X-FAB Expands Offering for Silicon-Based Microfluidics

www.eenewseurope.com June 2020

electronics europe News News

X-FAB expands offering for silicon-based microfluidics

Executive Interview: Tim Ramsdale CEO of Agile Analog Special Focus: ASICs, FPGAs & Accelerators

european business press November 2011 Electronic Engineering Times Europe1

200526_9_Mill_EENE_EU_BRC_Snipe.indd 1 5/26/20 12:25 PM 200526_TCRL_EENE_EU.indd 1 5/26/20 3:52 PM CONTENTS JUNE 2020

Dear readers, www.eenewseurope.com

June 2020 On the socially and economically devastating COVID-19 trail, European countries are electronics europe News trying to get their act together, companies share their workaround tips in time of the News coronavirus, alleviating what was initially seen as a huge supply chain disruption for European industries.

The June edition of eeNews Europe reflects some of the initiatives taken to tackle X-FAB expands offering for our dependence on China for manufacturing. More resilient product design is an silicon-based microfluidics emerging trend, somehow becoming a necessity for businesses to secure the stability of their operations, and distribution in general has never fared better than during the pandemic.

Other topics covered in your June edition include three technology focuses, namely: Sensors & Transducers, LEDs & Display Technologies, and the now Executive Interview: Tim Ramsdale pervasive ASICs, FPGAs and Accelerators, mostly concerned with bringing AI CEO of Agile Analog to the edge. Special Focus: ASICs, FPGAs & Accelerators

european business press Julien Happich

November Editor in Chief 2011 Electronic Engineering Times Europe 1

200526_9_Mill_EENE_EU_BRC_Snipe.indd 1

5/26/20 12:25 PM

4 - 49: OPINION 30 - 39: ASICS, FPGAS & ACCELERATORS Uncommon Market: Universal processor startup goes for split tape Keeping electronics product Semiconductor startup design on track during Tachyum Inc. will be coronavirus. using a split tape-out strategy for its Prodigy Last Word: processor before the end How to fight a virus: Les- of 2020 in an effort to get sons from cybersecurity samples into customers’ hands early in 2021. 6 - 23: NEWS & TECHNOLOGY 40 - 45: SENSORS & TRANSDUCERS TSMC dragged to the altar of US manufacturing Nano-optomechanical displacement sensor The double-whammy of reaches 45fm resolution enhanced restrictions on Researchers from Huawei and HiSilicon from the Eindhoven University US, plus the announcement of of Technology have TSMC’s plan to build a 5nm wafer fab in Arizona has drastically increased the the feel of stage management about it. resolution of such systems, leveraging a newly designed nano-optome- 24 - 29: LEDS & DISPLAY TECHNOLOGIES chanical system (NOMS) with unprecedented measurement resolution. Lumotive drives competitiveness of lidar sensors into a new dimension 49: DISTRIBUTION CORNER With the presentation of its first commercial solid-state lidar sensors, start-up company Lumotive has further advanced the development of its technology towards economically viable series application in driver assistance systems.

www.eenewseurope.com eeNews Europe June 2020 News 3 UNCOMMON MARKET COVID-19 Keeping electronics product design on track during coronavirus By Dunstan Power he COVID-19 pandemic has thrown up a number of chal- pieces of equipment or hardware-related issues that re- lenges for electronics teams, both in-house and within quire test gear. Even here though, with a bit of imagination, design consultancies. While resources for remote working the need for several visits the office can be reduced. For abound,T here we explore how electronics teams rise to the trials example, on one of our projects, the engineers developed of remote working and successfully deliver projects over the a work-around by using Arduinos to cycle power to boards next few months. and web cams to monitor the status of LEDs and displays on the PCBs. This has reduced significantly the number of Establish a framework people in the barebones office. Factors such as CAD tool licenses, ESD protection and test equipment access are just a few factors you need to check, to Ensure software development supported by ensure your electronics product design team can deliver what a secure IT infrastructure customers need on time. A solid IT infrastructure like a secure and good quality VPN needs to be in place. Take easily portable IT equipment home to Determine viability of home working for reduce VPN loading and improve ease of use, rather than Virtual electronics engineers Network Computing (VNC) and a remote desktop. Ensure you You also need to be confident that your electronics design en- have adequate server bandwidth for remote desktops and that gineers are able to continue working on your project from their your engineers are working with (preferably at least two) moni- own homes, so establish tors and are not just huddled over small laptop screens. Think about how your software engineers will communicate 1. If they have access to all the tools necessary for the job with customers. Some projects may have a screen that requires – for example soldering equipment, spectrum analysers, interaction via a mouse or touch inputs and this could be prob- Multimeters, PSUs, oscilloscopes and test equipment. lematic when working from home – particularly if the user can’t 2. How suitable their home working environment is. Electron- access the embedded device screen that they’re working on. ics and firmware engineers typically need more space as There are workarounds though, which allow the user to they will have hardware and test equipment being plugged remotely control a screen. Windows Remote Desktop or VNC into their desktop or laptop PCs. Some of our design con- are two such workarounds as both of those systems comprise sultants have repurposed whole rooms while working from two key components: a server which shares the screen of the their homes and some have considered ergonomics and device, and a viewer which displays the screen received form taken desk chairs home. the server at the remote end. Windows Remote Desktop uses 3. How they can maintain electrical safety, for themselves built in hardware acceleration on PCs and is the quicker of the and others in their household. This is particularly important two when available in a Windows environment. VNC is more when working with any equipment that is mains powered portable and can be used on Linux desktop and embedded or generates high voltages or heat. Prototype devices boards. There are multiple software tools that offer a remote should not be left powered and unattended in a domestic control service that enables users to control the remote device environment, but sometimes this can’t be helped dur- from any modern browser, including mobile devices. Although ing testing, so bear in mind that even the most harmless they are more portable, mobile devices may also need to be in looking board could become a fire hazard if it’s not treated close proximity to the embedded devices in the office, but there appropriately. are free tools such as Genymobile (https://github.com/Genymo- 4. How privacy and security of data will be addressed. Many bile/scrcpy) which enables users to view and control a con- projects mean working with sensitive data and this needs nected Android device. Again, here both your project engineers to be taken care of in the exact same way at home as it and those from your design partners all need to be well-versed would be at work. in your development tools, platforms and IDEs. 5. How products that are under development are being protected, in particular, whether ESD protection is in place. Adapt design tool licenses At ByteSnap we provided all of our engineers working on Remember to check your license servers, as you might need unenclosed PCBs at home with earthing matts. to switch from node-locked licence to a floating one. If you’re 6. Can they have access to a “bare-bones” shared office. sharing software that could previously only be accessed using If they are still able to access their regular place of work, a node-locked device (node locked is where the license of a while maintaining the required social distancing, this can CAD tool, compiler etc only allows the operation on a single be very helpful in sustaining project momentum, but you PC by the use of a MAC address or plug in dongle), that would would expect these visits to be limited to members of the normally be fine in a shared office. design team only who absolutely need to use the work- However, with your electronics design engineers now all place – such as those who are dealing with either large working remotely, you will probably need to check your licence terms and see if you can upgrade to a floating license that can Dunstan Power is Director of ByteSnap Design - be accessed across a VPN. Talk to your licence suppliers as www.bytesnap.com some of them may be offering concessions such as free or re-

4 News June 2020 @eeNewsEurope www.eenewseurope.com COVID-19

duced rate licence upgrades during the lockdown. For example, waiting for component deliveries. OrCAD are currently providing a 30 day free license to users to Liaise with your suppliers about how they will get any neces- help them work from home (see https://www.orcad.com/orcad- sary parts from you for your development. Will they deliver to work-from-home-program). your barebones office or directly to engineers’ homes? Sort those details out as soon as you can to avoid vital components Factor in supply chain disruption going astray, which will only add costly delays to projects. Given the disruption to various industry supply chains globally, you may need to build extra time into your project, for delays Keep in touch with regular communications around sourcing and shipping of parts/modules/components. We’ve found it very helpful to stick to regular, scheduled com- Schedule kitting of parts early on in a design process, as munications, both company-wide and project-specific. We’ve even for stock items, parts are on extended lead times. Plans also increased our team briefings to build on our sense of com- will need reworking to take this into account. For instance, munity and have more space to discuss operational matters and where we would normally kit each build of prototypes prior to well-being. Use common platforms as there are a large number the revision, we have now switched to kitting later builds as well available. Whichever one you choose, make sure everyone - as the first revision, once the BOM is ready. The small wastage from new starters and tech newbies to your team’s tech gurus from changes to the design after the first revision is easily offset – is able to use all of the features. This will keep all-important by the potential saving in time which would have been spent communications channels clear. Innovation during COVID-19: Digi-Key’s ultraviolet sanitisation tunnel By Ian Wallace very employee at Digi-Key is famil- UV tunnel concept on 26 March. After several iar with the phrase “a thousand little Digi-Key employees spent the weekend double- things done right every day.” This is checking the math and design concept, they aE guiding principle in our company – from were ordering parts by the following Monday. the warehouse to our offices throughout The result was a 4.5-metre ultraviolet saniti- Europe and around the world. While this sation tunnel built into the curve of the conveyor is an unprecedented time in history, the belt in record time. By using UV light, the inven- phrase still rings true for us. tion sanitises the 8,000 plus totes that travel Now, more than ever, we’re committed throughout our warehouse every day – drasti- to doing a thousand things right – and de- cally reducing potential virus contact for our veloping innovations and protocols designed to protect those who employees. We have been quick to share this potentially life-saving matter most - our employees, our customers and our community. invention with others in the industry. In addition to sharing the One of those innovations is an ultraviolet sanitisation tunnel inven- design, Digi-Key president Dave Doherty is actively working with tion. As COVID-19 uneasiness moved to the forefront of everyone’s other distributors on the Electronic Components Industry Associa- minds, we identified an opportunity to sanitise the 8,000 plus totes tion (ECIA) distribution council to share other ideas and best prac- that travel through our intelligent conveyor system each day. tices for helping to protect the safety of employees and customers. To ensure warehouse employees were kept as safe as pos- We’re all in this together, and by sharing innovations that will help sible, Digi-Key quickly assembled a six-person team dedicated to businesses protect their employees and communities, we all gain. solving this uniquely COVID era problem. At first, the team looked The COVID-19 pandemic has been a global crisis in every into machines that spray sanitising chemicals. These are typically sense of the word, but there’s no denying there have also been found in schools, medical settings, etc. With schools being closed, some bright spots. People are coming together even when they’re the team was able to borrow the machine from a local school. forced to be apart - and the electronics components industry is The team quickly discovered they would need more machines no exception. We’ve all needed to adapt quickly to this continually for all of the totes to be properly sanitised, but unfortunately, evolving crisis, and respond by doing what we do best: innovating. these machines were all on backorder. That didn’t hinder them The ultraviolet sanitisation tunnel is just one of many projects from quickly developing a brand-new design – one that effectively our company has been involved in during this crisis; we’ve also sanitised the totes, while also having the portability needed for the supplied components necessary for other medical devices used warehouse environment. In addition to spraying a sanitising chemi- on the frontlines of diagnosing and treating COVID-19 such as a cal, the team wanted to incorporate the added security of Ultravio- first-of-its-kind, low-cost ventilator, forehead thermometers and let Germicidal Irradiation (UVGI) – a technique used by many busi- touchless Purell sanitiser dispensers. nesses to kill and deactivate microorganisms by destroying nucleic Like so many, our employees have embraced their status as acids and disrupting their DNA. The team began discussing the essential workers and risen to the occasion to adapt, adjust, continue to thrive and meet customers’ needs, even in challenging Ian Wallace is Director, EMEA Business Development at Digi- circumstances. We’ve never been prouder of our employees and Key Electronics where he leads the company’s global expansion customers, who have pivoted to implement fast-changing safety by supporting new and existing customers and supplier protocols, and also helped distribute the essential products and partners in the region - www.digikey.com services for fighting against COVID-19. www.eenewseurope.com eeNews Europe June 2020 News 5 COVID-19

Data science in the times of Corona: (Some) reassembly required By Michael Berthold

he enormous impact of the current crisis is obvious. What a significant portion of the information that went into model many still haven’t realized, however, is that the impact on construction, the data science team is well advised to rerun the ongoing data science production setups can be dramatic, model selection and optimization process as well. too.T Many of the models used for segmentation or forecasting started to fail when traffic and shopping patterns changed, sup- Impact scenario: partial change ply chains were interrupted, borders were locked down, and just In many other cases, the base assumptions remain the same - in general the way people behaved recommendation engines, for ex- changed fundamentally. Sometimes, ample, still work very much the same, the data science systems adapted but only some of the dependencies reasonably quickly when the new extracted from the data change. This data started to represent the new is not necessarily very different from, reality. In other cases, the new real- for example, a new bestseller enter- ity is so fundamentally different that ing the charts, but the speed and the new data is not sufficient to train magnitude of change may be bigger: a new system, or worse, the base the way, for instance, health- related assumptions built into the system supplies jumped in demand outpaces just don’t hold any more, so the how a bestseller rises in the charts. entire process from data science If the data science process has been creation to productionizing must be designed sufficiently flexible, its built-in revisited. change detection mechanism should This article describes different identify the change and quickly trigger scenarios and a few examples of a retraining of the underlying rules - if what happens when old data be- it was built-in and the quality of the re- comes completely outdated, base trained system does achieve sufficient assumptions are not valid any more, quality levels. or patterns in the overall system change. I then highlight some of the challenges data science teams face when updating their Impact scenario: no change production system and conclude with a set of recommenda- This brief list is not complete without stressing that many tions for a robust future-proof data science setup. concepts remain the same: Predictive maintenance is a good example - as long as the usage patterns stay the same, engines Impact scenario: complete change will continue to fail the same way as before. But the important The most drastic scenario is a complete change of the underly- question here for your data science team is: Are you sure? Is ing system that not only requires an update of the data science your performance monitoring setup thorough enough that you process itself but also revising the assumptions that went into can be sure you are not losing quality? This is a predominant its design in the first place. This requires a full new data sci- theme these days anyway: Do you even notice when perfor- ence creation and productionization cycle: understanding and mance of your data science system changes? incorporating business knowledge, exploring data sources (pos- A little side note on model jump vs. model drift which is also sibly to replace data that doesn’t exist any more), and selecting often used in the context but refers to a different aspect. In and fine tuning suitable models. Examples of this are traffic the first two scenarios above (complete or partial change), that predictions (especially near suddenly closed borders), shop- change can happen abruptly (when borders are closed from one ping behaviour under more or less stringent lock downs, and day to the next, for example) or gradually over time (some of the healthcare-related supply chains. bigger economic impacts will be visible in customer behaviour A subset of the above is the case where the availability of the only over time; take a SaaS business, for example, customers data changed. A very illustrative example here is weather pre- will not cancel their subscriptions overnight but over coming dictions where quite a bit of data stems from commercial pas- months). senger aircraft that are equipped with additional sensors. Those aircraft remaining on the ground suddenly reduces the available What’s the problem? data drastically. Base assumptions about weather develop- In reality, one most often encounters two types of production ment itself remain the same (ignoring for a moment that other data science setups: the ones that were built once, deployed changes in pollution and energy consumption may affect the and have been running for a while without any further refine- weather as well), so “only” a retraining of the existing models ments, or the data science process may have been the result of may be sufficient. However, if the missing data really represents a consulting project or maybe even the outcome of a modern automated machine learning (AutoML) type of project. In both Michael Berthold is CEO and co-founder at KNIME, an open cases, if you are fortunate, automatic handling of partial model source data analytics company - www.knime.com change has been incorporated into the system, so at least some

6 News June 2020 @eeNewsEurope www.eenewseurope.com European Headquarters: www.keyelco.com • 33 (1) 46 36 82 49 • 33 (1) 46 36 81 57

EENewsEuro THiNK Fist+T-1SpacerMt_5-20prod.indd 1 5/4/20 10:42 AM European Headquarters: www.keyelco.com • 33 (1) 46 36 82 49 • 33 (1) 46 36 81 57

EENewsEuro THiNK Fist+T-1SpacerMt_5-20prod.indd 1 5/4/20 10:42 AM NEWS & TECHNOLOGY HIGH PERFORMANCE COMPUTING

of the partial changes are handled automatically. But since none engage that service, but especially in the case of the change in of the currently available AutoML tools allow for performance business dynamics, you should expect to be involved quite a bit monitoring and automatic retraining (and usually one-shot proj- - similar to the first time this project was run. ects don’t worry about that either), you may not even be aware One side note here: Be skeptical when someone is trying that your data science process has failed. to push for super cool new methods. In many cases, this is If it is more of a setup where the data science team has not needed, but one should rather focus on carefully revisiting made numerous improvements over the years, chances are the assumptions and data used for the previous data science higher that automatic model drift detection and retraining is process. Only in very few cases is this really a “data 0” problem built in as well. However, even then - and especially in case of where one tries to learn a new model from very few data points. a complete model jump - it is far more likely that the existing Even then, one should also explore the option of building on system cannot easily be recreated to accommodate the new top of the previous models and keeping them involved in some setup because all those steps are not well documented, making weighted way. Very often, new behaviour can well be represent- it difficult to revisit the assumptions and update the process. ed as a mix of previous models with a sprinkle of new data. Also, often the process relies on obscur code pieces, written by But if your data science development is done in-house, experts who have left the team in the meantime. The only solu- now is the time where an integrative and uniform environment tion? Start an entirely new project. that’s 100% backwards compatible comes in very handy: The assumptions are all modelled and documented in one environ- What’s needed? ment, and well-informed changes and adjustments can be Obviously, if your data science process was set up by an ex- made. It’s even better if you can at the end validate, test and ternal consulting team, you don’t have much of a choice other deploy this into production from that same environment without than to bring them back in. If your data science process is the the need for manual translation/interaction. result of an automated ML/AI service, you may be able to re- The impact of the new COM HPC standard By Christian Eder and Martin Frederiksen t’s been three years since the latest Computer on Module of PCIe Gen 3.0 with 8 Gb/s per lane, a COM-HPC module (COM) specification, COM Express 3.0, was released for in- achieves up to 32 Gb/s per lane via PCIe-5.0 — that’s four dustry use. Now, the embedded computer industry is on the times the data rate of COM Express. COM-HPC modules will vergeI of a new standard: COM for high-performance computing therefore be used in particularly performance-hungry applica- (COM-HPC), which is due to be ratified this year. In a joint Q&A tions, for instance to embed artificial intelligence with deep session, chairman of the PICMG COM-HPC technical subcom- learning in embedded systems, or even to implement tactile mittee and marketing manager at congatec, Internet at the edge server level.” Christian Eder, and Martin Frederiksen, managing director of embedded computing sys- On the subject of edge tems provider Recab UK, explain the impact servers, what performance this new standard will bring to the market. enhancements can be expected from COM-HPC modules in What are the main differences terms of Ethernet connectivity? between COM-HPC and COM CE: “The enormous speed increase has an Express? immense effect on the connectivity perfor- Christian Eder (CE): “Computer-on-Modules mance. Current COM Express modules (Type based on the new COM-HPC standard prom- 7) at edge server level offer a maximum of 10 ise considerably higher transmission perfor- Gb Ethernet per signal pair. COM-HPC, on mance, many more high-speed interfaces the other hand, specifies 25 Gb Ethernet, and and significantly faster network connection, more. besides other benefits. This is down to a With up to eight network connections, it completely redesigned, more powerful new then becomes possible to achieve transfer module to carrier board connector. While rates of 100 Gbit/s, and theoretically even COM Express establishes this connection 200 Gbit/s. Such rates are needed in the first with 440 pins, the COM-HPC specification instance for high-performance edge server so- provides 800 pins. This doubles the maxi- lutions at the edge of telecom networks. Here, mum number of PCIe lanes from 32 for COM fast up, down and crosslinks in all directions Express Type 7 to 64 for COM-HPC/Server. must be established: i.e. north in the direc- While COM Express supports a maximum Christian Eder tion of the central cloud; east and west in the direction of neighbouring edge fogs; and also Christian Eder is chairman of the PICMG COM-HPC technical south in the direction of industry 4.0 controls at process level.” subcommittee and marketing manager at congatec – www.congatec.com How will the changes presented by COM- Martin Frederiksen is managing director of embedded HPC impact board design, and what new computing systems provider Recab UK - www.recabuk.com system potential might this bring?

8 News June 2020 @eeNewsEurope www.eenewseurope.com HIGH PERFORMANCE COMPUTING

Martin Frederiksen (MF): “The most significant changes to baseboard design will relate to signal routing. With the COM-HPC standard, the higher speeds of PCIe-4.0 and 5.0 — which as Christian mentioned, can be up to four times faster than the PCIe-3.0 speed — will require very careful routing to ensure signal integrity. In addition, these signals will not travel as far as previous versions. There are several steps that COM system design and production companies such as Recab UK can take to mitigate for this. For example, PCIe retimer chips may be used to extend signal strength if the trace length requirements are long. We can also look at the material of the printed circuit board (PCB) itself. For many years, the base material of the PCB has been FR-4, which has been sufficient for the data transmission required for COM Express modules with PCIe-3.0. With COM-HPC, achieving the best performance will require changing to a base material with lower transmission loss, such as Megtron-4 laminates. However, these adaptions to baseboard design will Martin Frederiksen certainly pay off in the boundless system potential of the standard. Bringing high performance computing out of the environments and require the largest possible memory capac- server room and into the embedded market will open up new ity, a particularly powerful network connection and the option opportunities in areas such as artificial intelligence, 5G network- to provide many cores for consolidating high workloads. These ing and machine vision. Server-on-Modules will host the mentioned eight DIMM sockets For example, the capacity for high data throughput and low on a 200x160mm footprint, while the smaller 160x160mm latency will benefit imaging applications, such as 3D imaging in server modules will integrate up to four DIMM sockets. medical equipment and areas like 3D mapping of land for rail infrastructure and the seabed for subsea operations. Overall, The COM-HPC/Client modules have a slightly more com- this means a new generation of embedded computing applica- pact design, are also planned in two footprint variants — tions for a wide range of vertical markets, such as telecommu- 120x120mm and 160x120mm — and are designed for use in nications.” high-end embedded computing applications. Unlike the server modules, they provide a maximum of 2x GbE interfaces (via As an experienced developer and provider NBASE-T) for Ethernet connection. In addition, COM-HPC/Cli- of embedded systems, where do you see ent modules integrate video interfaces such as DDI and eDP/ the most challenges for baseboard designs MIPI-DSI, which – in contrast to COM-HPC/Server modules – using COM-HPC modules? can be used to control up to four independent high-resolution MF: “The higher data transfer speed of up to 100 Gbit/s and 65 displays.” PCIe lanes means that the baseboard design becomes really tricky and you need the right equipment to ensure good signal For end users, what will COM-HPC systems integrity. Here, the experience of Recab UK is invaluable, as we make possible? have made hundreds of baseboard designs and have good con- MF: “We anticipate that telecom and broadcasting will be an nections with congatec’s engineers to ensure we have access interesting area to watch for COM-HPC. This was where we to the best equipment. saw the COM Express Type 7 quickly accepted and where you One of the biggest challenges beyond that of maintaining have the need for 24/7 operation, high data throughput and low signal integrity will be thermal design. These systems will fea- latency. The COM-HPC will be used where you cannot use a ture processors operating in excess of 100 Watts in a compact temperature cooled data centre, such as edge cloud applica- footprint, so an effective means of cooling these processors will tions. be imperative to minimise the risk of overheating and compo- Therefore, I can see a need for these modules for faster nent degradation in these embedded systems.” telecommunications in rail and defence. In both these areas, this will allow closer to real-time communication of data be- Speaking of footprints, does the new COM- tween on-board or remote embedded systems and centralised, HPC standard specify several module manned computers. Naturally, faster transmission of data in classes? these critical sectors is highly advantageous. CE: “Yes, just as there are currently Type 6 and Type 7 high- Another area will be the emerging 5G network infrastructure, end specifications for COM Express, we have also planned two where fast computing and communications will be a neces- module classes for COM-HPC that address different application sity and the flexibility of scalable processor modules and 25Gb and performance requirements. In addition, there are two dif- ethernet will come to the fore.” ferent form factors within these two module classes, similar to The COM-HPC standard is expected to be fully ratified by COM Express Basic and COM Express Compact. To be precise, mid-2020, with the first COM-HPC modules beginning to come we currently distinguish between server and client modules, in to market. For further information on the COM-HPC standard, analogy to client/server computing. visit the PICMG website or contact Recab UK to see how the COM-HPC/Server modules are tailored for use in edge server latest COM technologies can benefit your project. www.eenewseurope.com eeNews Europe June 2020 News 9 NEWS & TECHNOLOGY EXECUTIVE INTERVIEW The importance of being agile By Peter Clarke

