Bits&Chips 1 | 28 February 2020 | Machine learning by Bits&Chips

28 FEBRUARY 2020 | 1 MAY 2020

Solaytec hedges its bets with SALD spinoff

BITS-CHIPS.NL

ASML feels the squeeze from US-China standoff

QUTECH READIES THE QUANTUM INDUSTRY FOR TAKEOFF

13 MAY 2020 IGLUU EINDHOVEN

PLATINUM SPONSOR

IDEA2INDUSTRY.COM

#BCI2I

pinion EDITORIAL Paul van Gerven is an editor at Bits&Chips.

Throwing money

he other day, Kees van der Lede, former CEO of Akzonobel and figurehead of Dutch business and industry association VNO, urged the right-of-center political establishment to be courageous. Governments backing off and leaving things to businesses and markets, he contends, just doesn’t cut it anymore. Hard-working people, stooped in job insecurity, barely reaping the benefits of economic growth, businesses failing to reinvest profits, monopolies cornering markets and many more wrongs – something needs to be done. Just as the left ditched dogma, in the 80s and 90s, to help usher in an era of prosperity, now it’s the turn of the liberal victors to move to the political center, argues Van der Lede. Van der Lede’s analysis reminded me of some reading I’ve been doing recently. Economist Mariana Mazzucato, too, has been talking up the state, but in a specific policy area: innovation. Her core argument is that primarily the state drives progress, not pioneers pushing through the thicket of government regulation and taxation as is commonly held today. Take the Iphone, says Mazzucato. Everything that made this phone revolutionary – features like GPS, the touchscreen and Siri – has a long history of government investment, she revealed. Only when technologies had matured and most of the risk had worn off, Apple swooped in and put them together in an – admittedly – very smart product. So why does Steve Jobs get all the credit? Surely, the government deserves some, if not most. It rarely gets any these days. In fact, governments are much more likely to be vilified. Good things are often considered to happen despite,

rather than thanks to, governments. This holds particularly true in Silicon Valley, which ironically owes its success to massive public spending, according to a study by historian Margaret O’Mara. Once Silicon Valley was able to carry its own weight and then some, it turned sour on governments. Only to scurry back to Washington when some foreign threat presented itself, like Japan in the

Risk is socialized and reward is privatized eighties (and, perhaps, China today). It’s all very reminiscent of banks: risk is socialized and reward is privatized. Nonetheless, it’s obvious that the public investments that led to the Iphone and Silicon Valley have paid off handsomely in the long term. Though, they would have done so even more if big tech wasn’t so averse to paying taxes. Silicon Valley is, therefore, not a triumph of the free market but of public-private collaboration. What got it where it is today wasn’t a government-led effort, nor was it the pioneering companies that didn’t need or care about government support. It was a bit of both. This insight neatly fits in with Van der Lede’s plea to reconsider and re-appreciate the role of governments. It’s also an excellent source of inspiration for revamping Dutch innovation policy, which since the introduction of the Topsectoren also primarily has been based on free-market mythology. Encouragingly, though, one can already see some

of the above insight shine through in recent policy changes, such as the emphasis on key technologies. Even if we might be on the right track already, however, there’s still going to be a huge obstacle: the money. Getting the Dutch government to invest in education and innovation has been like squeezing blood from a stone lately, and that doesn’t bode well: public largesse was a crucial element in the success of building Silicon Valley. As the Center for Strategic & International Studies argued recently, throwing money at innovation has become vastly underrated, even if that involves waste in the short term. “We should shift the debate away from whether to throw money at the problem and toward how best to foster the innovation and return on investment that federal money can enable. The crucial element of successful American money-throwing is its decentralized research and innovation model, in which the government provides competition-enhancing regulation, long-term investment and a market to acquire and use the emerging technologies,” wrote the CSIS. With the world in desperate need of green technology and Europe increasingly getting squeezed between the US and China, let’s take that to heart.

CONTENTS IN THIS ISSUE OF BITS&CHIPS

News

ASML: if we can’t ship it to China, we’ll ship it somewhere else

Solaytec looking to spread its wings with spinoff SALD

ASML is rather indifferent on the issue whether it should be allowed to sell EUV scanners to China.

ALD tool manufacturer Solaytec is starting a sister company to tap new applications for its technology.

11 News 7 8 9 10 11 14 17 52 54

ASML’s multi-e-beam metrology tool delayed

Noise ASML: if we can’t ship it to China, we’ll ship it somewhere else ASML is too convenient a target for the US EUV profitability moving towards DUV level ASML’s multi-e-beam metrology tool delayed Seecubic doubles down on glasses-free 3D in Eindhoven Solaytec looking to spread its wings with spinoff SALD Eindhoven startup Maxwaves beams 5G to the max Lightyears away from production

Doctors embrace AI: computer calculates best radiation treatment

Opinion 3 13 30 41 45 51

Throwing money – Paul van Gerven The headhunter – Anton van Rossum My algorithm is better than yours – Marco Jacobs Moving beyond standard AI solutions – Peter de With Digitalization: act now, act fast – Robert Howe Are we prepared for disruptive times? – Biba Visnjicki

Background

18 Qutech: building a quantum delta, one (qu)bit at a time 26 Ampleon: riding the 5G wave with a smart factory 46 Coaching in the third wave of Agile

2020

EVENT CALENDAR

11 MARCH, ’S-HERTOGENBOSCH

BITS&CHIPS

INDUSTRIAL 5G CONFERENCE

8 APRIL, EINDHOVEN

Background

Qutech: building a quantum delta, one (qu)bit at a time

13 MAY, EINDHOVEN

The Qutech initiative is leading the effort to bootstrap a quantum industry in the Dutch delta. 17 JUNE, ’S-HERTOGENBOSCH

Digitalization: act now, act fast

Theme Artiﬁcial intelligence

28 High tech prepares for AI 31 Doctors embrace AI: computer calculates best radiation treatment 34 Mastering machine learning with tunable capabilities for eliminating overﬁtting 38 Lack of funding leaves Dutch AI lagging

7 OCTOBER, EINDHOVEN

Interview

28 OCTOBER, EINDHOVEN

42 90 years of sensing and control – and now machine learning BITS&CHIPS

BENELUX R F CONFERENCE

2 DECEMBER, NIJMEGEN

bits-chips.nl/events

TESTEN EN METEN D e s k u n d i g e p a r t n e r s . M e r k l e i d e r s . O n e i n d i g ve e l ke u z e .

nl.rs-online.com

NEWS

NOISE 5G

Top-10 buyers of silicon, in million dollars Rank 2019 1 2 3 4 5 6 7 8 9 10

Rank 2018 Company 2 Apple 1 Samsung 3 Huawei 4 Dell 5 Lenovo 6 BBK Electronics 7 HP 10 Xiaomi 9 HPE 11 Hon Hai Others Total

2019 spending 36,130 33,405 20,804 16,257 16,053 12,654 10,428 7,016 6,215 6,116 253,224 418,302

Ericsson and Nokia actually could use a little help

2019/2018 (%) -12.7 -21.4 -1.8 -15.0 -9.2 -8.8 -9.0 1.4 -14.6 -7.1 -11.7 -11.9 Source: IC Insights

Apple reclaimed first position in the ranking of biggest buyers of semiconductors, says Gartner. Spending over 36 billion dollars last year, Apple bought 8.6 percent of the market. Samsung, which ranked first previously, is responsible for 8 percent. The switch at the top is mainly due to Apple’s success in wearable products, such as the Apple Watch and Airpod headphones, Gartner points out. Despite the US-China trade war, Huawei’s IC spending didn’t change much, allowing the company to retain third place. Overall, semiconductor spending was down significantly, however. This was due to said trade war, Brexit, a trade conflict between Japan and South Korea and the protests in Hong Kong. PvG

America’s attorney-general William Barr floated a rather peculiar proposal recently: the US should scoop up the European telecom equipment companies Ericsson and Nokia to help them – and the US – fend off Huawei. The White House quickly dismissed it, but Barr’s suggestion highlights an interesting point about the position of the

European duo in the 5G wars: they might indeed need US backing to stay afloat. Huawei already has a pretty commanding lead, enjoying the advantage of a huge domestic market, the largest market share worldwide and access to cheap state-backed financing. US backing taking shape not as an active intervention but as exclusive access to the American market – Huawei has been banned – might just be the thing that could give Ericsson and Nokia a sorely needed edge. If European policymakers, as promised, also start leveling the playing field with China, Ericsson and Nokia might actually become quite competitive. PvG

