology to enable everyone to produce Hollywood quality videos, announces a $3 million pre-seed fundraise from

the quality of music videos, TV ads and indie films is immense. The professional videos often have budgets exceeding $1 million, a large part of which goes into visual effects. Their production utilizes specialized 3D software requiring extensive expertise, are filmed using professional gear in a studio setting, and require massive

Credit: Voia

private investors. The company was founded in 2022 and employs 14 people across the US.

compute time to render. Voia uses AI to enable creators to achieve the Hollywood look using

Voia’s five founders bring together Hollywood and Silicon Valley. They include Avner Braverman, formerly co-founder and CEO of Reshuffle (acq Twitter); Mitch Singer, who served over two decades in leadership positions, including CTO, at Sony Pictures; Haim Helman, co-founder and CTO of Octarine (acq VMWare) and co-founder of XIV (acq IBM); Noam Malali, AI and computer vision expert; and Bryan Barber, formerly VP Digital Initiatives at Warner Bros. Voia’s app enables creators to dramatically upgrade the quality of their videos. The app employs advanced AI capabilities, producing visual effects with Hollywood quality. 100 million creatives are active on social video sharing platforms, like YouTube, TikTok and Instagram. But as these platforms continue to grow, most creator videos still look like home movies. The gap between video quality on social networks, and

only their iPhones. Noam Malali, co-founder and head of AI says: “We are at the forefront of AI. The challenges of achieving the Hollywood look using AI, require that we push the limits of our AI algorithms, and that we build a multidisciplinary product to enable creators to produce realistic visual effects. We need to produce cinematic videos at near real time speeds, so that users don’t have to wait hours to see their final videos. We will continue to push the envelope of technologies that enable every creator to produce Hollywood quality.” Avner Braverman, co-founder and CEO adds: “We founded Voia to bring Hollywood to every iPhone. We believe that when everyone can create such cinematic videos, we will see an explosion of creativity and engaging content in quantities and quality never before seen.”

New-Tech Magazine Europe l 15

A simpler method for learning to control a robot

Adam Zewe, MIT

Researchers develop a machine learning technique that can efficiently learn to control a robot, leading to better performance with fewer data. Researchers from MIT and Stanford University have devised a new machine learning approach that could be used to control a robot, such as a drone or autonomous vehicle, more effectively and efficiently in dynamic environments where conditions can change rapidly. This technique could help an autonomous vehicle learn to compensate for slippery road conditions to avoid going into a skid, allow a robotic free-flyer to tow different objects in space, or enable a drone to closely follow a downhill skier despite being buffeted by strong winds. The researchers’ approach incorporates certain structure from control theory into the process for learning a model in such a way that leads to an effective method of controlling complex dynamics, such as those caused by impacts of wind on the

trajectory of a flying vehicle. One way to think about this structure is as a hint that can help guide how to control a system. “The focus of our work is to learn intrinsic structure in the dynamics of the system that can be leveraged to design more effective, stabilizing controllers,” says Navid Azizan, the Esther and Harold E. Edgerton Assistant Professor in the MIT Department of Mechanical Engineering and the Institute for Data, Systems, and Society (IDSS), and a member of the Laboratory for Information and Decision Systems (LIDS). “By jointly learning the system’s dynamics and these unique control-oriented structures from data, we’re able to naturally create controllers that function much more effectively in the real world.” Using this structure in a learned model, the researchers’ technique immediately extracts an effective controller from the model, as opposed to other machine learning methods that require a controller to be derived or learned separately with

additional steps. With this structure, their approach is also able to learn an effective controller using fewer data than other approaches. This could help their learning-based control system achieve better performance faster in rapidly changing environments. “This work tries to strike a balance between identifying structure in your system and just learning a model from data,” says lead author Spencer M. Richards, a graduate student at Stanford University. “Our approach is inspired by how roboticists use physics to derive simpler models for robots. Physical analysis of these models often yields a useful structure for the purposes of control — one that you might miss if you just tried to naively fit a model to data. Instead, we try to identify similarly useful structure from data that indicates how to implement your control logic.” Additional authors of the paper are Jean Jacques Slotine, professor of mechanical engineering and of brain and cognitive