e asked Tim Ramsdale, CEO of Agile Analog Ltd. don’t solve that challenge but we have a way round that makes (Cambridge, England), to say more about the core of it become less important.” Agile Analog’s technology and also how he sees the It is worth noting that company founder and CTO Mike impactW of the Covid-19 pandemic. Hulse spent a large part of his career at leading semiconduc- On the telephone call Ramsdale was accompanied by An- tor companies as a design manager for analogue, digital and drew Farrugia, vice president of marketing, and they revealed a mixed-signal IC engineering. In these roles he introduced tech- Cambridge startup that has many software engineers seeking to nical innovations and commercial strategies to improve design model and accelerate the analog design process. productivity, design quality and schedule predictability. The company also has numerous senior executives with Ramsdale points out that Agile Analog is looking to fit in with experience gained at the such employers as ARM, Broadcoam, the engineering world as it already is rather than asking engi- Alphamosaic, CSR, STMicroelectronics, TSMC and Cadence. neers to overturn what they do. “We sit on standard EDA tools The company includes Sir Hos- and capture the recipe design sein Yassaie, former CEO of intent and make use of the PDK Imagination Technologies, as a [physical design kit] from the non-executive director. Ramsdale foundry.” The fact that Agile is himself a former vice president Analog works with a PDK as one of engineering and general man- of its inputs helps with its ability ager of the imaging and vision to target different manufacturing business at ARM. process technologies, a point Agile Analog has not given out the company makes in its sales much information about its in- pitch. ternal processes preferring to let So what are the limitations? its analog IP available for license “The classes of circuits we can do the talking. The company has address and frequencies attain- now started to bring out analog able do depend on the process- IP at a rapid rate and is also makes,” said Farrugia. “The tool is ing it configurable. still driven by an analog designer. One thing that Agile Analog It is not pushbutton – otherwise is not doing is analog synthe- mistakes happen – but what it sis, said Ramsdale. There have does do, is it addresses produc- been previous attempts to try and recreate the original success tivity.” of Synopsys with design compilation, but in the analog rather Ramsdale added: “Analog design is 10 percent inspiration than the digital world. However, the digital domain has a circuit and 90 percent perspiration. We value the analog engineer and isolation that allows blocks to be joined together easily and a look to speed up the 90 percent so we can get more done. We generality in terms of circuit equivalence that is just not present look at it from the point of view of an engineer starting a new in the analog world. project and look to optimize all the way through the process.” And even within classes of similar analog circuit, such as Pushing again on the limitations of the technology we asked families of amplifiers, there can problems. “Previous attempts how generally applicable it is in circuit design. Ramsdale said: to automate analog design have used a brute force explora- “There is a high general capability. Can it do RF? Yes. MEMS tion of the design space using genetic algorithms and similar is probably a bit of a stretch. But, in any case we are focused approaches. But the space is complex and this approach can on the analog circuit classes and foundry processes for which produce local minimum solutions that are not optimal,” ob- there is high demand.” Ramsdale added that its IP is also in served Farrugia. demand at a number of smaller foundries with unique process variants who otherwise find it hard to get IP core support. “Agile Analog looks at the whole way the analog designer So how far and widely has Agile Analog’s IP been licensed so works; a proper top down approach in analog; a good hierarchi- far? cal systematic approach. We have an in-house tool that can “We’ve engaged with the big-four foundries and a few take a recipe for a class of circuits from a designer, not just the smaller ones,” said Farrugia. The big four are TSMC, Samsung, essential physics but also the rules of thumb they have devel- Globalfoundries and SMIC. All of the circuits to date have been oped with experience. We have a very big software team work- developed for planar processes, and Furrugia includes fully- ing on the tool.” Farrugia noted. depleted silicon-on-insulator (FDSOI) in that category. “In 2019 The fact that analog circuits cannot be treated in isolation we started working with FinFET PDKs. Our design approach is the way digital circuits can, has always made their integration FinFET ready,” complex and the labour-intensive domain of skilled engineers. If When asked how the Covid-19 pandemic would affect the Agile Analog has an advantage here it is still something that the semiconductor industry and startups, Ramsdale answered company is not revealing. that Agile Analog is fortunate to be at a stage in its develop- The furthest that Ramsdale would go was: “Let’s say, we ment where it still has venture capital runway. This means that

10 News June 2020 @eeNewsEurope www.eenewseurope.com EXECUTIVE INTERVIEW

the company is not yet compelled to turn a profit – although it major global event such as the Covid-19 pandemic will change already has an income stream. It also means that development the world’s business landscape radically. Some companies that activity can continue and allows Agile Analog to prepare itself to were previously technology and market leaders will not adapt rise with the tide when business activity picks up. well, while others will – and some companies will spring up to Ramsdale added: “While Covid-19 is a personal tragedy for exploit the changed environment. the millions who have contracted it and those that have died “You have to look for the opportunity. You have to look from it, we have to look for the good that can come out of it.” outward and make some prediction about what the world will He continued: “Our industry is used to boom and bust and a look like in 12 or 24 months’ time and then make sure you are lot of chip companies can plan their way through recessions like serving the market.” this. The companies that will be most affected will be those with It is what a famous ice hockey player once described as a lot of human activity in the supply chains.” “skating to where the puck is going be.” In terms of market changes Ramsdale expects a new emphasis on industrial and health applications. “I think there will be a large increase in industrial automation and industrial IoT, which will be fantastic for robotics and chip companies. Also healthcare; I expect every wrist watch and smartphone company is wishing they could do SpO2 (saturated periph- eral capillary oxygen) measurements – an estimate of the amount of oxygen in the blood.” Ramsdale acknowledged that a

No Boundaries! When engineers need resistors for critical missions in a no-replace environment like Mars, they choose State of the Art. We are aboard three Mars orbiters: Odyssey, MRO, and Maven. We are aboard four rovers: Pathfinder, Spirit, Opportunity, and Curiosity, with another Tim Ramsdale, CEO of Agile Analog rover to be launched in 2020. And we are aboard the InSight lander Ltd. that is studying the interior of the planet. Working toward a manned mission to Mars, NASA chose State of the Art resistors. Whose resistors will you choose for next mission?

Mission Critical? Choose State of the Art resistors.

State of the Art, Inc. RESISTIVE PRODUCTS Made in the USA.

Andrew Farrugia, vice president of marketing, Agile Analog Ltd. www.eenewseurope.com eeNews Europe June 2020 News 11 NEWS & TECHNOLOGY ARTIFICIAL INTELLIGENCE

Deep learning gives self-driving cars better perception By Christoph Hammerschmidt eople, bicycles, cars or street, sky, grass: Which pixels of has now reached first place in Cityscapes, the probably most an image belong to people or objects in the foreground influential public benchmark for methods for understanding of the environment of a self-driving car, and which pixels scenes in autonomous driving. EfficientPS is also listed in other representP the urban background scenery? This task, known as benchmark data sets such as KITTI, Mapillary Vistas and IDD. panoptical segmentation, is a fundamental problem in many On the project’s website Valada shows examples of how the fields such as self-driving cars, robotics, augmented reality team has trained different AI models on different datasets. The and even in biomedical image results are superimposed on the respec- analysis. tive image captured with the camera, with Abhinav Valada, holder of the colours showing which object class the the junior professorship for model assigns the respective pixel to. For Robot Learning at the Insti- example, cars are marked in blue, people in tute of Computer Science at red, trees in green and buildings in grey. In the University of Freiburg, is addition, the AI model also draws a frame investigating this research around each object recognized as a sepa- question. He and his team have rate entity. The Freiburg researchers have now developed the innova- succeeded in training the model to transfer tive “EfficientPS” model, which uses artificial intelligence (AI) to the information learned from urban scenes in one city to other recognise visual scenes more quickly and effectively. urban backgrounds - in this case from Stuttgart to New York. The task of understanding scenes is usually solved with Although the AI model did not know what a city in the USA Deep Learning (DL), a technique of machine learning. This might look like, it was able to accurately identify scenes from involves artificial neural networks inspired by the human brain New York City. Most of the previous methods that address this learning from large amounts of data. Public benchmarks play problem require large amounts of data and are too computa- an important role in measuring the progress of these tech- tionally intensive for use in real-world applications such as ro- niques. “For many years, research teams from companies like botics, because these runtime environments are highly resource Google and Uber have been competing for the top spot in these constrained. “Our EfficientPS not only achieves high output benchmarks,” says Rohit Mohan from Valada’s team. However, quality, it is also the most computationally efficient and fastest the new method developed by the Freiburg computer scientists method,” says Valada. IMEC’s spiking neural network aids drone navigation By Peter Clarke he IMEC research institute has produced a spiking tion is too burdensome. In addition the time taken to move data recurrent neural networking IC suitable for use with from sensor and through the AI inference algorithm is too long. radar signals. Spiking neural networks (SNNs) are closer “Today, we present the world’s first chip that processes analoguesT of the biological neural networks than common radar signals using a recurrent spiking neural network,” said Ilja weighted-synapse neural networks. In SNNs data is moved as Ocket, program manager of neuromorphic sensing at IMEC, in a a series of electrical pulses over time but only when the sensory statement. Ocket made the point that SNNs can be connected input changes. IMEC’s chip consumes 1 percent of the power recurrently – turning the SNN into a dynamic system that learns of traditional artificial NNs while providing a tenfold reduction in and remembers temporal patterns. That avoids keeping lengthy, latency. Radar for military and avionic applica- data intensive training separate from tions has also had a traditional requirement for inference. “The technology we are high-speed and high-performance digital signal introducing today is a major leap processing and is increasingly being consid- forward in the development of truly ered for road traffic and autonomous driving. self-learning systems,” said Ocket. IMEC’s chip has 336 neurons. Part of this She said that by doing process- neuron fabric is fully feed-forward connected, ing much closer to the sensor the part has reconfigurable feed-back and feed- radar should be able to distinguish forward recurrent connections. It operates at more quickly and more accurately a nominal 1.1V and was implemented in 40nm between approaching objects. Kath- CMOS by TSMC. It can classify micro-Doppler radar signatures leen Philips, program director of IoT cognitive sensing at IMEC, using only 30-microwatts of power although the architecture added: “For [the chip’s] creation, we rallied experts from various and algorithms can be applied to variety of one- and two- disciplines within IMEC – from the development of training algo- dimensional sensor data, including electrocardiogram, speech, rithms and spiking neural network architectures that take neu- sonar, radar and lidar streams. The first use case for IMEC’s roscience as a basis, to biomedical and radar signal processing spiking neural network is an anti-collision radar system for and ultra-low power digital chip design.” drones. Although artificial neural networks are used in automo- The implication is that IMEC has used near-threshold voltage tive industry, IMEC points out that in power-constrained envi- circuit design in relatively-aggressive CMOS manufacturing ronments such as battery-powered drones the power consump- process.

12 News June 2020 @eeNewsEurope www.eenewseurope.com ELECTROMOBILITY

Continental joins forces with Pioneer on infotainment By Christoph Hammerschmidt utomotive electronics supplier Continental and the Japa- tainment, which we have gained through more than forty years nese consumer electronics giant Pioneer Corporation of global experience in both the OEM and aftermarket sectors,” have signed a strategic cooperation agreement, aiming said Naoto Takashima, Head of Mobility Products Company at Aat joint developments in the infotainment sector. With a jointly Pioneer Corporation. “Both components are essential ingredi- developed integrated infotainment ents for a new level of human-machine solution, the partners plan to create a interface that enables safe and intuitive holistic user experience. operation for increasingly comprehen- As part of the collaboration, Conti- sive functions. Continental and Pioneer nental will integrate Pioneer’s compre- have already begun sharing expertise. hensive infotainment subdomain into its Until now, the vehicle’s HMI has high-performance computer platform for consisted essentially of two separate vehicle cockpits. The automotive sup- segments that were implemented on plier hopes that this will bring long-term two displays: the actual driver informa- benefits for its customers: The integration in the instrument cluster and the tion of such extensive software packag- center console display, which is aimed es into a complex overall solution such at both the driver and front passenger. as Continental’s Cockpit High Perfor- The infotainment part was concentrated mance Computer (HPC) is intended to offer vehicle manufactur- on the center console display. As information, services and ers greater flexibility in the development of cockpit systems. functions become increasingly interlinked, this architecture is Such flexible solutions are essential for future developments no longer ideally suited. Instead, with the cross-domain solution in order to meet the trend towards a centralized vehicle archi- of the Cockpit HPC, all displays and control elements in the tecture as well as the increasing software complexity and the vehicle merge into a holistic user interface. rapidly growing range of cockpit functions, Content can thus be displayed inde- Continental said in a release. pendently of display limits and arranged In the future, this will increasingly require individually by the user. In potentially critical the strategic cooperation of multiple spe- situations, this ensures that only the informa- cialized partners. By integrating the entire tion that is really needed at that particular infotainment domain into a holistic system, moment is displayed with priority. This coun- Continental and Pioneer are demonstrat- teracts excessive demands or distraction of ing what such collaboration models can the driver. In the automated driving mode, on look like in the area of centralized cockpit the other hand, the driver also has access systems. Pioneer is one of the specialists in to all services and apps that are otherwise the field of car entertainment and makes this expertise available only available on the passenger side. This comfort and safety for the collaboration. In addition to Cockpit HPC, Continental is gain are made possible by an electronic architecture that allows contributing its significant integration know-how to the part- completely different systems and operating systems to be sepa- nership. The cooperation also reflects the growing importance rated on one hardware and continuously updated or supple- of an individual and customizable digital user experience for mented. To achieve this, it must be possible to host software vehicle users. from many sources on a single high-performance computer, “Continental’s broad knowledge of instrument clusters and independent of hardware. The two partners are using this flex- vehicle safety systems complements our expertise in car enter- ibility of the Continental cockpit platform for their cooperation. VisIC, ZF jointly develop GaN inverters for electric cars By Christoph Hammerschmidt he automotive supplier ZF Friedrichshafen AG and the marks nothing less than the breakthrough of GaN technology in Israeli power electronics company VisIC Technologies Ltd the automotive industry. “VisIC’s D3GaN technology is designed have signed an agreement to jointly to meet the high reliability standards of developT power electronics components the automotive industry and offers the for electric vehicles. For ZF, the focus is lowest losses per R ds(on). It also simpli- on a highly efficient driveshaft; VisIC will fies system solution development and contribute its GaN expertise. enables highly efficient and cost-effective The partnership is expected to powertrain solutions,” said Baksht. strengthen the partners’ development The adoption of wide band gap semi- efforts based on VisIC’s D3GaN semicon- conductor technology, which includes ductor technology. The activities will focus silicon carbide and gallium nitride materi- on 400-volt driveshaft applications cover- als, would significantly strengthen ZF’s ing the largest segment of the electric vehicle market. competitive position in the development of the most cost-effec- According to VisIC CEO Tamara Baksht, the development tive and efficient electric drive axles. www.eenewseurope.com eeNews Europe June 2020 News 13 NEWS & TECHNOLOGY ELECTROMOBILITY

Autonomous driving, electrification boosts market for test & measurement By Christoph Hammerschmidt s the global automotive industry is focusing on the devel- require testing methodologies such as hardware-in-the-loop opment of autonomous cars and electric vehicles (EV), (HIL),” said Rohan Joy Thomas, Industrial Research Analyst at which present immense opportunities for the test and Frost & Sullivan. “HIL simulates real-world traffic in the lab en- Ameasurement industry. Market research and consulting com- vironment and validates the performance of software embedded pany Frost & Sullivan therefore has dedicated a study to this inside the ECUs and DCUs against the stimuli.” industry segment. The analysis comes to the conclusion that if Thomas also noted the start-ups that continue to prolifer- solutions offered by pure-play test & measurement OEMs and ate in the automotive industry pose a challenge to traditional third party testing and certification service providers are com- automotive original equipment manufacturers (OEMs) in terms bined, the global autonomous car and EV test and measure- of innovation. ment market is estimated to generate revenues of $1.6 billion by As autonomous vehicles gain traction, testing solutions that 2025, up from $1.2 billion in 2019. can assure the security of the fully autonomous car will also The development of autonomous and electric cars does grow in popularity, presenting business prospects in the follow- represent huge op- ing areas: portunities not only Upcoming ADAS for automotive OEMs and infotainment use- and electronics provid- cases will further drive ers, but likewise to the the use of Automotive test and measurement Ethernet and the de- industry, the authors of mand for solutions that the study “Autonomous can test the technology Car and Electric Vehicle Testing vendors Growth Opportunities in should implement the Test and Measure- solutions that can test ment Market, Forecast levels of automotive to 2025” find. applications and pro- Among four types tect the autonomous of testing—advanced vehicle with adequate driver-assistance cybersecurity. systems (ADAS), EV, Testing vendors infotainment, and con- must provide testing nected car—ADAS is expertise such as mil- estimated to have the limeter wave testing, highest growth, reach- to validate 5G network ing $659.9 million by 2025. This expansion can be attributed for autonomous car use-cases, as well as for testing dedicated to the number of electronic control units (ECUs) and domain short range communications (DSRC) control units (DCUs) that are incorporated into a vehicle to make With the rise of Feature-on-Demand, Automotive OEMs it fully autonomous and reduce its production time. would want to monetize on features that their customers “There is a significant degree of overlap between infotain- subscribe to. Testing vendors should partner with providers ment, ADAS, and vehicle-to-everything (V2X); this overlap will of a wide range of in-car services that passengers or drivers increase as vehicle autonomy continues to grow, which will could subscribe to. 2020’s perfect storm: Wi-Fi 6, BLE, and AI? By Richard Edgar

he debut of Wi-Fi 6 and the explosion of AI hardware, net- significant boost to the Artificial Intelligence of Things (AioT). works, and tools will open new markets and spur future This is a win-win for those wishing to utilize these technolo- technology trends. gies, but not everyone will feel like a winner in 2020. The year TThe innovations of connectivity and AI are about to shift into could bring hardship to AI hardware startups that have risen full gear as new advances—edge computing, Wi-Fi 6, and Blue- up after years of long-term and highly intensive R&D. In many tooth Low Energy (BLE) v5.2, to name a few—arrive in full force. ways, this process has led to incredible results, including com- These developments are quite significant. Wi-Fi 6 will improve plex, high-value products. But those products also bring forth a robustness and performance, while Bluetooth audio sharing will strong patent portfolio, which can act as landmines to competi- make it possible for multiple consumers to personally enjoy the tors. Many firms have failed to keep up, inevitably leading to a audio of a single device. In addition, edge computing will give a decline—and soon, a contraction—within the space.

14 News June 2020 @eeNewsEurope www.eenewseurope.com ARTIFICIAL INTELLIGENCE

Let’s take a closer look at these and other notable innova- consumption. That means smaller devices with smaller batteries tions to watch for as 2020 unfolds. (such as hearing aids) now can take advantage of this impressive feature. The power of Wi-Fi Wi-Fi has changed the world as we know it, but the biggest crit- Computing at the edge of innovation icisms are that it often doesn’t work or it’s too slow. Networks At the same time, artificial intelligence will get a boost as edge are often strained by the number of users on board, a pain point computing becomes a major factor and a prominent focus that’s particularly prevalent at airports and other public ven- within the market. Powerful accelerators will enable devices to ues or events. Even the 2012 Olympic Games in London was run neural networks at the edge of the network. This could pres- bogged down by internet access of just 100 kB/s at the opening ent a host of interesting opportunities, particularly for AIoT. As ceremony. This was on a network that was supposed to offer low-cost edge compute nodes make it possible to do more on a several hundred megabits per second! But when overloaded, tighter budget, the potential will be limitless. the network failed to deliver a quality experience. One of the key challenges is simply the way Wi-Fi works. AI hardware startups face the music, When a mobile device connects to your router at home, for but AI apps will come to the rescue example, it does something called association. The access AI hardware startups have relied on a multi-year and highly point sends out messages every few seconds. In the case of intensive process involving near-endless research and develop- the Olympic Games opening ceremony, the devices were in fact ment. This has been effective in getting them to where they are “talking” to the access points, saying, “I’m here and I want to today, but it could come at a price. As every company under the send.” However, nothing could actually send, since so many sun rushed to take every dollar available from venture capitalists people were on the network simultaneously. Thus, the most exciting part of Wi-Fi 6 is that it will eliminate this pain point. It will effectively improve robustness and performance with two techniques: colors and orthogo- nal frequency-division multiple access (OFDMA). The former involves the use of different access points, which are a problem with current Wi-Fi. Now it’s possible to have different Basic Service Set (BSS) “colors,” or numbers between 0 and 7 that will allow devices to ignore signals from the AP it’s not associated with. In other words, if you’re in an apartment in New York City, your devices will be able to ignore what your neighbors are doing next door and provide a stronger signal. OFDMA helps speed up our con- (VCs), they may not have considered how difficult it would be nectivity by breaking the spectrum into smaller chunks, enabling to demonstrate the value of that investment. If they can’t show more devices to communicate. the fruits of their labor in 2020, VCs might not be interested in Wi-Fi 6 can also help reduce battery drain for small devices. returning for another investment round. This may not necessarily apply to mobile phones, but for sen- While AI hardware could be challenged this year, AI apps will sors in the home (such as a thermostat), it will be quite signifi- have their moment to shine. Two primary kits—Apple’s Core cant. Instead of being required to charge the device or change ML and Google’s ML—have opened the door to thousands its batteries on a regular basis, Wi-Fi 6 will make it possible to of developers, allowing them to incorporate machine-learning leave them be for a year or longer. models into their software. This will enable AI-centric apps to Finally, Wi-Fi 6 will get an additional boost now that the reach the mainstream, propelling the technology as users come 6-GHz band (1,200 MHz of spectrum) has been opened to to recognize its incredible value. This will surely get the interest unlicensed uses. The latest version of Wi-Fi, known as Wi-Fi 6E, of VCs, who might be intrigued by apps that can offer features uses the newly available spectrum to enable devices to connect that were once limited to very specific hardware. quicker and for data to transfer faster. It hasn’t received ap- proval just yet, but it looks like Wi-Fi 6E could become a reality The best is yet to come in 2021, if not sooner. This year will prove to be an outstanding opportunity for some of our most important technologies, such as Bluetooth and Sharing is caring (and aural bliss) Wi-Fi, to shine while giving future innovations like AI software The hidden brilliance of Bluetooth Low Energy (BLE) v5.2 can’t an opportunity to flourish. AI hardware companies could face be denied—it offers one of the most impressive changes since increasing pressure if they aren’t able to show that their VC the technology was introduced: audio sharing. While this has money was worth the investment, but it’s clear others will be been possible with Bluetooth Classic, which is quite power- waiting to take their place. hungry, this is the first time it can be done with BLE. As a result, This article first appeared on Electronic Design - you’ll receive the same benefits but with much lower power www.electronicdesign.com www.eenewseurope.com eeNews Europe June 2020 News 15 NEWS & TECHNOLOGY PROCESS

TSMC dragged to the altar of US manufacturing By Peter Clarke he double-whammy of enhanced restrictions on Huawei would greatly increase costs for TSMC compared with other and HiSilicon from the US, plus the announcement of locations. TSMC’s plan to build a 5nm wafer fab in Arizona has the It certainly makes sense that while Intel may be petitioning feelT of stage management about it. the US Department of Defense to help it create a wafer fab, There are also one or two indicators that TSMC’s plan is TSMC is the one being petitioned by the Department of Com- potentially provisional; at least to the extent that all such an- merce. The reason it makes sense if that Intel manufacturing is nouncements can be subject to change, or delay – or even underperforming while TSMC has established itself as the clear cancellation. technology leader. On Thursday, the Department of Commerce praised TSMC’s It is probably by a mixture of carrot and stick that TSMC has decision to build a wafer fab in Arizona, although time differ- been persuaded to say “yes” to the United States’ proposal. ences could have made that contemporaneous with TSMC’s The carrot will be the subsidies that are de rigueur for the set- press release dated Friday, May 15. TSMC stated that it plans ting down of any wafer fab. The stick will be the threat that if to spend $12 billion on a 5nm wafer fab in Arizona. And then on TSMC does not come to the US, Apple might start buying a Friday the DoC specifically called out Huawei as being the rea- larger proportion of its silicon from another source. About 23 son for enhanced restrictions on supplying chips that are made percent of TSMC’s revenue came from Apple in 2019, according with US software or manufacturing equipment – which applies to reports. The only company close to competing with TSMC in to almost all digital chips and anything made in advanced pro- leading-edge digital silicon is South Korea’s Samsung. cess technologies. Obviously, the US could easily print up In the case of Huawei, the argument of- an additional $4 billion or $5 billion if that’s fered up by the United States’ Department what it takes to persuade a blushing TSMC of Commerce seems contrived. The Depart- to come and stand alongside it. But it is also ment of Commerce argues that it has had to notable that TSMC has opted to set down enhance restrictions because, since being what is now a relatively small wafer fab that put on the “entity list” in May 2019, Huawei is not quite at the leading edge. The Ari- had been trying to undermine US export zona fab is denoted as a 5nm unit that will controls by using foundries to make its ultimately have a manufacturing capacity of chips. Huawei and its semiconductor sub- 20,000 wafer starts per month. That feels like sidiary HiSilicon have always been fabless. almost a minimal viable size for a production So, the use of foundries has always been fab at 5nm. a fundamental part of its business model. By the time the Arizona fab begins pro- TSMC which supplied a great deal of silicon duction in 2024, 5nm will seem a somewhat to HiSilicon in 2019 – reportedly worth 14 mature process compared with 3nm and percent of TSMC’s revenue – has always said it follows legisla- potentially some form of Forksheet transistor silicon at 2nm. Of tion and rules and clearly has felt it has been doing so up until course, fabs get upgraded and migrated during the course of now. their lifetime but the 5nm label sends a signal. It is also notable Be that as it may, there has been speculation for months that that the cost is in the semiconductor manufacturing equipment the US administration wants to thwart the progress of Huawei so TSMC’s commitment is not a major one until some time after while it still can. And this has been against the protestations of it has built the fab shell. Fab shells lying idle for years before national and international industry bodies. The Semiconductor being equipped or repurposed are a fact of life in the semicon- Industry Association’s CEO John Neuffer commented on the ductor industry that has always been a commercial rollercoaster development saying: “We are concerned this rule may create ride of boom and bust. uncertainty and disruption for the global semiconductor supply To be fair TSMC has been one of the most sure-footed of chain, but it seems to be less damaging to the US semiconduc- semiconductor manufacturers and rarely has to pull back from a tor industry than the very broad approaches previously consid- commitment. Again, it makes sense that the really big spend- ered.” ing on Gigafab production fabs remains in Taiwan but it feels Neuffer seems to be comparing some restriction that specifi- like TSMC is doing the minimum required to remain in US and cally mentions Huawei versus something more general but Apple’s good graces. it feels like he is whistling in the dark. This will make for big The final indicator is the use of a conditional verb in the last changes in the whole of the semiconductor supply chain. Indus- sentence of TSMC’s press release. “The Arizona facility would try body SEMI, which also protested against enhanced restric- be TSMC’s second manufacturing site in the United States.” It tions, has yet to comment. suggests the condition, “if it goes ahead.” The verb will is used But what of TSMC’s willingness to drop $12 billion on a US elsewhere in the press release. wafer fab, and those contra-indicators? First, there is the short The bottom line is that successful semiconductor policy is engagement period and the apparent haste of the wedding an- mapped out over several years while US administrations are nouncement. Only a few weeks ago – April 20 – TSMC’s chair- around for just four years at a time. So, there is no harm in man Mark Liu stated: “We are now actively evaluating the US TSMC planning for a $12 billion spend in the US for now, secure fab plan. But as I told the investors before, there is a cost gap, in the knowledge that such plans can be changed, delayed or which is hard to accept at this point.” At the time Liu seemed cancelled if they no longer make sense under different circum- to indicate that the lack of leading-edge ecosystem in the US stances.