R&D expenditure in the Netherlands 18

Private sector

Research institutes

Higher education

French bulldogs help predict property appreciation

Want to buy a nice house in a safe neighborhood but lack the necessary funds? AI is here to help. US startup Lofty AI has developed a system that predicts property appreciation, allowing users to pick a reasonably priced piece of real estate before its price skyrockets. It works for urban areas and whole neighborhoods, too. Sounds good, right? Where it gets interesting is what information is being used to make the predictions. One parameter, for example, is the number of Instagram posts about French bulldogs. If this is on the rise, chances are that wealthy people are moving into the area. Another indicator is the average wait time for upscale ride-sharing services. If that’s decreasing, relatively well-off people are probably flocking to the neighborhood. Tweets about coffee shops, Airbnb prices, postings for high-paying jobs – the possibilities are endless. One question, though. How can we be sure the predictions are fully based on data and not manipulated by one party or another to make a buck? PvG

x 1,000,000,000 euro

12 10

4.393 0.900

4.304 0.923

4.506 0.912

4.550 0.968

8 6 4

9.515

10.008

2015

2016

10.654

11.230

2 0

2017

2018 Source: Statistics Netherlands

In 2018, total R&D spending in the Netherlands amounted to about 16.7 billion euros – the highest ever recorded. As a percentage of the gross domestic product (GDP), however, it isn’t a record: the 2018 R&D intensity of 2.16 percent was down 0.02 percent compared to 2017. This R&D intensity is slightly above the European average of 2.11 percent. Only one region in the Netherlands features a significantly higher ‘local’ R&D intensity: the Noord-Brabant province, where it stood at 2.99 percent. PvG 1

NEWS SEMICON

ASML: if we can’t ship it to China, we’ll ship it somewhere else Whether ASML should be allowed to sell EUV scanners to China may be a geopolitical hot-button issue, the company itself is rather indifferent on it. Its scanners will sell out one way or another. Paul van Gerven

rguably, 2019 was an outstanding year for ASML. After two decades of development, EUV-made chips finally started powering electronic devices. The first full-fledged EUV scanners were shipped and EUV bookings now constitute half of ASML’s order book (in terms of dollars). The company grew 8 percent, despite the semiconductor equipment industry as a whole registering an overall decline of 10 percent. And net sales – yet again – set a record. None of this seemed of particular interest to the journalists who attended the annual results press conference. They made the trip to Veldhoven for one reason and one reason only: the US-Chinese power play over ASML’s EUV technology. With both the US and Chinese ambassador to the Netherlands weighing in, the issue has been dominating the news recently. CEO Peter Wennink was a good sport about it, though. None of the press material issued in advance mentioned the awkward circumstance the company finds itself in, 8

and Wennink didn’t say anything about it during his presentation. But once the floor was opened to questions and the elephant in the room was addressed, he presented his arguments on why the issue doesn’t impact his company all that much – despite the involvement of two superpowers. “EUV is a driver of so many technologies that we can be sure there will be demand for the chips around the world. If for whatever reason we’re not allowed to sell EUV systems to China, another company will jump in to satisfy the demand. We’ll just ship to that company. It doesn’t matter to us where the chips are manufactured,” Wennink explained. In other words, ASML doesn’t expect to lose a single order. “Financially, the impact is zero.”

Geopolitical arena

Wennink stressed that the risk of the Chinese using an EUV system to steal the technology is nil. “We’re the only company in the world capable of making these systems.

What do you think would happen if we catch them trying to copy it?” Wennink countered. The scanners themselves are well-protected, he assured, with ASML crews surrounding it 24/7 and sensors alerting Veldhoven even if someone removes a panel. Should the Chinese want to give it a try anyway: good luck to them. Wennink cited the (in high tech well-known) story of a Chinese university that took apart an ASML scanner years ago, copied every single part to assemble a new machine, yet never succeeded in getting it to work. “We’re a systems integrator in a collaborative knowledge network. Our knowledge is in the minds of people, not in patents.” So indeed, should the Dutch government OK the export license to China, ASML will be more than happy to ship the scanner. And if it shouldn’t, that’s fine too. In essence, Wennink tried to divert the spotlight away from his company and towards the geopolitical arena. Let governments worry about it, we have work to do.

ANALYSIS SEMICON

ASML is too convenient a target for the US ASML shouldn’t be singled out as the only semiconductor company to be restricted. If the US wants to cut off China, at the very least, ban American technology too. Paul van Gerven

intends to leverage to move up in the world. Shattering the previously widely held assumption that China would never be able to catch up to the West, or if it did, it would be to our advantage too – the government no longer considers China’s ascension on the world stage necessarily a good thing. Just as technology can change our future, so can a more powerful China. So, indeed, if the US government has been pressuring the Dutch, it might not have had to press all that hard. It’s not likely that the whole strategy was written merely to appease US wishes, either. Though only the US pursues an aggressive course of action, most Western countries have adopted a more stern stance on China in recent years. The US did – even before the stable genius became president. There’s much to be said for the Netherlands and Europe rethinking their approach to China, but the US’ pursuit of global dominance isn’t our fight. With that in mind, it’s interesting to note that ASML is a very convenient target for the Trump administration. Curtailing the company will have next to no impact domestically, unlike when chip sales to ZTE

were temporarily banned. So what about, say, Applied Materials? Will its sales be restricted as well? That company wants as much a piece of the action in booming China as ASML does. What about Intel? Surely its chips can be used to power advanced weaponry. ASML’s technology is of exceptional strategic importance, though. Cutting off the Chinese from EUV will leave them forever stuck at 7-ish nanometer process technology, preserving the country’s dependence on imports for the most advanced semiconductor technology. To cut it off completely, however, one would have to convince TSMC to stop supplying the mainland (the Taiwanese foundry recently shipped the first EUV-made chips to Huawei). Still, it doesn’t seem fair to single out ASML when loads of US tech companies are allowed to continue doing business there. So, unless the Dutch government itself feels it would be wise to cut off China from EUV or use it as leverage in dealings with China, it shouldn’t ban the export. The sad reality of it, of course, is that if the US would really flex its muscles on the issue, the Netherlands would most likely cave.

Credit: ASML

inally, ASML finds itself in the eye of a geopolitical battle. It was just a matter of time, once the Trump administration started to put the squeeze on China. The US is hell-bent on stopping, or at the very least slowing down, the Chinese advance on the world stage. Everything was fine as long as China was manufacturing the easy stuff, but now that it starts posing a threat – economically, technologically and therefore ultimately militarily – to the American hegemony, the fight is on. According to the reports, initially by Nikkei Asian Review and more recently by Reuters, US pressure led the Dutch government to withhold the license required for ASML to ship an EUV scanner to a Chinese customer. This is almost certainly SMIC, China’s most advanced semiconductor company. At this point, the foundry has no need for EUV in manufacturing. But, the same way Intel, TSMC and Samsung had machines to play with years before they even considered moving EUV into production, SMIC obviously will require some time to learn the tricks of the trade. I wouldn’t be so sure, however, that it was just US pressure that put SMIC’s order on hold. The Dutch government presented its own China strategy right before ASML’s export license wasn’t renewed. Called ‘A new balance’, it marks a new way of weighing national interests. Basically, short-term commercial interests are no longer drowning out all the other ones. Eyes have been opened to China keenly taking advantage of the possibilities that open, Western economies offer, while not returning the favor at home. The Dutch government acknowledges this asymmetry is both an economic and a national security threat. The document specifically mentions semiconductors and lithography as powering technological revolutions, which China

NEWS SEMICON

EUV proﬁtability moving towards DUV level Selling EUV scanners is quickly becoming ‘just another business’ for ASML. It will take another year or two before its proﬁtability is up to ASML’s typical level, however. Paul van Gerven

t doesn’t take an accountant to spot a bit of an oddity in ASML’s 2019 results. The company’s sales grew from 10.9 billion euros in 2019 to 11.8 billion in 2020, yet the net income was the same – 2.6 billion. What gives? The main culprit, it turns out, is EUV. The share of the next-gen lithographic technique in ASML’s sales is growing, but EUV is not yet as profitable as the other activities. ASML sold 18 EUV scanners in 2018 and 26 in 2019. As a share of system sales, revenue from EUV increased from 23 to 31 percent, and ASML now derives almost a quarter of its total revenue from EUV systems (apart from selling systems, ASML

also gains revenue from what it calls “Installed base management”). The technology everybody had to wait so long for has finally blossomed into a ‘regular’ business. Except for profitability, that is, which is still lagging. EUV gross margin increased from 20 percent in 2018 to 30 percent in 2019 and will continue to grow to an expected 40 percent this year. Only then will it start to approach ASML’s typical overall gross margin of 45-50 percent. And it will eventually match that, assures CFO Roger Dassen. “You’ll see an increase in EUV gross margin every time we launch a new model,” explains Dassen. “Every time we do, we

create so much more additional value for our customers that they’re willing to pay significantly more. If we can keep a good grip on cost at the same time, our gross margin will increase.” Last year, ASML introduced the NXE:3400C, which at around 130 million euros a piece costs 30 percent more than its predecessor. The company shipped some units of it already, but the bulk will follow this year, hence the significantly higher gross margin for EUV in 2020 compared to 2019. In 2021, ASML will launch the NXE:3400C’s successor, increasing EUV profitability yet again.