16 l New-Tech Magazine Europe

it achieved high performance even with few data. For instance, it could effectively model a highly dynamic rotor-driven vehicle using only 100 data points. Methods that used multiple learned components saw their performance drop much faster with smaller datasets. This efficiency could make their technique especially useful in situations where a drone or robot needs to learn quickly in rapidly changing conditions. Plus, their approach is general and could be applied to many types of dynamical systems, from robotic arms to free flying spacecraft operating in low-gravity environments. In the future, the researchers are interested in developing models that are more physically interpretable, and that would be able to identify very specific information about a dynamical system, Richards says. This could lead to better performing controllers. “Despite its ubiquity and importance, nonlinear feedback control remains an art, making it especially suitable for data-driven and learning-based methods. This paper makes a significant contribution to this area by proposing a method that jointly learns system dynamics, a controller, and control oriented structure,” says Nikolai Matni, an assistant professor in the Department of Electrical and Systems Engineering at the University of Pennsylvania, who was not involved with this work. “What I found particularly exciting and compelling was the integration of these components into a joint learning algorithm, such that control-oriented structure acts as an inductive bias in the learning process. The result is a data-efficient learning process that outputs dynamic models that enjoy intrinsic structure that enables effective, stable, and robust control. While the technical contributions of the paper are excellent themselves, it is this conceptual contribution that I view as most exciting and significant.” This research is supported, in part, by the NASA University Leadership Initiative and the Natural Sciences and Engineering Research Council of Canada.

Image: MIT News, with figures from iStock

Once they have modeled the dynamical system, many existing approaches also use data to learn a separate controller for the system. “Other approaches that try to learn dynamics and a controller from data as separate entities are a bit detached philosophically from the way we normally do it for simpler systems. Our approach is more reminiscent of deriving models by hand from physics and linking that to control,” Richards says. Identifying structure The team from MIT and Stanford developed a technique that uses machine learning to learn the dynamics model, but in such a way that the model has some prescribed structure that is useful for controlling the system. With this structure, they can extract a controller directly from the dynamics model, rather than using data to learn an entirely separate model for the controller. “We found that beyond learning the dynamics, it’s also essential to learn the control-oriented structure that supports effective controller design. Our approach of learning state-dependent coefficient factorizations of the dynamics has outperformed the baselines in terms of data efficiency and tracking capability, proving to be successful in efficiently and effectively controlling the system’s trajectory,” Azizan says. When they tested this approach, their controller closely followed desired trajectories, outpacing all the baseline methods. The controller extracted from their learned model nearly matched the performance of a ground-truth controller, which is built using the exact dynamics of the system. “By making simpler assumptions, we got something that actually worked better than other complicated baseline approaches,” Richards adds. The researchers also found that their method was data-efficient, which means

sciences at MIT, and Marco Pavone, associate professor of aeronautics and astronautics at Stanford. The research will be presented at the International Conference on Machine Learning (ICML). Learning a controller Determining the best way to control a robot to accomplish a given task can be a difficult problem, even when researchers know how to model everything about the system. A controller is the logic that enables a drone to follow a desired trajectory, for example. This controller would tell the drone how to adjust its rotor forces to compensate for the effect of winds that can knock it off a stable path to reach its goal. This drone is a dynamical system — a physical system that evolves over time. In this case, its position and velocity change as it flies through the environment. If such a system is simple enough, engineers can derive a controller by hand. Modeling a system by hand intrinsically captures a certain structure based on the physics of the system. For instance, if a robot were modeled manually using differential equations, these would capture the relationship between velocity, acceleration, and force. Acceleration is the rate of change in velocity over time, which is determined by the mass of and forces applied to the robot. But often the system is too complex to be exactly modeled by hand. Aerodynamic effects, like the way swirling wind pushes a flying vehicle, are notoriously difficult to derive manually, Richards explains. Researchers would instead take measurements of the drone’s position, velocity, and rotor speeds over time, and use machine learning to fit a model of this dynamical system to the data. But these approaches typically don’t learn a control-based structure. This structure is useful in determining how to best set the rotor speeds to direct the motion of the drone over time.

New-Tech Magazine Europe l 17

Hailo Expands Hailo-8 AI Accelerator Portfolio, Delivering Unprecedented Performance to a Diverse Range of Edge AI Applications

HAILO

The new Hailo-8 Century and Hailo-8L AI accelerators extend the Hailo-8 offering with a wide range of AI platforms to enable high performance AI in edge devices ranging from entry level to high-capacity deployments. Hailo, the pioneering chipmaker of edge artificial intelligence (AI) processors, today announced it has expanded its groundbreaking Hailo-8™ AI accelerator offering following hundreds of successful deployments in customer programs and products. The new high-performance Hailo-8 Century PCIe card line offers up to 208 Tera Operations per Second (TOPS) for most demanding applications, and the Hailo-8L makes advanced AI processing available for entry level applications. Both product lines are offered at a competitive price compared to the respective category leaders.

“The expansion of our Hailo-8 AI accelerator portfolio is unlocking new opportunities for our customers to harness real-time, power-efficient intelligence in a diverse range of applications and industries,” said Orr Danon, CEO of Hailo. “With the rise of generative AI driven applications, our powerful and cost-efficient solutions bring unmatched AI performance and power efficiency, enabling state-of-the art transformer-based models such as ViT, CLIP and SAM, at the edge.” The competitively priced Hailo-8L, with up to 13 TOPS, is designed to support entry level products requiring limited AI capacity or lower performance. The Hailo-8 Century line, comprising PCIe cards delivering 52 to 208 TOPS, enables demanding applications such as video management systems handling a large number of video streams. The Hailo-8 Century PCIe cards deliver best-in-class power efficiency at 400 FPS per watt on the ResNet50 benchmark model.