16 News June 2020 @eeNewsEurope www.eenewseurope.com FLEXIBLE ELECTRONICS

Stretchable fabric is self-charging to power smart garments By Julien Happich

esearchers from the Chinese Academy of Sciences and fabric (90% polyester, 10% Spandex) onto which conductive from the Georgia Institute of Technology have designed nickel electrodes are selectively patterned through electroless a stretchable fabric that incorporates not only micro- Ni deposition. supercapacitorsR (MSCs) but also triboelectric nanogenerators Subsequently, to create the MSCs, reduced graphene oxide (TENGs) so it can charge itself before powering smart garments. (rGO) films are deposited onto the conductive textiles by a hy- Both the TENGs and MSCs were fabricated through a resist- drothermal reduction of graphene oxide with Ni. A gel-type elec- dyeing method in a planar configuration, with their Ni-coated trolyte (PVALiCl) was applied to achieve the solid-state textile electrodes shown to maintain excellent conductivity at 600% MSCs which the researchers demonstrated, could be designed and 200% tensile strain along course and wale directions, into arbitrary-shaped logos or patterns for good aesthetics. respectively. Such microsupercapacitors were reported to reach a voltage up to 3.2V and discharge capacitances of 5.0, 4.9, 4.2 mF at a galvanostatic discharging current of 0.5, 1.0, 2.0mA, respectively. These textile-based planar MSCs were able to power a watch at a strain of 50% after being charged. As for the TENGs, they were formed as dual in-plane Ni- coated electrodes, with an elastomeric PDMS thin layer coated on top of only one of the electrodes. The stretchable fabric TENG then operates when a polyester textile comes in contact with the TENG textile. Contact electrification occurs at the PDMS−polyester interfaces, generating net negative charges in the PDMS and positive charges in the polyester; similarly, the polyester fabrics will be negatively charged and the Ni-coated fabrics will be positively charged at the polyester−Ni interfaces. When the polyester fabric is gradually separating away from the TENG textile, under movement or stretching, the unbalanced positive charges in the Ni fabric flow through the external circuit to reach the other Ni-textile electrodes, so as to screen the static charges in the PDMS (see fig. 2 (ii)).

The current flow stops when all of the static charges are screened and equilibrium is achieved (fig. 2 (iii)). If the counter Fig. 1: Resist-dyeing fabrication of stretchable textile polyester textile is approaching back to contact the TENG tex- conductors and MSCs tile, a reversed current flow in the external circuit is generated until local charge equilibrium is achieved again (fig. 2 (iv)). Described in more detail in a paper titled “Stretchable Copla- The repeated touching separating motions are then convert- nar Self-Charging Power Textile with Resist-Dyeing Triboelectric ed into pulsed alternative current (AC). Nanogenerators and Microsupercapacitors” published in the The fabric in-plane MSC with reduced graphene oxides as AC Nano journal, the self-charging fabric consists of a knitted active materials reached a maximum areal capacitance of 50.6 mF cm−2 at 0.01 V s−1 and showed no significant degradation at 50% of tensile strain. The stretchable fabric-based TENG was able to output a 49V open- circuit voltage and 94.5 mW m−2 peak power density. Because both the MSCs and the TENGs in the stretchable self-charging power textile can be designed coplanar with a one-batch resist-dyeing fabrication process, this approach is compatible with conventional textile processing. The authors anticipate that such self- charging power textiles could be used to power small electronics intermittently, Fig. 2: The working mechanism of the coplanar TENG. without extra recharging steps. www.eenewseurope.com eeNews Europe June 2020 News 17 NEWS & TECHNOLOGY FLEXIBLE ELECTRONICS Sweat-powered metabolic sensors wrap around the skin By Julien Happich esearchers from CalTech have designed a thin flexible set of metabolic sensors that are poweredR by sweat when wrapped around the wearer’s skin. The flexible and fully perspiration- powered integrated electronic skin Schematic illustrations of the flexible (PPES), as the authors describe it in BFC-biosensor patch. a paper titled “Biofuel-powered soft electronic skin with multiplexed and achieve efficient fresh sweat sampling wireless sensing for human-machine for stable BFC operation and accurate interfaces”, harvests energy from Schematic of a battery-free, biofuel-powered sweat analysis. human sweat through lactate biofuel e-skin that efficiently harvests energy from the The electronic components and the cells (BFCs). The battery-free e-skin interconnects of the PPES are then human body, performs multiplexed biosensing, and performs continuous monitoring of encapsulated with polydimethyl silox- wirelessly transmits data to a mobile user interface key metabolic biomarkers including ane (PDMS) to avoid sweat/electronic glucose, urea, NH4+, and pH, before through Bluetooth. contact. transmitting them wirelessly to a user interface via Bluetooth As for the actual biofuel cell, it consists of lactate oxidase low energy. (LOx) immobilized bioanodes that catalyze the lactic acid to The device is built on an ultrasoft polymeric substrate which pyruvate and Pt alloy nanoparticle-decorated cathodes that complies with the skin’s modulus of elasticity. Hence it can be reduce oxygen to water. These redox reactions on the BFC laminated conformally on different body parts for accurate bio- electrodes were proven to yield a stable current, up to 3.5 mW/ sensing. The PPES was designed in two main parts. First, the cm2, to power the metabolite sensors for up to 60 hours of nanoengineered flexible electrochemical patch contains a bio- continuous operation. The researchers demonstrated the PPES fuel cell array and a biosensor array for energy harvesting and could selectively monitor key metabolic analytes as well as the molecular analysis in human sweat, all on serpentine-connected skin temperature during prolonged physical activities. electrode arrays. The second piece is a flexible electronic patch Pushing their experiments further, the authors integrated soft that consolidates the rigid electronics on an ultrathin polyimide strain sensors to their perspiration-powered e-skin to moni- substrate through flexible interconnects for power management, tor muscle contraction. The strain signals were then sent to signal processing, and wireless transmission. a human-prosthesis, in effect, turning the PPES as a human- The researchers also integrated a skin-interfaced microfluidic machine interface. Even under mechanical deformation, with module with independent inlet-outlet design into the PPES, to a bending curvature of 1.5cm in radius, the PPES maintained consistent sensor readings, the researchers reported. Applying CNTs/PDMS elastomer-based strain sensors on the hand and the elbow, connected to the PPES, the e-skin was able to accurately monitor the bending of the finger and elbow (from resistance changes of the strain sensors). A robotic arm wirelessly fed with these signals could mimick the gestures of the wearer’s System-level packaging and encapsulation arm, approaching and grabbing a target of the PPES for efficient on-body biofluid object. Another practical use case envis- sampling. M-tape, medical tape. aged by the authors is robotic assistance in the rehabilitation settings. By extrapolation, they anticipate that the incorporation of more physical sensors for electroencephalogram and electromy- ography recording along with the continuous metabolic monitoring could make multimodal PPES useful for the design and optimization of novel prostheses that Schematic of a soft BFC array consisting Photographs of a PPES on a healthy bring the human into the loop of pros- of LOx-modified bioanodes and Pt alloy individual’s arm. Scale bars, 1 cm. thesis control, for real-time user-specific nanoparticle–modified cathodes. responses to human intent and behavior.

18 News June 2020 @eeNewsEurope www.eenewseurope.com FLEXIBLE ELECTRONICS

Stretchable and transparent electrodes scale up to large areas By Julien Happich esearchers from the Korea Insti- ire Networks for a Large‐Area Stretchable tute of Science and Technology and Transparent Electrode» published in the (KIST) have developed a process to journal of Advanced Functional Materials fabricateR large-area highly transparent and reduces the frictional resistance between stretchable electrodes out of wavy silver the individual nanowires in the nanowire Electroluminescent image of stretchable nanowires. networks as they become wet. Designed for use in stretchable displays, and transparent ZnS:Cu ACEL devices In particular, each silver nanowire can be the new electrodes were demonstrated to under diverse mechanical deformations, worked with in water and rearranged into a operate reliably even under 50% of stretch- including 40% stretching. curved nanowire structure with a large radi- ing or when twisted and rolled. The proof of us of curvature, so that a structure capable concept consisted of the KIST logo patterned as a transparent of stably stretching can be realized. Since the nanowires do ZnS:Cu alternating-current electroluminescent (ACEL) device not experience any unstable conditions, there are no nanowire formed with wavy Ag nanowire-based stretchable electrodes network fractures or nanowire layer peeling. connecting the two sides of the electroluminescent layer, on an By fabricating a silver nanowire network in this way, the elastomeric substrate the size of an A4 research team was able to stretch paper. the substrate and its nanowires by Here the silver nanowires, only tens of at least 50% of the initial length, nanometers in diameter, are first spread in stably maintaining transparency a random network of straight nanowires and conductivity for approximately on a pre-stretched sheet of polydimeth- 5,000 stretching-relaxing cycles. ylsiloxane (PDMS). But instead of releas- The team also found that this ing the strain and buckling or breaking type of material could be pro- some of the nanowires in the process, the Schematic image of a logo-patterned stretchable duced using an inexpensive and researchers wetted them with solvents and transparent ZnS:Cu alternating-current environmentally-friendly process (water). This solvent annealing step, as electroluminescent (ACEL) device with wavy Ag that uses ethanol and water as they describe the process in a paper titled nanowire-based stretchable and transparent solvents. «Buckling Instability Control of 1D Nanow- electrodes. Shapeable materials drive changing power designs By Nick Flaherty n important new trend is emerging where shapeable do more than just cut this sheet into fun or interesting shapes. materials are customised to function not just shape for It is thin and flexible. You can mold it around curved surfaces power systems, says Raghu Das at IDTechEx. Builders, such as bags and clothes. Our idea is anyone could transform Atextile manufacturers, those doing home improvement and oth- various surfaces into wireless charging areas,” he added. ers are starting to buy electrically-smart, shape- Johns Hopkins Applied Physics Labora- able materials direct from materials companies, tory has polymer-sheet batteries that can be bypassing the electronics industry, notes the cut to shape for chosen energy storage and market researchers at IDTechEx in a new report. locations and work even after being shot and For example, InfinityPV self-adhesive so- soaked. The University of Buffalo fabricates kiri- lar tape cuts to any length and that decides gami-inspired stretchable electronics that voltage and power produced, not just shape. also alters electric and mechanical properties, Layering gives other options. Increasingly, while Nanyang Technological University, Singa- other reconfigurable electrical and electronic material can pore shows a fabric-like wearable supercapacitor that can be be stretched, pressed, or cut to shape on arrival. Expect electri- cut, folded, or stretched and it even demonstrates customisable cally smart material fed into 3D printers to be another a huge stretchability of its wearable electronics. opportunity for the added-value materials industry, says Raghu Unrolling like a carpet, Renovagen photovoltaics will soon be Das, CEO of IDTechEx. available up to a huge 300kW output. Buyers choose power by “If you can get your battery, supercapacitor or solar power choosing length and width and the PV array will unroll across from area, you do not need the highest efficiency,” he said. a field as a temporary microgrid for outdoor events or charg- “This is often the logic behind the new plastic-film forms ing a farmer’s new robots in distant places. The basic copper of new thermoelectrics, piezoelectrics, triboelectrics, elec- indium gallium diselenide CIGS technology has been applied as trets and photovoltaics and of wide-area cuttable, printed or film to buses and building facades competing with the more co- painted sensors. Biodegradable papertronics with ink and pens lourful Heliatek organic photovoltaic film which both refreshes a using, resistive, conductive, light-emitting and semiconduc- building and makes it greener. tive inks are also part of this.” NYU Tandon, University of Buffalo, Peking University and He cites researchers at the University of Tokyo researchers others have also demonstrated painted photovoltaics and now who have demonstrated plastic film hybrid electronics. “You can seek efficiency high enough to commercialise them. www.eenewseurope.com eeNews Europe June 2020 News 19 NEWS & TECHNOLOGY BATTERIES Solid-state battery tech for electric cars: key to greater autonomy By Philippe Vereecken ith solid-state batteries, energy density as high as 1000 electric vehicles should Wh/l (or 500 Wh/kg). Since be able to achieve an au- today’s LiB cells can “only” de- tonomyW (driving range) matching— liver 700 Wh/l (or 230 Wh/kg), and eventually surpassing—that of a significant boost in energy cars with an internal combustion density is needed. engine. Battery-cell roadmaps The great thing about solid foresee that cells of 1000 Wh/l electrolytes is that they allow for should become available in the integration of different active 2030—in the form of solid- materials (such as a lithium metal state lithium-metal batteries. anode) and cell architectures (such We will come back to the as bipolar arrangements), mak- potential of those solid-state ing for a higher energy density at batteries, but let’s first discuss cell and/or battery-pack level. In an alternative approach to parallel with these all-solid-state increase the amount of energy batteries, developments such as that can be squeezed out of an “smart” battery cells with sensors Fig. 1: The energy density of the Li-ion battery (LiB) cell has electric car’s battery pack—an will increase the amount of usable more than tripled since its market introduction by Sony in approach leveraging so-called energy in the battery pack even 1991. Continuous improvements in LiB components with “smart” battery cells. further. LiCoO2-graphite chemistry resulted in an average increase of 25 Wh/l per year from 1995 to 2010. Introduction of new “Smart” battery cells: Today’s Li-Ion battery active cathode materials, such as the NiCoAl-based and a complementary cells are running out NiMnCo-based lithium-metal oxides (NCA and NMC), and approach of steam the gradual addition of silicon to the graphite anode, have Enlarging the energy density at For electric cars to really take off, cell level is one prerequisite to maintained the energy density increase ever since. However, they ideally should match—or increasing the driving range of it’s expected that with the materials known today, we will even exceed—the driving range of the electric car. But, next to the vehicles with an internal combus- reach a practical limit for wet LiBs around 800 Wh/l. Solid- many individual cells in large tion engine. The key to achieving state battery technology will be needed to break through this battery packs, a car battery this lies in the battery itself, more barrier and achieve an energy density of 1000 Wh/l—and module also contains electron- specifically in the use of solid-state more. ics and sensors to manage that battery-cell technology. battery’s usage. For example, At the heart of today’s EVs are lithium-ion battery (LiB) cells to ensure a long (enough) lifetime, the battery management that contain liquid electrolytes. The best-in-class of these “wet” system (or BMS) will typically only use part of the cells’ energy LiB cells have an energy density of around 700 watt-hours to avoid damage to the cell chemistry. per liter (Wh/l), accommodating a maximum driving range of In other words, the energy that can actually be used by an about 500 km. Yet, their energy density is expected to stagnate electric car for driving might only be 60-80% of the overall bat- at around 800 Wh/l due to the characteristics of their active tery’s capacity (depending on the type of electric car). materials (Fig. 1). A higher energy density can be expected from Improving the battery’s energy management can be another solid-state batteries, which contain a solid electrolyte instead of factor to enhancing the electric vehicles’ driving range. This a liquid one. In combination with new battery-pack and battery- could be realized by using “smart” battery cells, with micro-sen- module developments, the autonomy of electric cars could sors built into the cell to better monitor its state of charge and significantly be extended. state of health. At EnergyVille—an association of the Flemish research institutes KU Leuven, VITO, imec, and UHasselt in the 1000-Wh/l battery cells enable more field of sustainable energy and intelligent energy systems—re- autonomy search into smart battery cells with multi-array sensors and Today, the maximum driving range of an electric vehicle is integrated electronics for communication with the BMS is cur- determined by the amount of energy that’s contained in the rently under way. individual LiB cells of the car’s battery pack. Cells are connected in parallel and series arrangements to provide the high First-generation solid-state batteries: currents and high voltage needed to power the electric engine. cells with solid-state electrolyte To achieve a driving range of 700 km, cells are required with an The first generation(s) of solid-state LiB cells will have solid electrolytes instead of liquid electrolytes. But, at cell level, they Philippe Vereecken is Scientific Director, Electrochemical will not yet feature a higher energy density than LiBs contain- Storage, at imec - www.imec-int.com ing a liquid electrolyte, because they will initially contain similar

20 News June 2020 @eeNewsEurope www.eenewseurope.com BATTERIES

types of active electrode with graphite as a safe materials. solution, enabling the Why, then, develop commercialization of the solid-state batteries at all, current wet LiBs in the you may ask? Well, for one, early 1990s. solid-state batteries allow Inserting lithium-ions for different, more compact in between the layers arrangements in the battery of graphite was a safe pack. They can be built in compromise, but it bipolar arrangements pro- came at the expense of viding higher voltages at cell energy density and cell Fig. 2: Shown is a schematic of a solid-state lithium-ion battery with level. This, in turn, simpli- voltage. Several solid- fies connecting the cells graphite-silicon electrode and a solid-state lithium-metal battery with thin state electrolytes are and creates extra space in lithium metal as anode (Source: Xubin Chen, Philippe Vereecken, Fanny stable against metallic the battery pack to include Bardé). lithium; thus, the use of more cells. thin lithium as an anode becomes a possibility. In fact, lithium- Solid-state LiBs will also be more intrinsically safe. Thus, less metal batteries with solid polymer electrolytes are already com- safety monitoring electronics are needed in the periphery of the mercially available. battery module. Finally, solid-state batteries can have a larger However, these batteries only operate at temperatures of voltage window than wet LiBs, which means that the risk of 70°C and therefore aren’t suitable for an electric family car. The damaging the cell during charge and discharge is reduced and reason for this is lithium-ion conductivity is too low for these a larger part of the cell energy becomes available for use. solid polymer electrolytes. As a result, for many years, solid- Combining all of those factors, the amount of energy that’s state battery research has focused on finding solid electrolytes available in the battery pack will be higher for solid-state LiBs— with a high-enough ionic conductivity. even if the energy density of the first generation(s) of solid-state cells will be the same, or perhaps slightly lower, than the energy Exploring the options for a matching density of wet LiB cells. solid electrolyte The first electric cars equipped with these first-generation In the last few years, several good options have emerged with solid-state batteries are expected to launch somewhere in the lithium-ion conductivities matching, or even surpassing, the ion middle of this decade. conductivity of liquid electrolytes that are currently used for lithium-ion batteries. Toyota is the current champion with a sulfidic In pursuit of the holy grail: solid-state inorganic solid electrolyte with Li-ion conductivity about 3X that Lithium-metal batteries of liquid electrolytes. Another contender is LLZO, a garnet-type As described above, simply replacing a liquid electrolyte in oxidic inorganic electrolyte that, even though it has somewhat a LiB with a solid electrolyte doesn’t make for an increase in lower lithium-ion conductivity than liquid electrolytes, is of energy density. On the contrary, ceramic solid electrolytes in interest because of its large electrochemical window. On the powder form are likely to take up more space—and weigh downside, these inorganic electrolytes are extremely sensitive more—than their liquid equivalents in current LiBs. Hence, solid to moisture, which makes the assembly of the solid-state cells electrolytes would somewhat lower the energy density of a cell difficult (and potentially costly). In contrast, imec has developed with the same active electrode materials. a nanocomposite electrolyte that’s fabricated from liquid and The benefit lies in the fact that some solid electrolytes can turned solid once it’s inside the cell. This makes the material provide a larger electrochemical window. In other words, they compatible with current lithium-ion cell-fabrication processes, remain stable also at very high voltages (where current liquid and thus potentially low cost. electrolytes do not). This means that higher-voltage cathode materials can be used than those employed in 3.6- to 3.8-V Taking the next step: LiBs today, and a higher cell voltage leads to an increase in cell overcoming cell-assembly challenges energy, provided the lithium-ion capacity remains the same. Now that several options are available, solid-state battery

For example, LiMn1.5Ni0.5O2, or LMNO, has an electrode research has shifted its focus toward the assembly of the cells potential of 4.7 V and can’t be used with liquid electrolytes that and the integration of all components into a functional cell. One react at these high voltages. Solid electrolytes such as lithium issue is the reactivity of the materials forming resistive or block- lanthanum zirconium oxide (LLZO), however, are stable up to 5 ing layers between the functional components, severely inhibit- V and are thus compatible with these high-voltage cathodes. As ing the cell’s operation. Special thin-film coating (or so-called such, solid-state LiBs can break the 800-Wh/l barrier. “artificial interphase coatings”) are being developed to prevent The real game-changer is the use of metallic lithium as an these reactions from happening. anode. Lithium metal has the highest energy density as active At EnergyVille, upscale and cost-effective methods are devel- anode material and provides the highest cell voltage from the oped for the deposition of these sub-nanometer thin coatings anode contribution in the cell. To reach the much desired 1000- inside the thick battery electrodes. These nanoscale coatings Wh/l energy density—and to even go beyond—all-solid-state will be instrumental not only for the solid-state lithium-metal lithium-metal battery (LMB) cells are therefore targeted. batteries under development, but also for the extendibility of So far, the use of lithium metal didn’t work because charging LiBs with a liquid electrolyte. Indeed, much of the solid electro- of the battery leads to the formation of lithium-metal needles lyte know-how can be applied to wet LiBs as well. Who knows, (or dendrites). These short the battery internally, resulting in the solution may lie in a hybrid approach after all. thermal runaway and explosions. It was Dr. Akira Yoshino, one This article first appeared on Electronic Design - of last year’s Nobel prize winners in chemistry, who came up www.electronicdesign.com www.eenewseurope.com eeNews Europe June 2020 News 21 NEWS & TECHNOLOGY PHOTONICS