Because it makes perfect sense

with SMC pressure & flow sensors A A diverse range Precise sensors for a diverse range of fluids A Ease your control Three screens display to identify e.g. current and set value A Easy programming 3-step simple setting or via IO-Link communication A Get more from your sensor Advanced functions such as antichattering and power saving SERVICE

RESPECT

EXPERTISE

ATTENTION

CONTINUITY

HONOUR

www.smc.nl

NEWS SEMICON

ASML’s multi-e-beam metrology tool delayed After a complication in IP sharing, ASML was forced to start over developing a multi-beam e-beam metrology tool. As a result, the launch of the system was delayed by a year. Paul van Gerven

Real-time feedback

The integration of e-beam into the holistic litho suite still leaves a lot to be desired as far as speed is concerned. The obvious solution: multi-beam tools. This is another justification for the acquisition: ASML’s deep pockets to fund and accelerate the R&D effort to increase the number of beams. HMI and ASML have been working on their first multi-beam tool for a couple of years now, featuring nine beams in a 3x3 configuration. It’s mainly intended for process development R&D, but it could also be used in production applications whenever optical inspection just doesn’t cut it. The question is: where are these tools? The first ones were supposed to ship at the beginning of last year but weren’t. As it turns out, a setback in IP delayed the tool for almost a whole year. “Initially, we worked with Zeiss on multibeam, but they were sharing relevant IP with another company, which made it impractical to share with us as well. We would

have had to start a separate company to make that work. Instead, we decided to do it ourselves, even though that meant we had to start from scratch. Hence the delay,” ASML CFO Roger Dassen told Bits&Chips. Fortunately for ASML, early last year it had the opportunity to scoop up some excellent multi-e-beam expertise when Mapper went belly-up. ASML didn’t continue Mapper’s mission in direct-write e-beam lithography but was more than happy to offer its engineers a new job. Without referring to e-beam technology specifically, ASML CEO Peter Wennink recently said that “we’re very happy with our former Mapper people.” The first 3x3 tools are expected to be shipped in the current quarter. In the future, ASML intends to increase the number of electron beams, aiming to eventually build e-beam tools that can provide real-time feedback during IC manufacturing in a similar way the optical metrology tools in the holistic suite already do.

Credit: ASML

hen ASML acquired US-Taiwanese metrology specialist Hermes Microvision (HMI) in 2016, the Veldhoven company had a clear vision for its e-beam inspection tools: integration in the holistic lithography suite. This assortment of computational and metrology techniques in and around ASML’s scanners helps to make sure that chip patterns are printed correctly on the wafer. Over the past few years, the additional control has become a must-have in chip manufacturing at the leading edge. The addition of e-beam to the exclusively optical inspection methods in ASML’s suite makes sense. Much like lithography is limited by wavelength in what size features it can print, the minimum detectable defect size is capped for optical metrology. At the same time, as chip features get downscaled, the margins of error shrink and increasingly small defects can ruin entire chips. So helping chipmakers to find and analyze them thus presented ASML with a nice revenue-increasing opportunity. Electrons can help do that because it’s easier to produce low-wavelength electrons than photons. E-beam is notoriously slow, however. A single electron beam takes ages to trace the entire surface of a wafer. Unsurprisingly, HMI’s tools were originally used for process qualification and calibration. ASML and HMI figured that, together, they can speed up e-beam inspection significantly: when the information obtained by holistic lithography is used to direct the e-beam to ‘hot spots’ on the wafer, the search area is dramatically narrowed. So there’s a two-way synergy happening here: e-beam inspection expands the holistic litho suite and the holistic litho speeds up e-beam inspection. All the more reason why it made sense for ASML to acquire HMI.

1 11

ORDER NOW

“ASML’s Architects is an impressive book, a curious book and a book for the curious. (…) Clearly a labour of love by Raaijmakers but nonetheless an easy read.” Peter Clarke, eeNews, February 1, 2019 “Rene Raaijmakers’ book on the history of ASML is a monumental work in its depth and breadth from ASML’s beginning through 1996. (…) No tech company’s history has ever been covered to such a degree.” Dan Hutcheson, The Chips Insider, February 1, 2019

techwatchbooks.nl/architects

pinion

THE HEADHUNTER Anton van Rossum anton.van.rossum@ir-search.nl

Ask the headhunter K.R. asks: Just over a year ago, after a career of more than 10 years doing chip design at semiconductor companies, I switched to a customer-interfacing technical position in the industry. Although the technology hasn’t stopped fascinating me, I find that my qualities aren’t sufficiently utilized when I’m only sitting in front of a screen. Being in contact with other people is always very inspiring to me, especially with people in high tech. However, such roles are rare in the part of the country where I live. I was therefore very enthusiastic when I got the opportunity at a renowned international company. The position seemed to fit me perfectly. For a year, I did the job with great pleasure – I learned something new every day. I got along with my boss, although I didn’t see him often because he spent a lot of time abroad. I also had a great working relationship with my colleagues. A few months ago, my manager was given a new position at the head office and I got a new supervisor. Initially, my rapport with him was also excellent: we spoke only once, but the contact was good. A few weeks later, however, I was told that my contract wouldn’t be renewed because I wasn’t suitable for the position. This was a real slap in the face for me. In the year that I’d been working for the company, I’d never heard a word of criticism about my performance. Not that it matters much – the decision is irrevocable. How to proceed? I no longer want a purely technical function; I want my job to be more about people and

technology. Do you know companies where I can find such a position?

The headhunter answers: You were unlucky to get a new manager right before securing a permanent contract. While your previous supervisor was very content with you, the new guy has doubts. He asked you about your achievements in the past year and wasn’t satisfied with the answer. You haven’t really done anything wrong, but I’m guessing you didn’t show enough initiative towards customers. You should have asked him how he assessed your performance and how you could improve it. When you have a commercial position, it’s advisable to always get feedback from your colleagues and your

cation engineering). All of these roles put greater demands on your social and contact skills and – in case of a commercial position – on your commercial skills as well. In my view, you have the potential to grow in such a role, but you need guidance. Appropriate training also seems useful. Your previous employer has really missed an opportunity here. Especially since there are so few candidates in your field who meet the profile, it would have been beneficial to everyone if you’d been given a better chance of succeeding.

By failing to get feedback, you planted a seed of doubt in your manager manager. By failing to do this, you planted a seed of doubt in your manager over your abilities to grow in your position. That finally killed you. Looking for a position with a focus on people and technology in your field, you have the choice between organizational technical functions (such as project management) and commercial technical functions (such as sales and marketing or field appli1 13

NEWS DISPLAYS

Seecubic doubles down on glasses-free 3D in Eindhoven As 4K technology has become the new standard in video resolution, Seecubic’s glasses-free 3D solution offers a new dimension for OEMs. But after several years of maturation and small-volume production, the company is looking to get back to what it does best: development and innovation for the future. Collin Arocho

n late 2010, former Philips employees were developing a glasses-free solution to bring 3D to televisions. In December of that year, seeking investors to help them bring their innovation to the market, the group had a fortuitous meeting with Mathu Rajan, the founder of a newly started media and technology company called Stream TV Networks. As it turned out, Rajan was on the lookout for 3D technology and was interested in their solution. Within weeks,

the group set up a booth at the Consumer Electronics Show in Las Vegas to unveil their technology: Ultra-D – 3D content on any screen, without the need of glasses. Merely months later, the Eindhoven-based Seecubic was founded as an R&D subsidiary of Stream TV and tasked with developing the technology, components and services of Ultra-D. Today, some eight years later, Seecubic and its parent company seem to have hit their stride as a developer and supplier of

display technology. “We’re a technology provider, not a device maker. Our customers are the OEMs. We want to work with them on integrating our core technology to be used in their devices. They know how to make a computer. They know how to make a TV. They know how to make a smartphone. We don’t want to reinvent those wheels,” illustrates Bud Robertson, CEO of Seecubic. “It’s really about facilitating the adoption and integration of our optics, rendering and content format to participate in the overall content ecosystem.”

Global partners

Credit: Seecubic

It appears this business model is poised for success. Stream TV is currently in talks with a number of global brands, the likes of Lenovo and Huawei, that are interested in adopting its technology in their devices. Recently, China’s BOE Technology Group, which this year surpassed LG as the world’s largest flat-panel display manufacturer, announced it would be partnering with the US company to utilize the glasses-free 3D platform developed at the Eindhoven-based R&D facility. This surge in international interest stems from, at least in part, the advances in screen resolution technology – ie upgrading from 1080p to 4K and soon to 8K. The faster-than-expected rollout and market acceptance of 4K technology have had a major effect on Seecubic’s roadmap. “We already had a few customers interested at 1080p resolution, but it was clear that making a widespread consumer play was going to be difficult with that lower resolution,” recalls Robertson. “With the arrival of 4K a few years ago, everything really changed

Credit: Seecubic

for us. Suddenly, we started to generate significant customer interest. This rush of customer interest, however, really shifted the business activities of Seecubic. Suddenly, we were less able to spend time innovating and we spent much more time reacting to customer inquiries, proving that our technology could be adapted to fit their specific needs. We were sort of limited to smallscale production.”