They are also offering the highest cost efficiency, starting at $249 for the 52 TOPS card, representing up to 70% reduction in edge AI deployment costs. They feature a robust design that supports industrial temperature ranges and ensures compatibility with virtually any environment or application. “We see the Hailo-8 Century product line as an enabler for many market verticals, such as security, retail and smart cities, which show increasing demand for AI compute power in large camera deployments,” said Tom Larson, President of Velasea. “By offering Hailo-8 Century empowered edge boxes, we offer our customers a cost effective solution for high performance video analytics at scale.” Hailo-8 Century and Hailo-8L both boast exceptional low-latency, high-efficiency processing, capable of handling complex pipelines with multiple real-time streams and concurrent processing of multiple models and AI tasks. The expanded

18 l New-Tech Magazine Europe

About Hailo: Hailo, an AI-focused chipmaker, is developing specialized AI processors that enable data center-class performance on edge devices. Hailo’s processors are the product of a rethinking of traditional computer architecture, enabling smart devices to perform sophisticated deep learning tasks such as object detection and segmentation in real-time, with minimal power consumption, size, and cost. The processors are designed to fit into a multitude of smart machines and devices, impacting a variety of sectors including automotive, security, industry 4.0, and retail. For more information visit https://hailo.ai.

portfolio of Hailo-8 products features seamless scalability for future upgrades, as all products are compatible with the same field-proven and comprehensive Hailo-8 software suite. For reference, the Hailo-8L can run the benchmark classification model ResNet50 at 500 FPS, and the Century high performance PCIe cards can run this model at up to 10K FPS, outperforming Nvidia’s comparable products in performance, cost efficiency and power efficiency parameters. "With the power of Hailo-8 now accessible to a wider range of applications and products, catering to both demanding and entry-level use-cases, we are driving the industry forward in an era where efficient, scalable edge AI processing is increasingly important for businesses to achieve their targets using enhanced processing and analytic capabilities with unparalleled speed

and accuracy," Danon concluded. "This significant expansion marks a milestone in delivering seamless real time intelligence for diverse edge AI applications ranging from security and smart cities to transportation, smart retail, industrial automation, automotive, and beyond." The Hailo-8 expansion follows Hailo’s successful launch of the Hailo-15™ vision processor units in Q1 2023, which are currently sampling and integrate unprecedented AI performance directly into cameras, enhancing both video quality and video analytics. Both Hailo 8L and the Hailo-8 Century product lines are available for order. High AI capacity products similar to the Hailo-8 Century are available also from Hailo’s ODM partners Aetina, Lanner, and Mitac. For more information about Hailo's AI processors for edge devices, visit hailo.ai.

Read To Lead

www.new-techeurope.com

New-Tech Magazine Europe l 19

High-performance boost for modular systems

Christof Wilde, Content Manager at congatec

The PICMG created the COM-HPC Computer-on-Module standard with the specific aim to address the performance and bandwidth requirements of the digital transformation. Ranking well above COM Express, COM-HPC provides a highly sustainable long-term basis for all current and future applications. The COM HPC Mini specification facilitates the migration of existing designs to this new high-performance standard. To start with, it’s important to note that COM Express will continue to be actively supported with new modules and updates. After all, there are countless applications that do not require a performance boost above PCIe Gen 4 and GbE or that need significantly more interfaces than before. So, only applications that actually require these new features will migrate. After all, it

was the introduction of PCIe Gen 5, among other things, that necessitated the creation of a new specification that would open the potential of latest processor technology, such as 13th generation Intel Core processors, to Computer-on-Modules targeting cost sensitive applications requiring long term design security. Smaller always fits The new COM-HPC Mini specification, which is expected to be fully ratified in the first half of 2023, makes migration from existing COM Express designs comparatively easy. Compared to COM Express Compact, COM-HPC Mini has a smaller footprint and a flatter design, yet with 400 high-performance pins offers almost the same number of interfaces via a single connector, whereas COM Express needs two. So, from a mechanical perspective, integrating such modules into existing system designs doesn’t present a

challenge. All that needs adapting is the carrier board. OEM customers looking to design next-generation systems therefore face the question: Should they use 13th generation Intel Core processors on COM Express Compact? In that case, all they need to do is integrate new Computer-on-Modules. Or should they use 13th generation Intel Core processors to leverage all the new features that COM Express doesn’t support in these next-generation systems? Ultimately, and apart from long-term strategy decisions, the question boils down to whether the current performance range will be adequate for the new designs, or whether it’s better to opt for a high performance upgrade path. More powerful interfaces One of the high-performance reasons to opt for COM-HPC is certainly the ability to leverage the 2x higher