Solar water-splitter promises perpetual power By Peter Clarke esearchers from Rice University (Houston, Texas) have The authors state that the module, while suitable for further created a low-cost panel that uses solar power to efficiency optimization, is a self-sustaining producer of fuel that split water to produce hydrogen fuel. When there is no should be simple to mass produce. sunlight,R previously stored excess hydrogen, could be used The work was performed in the Lab of Rice materials chemist creating the prospect of a perpetual source of power that only Jun Lou at Brown School of Engineering. The lead author was requires water and intermittent sunlight. Rice postdoctoral fellow Jia Liang and the work is reported in One of the by-products of research into solar cells for elec- the American Chemical Society journal ACS Nano. tricity production has been the investigation of solar energy to The most efficient PSCs have efficiencies of more than 25 produce other sources of energy or useful products such as percent but the materials for those PSCs are expensive and hydrogen fuel. tend to be stressed by light, humid- In this case a panel made at Rice ity and heat, the authors observed. University integrates catalytic elec- “Jia [Liang] has replaced the trodes and perovskite solar cells more expensive components, like that, when triggered by sunlight, platinum, in perovskite solar cells produce electricity. The current with alternatives like carbon,” Lou flows to the catalysts that turn wa- said. “That lowers the entry barrier ter into hydrogen and oxygen, with for commercial adoption. Integrated a sunlight-to-hydrogen efficiency as devices like this are promising high as 6.7 percent. because they create a system that The device has two series- is sustainable. This does not require connected perovskite solar cells any external power to keep the (PSCs) and two cobalt-phosphide module running.” CoP catalyst electrodes. The use “With a clever system design, of PSCs and CoP-catalysed water splitting is not new but Rice you can potentially make a self-sustaining loop,” Lou said. has packaged the two together in a single module that can be “Even when there’s no sunlight, you can use stored energy in dropped into sunlit water and produce hydrogen without any the form of chemical fuel. You can put the hydrogen and oxygen other input. This is aided by the use of polymer insulating film products in separate tanks and incorporate another module like called Surlyn. a fuel cell to turn those fuels back into electricity.” Plasmonics on silicon promising for light-based computing By Julien Happich esearchers in Australia and Germany have developed in transforming computer processing to be based on light. It a modular method to design nanoscale devices that will also be very useful in the development of quantum-optical combine the best of traditional chip design with photonic information systems, a promising platform for future quantum architectureR in a hybrid structure. Their research published in computers,” said Associate Professor Stefano Palomba, a Nature Communications co-author from the University of Sydney and Nanophotonics under the title “Modular nonlinear Leader at Sydney Nano. hybrid plasmonic circuit”describes a “Eventually we expect photonic informa- device formed by two in-series plasmonic tion will migrate to the CPU, the heart of elements on a SOI waveguide: a mode any modern computer.” converter and a focuser. The focuser com- On-chip nanometre-scale devices that bines a taper with a sharp tip of nanometre use metals (known as “plasmonic” devices) scale, which functions as a nonlinear na- allow for functionality that no conventional noscale light source. The hybrid plasmonic photonic device allows. Most notably, they integrated circuit (HPIC) offers a bridge efficiently compress light down to a few bil- between industry-standard silicon photonic lionths of a metre and thus achieve hugely systems and the metal-based waveguides enhanced, interference-free, light-to-matter that can be made 100 times smaller while Schematic of our SOI-HPIC. It consists of interactions. retaining efficiency, explains lead author an industry-standard TE ridge waveguide “We have shown that two separate Dr Alessandro Tuniz from the University followed by two in-series plasmonic circuit designs can be joined together to enhance of Sydney Nano Institute and School of modules: (i) efficient TE-photonic to TM- a run-of-the-mill chip that previously did Physics. plasmonic rotator, and (ii), nano-focusing nothing special,” Dr Tuniz said. The researchers have shown that they tip. Below is the scanning electron This modular approach allows for rapid can achieve data manipulation at 100 times rotation of light polarisation in the chip micrograph of a fabricated device. Scale smaller than the wavelength of light carry- and, because of that rotation, quickly bar: 400nm. ing the information. permits nano-focusing down to about 100 “This sort of efficiency and miniaturisation will be essential times less than the wavelength.

22 News June 2020 @eeNewsEurope www.eenewseurope.com PROCESS

X-FAB expands foundry offering for silicon-based microfluidics By Julien Happich n order to address heightening demands, X-FAB Silicon The company is continuously expanding its portfolio of stan- Foundries has taken steps to simplify the integration of dard process blocks covering these capabilities. Rather than microfluidic elements with CMOS and SOI dies. Part of its beginning complex microsystem design projects from scratch, extensiveI MEMS-oriented technology offering, the company is it will be possible to re-use existing process IP, and thereby get now able to provide a large variety of process capabilities for a head-start. This will help to accelerate product development silicon- based microfluidic systems. Combined with a highly cycles, as well as facilitating the scale up to volume production. standardized foundry approach, this will Among the help to remove barriers to market entry and multitude of shorten development cycles. new opportuni- X-FAB has already established itself as the ties envisaged leading semiconductor foundry with regard to are develop- the integration of silicon-based microfluidics, ment of MEMS making $25 million worth of investment in this solutions for area over the course of the last 5 years. drug admin- In that time, the company has undertaken istering, flow numerous projects for medical and industrial metering and clients. These include applications such as pollution moni- lab-on-a-chip, DNA sequencing and synthe- toring. sis, cancer diagnostics, etc. Through these projects, its engineering team has gained un- “We are matched proficiency in areas like noble metal seeing a grow- processing, high aspect-ratio deep reactive ing number ion-etching (DRIE) and the deposition of both of requests organic and inorganic materials. The team to implement has also become highly adept at tailoring the interfaces needed silicon-based microfluidics. As a consequence, our engineering to connect CMOS dies and their accompanying microfluidics team is striving to bring two worlds together; so that medical elements, so the exact application requirements can be met. OEMs’ deep understanding of physiology and fluid analysis can be combined with X-FAB’s knowledge of high-volume semicon- Using all this expertise as a foundation, the company has ductor manufacturing technologies,” states Volker Herbig, VP of extended its process capabilities to now include: X-FAB’s MEMS business unit. • Noble metal electrode arrays • Inorganic layers for surface passivation and protection “We are excited to help our medical customers in developing • Deep silicon trenches and cavities their smart integrated microfluidic systems via the multitude of • Bonded silicon or glass lid wafers new process capabilities we can now offer.” • Polymeric fluidic structures Quantum entanglement aids radar detection By Peter Clarke hysicists in a team from across Europe and the US have the idler photons are measured in relative isolation, free from developed a form of microwave radar that makes use of interference and noise. When the signal photons are reflected quantum mechanically entangled photons. back, true entanglement between the signal and idler photons is PThey have called the detection lost, but a small amount of correla- technology ‘microwave quantum tion survives, creating a signature or illumination’ otherwise known as pattern that describes the existence quantum radar. The technique is or the absence of the target object able to detect objects in noisy – irrespective of the noise within thermal environments where the environment. The team is drawn classical radar systems often from the research group of Professor fail. The technology has potential Johannes Fink at Institute of Science applications for ultra-low power and Technology Austria (IST Austria) biomedical imaging and security together with Stefano Pirandola scanners, the researchers claim. from the Massachusetts Institute of Instead of using conventional Technology (MIT) and the University microwaves, the researchers of York and David Vitali from the entangle two groups of photons, University of Camerino, Italy. which are called the ‘signal’ and ‘idler’ photons. The signal “What we have demonstrated is a proof of concept for Mi- photons are sent out towards the object of interest, whilst crowave Quantum Radar,” said lead author Shabir Barzanjeh. www.eenewseurope.com eeNews Europe June 2020 News 23 DESIGN & PRODUCTS LEDS & DISPLAY TECHNOLOGIES Driving high brightness LEDs By Mark Patrick igh brightness light emitting diodes (HBLEDs) are bring- Remaining HBLEDs challenges ing about a revolution in lighting - with applications rang- HBLEDs are not necessarily the obvious solution for every lighting from the humble flashlight to car front lighting. There ing application. They have better efficiency than incandescents, areH still trade-offs, however, in relation to cost and benefit, with converting 6x as much electrical power into light and have as there being a need to have a long lifespan to fully justify their much as 25x longer lifespans, but are around 20x more expen- use. Apart from manufacturing quality, operating temperature sive. Fluorescents have about the same efficiency and longevity, is the main factor when it comes to longevity. This means that but are about a quarter of the price of their solid state equiva- careful heatsink design, along with a driver that keeps the cur- lents. That said, there are other factors that need to be consid- rent at an optimal level while contending with abnormal condi- ered - such as fitting/replacement costs, dimmability, mechani- tions (such as cooling loss - which could otherwise result in cal robustness and carbon footprint, as well as disposal costs. early failure and expensive replacement costs) will be mandat- Sometimes efficiency isn’t even an issue - if an incandescent is ed. To increase the benefit side of the equation, the driver can in a room that needs warming, the power dissipated as heat is also provide features such as dimming, fault protection and the utilised anyway. If the voltage is reduced a few percent, the life- ability to control multiple strings of LEDs. In this article, we look time extends dramatically as well. For example, a 10% reduc- at the background, the cost/benefit factors and how LED drivers tion will reduce luminous intensity by about 30% - the threshold are selected for different example applications. at which it just becomes noticeable, but lifetime increases by about a factor of 4.5 (colour temperature does change as well Dip don’t dazzle! though). Back in the 1970s, in the UK, there were regular public service announcements on the TV. They ranged from advice on shel- HBLEDs can win out in most applications, if their prospec- tering under kitchen tables in case of nuclear attack to how tive lifetime can be achieved - and this comes down to effective to cross roads safely (explained by David Prowse, who would thermal management. LEDs rarely fail abruptly under normal later be cast as Darth Vader in the Star Wars movies) and how conditions, but lose brightness due to propagation of lattice de- to drive safely. One of the campaigns was ‘Dip Don’t Dazzle!’ - fects or ‘dislocations’. A typical plot of lifetime against HBLED reminding that ‘full beam’ could blind the drivers of oncoming chip temperature is shown in Figure 1. An accepted value for traffic. It was hardly a big problem at the time - with the yellow- dimming with time is the ‘L70’ value, or when the lumen output ish light, cloudy lenses and tarnishing reflectors …but today, the has reduced to 70% of the initial value. The plot shows that this piercing HBLEDs of cars (and even pedal bikes) are a distinct can be from 20,000 to 70,000 hours with a junction temperature hazard. HBLED technology has made it possible to replace difference of just 20°C. incandescent bulbs, fluorescent tubes and sodium/mercury vapour lamps with far higher performance products. The market agrees, with a projected $22B global revenues by 2023 at 4.9% CAGR (according to Global Market Insights). HBLED basics LEDs exploit the characteristic of a semiconductor p-n diode junction - where photons are emitted by electroluminescence when electrons recombine with electron ‘holes’ crossing the semiconductor band gap as the junction is forward biased. The amount of doping of the semiconductor affects the band gap size which in turn affects the energy and hence frequency or perceived colour of the photons emitted. Modern HBLEDs are typically a high-power blue emitter behind a transparent melding that has been impregnated with a cerium-doped yttrium aluminium garnet phosphor (Ce3+:YAG), which emits yellow light. The combination of blue and yellow then gives white with good colour rendition. Semiconductor material and Fig. 1: Reduction in lumens over time at different HBLED chip packaging advances for better thermal performance have now temperatures [adapted from Philips Luxeon K2 data, when resulted in single HBLEDs that can produce over 100lumens/W. driven at maximum rated current]. Other LED phosphor combinations can give even better colour characteristics (but with lower efficiency), and white can also be derived from three separate red, green and blue LEDs (though HBLED driver performance is key with poorer colour stability in relation to temperature and time). Fundamental to maintaining chip temperature within design These RGB systems are useful, however, where dynamic colour limits is the driver IC specified, but the application also dictates change is required - in applications such as mood lighting or the type of driver needed. Many uncritical, lower power applica- stage illumination. tions might use a series resistor from a constant voltage source to set LED current. This is certainly simple, but also inefficient Mark Patrick is Technical Marketing Manager, EMEA at Mouser - with power lost in the resistor. If the source voltage is kept Electronics - www.mouser.com close to the LED forward voltage then dissipation in the resistor

24 News June 2020 @eeNewsEurope www.eenewseurope.com LEDS & DISPLAY TECHNOLOGIES

is lower, but LED current is less accurate - there can easily be Application example - architectural lighting an initial variation in forward voltage (VF) of 20%. A typical blue Architectural lighting is all about mixing colours to enhance HBLED VF can vary from 3.03V to 4.47V (with manufacturing appearance and create impact. The power source will be AC tolerances), so with a source voltage of say 5V current will vary mains, so an AC/DC converter can be used - providing a con- by a factor of about 3.7 to 1 - clearly not an optimal situation venient constant source voltage to the LED driver at some low, with large changes in LED dissipation and colour rendition. The safe value for distribution within the lighting fixtures. HBLEDs HBLED voltage itself also changes with junction temperature so will typically be in series strings with multiple channels for the actual current variation could be even greater. increased light intensity and different colours. An example driver A controlled current source is the best solution and again circuit is shown in Figure 3, based on the Micrel MIC3201 driver there are various options, the simplest being a linear constant IC. The input voltage is up to 20V, allowing 1A to be delivered current regulator. This type of driver can maintain LED current to a string of four LEDs. The circuit topology takes the form of a at elevated degrees of accuracy with no generated noise, but hysteretic buck converter, which is a variable-frequency topolo- dissipation is relatively high unless HBLEDs are selected (or gy maintaining better than 90% efficiency and ±5% LED current ‘binned’) to be within a small VF band and the source voltage with only a few external components required. The IC has a dim minimised. Analog dimming is possible but colour tempera- control, which can be driven with a low-frequency PWM signal ture varies with intensity. Figure 2 shows a circuit using the to vary the LED high-frequency PWM signal from 1% to 99%. PAM2808 from Diodes Inc. which is suitable for a single emitter. Selective dimming of drivers with red, green and blue LEDs can Pulse width modulated (PWM) dimming is possible if a signal then combine their projected light to give any colour. Current is is applied to the enable pin, but in simple applications a signal sensed with a convenient high-side resistor, which drops just may not be available. RS is a low value, typically 100mΩ for 0.2V at the maximum rated 1A current. a 1A LED. Quiescent current is also low, which is of greater importance in battery-powered applications.

Fig. 3: Typical driver for a colour channel of architectural Fig. 2: A linear LED driver for battery-powered applications. lighting.

Switched-mode regulators are the favoured solution for high Application example - horticultural lighting efficiency and when sophisticated control is needed. Source In horticulture, artificial lighting has long been utilised but cost voltage can be higher or lower than the total LED VF and a mul- has been a major concern in an industry that works within tight titude of features can be included (such as remote control and margins. Also, heat produced by incandescent lights has proven protection against over-temperature or other fault conditions). to be problematic meaning that light sources have to be kept

Toppan's “Blanview” TFT-LCD technology can TFT-LCDs upgrade your product capability  Sunlight Readable • High contrast image creation in any situations - Outdoor & Indoor  Ultra Low Power Consumption • Energy saving - Extend battery-life  Wide Product Lineup • From 2.2”, Up to 10.4”, Broad Size & Resolution choices  Wide Temperature - Op. -30 to +85°C / St. -40 to +95°C *Specific models only  Robust Mechanical Design - 6.8G Vibration *Specific models only  Small Batch Delivery - Small size business support  Long-term Continuous Availability

Ultra Low [email protected] | www.toppan.co.jp/electronics/ortustech/ Sunlight Readable Power Consumption The Toppan LCD Brand 2-8-7, Asahigaoka, Hino, Tokyo, Japan www.eenewseurope.com eeNews Europe June 2020 News 25 DESIGN & PRODUCTS LEDS & DISPLAY TECHNOLOGIES

a distance away from sensitive plants. LED lights, with their PWM dimming for constant colour rendition and analogue re- greater efficiency, have opened up the market, and growers duction of LED current in case of over-temperature conditions. have realised that the different colours of LEDs available can be For precision backlighting applications where flexibility is employed to promote different stages called for, a boost converter can of plant growth. For example, ‘deep be used - so that for a nominal blue’ and ‘hyper red’ are optimum for battery range, a high voltage will be photosynthesis and ‘far red’ controls generated (thereby allowing multiple germination, vegetative growth and series LEDs to be driven). The Sem- flowering. A mix of colours, along with tech SC5012 is an example which white, for human comfort is easily pro- operates from an input of 4.5V grammed to change a plant’s qualities. to 45V and can drive four strings LED manufacturers have also respond- simultaneously at up to 65V/150mA, ed by providing specific LED colours, each matched within ±1%. This optimised for horticulture. The driver device has an I2C interface - which arrangement shown in Figure 3 would can be used for fault monitoring, also be suitable in this application as well as detecting open/short with multiple channels for the selected LEDs and over-temperature, while colours. also providing external frequency synchronisation. Analog and PWM dimming is available with select- Application example - able 9/10-bit resolution. A particular automotive displays feature is ‘phase spreading’ where An automotive application for either in- each of the four strings is driven terior lighting or displays can use a sim- with pulses separated by 90° in the switching cycle. This reduces ilar circuit to that used in architectural Fig. 4: Typical application circuit for the Semtech lighting, except it is useful to be able to overall ripple current (requiring 5012 LED driver. operate to higher and lower voltages so less input/output capacitance) and as to at least partly meet the automotive specifications for tran- improves dimming linearity. The SC5012 is available AEC-Q100 sients and dips. Typically, an initial stage of protection is needed (Grade 2) qualified in a 4mm x 4mm 24-lead QFN package. for high energy ‘load dump’ transients, but this may already be Figure 4 shows a typical application circuit for it. present for protecting other electronics. A buck converter as in the circuit of Figure 3 can only reduce the voltage, so with a Summary nominal 12V system in a car, practically only one or two white There is increasing support for HBLED technology, with an ex- HBLEDs can be driven in series, with their worst-case voltage pansive range of drivers now available and an array of sophisti- drop being about 4.5V each. The Maxim MAX16832A/C drivers cated features offered. Whether it is for the simplest of lighting fit the bill here, with their systems or ones for the most demanding of applications, there -40°C to +125°C automotive temperature range and opera- are both high performance and cost-optimised solutions out tion from an input of 6.5V through to 65V. These ICs feature there that can meet whatever the required criteria may be. Increased hygiene using touchless control panels he use of touch panels for input and control is expected and easy control along with peace of mind. An original high- to grow, however, there are many situations where people sensitivity capacitive sensor detects the approach of a hand in feel either physically or psychologically uneasy about a series of steps – detecting the presence of a hand within 10 touchingT a panel because they won- centimeters from the panel; the der who else might have touched it, or actual position of the hand within because it is dirty, or their own hands 5 centimeters; and the positions of are dirty. fingers within 3 centimeters. Visitors to medical facilities, for Detected data is processed us- example, might be concerned about ing Alps Alpine’s original algorithm the risk of contracting a disease from to realize diverse forms of control an item because they do not know who based on gestures or finger posi- has touched it. Some people might tion. The panel also allows touch have a distaste for dirty surfaces on operation. ticket machines in train stations or flush The system is designed so buttons on toilets in public facilities. features and information on the Others might want to avoid operating display change with the distance a home appliance while eating sweets of the hand or finger and the type with their hands or while cooking in the of operation. It also incorporates kitchen. In situations like these, touch sound feedback, allowing intuitive panels could potentially lose their convenience factor. operation even by first-time users. Alps Alpine aims to commer- The touchless control panel proposed by Alps Alpine , cialize the touchless control panel in around 2021. incorporates the benefits of touch panels while providing safe

26 News June 2020 @eeNewsEurope www.eenewseurope.com LEDS & DISPLAY TECHNOLOGIES

Skin-friendly UVC light-emitting diodes to fight hospital germs By Julien Happich novel LED irradiation system developed by the Ferdi- ogy at Charité – Universitätsmedizin Berlin for skin examina- nand-Braun-Institut aims to kill microorganisms with tions. Another device will soon be delivered to the Institute for ultra-short wave UV light – without side effects. Hygiene and Environmental Medicine of the University Medicine AAccording to the Robert Koch Institute, 400,000 to 600,000 Center Greifswald to clarify the microbicidal effectiveness. infections with hospital germs occur in Germany every year VIMRE is funded by the German Federal Ministry of Education – about 10,000 to 20,000 people die from them. Since multi- and Research (BMBF) as part of the consortium “Advanced UV drug resistant (MDR) pathogens often cannot be treated with for Life” within the Twenty20 program. antibiotics, alternative approaches are needed. One promising Tests carried out by the two project partners with these physical principle is irradiation with UVC light, which can be devices are intended to show that UVC irradiation is suitable for used to destroy microorganisms without allowing resistances to killing microorganisms and especially MDR pathogens (eradi- develop. Within the framework of their Joint Lab GaN Optoelec- cation). At the same time, it is to be demonstrated that this tronics, the Ferdinand-Braun-Institut (FBH) exposure is harmless to humans as long as and Technische Universität Berlin (TUB) specific irradiation doses are maintained. have developed LEDs emitting in the far This will be verified using tissue samples ultraviolet (UV) spectral range. The LEDs of human skin as well as skin and mucosa emit at wavelengths around 230 nm and models, since the preferred habitat of micro- provide more than one milliwatt output organisms such as MDR pathogens is the power. Such UVC LEDs are not yet com- anterior nasal cavity and the pharynx. For mercially available worldwide due to tech- this purpose, the Charité conducts dose- nological challenges of the utilized material dependent investigations of possible DNA system aluminum-gallium nitride (AlGaN). damage to irradiated skin. The University Their light does not penetrate into the liv- Medicine Center Greifswald will determine ing layers of the skin because of their high how effectively the UV LED emitters kill MDR degree of absorption. It is therefore expected that the skin – in pathogens at 230 nm and compare the results with those of UV contrast to long-wave UVC radiation as emitted by mercury lamps with emission at 254 nm and 222 nm. vapor lamps, for example – will not be harmed at all or will be The researchers also envisage that such LEDs could be damaged so little that the natural repair mechanisms compen- used endoscopically. LEDs also emit only little heat and hardly sate for the effect. The researchers hope that this will help to put any strain on the skin. In addition, they do not require high kill MDR pathogens without any long-term side effects. Within voltage – an important safety aspect, since they are used on hu- the framework of the VIMRE project (prevention of infection with mans. The UV LED irradiation system is to be further developed multidrug resistant pathogens via in-vivo UVC irradiation), FBH in the future so that pathogens can be eliminated in places that has developed and produced an irradiation system comprising are difficult to access. The device might also be interesting for an array of 118 of these LEDs on a square area 8cm by side. corona viruses, as they can also be inactivated by short-wave It achieves a maximum irradiation power of 0.2 mW/cm2 with UVC light. Since SARS-CoV-2 replicates in the pharynx in the more than 90 % uniformity over an area of 6 cm x 6 cm. The first phase, it seems plausible to use such light sources in this first prototype was delivered to the Department of Dermatol- part of the body to prevent a COVID-19 disease. Porsche expands development centre with a light channel By Christoph Hammerschmidt ntelligent lighting is becoming an increasingly important ambient conditions. In addition to a 100 metre long, asphalted element for vehicles - for safety, but also for brand design. light evaluation track, the 2,700 m² building will include an This can be seen from an investment that Porsche has now analysis and testing area and a light measurement laboratory. announced:I The sports car manufacturer is expanding its de- According to current planning, it will go into operation at the velopment centre by adding a test bench end of 2022. building for headlamp development. The test bench building will be adjacent The Porsche designers not only want to the design studio, which will open in to improve visibility for the driver, but also 2014. Due to its location in the immediate the car’s recognisability for other road vicinity of the studio, functional and design users. So far, the Porsche Development development will in future be brought to- Center already has a light laboratory and a gether in one place. After consultation with light evaluation track of approximately 30 the building authorities, the light channel meters in length. For night-time measure- will be built to cover the ground. ments and to assess the distribution of the The investment in the new head- headlamps, an outside test track is used. Due to the develop- lamp test bench building is part of a strategic development of ments in lighting technology in recent years, engineers will in Porsche’s development capacities. Since 2010, the company future need a development and testing environment that is inde- has invested a high three-digit million euro amount in the con- pendent of weather and time of day and provides reproducible struction of new buildings and facilities. www.eenewseurope.com eeNews Europe June 2020 News 27 DESIGN & PRODUCTS LEDS & DISPLAY TECHNOLOGIES

Lumotive drives competitiveness of lidar sensors into a new dimension By Christoph Hammerschmidt ith the presentation of its first commercial solid-state and evaluation electronics. The building blocks are also avail- lidar sensors, start-up company Lumotive has further able for automotive suppliers and industrial sensor manufac- advanced the development of its technology towards turers who want to produce their own lidar sensors. The X20 en- economicallyW viable series application in driver assistance ables automotive applications with a range of over 120 meters systems. In the medium term, Lumotive is aiming at even more under full sunlight with a field of view of a 120° x 30°. The Z20 areas of application - outside the car. sensor will have a shorter range (~ 50 meters); however, with a Lumotive’s lidar sensors are based on a proprietary technolo- vertical aperture angle of 70° it opens up the prospect of vari- gy called Liquid Crystal Metasurfaces (LCM). The start-up uses ous industrial applications. Lumotive’s engineers do not intend this technology to steer the laser beam for its lidar sensors. The to limit themselves to the automotive market. Such sensors are technology enables the construction of lidar sensors without also well suited for use in drones, robots and industrial automa- moving parts. Therefore, these devices are more reliable, more tion in general, says CEO William Colleran. compact and above all Also, the automotive much more cost-effective market is not so homoge- than those that use a neous that it could manage mechanical beam steering with a single type of sen- system. In addition, Lumo- sor: In addition to sensors tive’s technology allows the for sophisticated ADAS products to be manufac- systems, which are intend- tured in a standard CMOS ed to relieve the drivers of process that has been proven millions of times over – plus, Size comparison: Prototype, conventional vehicles of it is cost-effective and also scales well. Lumotive also claims commercial product today, goal their tasks, but not replace an advantage over lidar sensors based on micro-electronic- (smartphone). them, Colleran sees a mechanical systems (MEMS), which are also semiconductor- separate market for com- based: LCM technology enables more efficient implementation pletely self-driving vehicles for taxi and delivery services. These and a longer range, they say. move more slowly and are subject to certain restrictions (route, After presenting the X10 prototype last year, Lumotive has surroundings), but without a human driver. The requirements for now introduced its first two lidar sensors, which are destined these sensors are therefore different from those for ADAS. In for commercial series production. The new X20 and Z20 sensor addition, Lumotive also intends to enter the numerically promis- models, which build on the architecture of the X10, are expecting market of consumer devices, especially smartphones, in the ed to be available in the fourth quarter of 2020. These sensors medium term. However, this would require a further significant integrate functional units for the laser transmitter, beam steering reduction in the size of the technology. Mojo raises $51 million for smart contact lenses By Peter Clarke ojo Vision, developer of a contact lens with an in-built overlays. These could be real-time contrast and scene enhance- display, has raised more than $51 million in Series B-1 ment to help with obstacle avoidance and navigation. equity investment. Mojo Vision Inc. (Saratoga, Calif.) Mojo Vision’s latest round of funding was led by New Enter- Mwas founded in 2015 and has now raised more than $159 mil- prise Associates (NEA) with participation from Khosla Ventures, lion in total although the company Dolby Family Ventures, Motorola is still in the research phase. The Solutions Venture Capital, KDDI Mojo Lens gives people access Open Innovation Fund and others. to additional information without Greg Papadopoulos, venture partner looking away to a smartphone or at NEA, will join its board of direc- computer screen. Mojo calls this tors. Prior to NEA, Papadopoulos “eyes-up experience” invisible was CTO of Sun Microsystems and computing. Although the company before Sun, he was an Associate has developed a prototype the Professor of EECS at the Massachu- company is expecting a couple of setts Institute of Technology (MIT). years of regulatory review by the “The potential applications of U.S. Food and Drug Administra- this innovative technology could tion (FDA) before a commercial fundamentally change the way we launch. Initial products are being targeted at those with vision work and view our world as well as improve the quality of life for impairment. An early application of the Mojo Lens will be to help people with vision impairment,” said Papadopoulos, in a state- people with vision impairment by providing image enhancement ment issued by Mojo Vision.