New mandate

Now, a few years further down the road, Seecubic and its parent company Stream TV are again adjusting course. With the technology now to the point where the customer inquiries are satisfied, Stream TV is ready to move its technology to a larger-scale production phase. To facilitate this, the company has added a new product team and facility in Silicon Valley. “The past couple of years have been somewhat restrictive. We’ve had to put a lot of focus on maturing our products and making a lot of demos – sample panels for tablets, phones, TVs,” expresses Robertson. “The new team will be tasked with taking what we’ve incubated in the lab here in Eindhoven and productizing that technology into components that can be used by OEMs. What that really means for us at Seecubic is that

we can get back to our roots, back into the innovation cycle.” Expectedly, the opening of the new facility and the transfer of this technology to Silicon Valley left some of the Eindhoven employees a little skittish, not knowing what to expect. But to Robertson, the future in the Netherlands is clear. “I have a mandate to double the Eindhoven team over the next year or two. We’ve got a lot of work to do and we’ve got enough innovation projects to keep us busy for many lifetimes. We just need to develop them and get them ready for market,” says Robertson. “Our new mandate is expansion and innovation. We have some pretty exciting evolutionary work ahead to establish our technology as the de facto standard over time.”

New flavors

Relieved by the spinout of the new group in America, the Seecubic team has wasted no time getting back to R&D. According to Robertson, there’s never been a shortage of projects and ideas, only a lack of time to focus on them. Robertson: “The new message from executive management in the US is to let the product team figure out integration; Seecubic’s focus should be on creating a new research roadmap and to get us working on some of the projects for

which we haven’t had the time. In all, we have more than a dozen new ventures coming up on the agenda for 2020.” One such project: integrated depth sensing. Currently, Seecubic’s Ultra-D format uses depth information plus video imaging. The thought is to collect depth information immediately at the camera, rather than processing the information after the fact. “We might not need real-time conversion. We want to develop real-time capture,” highlights Robertson. “That’s one project that we’re now exploring. Of course, it will largely depend on factors like whether current depth sensors are good enough to capture the quality of data that we need. Five years ago, they weren’t.” Another project that has caught the interest of Seecubic is providing a glasses-free 3D experience for the cinema. “That’s a very different kind of project, but it falls well within the scope of our technology and one where we’re going to make a very big push. In fact, we’re already in contact with a very high-profile 3D filmmaker who wants to release a 3D film without the need for glasses,” reveals Robertson. “Work like this really gives our guys a chance to put on their creative hats and the ability to focus on inventing new flavors instead of cranking out variations of what we’ve been doing to date.” 1 15

Digitally Enhance Your Analog Controls Combine the Speed of an Analog Controller With the Flexibility of a Digital Microcontroller

No system can perform without reliable power supplies. Our Digitally Enhanced Power Analog (DEPA) family of products combines the performance of an analog Pulse-Width Modulation (PWM) controller with the configurability of an 8-bit PIC® microcontroller (MCU). The combination of these methods allows the addition of digital features to a reliable, easy-to-implement analog control loop, including fast transient responses, high efficiencies, reliable gain and phase margins. Adding the ability to measure and respond to changes with tailored algorithms improves the robustness of the system, while offering diagnostic and communication options. The single-chip solution can accept a high-voltage input and regulate a wide output current or voltage range, which allows you to maintain robust operation within an unstable environment. Discover how the flexibility of our DEPA products can enhance your next design.

Key Features •

Fast and efficient power conversion with analog current-mode control loop

•

Flexible control with an integrated MCU

•

Dynamically settable hardware protections enable robust operation

www.microchip.com/FlexiblePower The Microchip name and logo and the Microchip logo are registered trademarks of Microchip Technology Incorporated in the U.S.A. and other countries. All other trademarks are the property of their registered owners. © 2019 Microchip Technology Inc. All rights reserved. DS20006124A. MEC2309A-ENG-12-19

NEWS EQUIPMENT ENGINEERING

Solaytec looking to spread its wings with spinoff SALD Almost ousted from the solar market by Chinese tools, atomic layer deposition tool manufacturer Solaytec is optimistic it can tap new applications for its technology. It’s starting sister company SALD to do just that. Paul van Gerven

High throughput

“The solar market, particularly in China, is currently driven by reducing cost, not by improving performance,” comments Verhage. “This could change, but there’s

Unlike the OEM Solaytec, SALD will focus exclusively on the deposition head and leave the equipment engineering to partners.

Credit: Solaytec

olar alone isn’t a viable market anymore for Solaytec. That’s the cold hard truth, explains Frank Verhage of the Eindhoven-based enterprise. Alarmed by slowing demand, then-parent company of Solaytec, Amtech, hired him in October 2018 to investigate whether the Dutch subsidiary could survive. Verhage’s answer: not without looking at new applications. Solaytec was spun off from research institute TNO a decade ago to commercialize in-house developed spatial atomic layer deposition (SALD) technology for the solar market. ALD, in general, involves sequentially exposing a substrate to different gases, which react with the surface in a self-limiting fashion, thus ‘coating’ it literally atomic layer by atomic layer. The difference between SALD and traditional ALD is that in SALD, the substrate moves past the gases, while in temporal ALD, it’s stationary while a reactor is repetitively filled and emptied. The continuous character should make SALD the superior technique for industrial applications. In solar, at least, it was for a while. Compared to other deposition techniques, an ALD’ed layer of aluminum oxide imparts additional efficiency to solar cells, which led several high-end manufacturers to buy Solaytec’s SALD tools (as well as those of Levitech, another Dutch company selling SALD equipment). Unfortunately, as the aluminum oxide layer became more and more mainstream in solar, so did much cheaper Chinese batch ALD tools.

no way to predict when that will happen. Similarly, perovskite solar cells present an excellent opportunity for us, yet there’s no way of knowing when they’ll emerge from development.” Fortunately, other high-tech industries have expressed interest in SALD: battery manufacturers, the flat-panel display industry and the printed electronics sector, among others. There are ample low-tech applications as well, such as airtight sealing of packaging materials and treating textile fibers. “As a high-throughput, roll-to-roll compatible technique, SALD is very suited for these applications,” says Verhage. Solaytec has been working with multiple companies and research partners to explore alternative applications for a while now. Verhage can’t go into detail at this point, but he assures that Solaytec will be ready to offer new solutions in the near future. “We’ve made great progress over the past six months. We just need to wait until the patents are in order.”

Sister company

In light of the expanding application scope, the “Solaytec” brand name is no longer satisfactory. It’s too closely associated with the solar industry, notes Verhage. That’s why Solaytec is starting a sister company, called SALD, which focuses on applications other than solar. SALD is not to be confused with Saldtech, another SALD company spun off from TNO, which is focused on OLED displays. Unlike Solaytec, SALD’s equipment won’t feature its logo. “Solaytec is an OEM: we design the whole system, and though NTS built it for us, we market it ourselves. SALD is about innovation: our core technology is the deposition head and our core business is how to use it in different applications. We won’t be engineering and marketing the complete tools ourselves. Partners such as VDL-ETG will do that. More generally, in the spirit of the Brainport area, we’ll be actively seeking out collaboration to make the most of this great technology.” 1 17

B a c kg r o u n d

Quantum technology

Qutech: building a quantum delta, one (qu)bit at a time Even though quantum technologies are only just starting to take shape, now is the time to consider the opportunities they present – both to enhance existing business and to generate new opportunities. In the Netherlands, the Qutech initiative is leading the effort to bootstrap a quantum industry in the Dutch delta. Paul van Gerven

he quantum computer, despite all the excitement surrounding it, is still a mysterious thing. Sure, the theory is solid: scientists have a pretty good idea of what’s needed to build one. But those are the contours, the basic principles. Few would dare to predict what the first fullfledged quantum computer would actually look like on the inside, what technology it would be based on. Even today, several candidates are being considered, and probably more options will pop up – or fizzle out – as time passes. Still, the quantum computer has moved from concept into reality: already some primitive ones have been built. To outsiders, these systems can’t do anything particularly useful, but the point is that quantum calculations are being performed, which means people are getting their hands dirty. These quantum builders need to think about the whole system: not just the quantum bits (qubits) that perform the calculations but also the electronics and software that are essential to manage and control them. Just like a PC needs a hard drive and an operating system in addition to a processor, a quantum computer needs peripheral electronics. Actually, it needs a lot of it. It takes a couple of pretty sizable ‘classic’ computers to run a quantum one. 18

The elements that constitute the modern computer were developed gradually, one at a time, until they all came together. It started decades ago with the invention of the transistor, then came the integrated circuit (computer chip), at some point, someone started working on the software, and so on.