20 l New-Tech Magazine Europe

PCIe Gen 5 bandwidths with 32 gigatransfers per second (GT/s). This makes it possible to use latest GPUs and SSDs in combination with latest processor technology, such as 13th Gen Intel Core processors, for optimum results. Higher bandwidth is immensely important because of the ever-larger AI and high-performance computing datasets, and because loading this data is crucial for the end-to-end performance of applications. If the systems are massively networked as edge nodes, native on-board support of a second Ethernet interface as well as higher Ethernet bandwidth are also essential. The leap from 1x GbE to up to 2x 10 GbE with TSN and 2x 10 GbE via Serdes offers an immense performance boost for such systems, along with the option to implement line or ring networking instead of star networking. Another useful feature is the support of up to 4x USB 4.0, 4x USB 3.2 x1 / USB 3.2 x1 + 8x USB 2.0 for numerous extensions or easy Thunderbolt 4 implementation. This allows transmission of power supply, user input, and high-resolution video signals to HMIs via a single cable, or to connect up to five additional devices in series: docking stations, hubs, external PCIe devices, external graphics, and up to two 4k displays. This and numerous other enhancements make COM HPC the standard of the future. And thanks to COM-HPC Mini, it can now be implemented with relatively little effort since in principle only the carrier board and the cooling solution for the module need to be adapted. embedded world highlight: COM-HPC Mini The flagships among congatec’s embedded world presentations are therefore early samples of a COM HPC Mini design. These first high performance COM-HPC Mini modules, which will be officially launched after the

Figure 1: 13th generation Intel Core processors: congatec provides them on COM Express Compact as well as COM-HPC Client Size C, A and Mini modules (from left). Credit: congatec

Figure 2: The many new high-speed interfaces provided on COM-HPC Mini’s footprint of 95x70 mm are very attractive for compact high-performance box

and DIN rail PCs. Credit: congatec

New-Tech Magazine Europe l 21

final ratification of the new specification by the PICMG, are equipped with the new 13th generation Intel Core processors (codenamed Raptor Lake), which set the latest benchmark for the high-end of embedded and edge computing at client level. Together with the recently introduced high-performance Computer-on Modules with 13th generation Intel Core processors on COM-HPC Client Size A and Size C, congatec now offers developers the full module bandwidth on COM-HPC for this new processor generation. Thanks to state-of-the-art connectivity, the COM-HPC standard opens new horizons for developers of innovative designs with demands that go beyond what is achievable with COM Express in terms of data throughput, I/O bandwidth, and power density. congatec’s COM Express 3.1 compliant modules with 13th generation Intel Core processors, on the other hand, primarily help to secure investments in existing OEM designs, for instance by providing a path to upgrade data throughput thanks to PCIe Gen 4 support. But how can OEMs accomplish a COM HPC implementation quickly? This is where manufacturers like congatec come to the aid of customers with a dedicated training academy, which offers carrier board design courses to provide developers with best-practice knowledge for the design-in. The goal is to give system architects a quick, easy and efficient introduction to the design rules of the new PICMG standard. For that purpose, the training courses guide developers through all mandatory and recommended design basics and best practice schematics for Computer-on-Modules to then be able to start their own carrier board design projects. The focus of the knowledge Add-on services and training to simplify design-in

transfer is on standard-compliant carrier board designs that are essential for building interoperable, scalable and durable custom embedded computing platforms. Operating globally, the academy is offering online and on site courses and targets developers at OEMs, VARs and system integrators. Customers who want to use COM HPC but do not have the resources to integrate the modules themselves, can avail of congatec’s design-in services to handle this for them. Carrier board design, optimized cooling solution design, and even complete system integration services are also available. With this full range of products and services, congatec continues to build on its customer promise: “we simplify the use of embedded technology”. By providing such an all-round offering – from modules, cooling solutions, application-ready evaluation and application carrier boards, to design in training and complete design-in services – congatec is creating an extensive COM-HPC ecosystem in which new designs and design sprints for migration from COM Express to COM-HPC Mini are no longer a major challenge.

Figure 3: “Our COM-HPC Mini carrier board design course also teaches developers how to migrate COM Express Basic designs to COM-HPC Mini.” Daniel Stadler, Manager Support & Design-In at congatec. Credit: congatec

Christof Wilde, Content Manager at congatec Credit: congatec

Figure 4: congatec’s Server-on-Modules with Intel Xeon D 2700 processors in the compact COM-HPC Server Size D form factor. Credit: congatec

22 l New-Tech Magazine Europe