28 News June 2020 @eeNewsEurope www.eenewseurope.com LEDS & DISPLAY TECHNOLOGIES

Perovskite project to combine battery-free wearable LiFi and lighting By Nick Flaherty n ambitious European project is aiming to combine a actually be used in commercial objects such as light panels and photovoltaic solar cell with a low power LiFi light-based wearables,” said Dr. Sylvain Nicolay of CSEM in Switzerland communications system for wearable designs and LED which is coordinating the project and working on the develop- Alighting panels all using the same low cost, flexible perovskite ment of the perovskite photovoltaic cells. semiconductor materials. VTT in Finland will develop a manufacturing process for The €5.6m PeroCUBE Large Area flexible PeroCUBE devices and the integration into project will use perokskite wearables. Aura Light Italia will act as the integrator of the materials to produce both the lighting applications in its first European project. “Perovskite flexible PV solar cell and flex- technology is undergoing a development that is as rapid and ible high speed LEDs and sen- fascinating as OLED technology. Therefore, we would like sors for the LiFi in a wearable to use our extensive know-how in the characterization and gadget such as a wristband. encapsulation of large-area and flexible OLEDs and achieve The project will also produce large LED lighting panels that in- a reasonable combination of these technologies,” said Dr. corporate the LiFi sensors to communicate with the wristband. Christian May, Head of Division Flexible Organic Electronics at 14 European partners from industry and science in ten the Fraunhofer FEP which will characterise and encapsulate the countries will work on the technology for the next 42 months devices. developing the devices and a roll-to-roll manufacturing process The University of Oxford in the UK will bring its expertise for high volume production for the panels. in the development of application-driven perovskite materials Organic-inorganic metal halide perovskite semiconduc- and the fabrication of efficient photovoltaic and light-emitting tors are already used for low cost, flexible solar cells and are devices to the PeroCUBE project, while in Greece IoT expert increasing in efficiency all the time. These are already in pilot Vodafone Innovus in Greece will developinpotron, 45 the x 136 LiFi mm Satzspiegelapplications production with Oxford PV in Germany, and the lab at the Uni- and Eulambia Advanced versity of Oxford where the technology was developed is part of Technologies will develop Professional Power the project. the LiFi transceiver while PeroCUBE takes perovskite technology several steps for- the system will combined ward by using the same materials for large panel LED arrays by Optiva Media in Spain. for lighting and for the next generation of LiFi and Visual Light The project also Communication (VLC) standards and devices that carry data includes TNO in the Neth- over light from LEDs. This links into the development of hu- erlands, CNRS in France, man centric lighting (HCL) that adjusts to the requirements of the Technical University the people in the room as well as data transmission, wearables of Vienna, Alpes Laser in and applications in the Internet of Things (IoT). “PeroCUBE will Switzerland and Noesis provide proof that the specific class of perovskite materials can Technologies in Greece. A new metric to quantify the naturalness of light

Bridgelux has developed a new metric to enable objective comparisons of light sources to natural light. With the growing market interest in human-centric lighting, the question of “naturalness” of light source spectra is becoming a frequently asked question. Lighting specifiers, designers, and Industrial applications luminaire manufacturers regularly consider color quality metrics such as CRI and TM-30, as well as individual Switch Mode Power spectral power distributions (SPDs), when making Supplies for industrial decisions regarding light source selection. However, requirements until now, there has been no metric that quantifies the naturalness of a light source. Standard lighting quality • customized development metrics such as CRI and TM-30 do not fully address the • efficiently • smart naturalness question. In order to address this shortcom- ing, Bridgelux has developed the Average Spectral Dif- ference (ASD) metric, providing an objective measurement of how closely a light source matches natural light over the visible spectrum, averaging the differences of the spectral peaks and valleys between a light source and a standardized natural light source of the same colour correlated temperature (CCT). The ASD value is a quantitative metric of the SPD differences. Unlike CRI and TM-30, where higher values indicate higher quality light, ASD represents the difference in spectra

compared to natural light, therefore the lower the ASD value, the closer the match to natural light. Customized Solutions Bridgelux Made in Germany www.bridgelux.com www.inpotron.com www.eenewseurope.com eeNews Europe June 2020 News 29 DESIGN & PRODUCTS ASICS, FPGAS & ACCELERATORS

Universal processor startup goes for split tape By Peter Clarke emiconductor startup Tachyum Inc. will be using a split 16 cores at 2GHz clock frequency up to 128 cores at 4GHz. The tape-out strategy for its Prodigy processor before the end specs are based on a single 64-core die that will be fault sorted of 2020 in an effort to get samples into customers’ hands and frequency binned for the lower spec chips and with the earlyS in 2021. higher spec being achieved with a two-die component. The company’s timetable has slipped from some earlier The core architecture is a combination of RISC, CISC and reports in 2018 and 2019 but now Tachyum (Santa Clara, Calif.) VLIW [very long instruction word] with the multiple cores con- reports the master chip layout has been completed and about nected via a synchronous mesh network. The ALU pipeline 90 percent of the physical design verified. The design targets depth is a relatively modest nine stages and there is some TSMC’s EUV-enabled 7nm FinFET manufacturing process and branch prediction and out of order execution but mainly through Tachyum’s CEO Radoslav Danilak said that he wants to have an compiler support, Danilak said. This is to keep the hardware FPGA-board emulation of the design in 3Q20 followed by tape- simpler and power consumption lower. In addition to the out before the end of the year. Samples with customers in 1Q21 conventional data types: the 8-, 16-, and 32-bit integers, the followed by production starting in 2Q21 and volume production standard IEEE floating-point formats Prodigy also supports in 2H21. the bfloat16 – a truncation of float32 to its first 16 bits. It has 8 bits for exponent and 7 bits for mantissa – first proposed by Intel in its now defunct Nervana processors. Prodigy also has its own format of 8bit floating point data type with 3 bits of mantissa and 5 bits of exponent. This restricts dynamic range, not necessarily a problem in neural network calculations, but allows Prodify to handle multiple FP8 numbers simultaneously using 32bit registers and data paths for enhanced efficiency. The Prodigy also handles a 4bit integer data type, Danilak said. Tachyum is also making is TPU [Tachyum Processing Unit] available for license. In the data center the processors will be used for training neural networks but at the edge a licensed core could be included in ASICs to perform inferencing. Tachyum is including extensive communications links on its chip and the company said that the simulation of multiple Given that silicon wafers on a leading-edge process can cores with DDR4/DDR5 DRAM controllers, PCIE 5.0, 112Gbps spend three months or more just moving through a wafer fab SERDES, USB, GPIO, PLLs and various I/Os indicate the die the timing still seems breathtakingly fast. “It’s tight, but not size is going to be within design goals and top-level clock fre- impossible. That’s why we will be going for a split tape-out,” quency could be better than expected. said Danilak. The 64-core flagship product is expected to outperform The fab will be able to start on bottom layers and then re- the fastest Xeon processors while consuming one tenth of ceive information about the upper metal connections at a later the power on data center workloads. In addition, Prodigy can date; a method used to overlap design engineering and manu- switch its workload and outperform Nvidia’s fastest GPUs on facturing cycles. It is probably not something that TSMC would neural net AI training and inference. It should be noted that do with every customer but Danilak has built up a track record Nvidia has just announced its 7nm A100 GPU, which it claims in previous employment with the likes of Nvidia, SandForce and can unify AI training and inference and boost performance by Skyera. up to 20x over its predecessors. Nvidia also describes A100 as The Prodigy processor is aimed at the Data Center but it in- a universal workload accelerator in that it can also perform data tends to beat out the competition by being a Universal Proces- analytics, scientific computing and cloud graphics. The A100 is sor that can do general-purpose work and high-performance in full production and shipping to customers worldwide, Nvidia computing work in the style of x86 and ARM and also artificial said. Which may explain Tachyum’s desire for haste. intelligence (AI), deep machine learning (ML), explainable AI, To address the legacy code of data centers in x86 and ARM biological AI and other AI disciplines within a single chip. There compiled forms format Prodigy will use software emulation of are considerable energy consumption, resource utilization and the x86 and ARMv8 instruction sets. The emulator lets Prodigy cost benefits that accrue from being able to simplify the archi- run x86 and other binaries at about 50 percent of native speed, tecture and network of the data center, Danilak argued and in which might be seen as a problem as it kills at least part of addition Prodigy will offer great scalability and PPA benefits at Prodigy’s claimed advantage over its competition. But Danilak the device level. claims that data center purchasers will be savvy enough to Tachyum has specificed four instantiations of Prodigy from realize that this is enough to get them going until they recompile

30 News June 2020 @eeNewsEurope www.eenewseurope.com ASICS, FPGAS & ACCELERATORS

apps into native Prodigy code and enjoy a speed up but more resides. Tachyum announced the raising of $25 million in Series importantly, a power down. A equity investment in July 2019 led by IPM Group and with the Danilak co-founded Tachyum Inc. in 2016 and the com- participation of the Slovak government. At the time Tachyum pany is headquartered in Santa Clara, California with a major said it would bring its chip to market in 2H20 so clearly there engineering site in Bratislava, Slovak Republic, where Danilak has been some slippage over the last few months. Danilak was born. Staffing of about 60 people is roughly split equally. agreed that Tachyum will need to raise more money in 2020. Bratislava is responsible for software and AI developments The cost of complete mask sets at 7nm alone is of the order of while Santa Clara is where the chip development engineering tens of millions of dollars. FPGA puts AI in space By Nick Flaherty

ilinx has launched its first FPGA with space qualification Protection against errors from single event upsets (SEUs) for a decade, designed to support machine learning and caused by radiation is key for space qualified parts. A triple artifical intelligence frameworks. module redundant (TMR) approach is used for some elements, XThe XQRKU060 Ultrascale Radiation Tolerant (RT) Kintex such as three versions of the embedded MicroBlaze proces- FPGA is built on a 20nm process, compared to the 65nm pro- sor. “You don’t have to TMR the whole chip,” she said. “Where cess used for previous space qualified Virtex devices. This gives we knew the circuit was more susceptible we beefed up those four times the logic capacity, with 633 logic blocks, and ten portions. The control circuits for the logic, DSP and DRAM we times the digital signal processing with 2760 DSP slices, as well beefed up with process techniques, duplicate or triple cells, in- as 32 Serialiser/deserialiser (SERDES) high speed interfaces. creasing the gate lengths etc. Then there is software mitigation “Machine learning in space is in its nascent stage but start- on top of that, with duplicate portions of the circuit in the logic ing to make ground in the next four to five years,” said Minal array.” The device is qualified from -55 to +125ºC and with- Sawant, Space Systems Architect at Xilinx. THe move to stands a total dose of 100Krad and single event latchup (SEL) 20nm enables up to 5.7 teraoperations per second (TOPs) of of >80MeVcm 2/mg. This is the requirement for a geostationary peak performance for 8bit integer frameworks (INT8), nearly 25 orbit, and exceed that for low earth orbit designs. It is packaged times that of the previous generation. in a 40 pin ceramic package. For machine learning, the architecture has been extended A development board uses SRAM and non-volatile memories with wide 27 x 18 from Infineon Tech- multiplier and extra ac- nologies, DDR3 and cumulator blocks which an MCU from Cob- otherwise would have to ham and high speed be implemented in the space qualified ADCs logic array. Initial sup- from Texas Instru- port is for open source ments and Teledyne machine learning com- e2V. A TMR Synthe- pilers and DNN array sis tool from Mentor such as TensorFlow and Graphics supports the PyTorch, but d the future design of redundant roadmap includes the logic, and Seakr En- Xilinx Vitis AI framework. gineering in Colorado “This will open up a has produced a rack whole new world of how system for the devel- to use machine learning opment of geostation- in space,” said Sawant. ary satellite communications systems using the RT Kintex. One initial use for the device is in large geostationary satel- “Seakr Engineering has collaborated with Xilinx for 15 lites that handle communication links, hence the 32 high speed years to consistently achieve challenging mission objectives SERDES interfaces. Another is for pattern recognition of camera for advanced space communications applications,” said Paul images. A key capability is that the parts can be upgraded in Rutt, CTO at Seakr. “We have baselined Xilinx’s 20nm Kintex space, for example to upgrade an ML inference framework, UltraScale FPGAs with 12.5 Gbps SerDes links enabling high- However the initial use is likely to be bug fixes, says Sawant. throughput, flexible and reconfigurable modulation, demodu- The FPGA can be reprogrammed by changing the 192Mbit lation, channelisation and routing capability in our newest bitfile that configures the chip. With larger 1Gbit non-volatile advanced RF reconfigurable processor, Wolverine. memories qualified for space use, up to three separate bitfiles The Processor leverages the 10x increase in DSP compute can be stored with completely new functions. capability for direct RF sampling compared to prior generation “The Ultrascale Kintex has true unlimited in-orbit reconfigu- systems.” rable capability – you can send a new bitfile and change the Flight units of the 20nm RT Kintex UltraScale space-grade design and there is no limit,” she said. “What we have seen is XQRKU060-1CNA1509 FPGA will bea vailable in Xilinx Class you can store the bit file in non-volatile memory before launch. B and Class Y test flows per the MIL-PRF-38535 beginning in The second way is via the onboard computer to do the change, September 2020. Mechanical samples and prototype units are for example for a bug fix.” available now. www.eenewseurope.com eeNews Europe June 2020 News 31 DESIGN & PRODUCTS ASICS, FPGAS & ACCELERATORS

AI to boost Earth observations onboard cubesats By Julien Happich uring an online press conference held by Teledyne e2v to promote the company’s latest product developments, theD group’s CMOS sensor division came into focus as being part of a new cubesat project launched by Grenoble’s University Space Centre (Centre Spatial Universitaire de Grenoble or CSUG). First, application engineer Thomas Porchez Roadmap for the Qormino platform. presented the on-board processing solutions tive project will leverage Teledyne e2V’s under development for such applications. 16MP Emerald CMOS sensor together With the 2-in-1 Qormino processing module with the radiation tolerant Qormino pro- soon to be launched, the company is integrat- cessing module. ing its radiation tolerant LS1046 processor Mathieu Barthelemy, Director of the with 4GB of radiation tolerant (RadTol) DDR4 CSUG, presented QlevEr-Sat as an Ar- memory on a single 44x26mm substrate with tificial Intelligence (AI) Earth observation a 2.1GT/s built-in DDR4 bus (featuring ECC). demonstration satellite designed to test AI Built around a quad-core ARM Cortex A72 routines on board and drastically reduce running at up to 1.6GHz, the Qormino offers Teledyne e2v’s space grade Emerald ground data transmission. To illustrate 30 000 DMIPS of computing capability, Por- CMOS sensor. the benefits that AI could bring in image chez highlighted, noting that a development analysis and data reduction, Barthelemy kit was already available while the standalone LS1046 proces- noted that a satellite taking one 16Mpixel image every second sor and DDR4 memory module were being tested for space (coded in 12 bits), ground data transmission amounted to 16 grade reliability before the Qormino module would undergo Tbits/day (2.03 TB/day). such tests by the end of 2020 or early 2021. Next Eric Chevallier, Bid Manager at Tele- dyne e2v discussed the evolution of image sensors for space applications, acknowledging a strong shift from CCD devices (declining) to CMOS devices. The market shift leverages the availability of low-cost mass-produced Com- mercial Off-The-Shelf (COTS) CMOS sensors (including those from Teledyne but also some from third party manufacturers) up-screened and tested according to space grade standards. Among Teledyne e2v’s image sensors getting traction for space applications, Cheval- lier highlighted the Emerald CMOS sensor family, with parts ranging from 8Mpixels at a The QlevEr Sat aims to bring AI onboard 4Kx2K resolution with pixels 2.8µm per side to a cubesat for Earth observation. a 67Mpixel device delivering 8K UHD resolution (with 2.5µm pixels) and capable processing 60 frames per second at full resolution. With AI and on board processing, The Qormino processing platform In the scope of the new cubesat project the QlevEr-Sat could make autonomous developed by Teledyne e2v. dubbed “QlevEr-Sat”, it is a 16Mpixel 4Kx4K decisions. With the Emerald CMOS (4096x4096) device that has been selected imager selected for this project, the ultra (with 2.8μm pixels) which can also be used as a 2kx2k matrix. small 2.8μm pixels provide a ground based resolution of 4.8m at The image sensor is available in B&W and color versions with 8, 500km. For a 16Mpixels sensor, it means the satellite’s camera 10, or 12 bit resolutions and has a throughput of 47fps at 10 bit covers an 18 kilometers-wide swath of land below its orbital or 30fps at 12 bit. path. The Qormino module onboard the QlevEr-Sat will run The partnership between the CSUG and Teledyne e2v is not machine learning algorithms developed by Grenoble’s Multidis- new, the two groups started their collaboration in 2017 with the ciplinary Institute in Artificial intelligence (MIAI) and trained for ‘AMICal-Sat’ project, a 2U nanosatellite whose mission is to different applications. observe auroras and perform image captures. The imaging pay- Due to be launched circa Q1 or Q2 of 2022, the QlevEr-Sat load of this satellite is based on Teledyne’s Onyx 1.3MP CMOS is expected to operate for a year or two, surveying a specific sensor. While the nanosatellite is now ready for launch into region for damages from natural disaster such as volcanoes, orbite with the ArianeSpace VEGA launcher, its launch initially earthquakes or tsunamis. The cubesat could carry out forest scheduled for the 24th of March 2020 has been delayed due to surveys on wild fires and deforestation or it could monitor urban the coronavirus pandemic. expansion or other human activities associated with societal Started in the first quarter of 2020, the QlevEr-Sat collabora- challenges.

32 News June 2020 @eeNewsEurope www.eenewseurope.com A50E_EETimes_2_375x10_875.qxp_A45.qxd 4/24/20 2:50 PM Page 1

ASICS, FPGAS & ACCELERATORS

Sparsity augments AI acceleration on Nvidia’s A100 GPU By William G. Wong he A100 platform pushes the limits of machine learning from the edge to the en- terprise. Peak performance for 64-bit floating point is 9.7 teraFLOPs (TFLOPS). At the other end of the spectrum is the INT4 performance of 1248 TOPS. DC-DC Converters TThe big news at this year’s virtual GPU Technology Conference (GTC) was Nvidia’s release of the A100 (Fig. 1). The A100 GPU with its 54.2 billion transistors can get a bit toasty with a max thermal design power (TDP) of 400 W, and a large array of mod- Transformers & ules can be connected using the built-in NVLinks. Each module supports 600 GB/s of NVLink bandwidth in addition to the 64 GB/s PCI Express (PCIe) Gen 4 interfaces. Inductors The PCIe interfaces support SR-IOV. The A100 GPU is based on the company’s new Ampere architecture that DC-DC Converters provides a significant performance boost 2V to 10,000 VDC Outputs compared to the earlier V100 Volta and Turing architectures. Included is 40 GB of 1-300 Watt Modules HBM2 on-module memory for the A100 • MIL/COTS/Industrial Models with a memory bandwidth of 1555 GB/s. • Regulated/Isolated/Adjustable Programmable Standard Models There’s also a 40-MB L2 cache that’s • New High Input Voltages to 900VDC almost seven times that of the V100. • AS9100C Facility/US Manufactured Seven GPU processing clusters (GPCs) • Military Upgrades and and seven texture processing clusters Custom Modules (TPCs) are incorporated into the A100 GPU, along with 16 streaming multipro- Transformers cessors (SMs) per GPC (Fig. 2). The ten Fig. 1: The A100, based on Nvidia’s 512-bit memory controllers support five & Inductors Ampere architecture, has a bandwidth of HBM2 memory stacks. The SMs support Surface Mount & Thru Hole 1.5 TB/s. all data types. A new shared-memory- • Ultra Miniature Designs based barrier unit provides asynchronous • MIL-PRF 27/MIL-PRF 21308 barriers to handle new copy instructions. The systems support 32 threads/warp and • DSCC Approved Manufacturing 64 warps/SM. • Audio/Pulse/Power/EMI Usually, multiple A100s are tied to Multiplex Models Available the NVLink interface, enabling very large • For Critical Applications/Pico models to be run across an array of Modules, Over 45 Years’ Experience chips. A new feature is Multi-Instance For full characteristics of these and the GPU (MIG), which allows the opposite to entire PICO product line, see PICO’s Full line catalog at occur by splitting up the GPU resources into dedicated and protected islands of www.picoelectronics.com computation. Up to seven instances can be defined running CUDA applications. PICO ELECTRONICS, Inc. CUDA 11 is Nvidia’s latest programming 143 Sparks Ave., Pelham, New York 10803 s environment. Call Toll Free 800-431-1064 Each MIG instance has separate and FAX 914-738-8225 isolated paths through the entire memory E Mail: [email protected] system. Other resources like the on-chip crossbar ports, L2 cache banks, memory Pico Representatives controllers, and DRAM address buses England Solid State Supplies/Ginsbury are also allocated to these logical islands. E-mail: [email protected] Phone: 44 1634 298900 Fig. 2: The streaming multiprocessor This provides predictable throughput and Fax: 44 1634 290904 (SM) is the building block for the A100 latency. L2 cache allocation and DRAM Belgium/Luxemburg/France system. utilization will not be affected by the op- Netherlands/Scandinavia eration of other instances. Error and fault ELCOS/BVBA E-mail: [email protected] isolation are maintained within each instance. Phone: 32 3 218 20 73 www.elcos.be The A100 GPU supports a range of numeric formats, including Tensor Float 32 Germany (Fig. 3). It provides the range of FP32 but trims the number of significant bits, thereby ELBV/Electra Bauemente Vertrieb allowing A100 designers to implement matrix calculations more efficiently. The system E-mail: [email protected] Phone: 49 089 460205442 also supports FP16 and BFLOAT16 along with a host of integer formats. All of these Fax: 49 089 460205442 are available and optimized for machine-learning (ML) model acceleration. Peak performance for 64-bit floating point is 9.7 teraFLOPs (TFLOPS). At the other www.eenewseurope.com eeNews Europe June 2020 News 33 DESIGN & PRODUCTS ASICS, FPGAS & ACCELERATORS

form factors, such as the DGX A100 (Fig. 4). The box delivers 5 petaFLOPS of performance in a single box. The DGX A100 consists of eight A100 modules, six NVSwitches with 4.8 TB/s of bandwidth, nine Mellanox ConnectX-6 200-Gb/s interface cards, dual 64-core AMD CPUs, and 15 TB of Gen 3, NVMe SSDs with a peak bandwidth of 250 GB/s. It’s based on the HGX A100 motherboard.