Quantum Delta NL

Qutech may be the front-runner, but many more Dutch organizations are involved in building the Quantum Delta NL. Most universities have some kind of quantum research program, Qusoft from Amsterdam focuses on application software for quantum computers and the quantum Internet, and the collective of Dutch high tech companies called the top sector High Tech Systems and Materials (HTSM) is keen to stay involved. In September 2019, these organizations presented their joint goals and ambitions in the National Agenda Quantum Technologies. Apart from the commercial ecosystem, the agenda also addresses educational and human capital issues as well as informing and engaging the general public.

That’s not how the quantum computer is being built, explains Ronald Hanson, scientific director of the Dutch quantum research center Qutech in Delft. “Already at this early stage, we have many disciplines working together, both scientists and engineers. Making progress in one particular area depends on progress being made in all others as well. It’s not a step-by-step evolution of isolated technologies.” Qutech is one of the few places in the world where such a concerted effort is taking place. Google and IBM are the most well-known organizations building complete quantum computers, along with some lesser-known companies, while most research groups focus on a single problem or a few at most. But Qutech is neither a company nor a research group. It’s a public-private collaboration trying to leverage its efforts into establishing a quantum industry in the Dutch delta. How has that been working out, so far?

Imperfect

Qutech was founded in 2013 by TU Delft, a university of technology and engineering, and TNO, an applied research institute. TU Delft had world-class quantum research to offer, as well as expertise in electrical engineering, physics and computer sciences

Credit: Marieke de Lorijn

A quantum computer in progress at Qutech. The big vessel (a cryostat) is where the magic happens.

– all very relevant for quantum technologies. TNO knows a thing or two about those things too, having assisted the international high tech industry in innovation for decades. Another role of Qutech, however, is to reach out to companies and try to get them involved with quantum technology. Two big fish already signed up: Microsoft, which built its own lab next to Qutech’s premises, and chip manufacturer Intel, which entered into a long-term partnership. Of course, these are the kind of

companies that don’t need convincing that quantum tech is a paradigm-shifting opportunity. “Getting firms on board whose core business doesn’t involve computing as much is a lot harder,” admits Rogier Verberk, director Semiconductor Equipment Industry at TNO and the organization’s point man for Qutech. “But we’re getting more and more traction.” Roughly three types of companies are potential Qutech partners. First, perhaps surprisingly, there are the end users – the

companies that could take advantage of quantum technology. That may sound strange, given the primitive state the technology is still in, but that shouldn’t deter anyone, Hanson assures. “The perfect quantum computer may be years or even decades away, but before we get there, we may be able to solve real-world problems with imperfect quantum computers.” Hanson refers to what are called noisy intermediate-scale quantum (NISQ) machines. These are modestly sized systems in terms of computing power and furthermore prone to making errors. Even if lacking compared to the real thing, NISQs could still prove useful for a range of problems that are out of reach of regular (super)computers, because methods can be devised to compensate for the errors. “The problem is that while we know what those problems look like, we don’t know many that would actually be useful to solve. In fact, it has become a research field on its own to find those,” says Hanson. “Surely, though, many such problems are to be found at companies. At banks perhaps, to sift through data. Or at pharmaceutical companies, looking to design a new drug. These companies have no idea what quantum computing could do for them, however.” Which means there’s all the more reason for Qutech to actively reach out and get companies involved. Already Qutech joined forces with a bank, ABN AMRO, though not for quantum computing but for quantum communication, a branch of quantum technology that’s a little farther along than computing. The project involves exchanging extremely secure cryptographic keys between users in a way that’s practically impossible to eavesdrop on. In the future, these keys could be used to secure Internet and mobile banking. Taking this one step further, Qutech is also working with telecom company KPN to make the quantum Internet a reality.

Business opportunities

The second kind of company that fits in well in the nascent Dutch quantum delta is one that’s involved in the quantum business itself. Qutech is the reason Finnish company 1 19

Building a foundation for the Dutch high tech ecosystem Despite competition from China and the US, the Netherlands continues to play a major role in the world of high tech. Patrick Strating of NTS believes it starts with high tech companies that have close ties to top-notch technical universities and continues with ambitious workers that thrive on life-long learning through training. To establish and preserve their expert knowledge, the workers at NTS often attend technical trainings in optics, mechatronics and systems development. Perhaps somewhat surprising, however, is the benefit the company sees by emphasizing social trainings like soft skills and sales.

nts-group.nl hightechinstitute.nl/consultative-selling

B a c kg r o u n d

Credit: Marieke de Lorijn

Quantum technology

Bluefors decided to open an R&D office on the TU Delft campus, where Qutech resides. Bluefors designs cryogenic systems for quantum computers. “It’s important for us to be able to design new specifications with leading users and to benefit from each other’s knowledge,” commented Bluefors CEO Rob Blaauwgeers when the Delft office was announced in 2018. More companies might want to follow this example. Qutech itself gave birth to two quantum startups already, both of which are now operational in Delft. The Qblox spinoff focuses on the previously mentioned spe-

Qutech Qutech is one of many research initiatives co-funded by the Top Consortium for Knowledge and Innovation (TKI) High Tech Systems and Materials. Through the Ministry of Economic Affairs and Climate, TKIs provide additional funding whenever industry invests in long-term innovation projects with public research organizations.

Insides of the cryostat.

cialized electronics and software required to run a quantum computer. Hanson: “You can’t operate a quantum computer using a normal PC. At Qutech, we’ve put a lot of work into electronics that satisfies the specific demands for quantum computing. Our results are so good that it makes sense to start selling.” The other spinoff, Delft Circuits, focuses on dedicated electronics wiring for quantum computing at ultra-low temperatures. A third spinoff is in the making, but Hanson can’t tell anything about that one yet. The third and final type of ‘Qutech company’ may be considered both a potential user and a technology supplier. Many high tech companies will be able to take advantage of quantum computing once it reaches a certain performance level, enhancing their toolkit for innovation. At the same time, quantum computing will present new business opportunities. As the technology evolves, there’s likely more revenue to claim. The supply chain will increasingly become more sophisticated, just like that 1 21

B a c kg r o u n d

Quantum technology

of – say – the PC is very large and complex these days. “We’re basically shooting for something very similar to the semicon ecosystem we already have here in the Netherlands, with companies like ASML, NXP and their supply chains. And we want to be the best in Europe,” explains Verberk. Wait, aren’t you the best in Europe already? Verberk, smiling: “We don’t like to be immodest, but yes, there’s something to be said for that. The large number of people that work here, the world-class quality of

the research, the fact that we successfully brought together scientists and engineers under one roof, and that Microsoft and Intel chose to join ... let’s just say many people admire what we’ve done here.” Still, there are quite a few places in Europe that aspire to become a quantum hub. That’s a lot of competition. Hanson: “Ultimately, a quantum hub needs critical mass to succeed. Clearly, we can’t have twenty hubs all over Europe. I suspect we’ll see consolidation over the next few years. In fact, I hope the European Union

What makes a quantum computer so great?

In a regular computer, information is coded into a series of 1s and 0s, the bits. These, in turn, correspond to physical states. The 1s and 0s on a hard disk, for example, correspond to the direction of a magnetic field (up or down) on a large number of distinct areas on a surface. A computer essentially performs operations (manipulations) on a series of 1s and 0s according to the instructions given by the user. A quantum computer, too, uses physical states to represent 1s and 0s, but – crucially – these physical states are small enough to obey the laws of quantum mechanics. It’s best for anyone without a degree in physics not to try and understand quantum mechanics, so you just need to know that quantum mechanics allows physical systems to be in two states at once. So, in a quantum computer, where a bit is called a qubit, there’s the 1s, the 0s and the both-1-and-0s. The both-1-and-0s are where the magic comes from: you can perform two operations for the price of one, ie a single operation can be performed on both a 1 and a 0 at once. A normal computer needs two operations for that. One or two operations – it may not seem like a big deal. But the difference between quantum and classical computing grows exponentially. A 10 qubit quantum computer can already perform 210 = 1024 operations at once, so imagine the power of a quantum computer with hundreds of qubits! There is a caveat, however: qubits are extremely fragile and frequently ‘flip’ their value. This means extensive error correction is required, which is made possible by using many physical qubits to make a single ‘real’ qubit. Currently, researchers expect the ultimate quantum computer will need millions of physical qubits, whereas the biggest one now only has tens of (still very fragile) qubits. Another important thing to note is that not all types of calculations benefit from quantum power. A quantum computer will be great at finding out that 268,568 is 569 times 472, for example. A classical computer has to check every reasonable combination of numbers until the solution is found. Not every problem is necessarily of this nature, though. Still, there are plenty of applications for quantum computers. They could be used for machine learning, in financial modeling and for logistics, for example. Another great application for quantum computers is in chemistry. Molecules are quantum systems and chemical reactions are dictated by quantum mechanics. What could better simulate these systems than an actual quantum computer? Imagine being able to design new drugs or materials without ever setting foot in a lab. 22

will encourage this. And I’m confident that Qutech will be one of the ecosystems to make the cut.” Are companies already lining up to start working with you? Hanson: “Compared to other countries, such as Germany and the United States, I must say that Dutch companies have been quite cautious. I’d really like to see things speed up. We need companies to get on board and, reversely, I don’t think it’s an exaggeration to state that this technology is too important to miss out on.” “On the other hand, we have some improvement recently. Before we started Qutech, I invited twenty companies or so, gave them a tour and told them what we were setting up. Afterward, I asked them whether they supported our concept, whether they would like to stay involved and whether they would like to invest in quantum technology. They all liked our idea and wanted to stay involved, but that was it. Nowadays, their interest is much more tangible. Only a few years ago, ABN AMRO would never have invested in quantum technology. And recently, we signed a long-term collaboration agreement with KPN.” Verberk nods: “Not that long ago, quantum technology wasn’t an item at trade conferences, not even at technology conferences. Today, dedicated parallel sessions are popping up all over the place. Slowly but surely the awareness grows. Our meetings with companies used to be generally exploratory in nature. These days, we dig a lot deeper.” It must be hard to convince companies without having something to show them, though. Hanson: “We have demonstrator programs! We’re building a small quantum Internet and we have a quantum computer that outsiders can use for getting to know the technology and its possibilities. Right now, it’s a simulation running on a supercomputer, but we want to start offering real quantum hardware in 2020. Having such goals is an excellent way to drive our own efforts, by the way. Because everything will need to be