The HGX A100 motherboard hosts Fig. 3: The A100 supports Tensor Float 32, which allows for eight A100 mod- more efficient implementationof AI matrix operations. ules connected by a new, faster end of the spectrum is the INT4 performance of 1248 TOPS. NVSwitch matrix These numbers are impressive, but Nvidia’s platform also imple- (Fig. 5). The plat- ments sparsity optimization. form also has PCI ML and artificial-intelligence (AI) applications are a prime Express switches target for the A100. The that support 200- Fig. 5: The eight-module, A100-based neural-network models Gb Ethernet NICs HGX A100 motherboard is available for for ML/AI applications like the Mellanox hyperscalers that design systems using typically utilize sparse ConnectX-6 as well open platforms. matrix operations. The as NVMe storage. A sparsity support allows pair of high-performance CPUs can be connected to the PCIe these operations to be switches. A four-A100 version is also available. Both support performed more quickly. GPUDirect Storage, which lets the GPUs work with the NVMe For the Tensor Float storage, bypassing the CPU. 32, the standard 156 Nvidia continues to deliver more performance by orders or TFLOPS doubles to 312 magnitude every couple of years. That alone would be impres- TFLOPS with spar- sive, but it’s really the software infrastructure that the company Fig. 4: The DGX A100 holds sity support. FP16 and has built around its hardware that makes it work. This starts eight A100 GPU modules (1), six BFLOAT16 also benefit with CUDA and includes everything from the Isaac SDK for ro- NVSwitches (2), nine Mellanox in a similar fashion, in botics, to cuDNN, to its TensorFlow support. Even the new MIG ConnectX-6 200-Gb/s interfaces (3), addition to the integer support is integrated with its Kubernetes container manage- dual 64-core AMD CPUs (4), and 15 support. ment system. The A100 is being This article first appeared on Electronic Design – TB of Gen 3, NVMe SSDs (5). used in a number of www.electronicdesign.com How to overcome high latency in the 5G era? By Robert Barnes

rom video conferencing to multiplayer online gaming, all In the standards for 5G many innovations have been included of us will have experienced latency issues at some point. with the target of reducing latency by 10 or 20 times compared But this isn’t a new phenomenon that’s tied to the unique to 4G. The degree to which these are delivered on will in a large Fcircumstances in which many find themselves today. It is a chal- part come down to whether we, as an industry, let latency get in lenge that goes to the heart of how communications technolo- the way of a supercharged future. gies work. Channel coding as a latency challenge Low latency is a challenge As mobile devices of all shapes, sizes and form factors com- to be overcome in the 5G era municate with fixed infrastructure, wireless communications are Higher latency or “ping time” means that in Fortnite you’ll find susceptible to noise and interference. If uncorrected this will yourself “killed” before you’ve ever seen your opponent; in a lead to the received message differing to the transmitted mes- game of FIFA, you’ve conceded a goal before you’ve seen the sage and therefore requesting a retransmit. ball kicked. High latency renders games unplayable, causes To address this, before sending a message, transmitters use people to talk over each other in video conferences, and risks a channel encoder to protect the communication from these killing the market for consumer applications such as Virtual and transmission errors by adding additional bits, which enable Augmented reality (VR and AR) that many hope will be major 5G the receiver decoding the transmitted signal to recompile the revenue opportunities but are highly latency sensitive. message correctly, a process sometimes called forward error correction. Robert Barnes, VP Sales and Marketing, Accelercomm Different types of channel codes exist, each coming with its www.accelercomm.com own benefits and challenges. For 5G, Low Density Parity Check

34 News June 2020 @eeNewsEurope www.eenewseurope.com ASICS, FPGAS & ACCELERATORS

(LDPC) coding has been chosen for the data channel, where the ment. Channel coding has to be addressed with configurabil- user’s information is transmitted, and Polar Coding has been ity at the design stage to enable optimal cost-effective design chosen for the control channel, where the mobile and the base choices across not only FPGA and ASIC implementations, but station exchange control information. also software-only virtualised network infrastructure. Innovation in this area includes patented encode and decode Delivering channel coding performance IP for both Polar and LDPC coding. For example, AccelerComm Although the coding schemes themselves have been stan- has collaborated with Intel to develop a new, highly optimised dardised, it’s the implementation techniques and algorithms of LDPC decoder in software for its FlexRAN Reference Software. these channel codes, which are outside of the standardisation This reference architecture can be implemented in software- process, that can greatly reduce latency and improve Quality of based radio stations, which can sit on any part of the wireless Service (QoS). networks from edge to core. Overcoming the latency challenge and unlocking the full Integrating LDPC codes within a FlexRAN-based software- power of 5G is the focus of a growing band of channel coding based radio station can increase throughput by up to 3X whilst experts, combining in-depth knowledge with unique insights as increasing network power efficiency and reducing latency – to the approach the industry should implement when it comes ultimately delivering a better user experience and potentially to LDPC and Polar coding. reducing system costs. The same designs also deliver market This expertise is critical because, whilst LDPC codes have leading performance when implemented in FPGA and ASIC. been used in Wi-Fi and satellite communications in the past, 5G has adopted the most flexible and most capable LDPC code Taking a new look at channel coding ever standardised. This flexibility is key to allowing New Radio McKinsey has forecast that low latency 5G networks in health- to meet the requirements of the diverse range of 5G use cases care could free up additional investment capacity in the sector but presents its own challenges. and generate $250 billion to $420 billion in global GDP impact 4G has a fixed timing for the radio whereas 5G supports up by 2030, while in manufacturing they see ultra-reliable and to 5 timing options (re- low-latency commu- ferred to as numerolo- nications (uRLLC) 5G gies). 5G also supports networks potentially a much wider range of having a GDP impact frequencies, including of $400 billion to $650 those above 30HGz billion by the decade’s referred to as mmWave, end. At the most recent and throughputs up to count, published by the 20X those of LTE. The Global mobile Suppliers channel coders and de- Association (GSA) at the coders must deliver the end of March, 70 opera- required performance tors have now launched for all these different commercial 5G services options. in 40 countries. How- For the industry to ever, this is just the tip go forward and unlock of the iceberg, with 381 5G’s full potential, the channel coding challenges of LDPC and operators in 123 countries now investing in 5G. Polar have to be overcome if we are to solve the 5G latency In a research note published in December 2019, Gartner challenge. Implementing flexible, high performance decoders is forecast that by 2021, investments in 5G NR network infrastruc- possible – but it requires a fresh approach. ture will account for 19% of the total wireless infrastructure revenue of communications service providers (CSPs), up from 6% For example, LDPC decoders have to have many elements in 2019. While early 5G applications will focus on fixed wireless working in parallel to meet the performance requirements and access (FWA) and enhanced mobile broadband (eMBB), the there are two different established architectures. One of these analyst firm is expecting first uRLLC private networks and then can efficiently achieve the throughput requirements, but at massive machine-type communications (mMTC) to be deployed relatively high latency whereas the other achieves lower latency between 2021-2025. These later use cases will really stretch the with reduced efficiency at some throughputs. AccelerComm has flexibility of 5G. developed a unique approach that efficiently delivers on both As more 5G networks are deployed, devices launched, and requirements. services commercialised, so the pressure to deliver on the 5G Compared with LDPC codes, Polar codes are far less mature subscriber promise will become more acute. Low latency is not and a conventional approach to their implementation has not a ‘nice to have’ or something that is the sole preserve of niche yet emerged. The encoder and decoder of a Polar code have ultra-low latency use cases. Rather, it is critical for both today’s regular but intricate structures. AccelerComm’s patented solu- mass market applications and tomorrow’s future use cases, be tions deliver a highly efficient complete 3GPP compliant solu- they autonomous vehicles, remote surgery or smart manufac- tion to simplify integration. turing. With a relentless downward pressure on revenue and margins, performance must be delivered efficiently. Integrating the channel coding By addressing latency at the design stage and integrating the building blocks latest channel coding technologies and innovations into both You can’t build a “one size fits all” when trading off latency, physical and software-defined network elements, mobile opera- area, throughput and power for multi-market solutions from tors and equipment manufacturers can design-in low latency handset modems, small cell and vRAN infrastructure develop- performance that is fit for the 5G era. www.eenewseurope.com eeNews Europe June 2020 News 35 DESIGN & PRODUCTS ASICS, FPGAS & ACCELERATORS FPGA power system management By Pinkesh Sachdev ield-programmable gate arrays (FPGAs) trace their origins On the other hand, a less accurate ±2% dc supply leaves less back to the 1980s, evolving from programmable logic room (±1%) for transients, requiring more bypass capacitance devices (PLDs). Since then, FPGA resources, speed, than the previous case. Fand efficiency have improved rapidly, making FPGAs the go-to Tweaking or trimming of the FPGA supply voltage level solution for a wide variety of computing and processing ap- around the default setpoint is required for last-minute design plications, especially when production volume doesn’t justify changes, design reuse in another application, board margin application-specific integrated circuit (ASIC) development costs. testing, and dynamically optimizing system power consump- FPGAs have advanced to such an extent that they’ve also tion during development or field operation. Soldering in different found homes in large-scale deployments. For instance, after resistors in the supply’s feedback network isn’t the quickest successfully speeding up the Bing search engine with FPGAs or most feasible solution for such situations. One method to in a 2013 pilot program, Microsoft expanded FPGA-equipped achieve voltage trimming is with a digital-to-analog converter server usage to its cloud data centers. (DAC) driving the feedback network of a voltage regulator (see Figure 2). Software code needs to be written for the trim routine FPGA power system requirements to obtain supply voltage measurement data from an analog- FPGAs require a few different low voltage supply rails, each with to-digital converter (ADC), to compute the correct DAC code, its own voltage and current specification, to power their internal and then slowly adjust the DAC output to the computed code core logic, I/O circuits, auxiliary logic, transceivers, and mem- for smoothly ramping the supply voltage, without glitches or ory. These rails may need to turn on and turn off in a specific overshoots, to the target level. This trim routine needs to be sequence to avoid damaging the FPGA. Point-of-load (POL) repeated over time to ensure that the supply doesn’t walk away regulators step down the board’s higher input supply voltage to from the target voltage due to components drifting with time or the multiple lower rail voltages required by the FPGA. Switching temperature. regulators are used as POL regulators when power conversion efficiency is paramount, whereas linear regulators—for example, low dropout (LDO) regulators—are employed for noise-sensitive circuits such as PLLs and transceivers. Typical board input voltages are 5 V, 12 V, 24 V, and 48 V, while FPGA rail voltages range from below 1 V to around 3 V. For high input voltages (12 V, 24 V, 48 V), an extra step-down may be needed to generate an intermediate voltage bus that feeds the POL regulators (see Figure 1). Among the FPGA rails, the core supply requires the lowest voltage (around or below 1 Fig. 2: POL supply output voltage trimming to target voltage V) and highest accuracy (±3% or better), with current levels in with DAC and ADC. the tens of amperes depending on FPGA resource utilization. To prevent logic errors, the supply variation needs to be limited to Monitoring FPGA supply voltages, currents, and faults is tens of millivolts, as dictated by the FPGA rail tolerance specifi- essential to understanding system health and power consump- cation, not just under dc conditions but also during FPGA cur- tion under different scenarios because the FPGA is the brain rent transients. The worse the power supply’s dc accuracy, the of the electronic system. Such an understanding, coupled with more bypass capacitance is needed to maintain an acceptable trimming capability, avoids designing supplies for the worst supply voltage under transient conditions. For example, assume case, saving cost and power. Moreover, an upcoming system a ±3% core voltage tolerance specification. Using a ±1% ac- malfunction could show up as an abnormal trend in FPGA curate dc supply leaves a good ±2% allowance for transients. power consumption, alerting the host controller or service staff before the board or system goes down. Voltage monitoring requires ADCs, whereas current monitoring also requires level- shift circuits to translate the high-side current sense voltage to a ground-referenced voltage; for example, with a transconduc- tance amplifier, as shown in Figure 3.

Fig. 1: One possible FPGA power tree: a high voltage input supply (for example, 12 V, 24 V, or 48 V) is stepped down to an intermediate voltage bus feeding the POL regulators that power the FPGA.

Pinkesh Sachdev is an applications engineer for power system Fig. 3: One possible discrete circuit for monitoring POL supply management (PSM) products at Analog Devices – output voltage, current, and power. www.analog.com

36 News June 2020 @eeNewsEurope www.eenewseurope.com ASICS, FPGAS & ACCELERATORS

One’s head may be spinning after reading this long laun- up board space. Supply output currents are measured using a dry list of requirements although we have not discussed fault sense resistor, inductor DCR, or the IMON output of a power management. What should happen when a POL output goes supply. The voltage and current measurements are multiplied undervoltage or overvoltage—that is, outside the valid volt- internally to provide a convenient POL power output reading. age window? Should only the faulting supply be turned off, or Supply sequencing, supervision, and EEPROM fault-logging should other supplies be turned off too? How does one debug a are built into the LTC2972. Sequencing is accomplished with fault that has shut down the board? time delays written to an internal register or with cascaded As one can see, managing an FPGA’s power system can power-good signals. Dedicated fast comparators signal faults become complicated very quickly, distracting from the essential when POL input voltage, output voltage, and temperature stray FPGA application. Remember that the FPGA’s power tree is outside digitally settable low and high thresholds. Faults trigger just a portion of the overall power system on a digital process- EEPROM black box recording, simplifying failure analysis while ing board. Most of the above requirements also apply to other providing insight into future system improvements. A first fault digital devices such as ASICs, DSPs, GPUs, SoCs, and micro- command provides additional insight into causes of system processors. What is needed is a power system management failure. Faults can be flexibly propagated to other supplies or solution that is simple, scalable, and flexible. other DPSM devices. The LTC2972 features voltage, current, power, and energy Digital power system management monitoring of the intermediate bus input to POL converters. Analog Devices provides a portfolio of digital power system Monitoring circuit board power and energy use is a prerequisite management (DPSM) devices to deal with the complex power for managing, optimizing, and reducing their consumption in systems found on digital processing boards. DPSM devices order to lower server and data center cooling and utility costs. are available with and without integrated dc-to-dc conversion The LTC2972 relieves the host of burdensome polling and com- to either replace POL regulators or work with existing POL putation by conveniently providing the input energy, reported regulators. Power system managers—that is, without dc-to-dc in joules, and the elapsed time through a PMBus interface, conversion—add digital monitoring and control to any existing the industry standard for communicating with power manage- analog power system, whether made up of switchers or LDO ment and conversion devices. When combined with its digital regulators. A single device such as the LTC2980 trims, margins, measurements of POL output voltages, currents, and power, monitors, sequences, supervises, fault logs, and fault manages the LTC2972 enables long-term monitoring of a power system’s 16 POL regulators. Differing channel-count devices (2, 4, 8, or conversion efficiency. 16 channels) can be mixed and matched to manage hundreds Programmable power-good, or general-purpose input/out- of rails. The 2-channel LTC2972 is the latest addition to this put (GPIO), pins are available with each channel. The LTC2972 portfolio, providing a simple introductory solution for monitor- connects with other power system managers to coordinate ing and controlling the two most critical rails in such a power sequencing and fault management of more than two rails. system; for example, the FPGA core and auxiliary rails. PMBus-compatible commands over an I2C/SMBus interface are used for flexible programming and data readback of the 2-channel power system manager power system. Configuration is accomplished through the The LTC2972 is a 2-channel power system manager, adding LTpowerPlay development environment, which supports all comprehensive software-based monitoring, control, and black of Analog Devices’ DPSM products (see Figure 5). Once the box fault recording to the power systems of FPGA, ASIC, and internal EEPROM is programmed with the needed application- processor boards, accelerating time to market, enhancing specific configuration, no further software coding is required for system reliability, and optimizing board energy consumption autonomous operation. (Figure 4). POL supply output voltages are trimmed, margined, and monitored using a best-in-class 16-bit ADC with 0.25% Conclusion total unadjusted error (TUE), improving board yields and long- FPGAs are spreading to all kinds of electronic systems, even term performance. The ability to tighten POL output voltage to taking over the job of ASICs, but they come surrounded by a ±0.25% accuracy leaves plenty of room for it to move during complicated power system. Analog Devices has a wide range of load transients (±2.75% for a ±3% FPGA rail specification), DPSM products to help manage this complexity. If you’ve never significantly reducing needed bypass capacitance and freeing tried DPSM before, the LTC2972 provides an easy introduction to their ability to remove the power system complexity on digital processing boards.

Fig. 4: LTC2972, a 2-channel power system manager with intermediate bus energy monitoring and POL output power Fig. 5: LTpowerPlay development environment for DPSM monitoring. products: eliminates coding for autonomous operation. www.eenewseurope.com eeNews Europe June 2020 News 37 DESIGN & PRODUCTS ASICS, FPGAS & ACCELERATORS

Startup launches near-binary neural network accelerator By Peter Clarke P company Leapmind Inc. (Tokyo, Japan) has introduced the 6.55TOPS at a 800MHz clock frequency. This represents an ef- Efficiera neural network accelerator IP core. Efficiera specifi- ficiency of 27.7 TOPS/W. The company is seeking applications cally targets inference processing for Efficiera in edge devices that ofI convoluted neural networks and are power- and cost-constrained can be implemented in FPGA or ASIC such as household electrical forms. The company claims its solution goods, industrial machinery, is “ultra-low power” due to its use of construction equipment, surveil- single- or two-bit quantization. lance cameras, and broadcasting Typically processing is done on equipment as well as miniature 8- or 16- or 32-bit data types depend- machinery and robots with limited ing the accuracy of result required. heat dissipation capabilities. However, for neural networks results Alongside the core available high resolution processing can be a for license LeapMind is launch- waste of area and energy for the reso- ing a software development kit to lution required. Binary neural networks provide a dedicated learning and can be used to reduce the processing development environment. There burden but at a loss of accuracy. is also the Efficiera Deep Learning Model and Efficiera Profes- LeapMind claims that by using 1-bit weight coefficients and sional Services. This combination allows customers to build 2-bit activation values for intermediate data it is possible to binary or near-binary deep learning models applicable to their maintain accuracy while achieving a significant reduction in the own unique requirements. model area, thereby maximizing speed, power efficiency, and LeapMind was founded in December 2012 as AddQual- space efficiency. The company has benchmarked an instantia- ity and since its formation has raised 4.99 billion yen (about tion of its core as achieving occupying 0.422 square millime- US$46.5 million). The company plans to ship Efficiera later in tres in a TSMC 12nm manufacturing process and achieving 2020. RISC-V system-level architecture exploration SDK and neural network IP

Mirabilis Design has released the VisualSim RISC-V system for low-power FPGA-based smart vision modeling and simulation environment. With this release, VisualSim Through its Smart Embedded Vision initiative, Microchip Technol- provides a complete RISC-V modeling package for developers of ogy aims to address the growing need for power-efficient infer- RISC-V IP, designers of RISC-V encing in edge ap- processors and systems engi- plications by making neers developing applications it easier for software using RISC-V components. The developers to imple- package contains configurable ment their algorithms RISC-V core, vendor-specific in the company’s RISC-V processors, TileLink, PolarFire FPGAs. Mi- DMA, peripherals, RTOS, crochip’s VectorBlox memory modules and over Accelerator SDK is 20 application templates in IoT, networking, wireless, consumer designed to enable developers to code in C/C++ and program electronics, automotive and high-performance computing sys- power-efficient neural networks without prior FPGA design experi- tems. The library components are compatible with the rest of the ence. The highly flexible tool kit can execute models in Tensor- hardware, software, schedulers and network library from Mirabilis Flow and the open neural network exchange (ONNX) format which Design. The VisualSim system model provides the first clear view offers the widest framework interoperability. ONNX supports many of your product using RISC-V. These models are constructed frameworks such as Caffe2, MXNet, PyTorch, and MATLAB. Un- early in the design phase and far before development has started. like alternative FPGA solutions, Microchip’s VectorBlox Accelera- This solution enables product designers to evaluate feasibility, tor SDK is supported on Linux and Windows operating systems. It eliminate risk and identify system bottlenecks of products using also includes a bit accurate simulator which provides the user the RISC-V. In the VisualSim environment, the user constructs a mod- opportunity to validate the accuracy of the hardware while in the el of the product, add the traffic and sensor interfaces and defines software environment. The neural network IP included with the kit the user-cases. This model is simulated with different parameter also supports the ability to load different network models at run values and scenarios. The generated reports provides visibility time. For inferencing at the edge, PolarFire FPGAs deliver up to 50 into timing deadlines, power consumption, resource efficiency, percent lower total power than competing devices, while also of- deadlocks, buffer occupancy, data overflow, quality of service fering 25 percent higher-capacity math blocks that can deliver up and functional correctness. RISC-V library from Mirabilis Design to 1.5 tera operations per second (TOPS), says the company. The contains all the building block required to assemble a product that PolarFire FPGA neural network IP is available in package sizes as incorporates a RISC-V. small as 11x11mm. Mirabilis Design Microchip Technology www.mirabilisdesign.com www.microchip.com

38 News June 2020 @eeNewsEurope www.eenewseurope.com ASICS, FPGAS & ACCELERATORS

Nvidia releases new GPU Low power 28nm FPGAs based on Ampere architecture deliver on-device AI processing The Nvidia A100 GPU based on the company’s Ampere GPU Lattice Semiconductor has launched the latest version of its architecture, is fabricated on the TSMC 7nm N7 manufacturing complete solutions stack for on-device AI processing at the process. The device includes more Edge, Lattice streaming multiprocessors (SMs), larger sensAI 3.0. The and faster memory, and interconnect new stack includes bandwidth with third-generation NVLink support for the to deliver massive computational CrossLink-NX throughput. The A100’s 40 GB (5-site) family of FPGAs for high-speed, HBM2 memory has a band- low power smart width of 1.6 TB/s, which is over 1.7x vision applica- faster than V100. The 40 MB L2 cache tions and features on A100 is almost 7x larger than that of Tesla V100 and provides customized convolutional neural network (CNN) IP, a flexible over 2x the L2 cache-read bandwidth, the company claims. accelerator IP that simplifies implementation of common CNN Nvidia also released CUDA 11, the latest version of its Compute networks and is optimized to further leverage the parallel Unified Device Architecture parallel computing platform. CUDA 11 processing capabilities of FPGAs. For applications like smart provides new specialized L2 cache management and residency vision that require higher Edge AI performance, CrossLink-NX control APIs on the A100. The SMs in A100 include a larger and FPGAs running sensAI software deliver twice the performance faster combined L1 cache and shared memory unit (at 192 KB at half the power when compared to prior releases of the per SM) to provide 1.5x the aggregate capacity of the Volta V100 solutions stack. Updates to the NN compiler software tool let GPU. The A100 GPU comes equipped with specialized hardware developers easily compile a trained NN model and download units including third-generation Tensor Cores, more video decoder it to a CrossLink-NX FPGA. When running on a CrossLink-NX (NVDEC) units, JPEG decoder and optical flow accelerators. All of FPGA, the sensAI solutions stack offers up to 2.5 Mb of distrib- these are used by various CUDA libraries to accelerate HPC and uted memory and block RAM and additional DSP resources for AI applications. The Multi-Instance GPU (MIG) feature can physi- efficient on-chip implementation of AI workloads to reduce the cally divide a single A100 GPU into multiple GPUs. need for cloud-based analytics. Nvidia Lattice Semiconductor www.nvidia.com www.latticesemi.com