Credit: Marieke de Lorijn

on point. Every single part of the computer will need to be up and running and working well with other parts.” IBM has had a similar program for a long time, on real hardware. Google recently had its quantum computer outperform a supercomputer on a specific problem for the first time in history. It seems like you’re falling behind. Hanson: “What Google did, is a breakthrough. No doubt about that. Yet, I don’t think we can compare what we are and what Google is trying to do. Google chose a very specific challenge five years ago: to demonstrate quantum supremacy. They succeeded,

which is fantastic. But that result isn’t very relevant to Qutech’s mission.” “As for IBM, it did very well to start offering quantum computing to the public. That’s necessary for educational purposes and for community building – two things that are really important for a radical new technology. That’s very much like Qutech’s mission: educating people – TU Delft started a new series of master courses Quantum Technology – and building an ecosystem.”

High Tech Highlights A series of public-private success stories by Bits&Chips

Verberk: “We’re not competing with Google and IBM, and it’s not our mission to outdo Google and IBM. What matters is what we do to bring quantum technology to fruition in the Netherlands.” There’s a long road ahead of you. What are the biggest obstacles? Hanson: “Apart from getting companies involved, the biggest bottleneck is talent. Quantum technology needs a lot of qualified people, but there simply aren’t that many of them around. There’s a reasonably steady supply of young researchers, but experienced talent is hard to come by. That’s definitely going to be our biggest problem in the coming years.” 1 23

INVENTED BY ALL

A profession in design or technology isn’t typically what most youngsters aim for when they’re in high school. Partly this is because of the lack of realization of what design and tech mean for the world. Actively confronting them with their own ability to design something that means a lot to others increases their motivation for an innovative study or job. This is exactly what the project “Invented by All” does.

iving at home with dementia Everyone has heard of it, lots of people suffer from it: dementia, or Alzheimer’s disease. It not only affects the patients but their families as well. For people developing dementia, living on their own often becomes more difficult. Together with October – a health facility near Eindhoven – the Discovery Factory engaged (14-year-old) youngsters in high school, asking them if they could come up with solutions. This generated lots of innovative ideas helping patients with devices for daily needs or security. The kids also looked into the social element of bonding with patients. A whole new game At the Rythovius College in Eersel, one group developed a board game especially for families having a member suffering from memory loss. The game aims to stimulate conversation and interaction between patients and caretakers.

A group from Sint-Lucas – a school for the creative – further developed the concept. At the end of their project, they presented a prototype, well tested with the target group. Their “Talktogether” game is now ready for use, complete with a wooden board, appealing questions and a format for playing a game together. The whole point of the game? All win! Your own “Invented by All” project The real winners are the kids who developed the game. They found out that what they design and make really matters to people! If you have aspirations in enthusing school children for science and technology as well, or the company you work for has, the “Invented by All” project could be your starting point. The context and societal impact of your company are the inspiration for the pupils. If you need assistance, you can always call on our professionals for help.

The Discovery Factory is there to inspire youngsters for a future in design and technology. Projects are supported by tech companies such as ASML, Brainport Industries, Daf Trucks, Frencken Europe, Hager, NTS Group, Philips, Stam en De Koning and VDL Group, and by Bits&Chips as the media partner.

discoveryfactory.nl

B a c kg r o u n d

Industrial automation

Ampleon: riding the 5G wave with a smart factory A leading global partner in RF power products, Ampleon works on developing new 5G technologies, setting up the intelligent factory and professionalizing its supply chain. The Factory Intelligence program is a roadmap to bring factory automation to the next level and to use data to increase performance and predictability levels. Ampleon’s Head of Global IT, Lody Hoekstra, talks about the journey, use cases and challenges of modernizing the company’s factory. Rolf Naberink

eadquartered in Nijmegen, Ampleon is a global player in the area of RF power. Its core business is the production of technology that amplifies communication signals for a wide range of applications, such as mobile broadband (3G, 4G, 5G), radio and television broadcasting. The company was established in 2015 after NXP was required to carve out its RF Power business. This carve-out meant that the new organization needed to establish a new IT environment within a short period. “An enormous challenge, but it also offered us a chance to improve and prepare for the future,” explains Lody Hoekstra, Head of Global IT at Ampleon. Hence, the company decided to build a scalable digital platform.

Ramping up 5G

The completion of Ampleon’s new digital platform marked the start of a new chapter: a rapid ramp-up. Hoekstra: “5G technology was coming fast. In a short time, we needed to more than double our production capacity. This resulted in several challenges, with two key areas that we needed to focus on.” “First, the supply chain. As we were scaling up our production, the demand for flexibility grew. Time-to-market needed to be reduced. We also required more insights and more control over the process.” The second area that Ampleon needed to focus on was the factory itself. “We operate a factory that employs 1,200 people 26

in the Philippines. And when you need to increase your manufacturing output that much, using new technologies, launching new products at a faster rate than ever before, this creates a tremendous challenge for the organization.”

Developing Factory Intelligence

Ampleon’s management posed the question: how can we modernize our factory? This produced a list of conditions and goals, such as becoming paperless, decreasing the man-machine ratio and full traceability of

products. Hoekstra: “80 percent of these initiatives are IT driven and we recognized that we needed to prepare for this.” Ampleon teamed up with the digital consultants of Itility to embark on the development of a new program: Factory Intelligence 2022. The starting point was to define concrete use cases that provide business value for the company and translate those into a roadmap. “Building a smarter factory involves using data. Lots of data,” says Hoekstra. The building of a data factory was, therefore,

view. Change management is a big part of innovation.

Use case: predictive maintenance

one of the first steps to take. Based on a central data lake, in which all Ampleon’s manufacturing data is collected, governed and made available.” “You need a solid architecture, a technical configuration and full integration. This process requires not only IT but domain expertise as well, combining the efforts of infrastructure engineers, data engineers, and business analysts.” When building a data factory, it can be difficult to keep stakeholders involved. This was solved by using a roadmap to build for the future while focusing on small deliverables. Like a single data source or dashboard: things that immediately make people happy.

Use case: real-time OEE monitoring

Gathering data is step one, but the real goal is to use that data to improve daily operations. The challenge is to balance ambitious, long-term improvements with deliverables that immediately provide value. One example of this is real-time monitoring of the overall equipment effectiveness (OEE). “As a first step, we built a new factory-wide yield report and real-time monitoring of our OEE. Presenting data on machine uptime,

usage and performance in a way we can analyze it and take action.” Hoekstra continues: “This tool is a great improvement compared to our previous systems. It provides immediate, actionable insights that are easily accessible for our entire organization. We now have a single source of truth for OEE information.”

Use case: traceability

End-to-end traceability is important, both for customers and Ampleon itself. “If a product is returned to us and we suspect it to be flawed, we want to be able to completely backtrack that product through our factory. Which machines did the product go through, in which phase? What values were measured at that time? If the cause of a flaw is confirmed, we can immediately identify which other clients might be impacted.” Achieving this kind of traceability takes a big effort. First, the manufacturing equipment itself. Ampleon uses industrialized IoT devices and smart machines to extract data. After disclosing this data, it’s vital to know that people need to adjust as well. Operators who are used to look at a single machine need to adopt a holistic

Another important goal in Ampleon’s Factory Intelligence program is predictive maintenance. “We’re working towards a scenario in which we know exactly when each machine was serviced, which components were replaced and what the status is of components that are being used right now. We’ll use this information to predict when maintenance is necessary and optimize maintenance actions. This helps us to minimize a machine’s downtime, which creates a big cost saving. If minimizing downtime in combination with an optimized factory scheduling means we can deliver the same output with 80 percent of the machines previously required, that saves millions of euros.” The first step in predictive maintenance is to make basic but important information available. For example: how many strokes have been made with a specific tool? This helps to replace the tool before it breaks down. Instead of basing maintenance on the time that has passed, you can use the real usage of a tool to replace it just in time, thus preventing replacements that are unnecessary or too late. The next step is to actually use machine learning to even better predict when a replacement is needed.