FPGA development suite FPGA accelerator board for Talon RF/IF signal recorders targets HPC, HFT and prototyping The ArchiTek FPGA development suite announced by Pentek Aldec has launched a new FPGA accelerator board for high enables engineers to add FPGA IP to recording systems, such performance computing (HPC), high frequency trading (HFT) as threshold detec- applications and high speed tion, spectral filter- FPGA prototyping. The HES- ing, digital down- XCKU11P-DDR4 is a 1U form conversion, signal factor board featuring a Xilinx classification, de- Kintex UltraScale+ FPGA, a modulation and PCIe interface and two QSFP-DD many other digital connectors (providing a total of up to signal processing 400Gbit/s bandwidth), and which hits the ideal techniques. De- sweet spot between speed, logic cells, low power draw and veloping custom IP for an FPGA requires an architecture that price. The new product, which joins Aldec’s popular range of protects the user from custom IP development pitfalls such as FPGA accelerators and prototyping boards also features an breaking the existing IP and corresponding recording software. FMC HPC connector for interfacing with Aldec’s FMC daugh- ArchiTek harnesses Pentek’s Navigator FPGA Development Kit ter cards. In addition, the HES-XCKU11P-DDR4’s FMC HPC (FDK) and Board Support Package (BSP) to provide a develop- connector is compliant with the ANSI/VITA 57.1 standard and ment environment that steps engineers through the process provides easy extension to similarly compliant peripherals. The of integrating custom IP into the recorder. Along with the ‘DDR4’ in the new product’s name reflects its ability to con- Navigator FDK, ArchiTek provides the foundation and example nect with an external DDR4 memory module via a SO-DIMM projects for adding IP to user blocks and creating additional memory socket while latest generation QSFP-DD connec- data-path branches from existing data streams. The structured tors enable network acceleration and wired communication design protects the recorder’s standard functionality, reducing projects. As for the Xilinx Kintex Ultrascale+ device at the heart development time and risk. Customers can now add FPGA IP of Aldec’s new board it is the XCKU11P-FFVE1517 (-3 speed to a recorder for real-time, on-the-fly digital signal processing grade) and has 653K system logic cells, 597K CLB flip-flops, during the data acquisition process, greatly reducing the time 299K CLB LUTs, 21.1Mb total block RAM, 22.5Mb UltraRAM, associated with post-processing recorded data. and 2,928 DSP slices. Pentek Aldec www.pentek.com www.aldec.com www.eenewseurope.com eeNews Europe June 2020 News 39 DESIGN & PRODUCTS SENSORS & TRANSDUCERS

Design of square PowerHap Touch me – feel me! actuators. This family of products includes four types By Christoph Jehle with edge lengths of between hether for smartphones, vehicles, industrial controls, 9.0 mm and 26.0 mm and domestic appliances or gaming consoles, the future insertion heights from 1.1 belongs to piezo-based push-button solutions with mm to 2.3 mm. Thanks to hapticW feedback. the multilayer technology, As a result of the demand for greater convenience and forces in excess of 25 N and safety, multi-functional human-machine interfaces (HMIs) such displacements of 230 µm can as touchscreens and other touch-sensitive surfaces can now be be achieved. found almost everywhere. Existing solutions for haptic feedback such as eccentric rotary mass (ERM) or linear resonant actua- piezo effect it contracts simultaneously in both the x and y axes. tors (LRA) have a number of drawbacks. These include their Two titanium cymbals bonded to either side of the plate amplify relatively large dimensions, the insertion height being particular- this contraction by a factor of 15 in the z-axis, achieving a large ly relevant for display controllers, and a high demand for power displacement of up to 230 µm for the largest 2626H023V120 in battery-operated devices such as smartphones or tablets. type. The high level of linearity between applied voltage and Furthermore, the frequency or the strength of the feedback are displacement is advantageous. For this type it is 1.8 µm/V, at best only variable within very narrow limits, and they do not enabling precise and, at the same time, variable control of both have any sensory capabilities. amplitude and signal form. A considerably better alternative is offered by TDK’s Power- Thanks to the multilayer pi ezo technology PowerHap offers Hap and PiezoHapt families of piezo-based actuators. These unprecedented performance in terms of response time and ac- components are a combination of pressure-sensitive sensor celeration. For example, with a rise time of just 1 ms the accel- and actuator in one. With the two product families, TDK offers eration of the cymbals is already 15.0 g and attains a maximum the widest range of actuators for haptic feedback. value of 35 g when a mass of 100 grams is applied, generating TDK has developed the L8060 and L3015 types specially for a very large force of up to 25 N. Comparable values cannot be use in vehicle and smartphone displays. They have impressively achieved using conventional solutions. Figure 3 illustrates the low-profile insertion heights of just 0.35mm and 0.3mm with response time and acceleration for type 2626H023V120. areas of 80x60mm (L8060) or 30x15mm (L3015). The unimorph design of the PiezoHapt is based on a flat multilayer piezo element that is connected to a vibration plate on one side and can be mounted directly underneath displays.

Fig. 3: Typical acceleration g as a function of the voltage Fig. 1: At no more than 35 mm, the TDK PiezoHapt types are when a mass of 100 grams is applied. The voltage signal here extremely thin and are suitable for smartphones, tablets and is a half-sine with a peak of 120 V and a pulse length of 5 ms, vehicle displays. corresponding to a frequency of 200 Hz. The actuators are designed for operating voltages of ≤24 PowerHap actuators are suitable for a wide stimulation range V (L8060) or ≤12 V (L3015). Depending on the amplitude and from 1 Hz to 500 Hz with variable amplitudes, durations and frequency of the applied voltage, a wide range of vibration signal shapes, such as sine, triangle or rectangle. Depending on patterns can be generated with PiezoHapt. These types are the type, the drive voltage is between -20 V and 120 V or -10 V especially well-suited for OLED displays. and 60 V. These actuators are therefore able to generate a customized haptic feedback to human mechanoreceptors. Corre- Acceleration, force and response time spondingly, these actuators enable designers to create specific, The actuators of the PowerHap family are based on multilayer high-definition haptic feedback profiles that users expect from piezo plates with copper inner electrodes. Figure 2 illustrates cutting-edge HMIs in automotive and industrial applications. the basic design. The combination of low-profile designs no more than 2.3 mm If a voltage is applied to the piezo plate, it only expands thick and strong feedback allows these actuators not only to be minimally in the z axis, but due to the constant volume of the integrated in displays, but also to be placed directly below flat surfaces. This enables solutions with a high degree of leak- Christoph Jehle is Manager Technology & Product tightness which is just as necessary in harsh industrial environ- Communications at TDK Electronics - ments as in applications with stringent hygiene requirements, www.tdk-electronics.tdk.com such as in the food industry or medical technology.

40 News June 2020 @eeNewsEurope www.eenewseurope.com SENSORS & TRANSDUCERS

Excellent sensory properties board is based on the haptic driver architecture from Boréas Apart from the excellent actuator performance, the Power- Technologies. The main features are the compact dimensions, Hap components also feature outstanding sensor functional- low current consumption and fast response times. In addition, ity thanks to the multilayered piezo technology. In contrast to the board offers extensive adjustment options with regard to the conventional switches or pushbuttons that, as is well known, voltage amplitude, frequency, pulse repetition rate, signal shape can only provide the On and Off switching states, PowerHap and sensor functionality. delivers an output voltage that is proportional to the pressure The second evaluation kit is suitable for both 60 V and 120 applied. Figure 4 illustrates this behavior. V PowerHap types and is available in two versions: A driver for one actuator and a driver for up to 5 actuators. These kits contain PowerHap actuators of the types 0909H011V060, 1313H018V120 and 2626H023V120.

Fig 4: When functioning as a sensor, the output voltage of PowerHap is to a great extent proportional to the force applied.

As a result, and unlike conventional mechanical push-buttons, it is even possible to create solutions based on PowerHap Fig. 6: The evaluation kits provide an initial impression of that trigger different actions and haptic feedback, depending on haptic feedback with PowerHap actuators and allow users to the force applied. evaluate design possibilities. Slim design for lateral installation In addition to the square PowerHap types that are particularly Competent design support through suitable for use under flat surfaces, TDK has also developed cooperative ventures four slim, rectangular PowerHap types that are designed to Systems with haptic feedback are relatively new technolo- provide lateral haptic feedback in addition to vertical haptic gies. Accordingly, no standard solutions yet exist which can be feedback. The two smallest versions only 9mm or 12mm long, drawn upon. In order to enable customers from every industry are particularly suitable for handheld-held devices. The two to complete a fast and cost-efficient design-in of haptic solu- larger versions with side lengths of 60 mm are currently the tions, TDK has entered into joint ventures with three leading most powerful PowerHap types, as they enable forces of up touch response companies. to 50 N to be generated, enabling masses of up to 1 kg to be Aito (www.aito-touch.com) offers controller solutions with moved. These powerful packages can, for example, be installed its HapticTouch hardware and software, which convert piezo at the side of displays in order to achieve a horizontal haptic actuators into precise touch sensors with haptic feedback. feedback, as shown in Figure 5. These enable tactile switches or pushbuttons to be simulated on smooth surfaces. Using the combined haptic technologies of TDK and Aito, touch response solutions can be implemented on smartphone touchscreens, or conventional switches and pushbuttons in vehicles can be replaced. Boréas Technologies (www.boreas.ca) has developed the BOS1901 driver IC on the basis of its patented CapDrive technology. It is particularly energy efficient, also offering sensor functions, and is especially suitable for controlling TDK PowerHap actuators with an operating voltage of up to 60 V. In a further stage of development, Boréas will develop the first low-power piezo driver IC for the larger TDK PowerHap models with a maximum driver voltage of 120 V. Fig. 5: Rectangular PowerHap types are suitable for the lateral Immersion (www.immersion.com) is a leading developer and control of displays. Typical applications are vehicle displays. licensor for touch -feedback technologies. TDK has struck a co- marketing agreement with this company for its PowerHap and To enable users to gain initial impression of the varied pos- PiezoHapt actuators. Immersion certifies these actuators for sibilities of haptic feedback with PowerHap actuators, TDK use with its software products and also includes the actuators offers two evaluation kits. The BOS1901Kit development board in its reference designs. This enables customers to implement is designed for the 60 V actuators that are contained in the kit advanced haptic solutions with the world’s thinnest and most (0909H011V060, 0904H014V060 and 1204H018V060). The powerful actuators. www.eenewseurope.com eeNews Europe June 2020 News 41 DESIGN & PRODUCTS SENSORS & TRANSDUCERS

Nano-optomechanical displacement sensor reaches 45fm resolution By Julien Happich ptical read-out sensors The sensor is built out based on optomechanical of an indium phosphide systems are frequently (InP) membrane-on- Oused in sensing applications such silicon (IMOS) platform, as in atomic force microscopes fabricated via a series (AFMs), able to measure the of lithography steps to smallest of surface variations, define the waveguides Scanning electron microscope (SEM) forces and movements. These and cantilever, while the devices generate sub-nanometer image of the nanomechanical directional final sensor integrates the Simulated electric field resolution images by measur- coupler used as a transducer (a). (b) transducers, actuator, and ing the laser light reflected by Schematic illustration of light passing distribution (|E|) before photodiodes. Compared the deflection of a cantilever through the direction coupler before and after displacement to silicon photonics, this over a surface of interest. Now, (up) and after (down) actuation. Credit: (55 nm) of an ideal device. platform allows integration researchers from Eindhoven Eindhoven University of Technology of passive components, University of Technology have lasers and detectors in a micron-thick and high-confine- drastically increased the resolution of such systems, leveraging ment InP membrane. a newly designed nano-optomechanical system (NOMS) with The waveguides are based on two InP membranes separated unprecedented measurement resolution. by 220-nm-thick InGaAsP (i-Q1.58) sacrificial layer, which is etched The transducer described in a Nature Communications paper in the sensing region to suspend the top membrane, while it is kept titled “Integrated nano-optomechanical displacement sensor and used as absorbing layer in the photodetector section. The with ultrawide optical bandwidth” consists of four evanescently- top, 330-nm-thick InP membrane is p-doped, with a 20-nm-thick coupled waveguides, with two waveguides suspended above two p-InGaAs contact layer, while the bottom membrane consists of a output waveguides. top 50-nm-thick n-doped InP layer and a 220-nm-thick undoped The structure is designed in such a way that, before displace- layer. The corresponding p-i-n junction is used to electrostati- ment, light coming from one input waveguide excites a superposi- cally actuate the inter-membrane distance in the sensing section, tion of symmetric and anti-symmetric supermodes which, after and forms the photodiode in the detection section. This set up traveling for a beating length in the directional coupler, interferes was demonstrated to yield a resolution of 45 femtometers (about constructively at the “cross” output port. A displacement of one 1/1000 the size of the smallest atom) in a measurement time of a suspended waveguide changes the propagation constants of the fraction of a second. supermodes and makes the interference destructive, resulting in One of the key advantages of this sensor is that it operates in increased transmission from the other waveguide. These change a large range of wavelengths, which eliminates the need for an in the relative transmission from the two output waveguides result expensive laser on the device. In terms of cantilever deflection, the from a combination of vertical and horizontal evanescent coupling, sensor also replicates the resolution of cantilevers in traditional, but the authors explain. bulky AFMs. Gas sensor breakthrough could miniaturize analytics By Peter Clarke esearchers at General Electric (Niskayuna, NY) have tion. The results have been reported in Nature Electronics and found a way to significantly improve gas sensing using the paper was featured as the cover story of the May edition. semiconducting metal oxides. The authors stated that dielectric excitation approach could RConventional metal-oxide gas sensors have non-linear be applied to a variety of n- and p-type SMOX materials for ap- responses and are subject to signal fluctuations and gas cross- plication as gas sensors within wearable monitors, autonomous sensitivities. These chemiresistors are suitable for general robotics, home health and sensor networks. detection they do not provide precision. The researchers used Within the paper the authors reported on wireless sensor a dielectric excitation technique to achieve SMOX- nodes based on the technology employed for based gas sensor with a linear response, an enhanced unattended, drone-based and wearable environ- dynamic range and significantly reduced humidity and mental and industrial gas monitoring. ambient temperature effects. “We have found that by providing electrical The SMOX material was used with conventional n- excitation of conventional sensing materials in type tin-oxide. It was found that the imaginary part of very specific and targeted ways, we can achieve the a.c. impedance at a certain frequency range pro- highly desired sensor performance characteris- vided a linear sensor response over a large range of tics, unavailable using conventional techniques,” gas concentrations. The technique was tested against a broad said Radislav Potyrailo, a GE Principal Scientist and the lead range of volatile gases including: benzene, toluene, hydrogen author of the Nature Electronics article. sulphide, hydrogen, carbon monoxide, methane, ethane, pro- The team of researchers will carry on working to better pane, acetylene, methanol, ethanol, acetone and formaldehyde. understand the full potential of the dielectric excitation sensor As a result, the researchers were able to produce a small-die design as well as exploring commercialization opportunities gas sensor that competes with desktop analytical instrumenta- with prospective partners for diverse markets.

42 News June 2020 @eeNewsEurope www.eenewseurope.com SENSORS & TRANSDUCERS

IMEC partners with Roswell on biosensor chips By Peter Clarke oswell Biotechnologies and European research insti- consumer surveillance and diagnostics tools has been made tute IMEC have teamed up to develop extremely clear in the current Covid-19 pan- biosensor chips for infectious disease demic,” said Paul Mola, CEO of Roswell, in a surveillanceR and DNA storage. statement. “In that area, the Roswell molecu- Roswell Biotechnologies Inc. (San Diego, lar electronic platform will transform the way Calif.) was founded in 2014 and has used infectious diseases are detected, with powerful advances in semiconductor technology to new capabilities that enable, rapid screening of integrate sensor molecules into the CMOS in- many infectious diseases at once, or many viral tegrated circuits. Roswell’s firt chip is designed strains, with portable or handheld devices.” to read DNA; with future chips being planned The Roswell platform will be able to support for protein detection and DNA storage. tests for the detection and containment of However, while the first chips are break- infectious diseases, such as Covid-19, includ- through developments higher volume produc- ing sequencing, nucleic acid detection, antigen tion requires customization of in-situ equipment and/or the detection and antibody detection. The technology could also refinement of manufacturing processes. IMEC and Roswell have have more direct application in the information technology as completed the proof-of-concept work and are now focused on molecular storage of data is a potential application. final process development. The initial products are expected to “One of the significant hurdles to commercializing molecu- be commercially available in 2021. lar electronics is the need for costly customized solutions for Roswell’s first chip is designed to read DNA; future chips will large scale manufacturing,” said Barry Merriman, chief scientific be designed for protein detection and other diverse bio-sensing officer at Roswell. “IMEC has overcome those challenges by applications. The DNA reading chip is at the core of Roswell utilizing state-of-the-art semiconductor manufacturing technol- Technologies’ platform for DNA sequencing, which could then ogy coupled with its deep experience in biosensor technology be used for other precision medical applications. to commercialize molecular electronics using standard tools. “The urgent need for a new generation of rapid, low-cost, We are excited to be partnering with IMEC on this effort.” Intelligent road-side radar detects wild animals By Julien Happich he Universities of Applied Sciences of Ulm and Heilbronn The technical demonstrator system will be able to measure along with industrial partners have created “SALUS”, a simultaneous data from three sources – radar, optical cam- machine-learning system designed to be era and infrared camera. Additional sensors ableT to differentiate between pedestrians, cars, could be integrated, for example, to measure bicyclists, motorbikes, deer, foxes, wild boar pollution levels, which has already generated etc. predicting the behaviour of these objects. commercial interest from companies in the Combining a radar, optical cameras and Project’s consortium. The project envisages infrared sensors with neural networks, the wide-scale deployment of self-contained units system sends warnings to car drivers and as posts beside roads across Germany, which other road users to prevent accidents. The means that they must be inexpensive and solar data of a micro-Doppler radar is gathered by a powered. The latter is particularly important for Spectrum Instrumentation PCIe digitizer card M2p.5926-x4 that rural areas where mains electricity is not readily available and provides the required number of channels and bitwidth. the need for such a warning system is greatest as street lighting “Car manufacturers are installing driver assistance sys- is scarce. tems starting with high end models but this This also means that the communication will take time to filter down to other models. system between units to form an intelligent Similarly, high end motorcycles are starting to transport infrastructure must be low power so have such functionality but limited installation Long Range Wide Area Network (LoRaWAN) is space will mean that it will be very difficult to used as it can reach up to 40 km in rural areas. achieve anything comparable to a car-based “We are using neural networks to develop alert system. The aim of our project is to have the machine learning that enables the system small installations by the roadside that detect to differentiate between e.g. bicyclists, cars hazards and communicate this to vehicles as or deer. This takes it far beyond pure motion they approach. In addition, for road users with- detection,” added Professor Mantz. “We are at out in-car warning displays, road lights could the critical part of the project which is the clas- be turned on to highlight the area of the hazard and/or warning sification of detected objects, which has never been done be- notices illuminated. Project SALUS detects and warns of the fore. With this, the system will be able to predict the movements harder-to-see hazards and should significantly improve road of objects which adds incredibly useful, real-time intelligence safety”, said Professor Dr. Hubert Mantz from the Ulm Univer- to the system enabling it to foresee how a hazardous situation sity of Applied Sciences, one of the leaders in the project. could develop.” www.eenewseurope.com eeNews Europe June 2020 News 43 DESIGN & PRODUCTS SENSORS & TRANSDUCERS

Vibrotactile display team wins 2020 MEMS design contest By Julien Happich -Fab, along with technology partner Europractice have ume production. The MPW fabrication run will commence in late announced the winning project in their competition to May. encourage further MEMS-based innovation. “Our design comprises a 3 degrees-of-freedom tactile sens- XThe triumphant project, titled “Capacitive MEMS Sensors for ing mechanism that relies on MEMS-based operation. Incor- High-Resolution Interactive Vibrotactile Displays”, was submit- poration of the X-Fab/Europractice fabricated sensor ICs will ted by a team of engineers from the University of Bath. Entrants add a whole new dimension to our prototype. It will mean that from the University of Malta and Delft University of Technology unique features which have not been possible using current (DUT) came second and third respectively. As a reward, the tactile displays can now be fully explored, and major benefits University of Bath project will now be included derived”, said Dr Ali Mohammadi, the Direc- in a multi-project wafer (MPW) run that utilizes tor of Microsystems Research Laboratory at X-Fab’s XMB10 MEMS process (which is University of Bath. highly optimized for multi-axis sensor imple- “Among the primary criteria for entries in mentations). this competition was that they should be of real Vibrotactile displays have been proposed value to our society, and the winning project as an assistive technology via which it would clearly does this to a very high degree, as well be possible for people with visual impairments as having a great deal of commercial viability to perceive graphical information using touch. The members too” said Volker Herbig, VP of X-Fab’s MEMS Business Unit. of the University of Bath team, who are based in the Faculty of “Once again this MEMS design contest has been successful in Engineering & Design, have been focused on this project for highlighting the wealth of engineering talent that Europe has to the last 18 months. Working in conjunction with several other offer, both in industry and academia.” departments within the university (Computer Science, Psychol- “Our hope is that through this contest, dissemination of ogy and Education), they have been able to develop a high- the associated project results can stimulate the wider uptake resolution vibrotactile display prototype. of MEMS by European academic institutions and companies, With access to a pilot run, which will be undertaken at X- and prove the inherent versatility of the XMB10 MEMS pro- Fab’s Erfurt production facility, they will now gain access to cess offered within Europractice portfolio,” added Dr Romano cutting-edge sensor IC devices that will allow them to further Hoofman, Program Director at imec and General Manager of enhance their design concept, and take it closer to future vol- Europractice. X-ray TDI camera boasts enhanced sensitivity to low energy X-rays By Julien Happich

amamatsu Photonics has developed an X-ray Time design from scratch, Hamamatsu Photonics is now able to cap- Delay Integration (TDI) camera with ture high-contrast images from low-energy enhanced sensitivity to low energy X-rays that do not easily pass through thin X-rays.H Model C15400-30-50A gives sharp objects and light-element materials, improv- high-contrast X-ray images of thin objects ing inspection accuracy in detecting tiny such as thin aluminum sheets and light- contaminants such as hair in thin aluminum element materials including carbon fiber food packing, irregular thickness of alumi- reinforced plastic (CFRP), and is expected num sheets, or uneven adhesive coating on to improve accuracy for in-line non-destruc- transportation equipment parts using CFRP. tive inspection of foodstuffs and industrial An image captured Moreover, employing a high-performance products. The X-ray TDI camera captures with the X-ray TDI camera photodetector developed in-house, the X-ray images with high contrast according company has boosted the signal readout to the difference in the amount of X-ray transmission. Areas speed to 1.8 times the ordinary speed to allow streamlined with high X-ray transmission are bright and areas with low X-ray in-line 100% inspection of products to about 154 meters per transmission are dark. Thick materials and minute. Another feature is that this X-ray substances with a high atomic number tend TDI camera also captures high-energy to block X-rays. If they are present in an X-ray images and so can simultaneously object inspected with X-rays, the captured inspect for contaminants in objects con- X-ray image will have a contrast accord- taining multiple materials. High-contrast ing to differences in their thickness and X-ray images can captured at an energy materials, allowing accurate detection of of about 20 keV, which is lower than the contaminants that may get into foodstuffs or energy level of 40 keV used for ordinary industrial products, preventing the intrusion X-ray inspection. The 450×1000×210mm of stones or metals The camera could also unit operates from +24V DC and has a X-ray TDI camera C15400-30-50A be used to detect poor package sealing and pixel resolution of 4096×128 (horizontal x prevent product deterioration. By utilizing a new technique to vertical), it supports a detection width of 300mm. detect low-energy X-rays with high sensitivity and making the