Moving forward

Over the past years, Ampleon has successfully taken several major steps in their smart factory journey. As a result, the company is already seeing a significant increase in its efficiency, quality and output. “We would never have succeeded in multiplying our production in only a few years using our old way of working. Digitalizing our factory wasn’t just a luxury, it’s a necessity to prepare for Ampleon’s future.” Looking at the future, Hoekstra expects to see a lot of developments in the area of artificial intelligence. “There’s a lot of potential if we can enable our software to identify risks and opportunities by itself. We’re now working with data reporting and analysis. The next step is to automate the data analysis as well and use more complex algorithms to start predicting and prescribing.” Rolf Naberink works at Itility. Edited by Nieke Roos

1 27

THEME ARTIFICIAL INTELLIGENCE

HIGH TECH PREPARES FOR AI Artiﬁcial intelligence promises mountains of gold. As smart algorithms are spreading like wildﬁre in the high tech industry, engineers are facing all kinds of obstacles. AI specialist Albert van Breemen tells us what tech companies have to take into account if they want to master the technology. Albert van Breemen

rtificial intelligence is a technology that, due to its versatile applicability, enables radical changes in many industries. AI has developed rapidly over the last eight years – even faster than experts predicted. This is mainly due to deep learning, an AI technology requiring a huge amount of data to work properly. The first wave of commercial AI applications has emerged at companies that already have lots of data, like Amazon, Facebook, Google and Uber. Currently, we’re seeing a second wave in which AI algorithms are being trained with sensor data. AI is entering the

engineering world and traditional machine builders are faced with the question of whether and how to invest in artificial intelligence. Designing advanced systems is getting ever more challenging. Each generation, their complexity grows. There’s an increasing demand for performance, intelligence and interoperability. Engineers regularly reach the limitations of their current toolbox, in which modeling and analysis from so-called “first principles” are standard. Adding AI gives them technologies to work with larger datasets (big data) and smart AI algorithms. Both elements form the ba-

sis for finding new solutions to the aforementioned challenges. In practice, however, companies are finding it difficult to embrace artificial intelligence. A good case study could convince management to invest more in AI. Unfortunately, the people who have potential cases within a company have no experience with the technology and fail to recognize the opportunities. This makes it difficult to demonstrate the benefits and secure an innovation budget. In addition to business-driven barriers, we also see many operational obstacles in practice, such as the complexity of the AI software and hardware,

the great diversity of algorithms (when to use which technique), the computing power needed to train models and finding talent.

Barrier 1: the AI technology stack

The technology stack for artificial intelligence is a complex set of hardware and software components. Model training requires an incredible amount of computing power because it uses computationally intensive gradient-descending and backpropagation techniques. This calls for special hardware, such as CPUs, tensor processing units (ie AI-specific computing cores), FPGAs or ASICs. Applying a trained model to an embedded system requires different hardware, designed with energy consumption and system resources in mind. The AI software stack is built up of a series of modules, most of which are developed independently of each other. Updating one of the modules often causes it to stop working smoothly with the rest. In addition, server-side software libraries, such as Tensorflow, aren’t always compatible with the libraries used on embedded systems.

Barrier 2: algorithm diversity

Deep learning algorithms can be subdivided into a number of categories, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs) and reinforcement learning. Within each of these, there are a number of more specific algorithms. Which one to choose isn’t only determined by the application, like object classification or object detection in CNNs. It also depends on the software stack and the hardware used, as well as the model size and accuracy and the type of input data,

for example. Optimizing a model is a time-consuming process.

Barrier 3: computing power

The demand for computing power to train modern AI algorithms doubles every three to four months. Training an algorithm like Deepmind’s Alphagozero requires more than a million teraflops of computing power. To put that into perspective: a modern GPU does about fifteen teraflops. According to a conservative estimate, training an AI algorithm for the Starcraft game will soon cost several million euros. Most companies lack the resources for this. Fortunately, many other useful AI algorithms can be trained with less computing power, but the standard IT infrastructure of most companies is often inadequate for this as well.

Barrier 4: talent

Many machine builders have recently experienced the introduction of artificial intelligence within their company. They have to catch up because they often are unaware of the value that data and AI can bring to their products. In addition to the challenge of bridging the gap between engineers and AI experts, they’re also in a fierce competition with larger and well-known AI companies to at-

Albert van Breemen is speaking at the second edition of the Machine Learning Conference, 11 March in ’s-Hertogenbosch. mlcon.nl

tract AI specialists. Universities are currently working hard to train new talent. Eindhoven University of Technology’s Eindhoven Artificial Intelligence Systems Institute (EAISI), for example, is starting a number of new master’s programs this year that close the gap between engineering and AI.

Companies at bat

Current developments in artificial intelligence will have a major impact on products and industries. Occasionally, you can hear people saying that we’re too late and everything happens in the US and China. But that’s not true. The success of an AI application in engineering strongly depends on domain knowledge. And that’s just where Dutch companies excel, for example in mechatronics and human-machine interaction. Knowledge institutions and organizations such as the High Tech Systems Center and EAISI are working hard to remove the barriers. Collaboration with companies is key to achieve this and to make a new voice heard: AI and engineering takes place in Brainport. Albert van Breemen manages the AI program at the High Tech Systems Center (HTSC) and the Eindhoven Artificial Intelligence Systems Institute (EAISI) of Eindhoven University of Technology. He founded the AI Engineering Lab and is working together with FME, the Dutch employers’ organization in the technology industry, to make AI more accessible for SMEs. He’s also the owner of the AI consultancy agency VBTI. Together with High Tech Institute, he teaches the masterclass “Introduction to deep learning”. Translated by Nieke Roos

1 29

pinion

ARTIFICIAL INTELLIGENCE Marco Jacobs is a marketing and strategy consultant.

My algorithm is better than yours

n 20 years in developing electronics, I’ve heard it a lot: “Our product is so much better! The algorithms our engineers developed are way beyond what our competition has.” In the 1990s, I saw it with my own eyes at Philips. Their TVs had the best deinterlacing algorithms in the world, resulting in superior picture quality. In the 2000s, I worked on audio enhancement: turning on sound processing algorithms gave quite a stunning effect. Tiny speakers would suddenly create bass and provide a 3D sound stage. Over the next years, I’ve seen cameras getting a much better picture quality, virtual reality headsets providing a much smoother experience and compression algorithms becoming a lot stronger. The before and after effect is a very strong sales tool. Now, we’ve entered into an era where it’s not about picture or audio quality anymore. Instead, our electronics need to become smart and adopt AI. Andrew Ng from Stanford put it well: “If a typical person can do a mental task with less than one second of thought, we can probably automate it using AI either now or in the near future.” Even though AI can only solve fairly simple tasks, this presents a lot of business opportunities. The market responds and big corporations acquire AI teams for large sums of money. In AI, I’m seeing the same “my algorithm is better than yours” claims. Many companies state their AI is better than everyone else’s and are showing the before and after effect. Kudos to these companies, for all having managed to hire the smartest AI algorithm engineers? No, not really.

With AI, for one, it’s actually fairly easy to build your own algorithm. Instead of having engineers that skillfully craft and program the algorithm, AI is simply trained. You give the AI engine lots of examples over and over again and it keeps adjusting itself until it doesn’t make any mistakes anymore. Another problem is that it’s hard to measure ‘better’ in AI land. There’s a famous saying that states there are lies, damn lies and then there are

AI doesn’t seem like a great foundation on which to build a solid company benchmarks. In AI, it’s no different. In automotive, for instance, you can measure false negatives, where you don’t detect a person in front of the vehicle. But a false positive, where the AI brakes for a pedestrian who’s not there, is almost equally bad. The size of the data set is something to consider as well. It sounds impressive when an AI algorithm scores perfectly on 10 million kilometers of automotive test data, but since we have one billion vehicles on the road that each drive 10,000 kilometers a year or so, the AI probably still only covers a fraction of the real-world situations that can occur. Even perfect scores can be meaningless in such situations.

A final complicating factor is that the algorithms are still rapidly changing. There are many competitions where universities and corporate research centers battle it out and continuously introduce new algorithms. The winning neural networks are freely made available on the web, as are the tools to adapt and train them. Download and retrain the model for your target application and you’re done. Thus, we come to the conclusion that AI algorithms are easy to develop, hard to benchmark and ever-changing. That’s a problem because it doesn’t seem like a great foundation on which to build a solid company. Strong companies typically operate in markets that have high barriers to entry, making it difficult for new entrants to come in and compete, which isn’t the case with AI. My advice: don’t rely solely on AI in isolation, but use it as an enabler for your business and closely integrate it into your products. Simply focusing on “my algorithm is better than yours” won’t give you a sustainable competitive advantage. When your engineers tell you that their algorithms beat the competition, congratulate them, but ask them right away how they’re planning to maintain that advantage.