44 News June 2020 @eeNewsEurope www.eenewseurope.com SENSORS & TRANSDUCERS

SkyWater to make temperature sensor ToF sensor enables multi-object ranging for Covid-19 patch from 2.5cm to 3m Specialist foundry SkyWater Technology Inc. (Bloomington, STMicroelectronics has extended the capabilities of its Flight- Minnesota) is going into production with a temperature sensor Sense ToF ranging sensors with the VL53L3CX, leveraging chip for use in a smartphone- patented histogram enabled patch. The patch can algorithms that allow be used to detect early-stage measuring distances symptoms of Covid-19. Sky- to multiple objects as Water is working with developer well as increasing ac- Linear ASICs Inc. (Tallmadge, curacy. The VL53L3CX Ohio), a fabless analog and measures object mixed-signal semiconductor ranges from 2.5cm to company and investment firm Asymmetric Return Capital. The 3m, unaffected by the patch is worn like a band-aid, paired wirelessly to a mobile target color or reflectance, unlike conventional infrared sen- device such as a smartphone where it records body tempera- sors. This allows designers to introduce powerful new features ture. The health monitor is being developed through additional to their products, such as enabling occupancy detectors to partnerships with SensiML, a subsidiary of QuickLogic and provide error-free sensing by ignoring unwanted background or Upward Health, an in-home and virtual care service provider. foreground objects, or reporting the exact distances to multiple A 130nm mixed-signal ASIC process is being used for the targets within the sensor’s field-of-view. The ST patented his- temperature sensor. “Our Technology Foundry business model togram algorithms increase cover-glass crosstalk immunity and enables us to provide volume manufacturing with IP security in allow real-time smudge compensation preventing external con- a flexible and agile environment which facilitates rapid time to tamination from adversely affecting the ranging accuracy of, for market and minimizes supply chain disruption risks by offer- example, vacuum cleaners or equipment that may be used in a ing a domestic sourcing option,” said Thomas Sonderman, dusty industrial environment. Ranging under ambient lighting is SkyWater president, in a statement. also improved. In addition, the VL53L3CX has superior linearity SkyWater Technology Inc. that increases short-distance measurement accuracy enhanc- www.skywatertechnology.com ing wall tracking, faster cliff detection, and obstacle avoidance in equipment such as service robots and vacuum cleaners. Like all FlightSense sensors, the VL53L3CX features a compact, all- Dual angle magnetic sensor demo board in-one package design that eases integration. STMicroelectronics The PVL3M development board provides an XMR sensor (AMR www.st.com angle sensor AA746C from Sensitec) in combination with the multiturn Hall sensor iC-PVL from iC-Haus for on-axis absolute angle measurement and TOF modules and imager chips revolution counting. Evaluation of the two sen- for non-contact sensors sors is performed by the With TOF modules and the Espros epc6xy imager chips, intel- new 26-bit encoder signal ligent, non-contact sensors can be implemented for a wide processor iC-TW29 with variety of integrated EEPROM. The applications. sine/cosine signals from The spread the XMR sensor are proof bacteria cessed in the chip, interpolated, and output with an arbitrarily and viruses is selectable single-turn resolution. The iC-PVL provides battery- largely done backed multiturn position to the iC-TW29 via its Absolute Data by hands. Interface (ADI). This combination can be used to make a mag- The danger- netic multiturn serial absolute encoder. For example, a BiSS ous patho- encoder with an angular resolution of 17 bits and 24 revolution gens stick to every- day things such as door handles, light counting bits is possible. Signal processing time is only 5 µs, switches and toilet flushes. Handles and buttons on a wide va- which is very close to that of optical encoders. Simultaneously, riety of devices and installations thus become virus spreaders! Auto Calibration in the iC-TW29 corrects the sine/cosine signal The perfect solution for public transport, for doctor’s surgeries, errors ensuring high measurement accuracy over temperature hospitals, restaurants or even the food processing industry of- or with axial play of the motor shaft. Position output via SPI fer touch-free buttons and switches. Even if the hands are busy or incremental quadrature signals with index at any arbitrary carrying objects, approaching the switch is enough to trigger resolution (FlexCount) from 1 to 32,768 CPR is also possible. a function such as opening a door. In addition to avoiding skin Requiring a single-sided supply of +3.3 V, the sensors as well contact with the switching element, various functions can even as the iC-TW29 operate over an industrial operating tempera- be simplified. While Espros does not produce disinfectants or ture range of –40 to +125°C. The iC-TW29 comes in 5x5mm protective clothing, the company can certainly help prevent or 32-pin QFN package. manage pandemics such as the current corona crisis. iC-Haus Espros www.ichaus.de www.espros.com www.eenewseurope.com eeNews Europe June 2020 News 45 DESIGN & PRODUCTS MISCELLANEOUS

Miniature programmable smart speaker Four-legged metal domes is only 24x24x17mm with better assembly tolerances The M5Stack ATOM ECHO is a tiny programmable smart The RCG series of four-legged metal domes from Snaptron speaker based on the M5ATOM series of products. The slim come with a unique ring in the center. This feature helps reduce design measures 24x24x17mm and impact from contamination, as- comes with a premium finish for a sembly tolerances, or off-centered professional look. The unit can be presses, making it the most versatile programmed to be voice-controlled, metal dome switch on the market. which means that when it is pow- The RCG series dome addresses a ered on, the program is uploaded. crucial pain point for customers, loss You can simply ask for things like of electrical contact between the music, news, information, and more. dome and board due to the buildup Input your voice through the microphone of M5Stack ATOM of debris. The “ring” feature provides ECHO, and output the text in your device using text to speech a 360-degree ring of contact, which minimizes the impact from APIs. It could also be programmed to control compatible smart contaminants. When debris obstructs the circuit in one area, devices in the home, voice-activated. The device can also be the ring still makes contact in another area. This feature on the programmed to access AWS, and other cloud platforms. The dome also ensures that electrical contact is maintained even speaker has an integrated RGB LED light (SK6812), which can with off-center or off-angle actuations. The RCG dome can visually display the connection status. As a smart speaker, activate multiple independent circuits. A fingered center pad al- it can control the ATOM series. Devices can be connected lows for two or more circuits to be contacted at the same time through the grove interface. Pins G21/G25 can be used for by multiple parts of the ring so that multiple circuits can be general purpose I/O, they do not support I2C and UART how- activated, even if the actuator makes uneven contact. The RCG ever. Screw holes on the back are convenient for users to affix dome offers a low profile, high life cycle rating, multiple custom the device in place. The factory default firmware is a bluetooth options, tactile and acoustic feedback, and a comprehensive speaker, which uses A2DP protocol to transmit audio data (call range of forces compared to other switch technologies. It reception is not supported). The firmware is compiled on the comes standard in 5.3, 6.3, 7, 8.5, 10, and 12mm sizes with ESP-IDF platform. forces ranging from 200g to 500g, depending on diameter. M5Stack Snaptron https://m5stack.com www.snaptron.com

1.2μm 50Mp dual pixel image sensor 4+16 channel MSOs for low-light photos for cost-effective test and debugging Samsung Electronics’ ISOCELL GN1 is a new 50-megapixel im- Pico Technology has launched three new 4-channel models in age sensor with large 1.2μm-sized pixels. The ISOCELL GN1 is its PicoScope 6000E Series oscilloscopes, each of which can Samsung’s first image sensor to be configured with 16 optional digital offer both Dual Pixel and Tetra- channels. With bandwidths of 300 or cell technologies. Thanks to the 500 MHz, 8-bit or 8/10/12-bit Flex- relatively larger pixel size, the GN1 ible Resolution and up to 4 GS deep brings image sensor performance capture memory, these products en- to a new level with a combina- hance the existing PicoScope 6000E tion of elevated light sensitivity for portfolio, adding to the 8-channel low-light photos and DSLR-level models that were introduced earlier auto-focus speeds, optimized for in the year. Coupled with the proven more dynamic picture-taking ex- PicoScope 6 application software, the USB units provides periences. The GN1 brings best-in-class auto-focusing with 100 advanced debugging tools to accelerate development of high- million phase detection auto-focus agents. The Dual Pixel technol- end embedded systems, signal processing, power electronics, ogy places two photodiodes side-by-side within a single pixel that mechatronics, and automotive designs. The PicoScope 6000E can receive light from different angles for phase detection. With all Series incorporates many unique features as standard, such as of the sensor’s active pixels working as auto-focusing agents, the 21 serial protocol decoder / analyzers, spectrum analysis capa- GN1 can detect and focus onto a desired still or moving object bility, a 50 MHz arbitrary waveform generator, and user-defined from every corner in an instant, even in low-light conditions. When math and alarm functions for in-depth analysis and long-term capturing images, a single pixel output is created by merging the unattended testing. Measuring only 245x192x62mm, the 6000E outputs from the two photodiodes within the pixel. Samsung also Series oscilloscopes fit easily on any workbench. A free-of- provides a software algorithm that takes light information from charge Software Development Kit (SDK) is available to down- each photodiode to produce image resolutions comparable to load that enables users to write their own applications around 100Mp. By binning four pixel signals into one, Tetracell doubles the PicoScope 6000E hardware. The SDK includes instrument the image sensor’s pixel size to 2.4μm and quadruples the light drivers for Windows, macOS and Linux, and code examples in sensitivity to take brighter 12.5Mp photographs. C, C#, C++ and Python. Samsung Pico Technology www.samsung.com www.picotech.com

46 News June 2020 @eeNewsEurope www.eenewseurope.com MISCELLANEOUS

Intelligent surveillance kit for the edge Spring-loaded pins designed

UP Xtreme, from Aaeon’s UP Board division, is being deployed for sliding and rotational connections along with technology from partners including Intel, Milestone Mill-Max Manufacturing has developed a superior interconnect and Saimos to bring Smart Surveillance to the edge with the product to address the challenges of making sliding or rotating UP Xtreme Smart Surveillance kit. Smart connections. The unique Omniball surveillance adds intelligent processing spring-loaded contact (patent pend- of data that’s already being gathered by ing) features a rolling ball interface, surveillance cameras to analyze traffic enabling mating components to with heatmaps, monitor areas with virtual engage in lateral, rotational and fences, or reacting to data in other ways angular alignments while providing to allow for automated control or alert- optimal electrical, mechanical, and ing staff on premises. UP Xtreme Smart structural reliability. Here, the traditional plunger of the spring- Surveillance works as an AI-enabled loaded pin has been replaced by a gold-plated ball. When NVR solution to proactively monitor video streams support- engaged, the ball compresses and rolls, allowing the mating sur- ing up to 32 cameras simultaneously (Intel Core i7 model with faces to make contact and then easily slide parallel to each other Milestone XProtect Express+). The kit integrates video man- while spring force acts to ensure consistent electrical contact is agement software (VMS) from Milestone with video analytic maintained. This rolling action alleviates the concerns of connec- software from Saimos, then adding a deep learning AI edge tor damage such as binding, premature wearing and structural inference through the Intel Distribution of OpenVINO toolkit, failure that may occur when using traditional plunger style spring to provide a total package solution that saves deployment pins in these types of applications. Features of the Omniball time and cost for end users, as well as overcoming challenges contact include a 0.762mm maximum stroke; an overall height such as storage issues. The AI inference is also accelerated of 6.76mm; gold plating on all components and surface mount with two Intel Movidius Myriad X VPUs, helping provide even termination. These spring-loaded pins are durable, they have more power alongside the 8th Generation Intel Core proces- been tested to 1,000,000 compression and rolling cycles at half sor that comes with the UP Xtreme system. Once UP Xtreme stroke while still meeting specifications for contact resistance of Smart Surveillance is installed, the platform is easy to use and 20 mili-ohms max., current rating of 3.5 A and spring force of 55 provides intuitive analytic parameter settings for users to have grams at mid stroke (0.381mm). a comprehensive overview of the entire surveillance installation. Mill-Max Manufacturing UP Xtreme Smart Surveillance provides key functions to sup- www.mill-max.com port applications such as people/object counting, heatmap- ping, virtual fence and perimeter protection, dynamic blurring and object detection. The system uses data analytics and IoT rapid prototyping: 10mn edge computing to provide real time analysis, and can provide proactive measures through alerting relevant staff as soon as from sensor to dashboard an incident is detected. Seluxit has released IoT rapid prototyping kits based on it’s sin- Aaeon gle-board computer, SLXPorcupine. If you have a smartphone, www.aaeon.com an internet connection and Seluxit’s IoT rapid-prototyping solution, then 10 minutes is all you need to create IoT prototypes that Dual silent switcher feel at home even in industrial settings. Claiming your device is as seamless as delivers low EMI operation scanning a QR code, and adding plug-and- The LT8650S, LT8652S and LT8653S dual-channel 4A/8.5A/2A play sensors can now be done by using a synchronous step-down Silent Switcher converters launched wizard and hitting a “deploy” button. SLX by Analog Device are aimed at automotive, Porcupine is an industrial-grade, single- communications and solid state drive pow- board computer with a Raspberry Pi com- er supplies. The proprietary Silent Switcher patible form factor,running Linux. That makes it possible to use 2 architecture along with spread spec- hats and open-source code from the Raspberry Picommunity. trum frequency modulation ensures PCB SLX Porcupine automatically connects to Viasens by Seluxit; layout robustness to achieve ultra-low EMI a cloud-based, secure IoT Infrastructure.Viasens has secure performance that easily meets CISPR25 connectivity, over-the-air firmware update, and a configurable emissions. The devices are also differenti- dashboard where you can see and analyse your data. Viasens ated with their stackable-friendly outputs also features data logging and extraction, reference code for for higher current applications. Outputs can be paralleled to your own iOS and Android native applications, as well as web scale up or down the output current based on the system need interfaces. The SBC features a 4GB on-board flash and 512 while optimising thermal performance. This also helps to mini- MB RAM. It also has 3 additional 20-pin headers, featuring mise inventory and qualification time with a single device. For functionalities such as Modbus, CAN bus and up to 8 channels example, the LT8652S provides 8.5A DC from each channel of 12 bit ADCs. It can handle the temperature range often re- simultaneously or up to 12A on either channel. Synchronising quired in an industrial setting, from -40 to +85°C. Secure boot two LT8652S chips can achieve a 4-phase, 34A supply. effectively prevents tampering on the device. Analog Devices Seluxit www.analog.com www.seluxit.com www.eenewseurope.com eeNews Europe June 2020 News 47 DISTRIBUTION CORNER

Multi-sensor IoT modules Farnell launches Raspberry Pi camera

deliver intelligence at the edge Farnell has released a new Raspberry Pi high quality cam- Arrow Electronics, Panasonic Industry, and STMicroelectronics era, complete with a 12-megapixel ultra-definition-resolution have introduced a low-power wireless multi-sensor edge-in- camera and interchangeable lenses. The Raspberry Pi High telligence solution for Quality Camera is the first-of-its- smart factory, smart kind to hit the market and is ideal home, and smart life for professional and consumer applications. The applications which require the IoT Solution Module highest levels of visual fidelity and combines Arrow’s integration with specialist optics. engineering and global The new camera can effectively distribution capa- capture still images and video bilities with Panasonic footage. It is compatible with all Industry’s IoT modules based on the ST BlueTile (STEVAL- models of Raspberry Pi boards from the Raspberry Pi 1 Model BCN002V1B) multi-sensor development kit. This combination B onwards and can be used as a desktop camera. The unit is enables customers to test their ideas easily and bring new IoT suitable for a wide range of professional applications including products to the market faster. machine vision, robotics, industrial and agriculture. Arrow Electronics Farnell fiveyearsout.com www.farnell.com

Online calculators help speed electronic People counter maker project helps design process with social distancing Mouser Electronics’ free Technical Resource Center now of- RS Components has joined forces with miniature PLC manu- fers an extensive lineup of online calculators designed to help facturer Barth Elektronik GmbH to develop a simple maker engineers project that aids social of all skill distancing in the effort levels save to prevent the spread of time figuring COVID-19. The project, out anything called ‘People Counter’, from power can be assembled in less conversion than an hour and records computations the number of people en- to resistor color codes. “Part of delivering exemplary service tering and exiting a room, to our customers means providing them with the tools and in- indicating when it is safe to allow more people to enter while formation they need to order and design with latest products,” still maintaining the recommended distance from others. This says Kevin Hess, Mouser Electronics’ Senior Vice President of is particularly effective in retail environments where safe social Marketing. distancing can be difficult to control. Mouser Electronics RS Components www.mouser.com www.rs-online.com

TACT switches provide joystick functionality Anglia stocks Bivar LED light pipes

Offering a flexible solution to HMI control functionality the MEC and indicators Navimec switch concept now available at Foremost Electronics Anglia announced a considerable expansion in its range of LED offers product designers a circular soft-touch TACT switch with optics and PCB accessories by signing of a distribution agree- joystick functionality. Navimec switches ment with Bivar Inc., cover- make it possible to navigate up, down, ing its light pipes, indicators left and right with the provision of a cen- and other components. tre function that will typically be used as “Bivar’s LED offering is highly an “enter” key. Navimec switches are complementary to our exist- available with or without illumination and ing range of LED indication the non-illuminated versions may be solutions. We are particu- over-printed with appropriate legends. larly excited to introduce the The five position navigational switch is ideal for data entry and range of LED Light Pipes, motion control devices where a joystick function is needed. which is a new range from Anglia and expands the range of The switch is available as a complete module or may be sup- devices we offer that fit around the panels and enclosures plied as single parts for custom PCB mounting. housing our customers’ PCBs. Foremost Anglia www.4most.co.uk www.anglia-live.com

48 News June 2020 @eeNewsEurope www.eenewseurope.com LAST WORD How to fight a virus: Lessons from cybersecurity By Yotam Gutman

here has been a great deal of conversation around the similarities between the spread of the Covid-19 virus and that of computer viruses. And indeed, as the first global pandemicT to occur during the age of connectivity, this comparison is valid. But while most focus on how we can leverage the knowledge gained in the “real world” in identifying and stopping the spread of plagues in the virtual world, I would like to offer another perspective.

Perhaps we in cybersecurity can return the favor. Perhaps the medical world can take the lessons learned in three decades of fighting “cyber viruses” and implement these in their fight to mitigate the Coronavirus? History Originally, the type of computer software described as “a program that can infect other programs by modifying them to include a, possibly evolved, version of itself” was named “Virus” by Fred Cohen in his 1986 Ph.D. thesis. Another biological reference made its way into the computer lingo when the first worm was unleashed (although the phrase was used in an earlier sci-fi novel).

In the last couple of years, computer viruses, or more widely Adopting cybersecurity response the panoply of malware as we think of cybersecurity today, have to fight Covid-19 undergone rapid evolution that has made them much more dif- To mitigate today’s plethora of rapidly evolving cyber threats, ficult to identify and mitigate: the cybersecurity industry has developed several methodologies. These (after adaptation) could be used to reduce the • More variants: 439,000 new malware variants were detected spread of malicious software and to mitigate its effects. I will in 2019. That’s a 12.3% increase over the previous year. refrain from discussing the obvious virus/Anti-virus analogy. • More capable: Modern malware threats are far more capable Obviously, a vaccine for a computer “virus” would be the an- than the old viruses spreading through illegal copies of soft- swer, but estimates suggest that such a vaccine would not be ware distributed via floppy-disks. Today’s malware can steal available in the next 12-18 months, and there’s a lot we can do passwords, exfiltrate sensitive data, encrypt and delete data, until then: and much more. • Zero trust policy- A methodology that defies the traditional • Harder to detect: Malware authors work hard to make their security assumption that everything inside the perimeter (pro- software difficult to detect. This includes hiding it in legitimate tected by the firewall) is trusted. The main principle of Zero documents (aka “weaponizing” Word, PDF and Excel docu- Trus is “never trust, always verify”. This means that every ments), utilizing detection-evasion mechanisms (like avoiding user is asked to verify their credentials every time they wish execution in sandboxed environments), and using legitimate to “enter” the organization and that every file and process are software update mechanisms, all to make the work of the being constantly monitored – even if they have been “autho- defenders harder. rized” to run on the computer. • More aggressive: Some malware types are extremely aggres- In a similar manner, humans should consider that other humans sive; they scan for open RDP ports, brute-force their way are carriers, and only “trust” them after they have been tested onto a device, and then move laterally within the organiza- negative (or at the minimum, have had their temperature taken). tion’s network, abusing password-protected servers and • Detection beats prevention: following a similar line of thought, seeking sensitive data, all without the knowledge of the most organizations today operate under the “Assume a victim. Breach” paradigm. Instead of striving to identify and mitigate • Fast: contemporary malware is extremely fast and works 100% of threats 100% of the time, they assume that some at machine-speed to bypass protection mechanisms and threats would be able to infect them and concentrate their ef- achieve its goals—ransomware like “Wannacry” disabled forts on quickly finding these and stopping them before they entire organizations in minutes. could do more harm.

Yotam Gutman is Director of Marketing at SentinelOne - Similarly, it is prudent to assume that humanity would not be www.sentinelone.com able to vanquish this virus, and we will be playing “whack-a- www.eenewseurope.com eeNews Europe June 2020 News 49 News LAST WORD

PUBLISHER mole” with it for the foreseeable time. what to expect. Security professionals, André Rousselot Given that this is the case, it’s prudent governments, and those in the secu- +32 27400053 to invest in rapid detection of the infec- rity industry have been formally and [email protected] tion (quick detection kits, even home informally sharing information about EDITOR-IN-CHIEF detection kits), ensure those that are malware, cybercrime groups, and data Julien Happich sick are given quick treatment, and leaks for a long time. This has proved +33 0534655306 continue to monitor the entire popula- to be immensely helpful in fighting and [email protected] tion for outbreaks. defeating cybercrime rings. • Segmentation; an important principle EDITORS that limits the “movement” within the Such collaboration should also be Christoph Hammerschmidt organization, so that intruders cannot adopted by global scientific, medi- +49 8944450209 move freely and infect other parts of cal communities, governments, and [email protected] the organization. healthcare organizations. As this threat Peter Clarke The real-life manifestation would be is new to humanity, we should all share +44 776 786 55 93 to identify infection “hot-spots”, lock information about detection and treat- [email protected] these down and then tend to these in- ment mechanisms, and notify others Nick Flaherty fected rather than to lock-down entire when we think we’ve made break- +44 7710 236368 countries. throughs in finding a cure or a vaccine. [email protected] • Risk modeling: it might be possible, Jean-Pierre Joosting perhaps, to provide 100% security, Conclusion +44 7800548133 100% of the time, but the cost to the We can debate the similarities between [email protected] organization would be detrimental; biological and computer “Virus” (which, Alistair Winning either the security costs would be some believe, more resembles a Bacteria +44 7834 770342 through the roof, or the security restric- than a virus), but the analogy is, for the [email protected] tions imposed to maintain 100% secu- most part, correct. Viruses are dangerous CIRCULATION & FINANCE rity would cause the business to stand to the victims, and they spread quickly Luc Desimpel still. Instead, a CISO conducts risk through the population until a cure, or a [email protected] assessments and prioritizes security vaccine is found. spending to mitigate the most acute The spread of the Coronavirus ADVERTISING PRODUCTION & threats and secure the most valuable pandemic and its impact on our lives is REPRINTS assets. nothing like the world has seen before. Lydia Gijsegom Healthcare officials should do the It spread almost at machine speed and [email protected] same and ensure that the most sensi- overwhelmed countries and healthcare ART MANAGER tive segments of the population (el- organizations. We believe that utilizing Jean-Paul Speliers derly, sick) are being shielded from the the lessons learned by the cybersecurity disease and if need be, are provided industry in the past 3 decades could help ACCOUNTING with better care. to thwart the Coronavirus pandemic. Ricardo Pinto Ferreira • Intelligence intake: fighting a stealthy enemy is hard because you don’t know REGIONAL ADVERTISING REPRESENTATIVES Contact information at: http://www.eenewseurope.com/content/ SkyWater to make temperature sensor sales-contacts chip for COVID-19 patch eeNews Europe is published 10 times in 2020 By Peter Clarke by European Business Press SA pecialist foundry monitor is being developed Chaussée de Louvain 533, SkyWater Tech- through additional partner- 1380 Lasne, Belgium nology is going ships with SensiML, a sub- Tel: +32-2-740 00 50 european intoS production with a sidiary of QuickLogic and Up- business press Fax: +32-2-740 00 59 temperature sensor chip ward Health, an in-home and electronics europe.net email: info@ for use in a smartphone- virtual care service provider. www.eenewseurope.com enabled patch. The patch A 130nm mixed-signal ASIC VAT Registration: BE 461.357.437. can be used to detect process is being used for the Company Number: 0461357437 early-stage symptoms of Covid-19. Sky- temperature sensor. “Our Technology RPM: Nivelles. Water is working with developer Linear Foundry business model enables us to Volume 22, Issue 6 eeNews P 304128 It is free to qualified engineers and managers ASICs Inc. (Tallmadge, Ohio), a fabless provide volume manufacturing with IP se- analog and mixed-signal semiconductor curity in a flexible and agile environment involved in engineering decisions – see: http://www.eenewseurope.com/subscribe company and investment firm Asymmet- which facilitates rapid time to market and ric Return Capital. The patch is worn like minimizes supply chain disruption risks © 2020 E.B.P. SA a band-aid, paired wirelessly to a mobile by offering a domestic sourcing option,” All rights reserved. P 304128 device such as a smartphone where it said Thomas Sonderman, SkyWater records body temperature. The health president, in a statement.

50 News June 2020 @eeNewsEurope www.eenewseurope.com News

News News MW Embedded RF - Microwave

european business press News LED LIGHTINGN ews& DISPLAYS News News

House Ad eeNews Automotive 2017.indd 1 2/04/19 12:06 POWER NMANAGEMENTews

www.eenewspower.com

News News MW Embedded EUROPE RF - Microwave

european business press ANALOGNews LED LIGHTINGN ews& DISPLAYS POWER NMANAGEMENTews TEST & MEASUREMENTNews

House Ad eeNews Power 2018.indd 1 10/02/20 16:12