Credit: Ivo van der Bent

THEME ARTIFICIAL INTELLIGENCE

DOCTORS EMBRACE AI: COMPUTER CALCULATES BEST RADIATION TREATMENT A computer that can calculate an entire series of optimized treatment plans for prostate cancer in 30 seconds. That took some getting used to for medical specialists. Using artiﬁcial intelligence, the computer presented better plans and more insight than the doctors thought possible. The ﬁrst patients are due to be treated accordingly already this year. Aschwin Tenfelde

urprised and amazed, that’s how radiotherapists felt at Amsterdam University Medical Center, AMC location. A computer using artificial intelligence generated an entire range of plans for the radiation of prostate cancer in no time. Not only that, but these plans were also completely optimized and even specified what the outcome would be of employing a lower or higher dose of radiation. It usually takes even the most experienced radiotherapist with a lab technician a great deal of time to determine the best radiation treatment for a patient. So how good can a radiation plan be if a computer seems to generate it so effortlessly?

Radiation dilemmas

Nevertheless, the need for improvement was felt strongly by doctors. For them, it’s always a race against the

clock to decide on the right radiation plan for brachytherapy, or internal radiation therapy. This kind of treatment is a weapon doctors often use to fight prostate cancer. Radiation is administered through ten to twenty catheters that are inserted into the patient, penetrating the tumor. They then deliver a source of radiation, which can briefly stop at certain places. The longer the source stops, the more radiation the tumor receives from that place. As catheters are extremely uncomfortable for patients, doctors give themselves an hour’s time at most to devise the best possible radiation plan. The difficulty lies in the considerations doctors are constantly having to weigh. Preferably, they would simply attack the tumor with a hefty dose of radiation. On the other hand, you have to minimize the damage to the surrounding healthy tissue. Unfortu-

nately, there’s a limit to the precision of the radiation beam. It weakens as it penetrates tissue, but it essentially will go through everything. As a result, some radiation inevitably ends up in healthy areas, where it does more harm than good. This presents physicians with dilemmas: how do you plan to deliver radiation accurately through catheters? And how high should the minimum dose be? How much radiation will ‘leak’ to the surrounding tissue, and is that acceptable? The ideal plan is different for every patient. With sufficient experience and practice, radiotherapists and lab technicians can usually create a plan together within an hour – one single plan, that is. With the current software they use for this, and in light of the time pressure, it’s not feasible to also give extensive thought to alternative radiation plans. 17 31

THEME ARTIFICIAL INTELLIGENCE

Self-learning software

Isn’t there a better way to do this? That’s the question radiotherapists, clinical physicians and researchers at the AMC put to Peter Bosman in 2014. Bosman, who’s affiliated with the Centrum Wiskunde & Informatica (CWI) in Amsterdam, was studying AI techniques at the time. He was already focusing on the question of how to use artificial intelligence for medical applications, especially new ways of doing medical image analysis. The fact that doctors posed this question is unusual, according to Bosman, because it doesn’t reflect how things normally transpire in the medical world. “Doctors are used to working within the possibilities provided by the software they use. They know this software like the back of their hand. It’s usually the medical companies that bring innovations to the clinic.” Bosman accepted the challenges and put together a research team with the AMC and Elekta, which develops radiation equipment and software for hospitals. The team decided to develop software that’s centered on a kind of AI called evolutionary algorithms. These algorithms are suitable for effectively and efficiently finding good solutions to difficult problems, especially when there are multiple conflicting goals at play. The team focused mainly on a type of evolutionary algorithm that displays ‘intelligent search behavior’: it can learn about the nature of a particular problem and find better solutions more rapidly. Bosman’s team adapted the algorithms so they could be used for brachytherapy in cases of prostate cancer. They did this by allowing them to use knowledge about how the dose of radiation builds up in the administered catheters. This made it possible to achieve far superior results than with other algorithms. 32

Pepsi challenge

“Initially the AMC radiotherapists reacted reluctantly to our proposed approach,” recalls Bosman. “Which is completely normal. Imagine, a couple of mathematicians and computer scientists drop by telling that AI can probably do a better job than doctors and their current software, despite their years of experience. We needed time to persuade them.” The major turning point came two years ago. At that point, the team had made enough progress to do a kind of ‘Pepsi challenge’ with the doctors. “We had the computer create radiation plans for patients who had already been treated in the past at AMC,” explains Bosman. “In other words, the doctors had already made plans for these patients that had been approved.” Subsequently, all plans were anonymized, so the doctors couldn’t check who had developed them. “Then we asked several doctors: which plan would you prefer to use with this patient? In 98 percent of the cases, they chose the plan generated by the computer.” Bosman’s team also demonstrated that the computer could immediately present an entire range of alternative radiation plans that used lower or higher doses. “The doctors viewed this insight as a unique added value,” says Bosman, “because it gives them time to think about which radiation plan they want to choose, instead of devoting all of their time to developing a single plan. They can now decide how much radiation to deliver to the tumor and how much the healthy tissue can receive.”

From CPU to GPU

The doctors fully embraced the idea and were in favor of using it in the clinic. To reach that point, the team first had to augment the software’s processing speed. At the time of the Pepsi challenge, it still took about an hour to

calculate the range of radiation plans for one patient – and that’s not quick enough for clinical practice. Bosman’s team eventually managed to reduce that time to a mere 30 seconds. This huge gain in speed was mainly achieved by no longer processing the calculations on the system’s central processor (CPU) but on the graphic processor (GPU). It’s a trick that many researchers apply to get more computing power in simulations, graphics and AI. The gain in speed is derived from the fact that the average GPU has thousands of cores that can carry out calculations simultaneously, whereas a high-end CPU has to make do with eight cores. You can’t simply ‘switch’, though. Algorithms made to run on CPUs don’t automatically run on GPUs. That’s because the cores of a GPU are far simpler than those of a CPU, and you often have to use them in blocks. Bosman: “So you need to look carefully at which calculations really benefit from being run on a GPU. Essentially, they all have to be really simple and you have to want to carry out a large number of them simultaneously. In our case, for example, you want to know how much radiation will be delivered to a huge number of areas. Also, you want to be able to update that frequently as you search for good radiation plans.” It’s striking that Bosman’s team gets by with a relatively simple consumer GPU, which costs around 1,200 euros. Although these GPUs lack error checking in the accompanying internal memory on the card, the margin of error turns out to be extremely small. This leads to margins of uncertainty in the generated treatment plans that are completely negligible. That’s also why everyone in AI uses these kinds of cards instead of the more expensive ones that do have error checking, according to Bosman.

Credit: Ivo van der Bent

Good rapport

The collaboration between doctors, researchers and Elekta was excellent. But that’s certainly not something you can take for granted, says Bosman. “It requires a lot of coordination. Our team’s work is spread across CWI, the AMC and the Elekta office in Veenendaal. To make sure that everyone’s on the same page and familiar with each other’s jargon, all researchers get together twice a week, alternating between CWI and the AMC’s premises. In addition, we go to Elekta every four weeks, and every six weeks, we have an update meeting that everyone attends, including the doctors and clinical physicians. This allows us to go over the results together and discover what’s important right now. It has enabled us to become a genuine team.” Bosman strongly advises researchers to continuously seek a good rap-

port, if they want to achieve innovation in a similar way. “I’ve also seen examples in the past of how things can go wrong. Despite making clear agreements and project plans in advance, researchers have the tendency to go their own way sometimes. This can cause lines of research to be cast adrift from the practical use that was the point of it all in the first place.” In the meantime, the partnership between CWI, the AMC and Elekta is thriving. The hospital is hoping to treat its first patient with the radiation plan created with the new AI software before this summer. “Elekta is also interested in commercializing the results of the project,” Bosman says. “And that could mean worldwide clinical application.”

Follow-up

The big advantage of the AI system is that it’s relatively easy to extend to

other types of cancer. Indeed, a follow-up project was inevitable. With a grant from KWF Kankerbestrijding (Dutch Cancer Society), Bosman and his colleagues are now going to focus on cervical cancer. The research will be conducted by a consortium that also includes Leiden UMC and almost all Dutch hospitals that treat cervical cancer with brachytherapy. They’ll test and evaluate the new software that emerges from the project. Elekta is on board again as an industrial partner. And so it’s probably only a matter of time before artificial intelligence can be used as a weapon against this form of cancer as well. Aschwin Tenfelde is the communications manager at Centrum Wiskunde & Informatica. Edited by Nieke Roos

1 33

THEME ARTIFICIAL INTELLIGENCE

MASTERING MACHINE LEARNING WITH TUNABLE CAPABILITIES FOR ELIMINATING OVERFITTING By integrating scalable software tools with tunable machine learning capabilities, engineers and scientists can efficiently identify the most suitable model that fits the specific industrial data and meets the model objectives, while protecting against overfitting. Paola Jaramillo Garcia Mohamed Anas

ngineers and scientists are building smarter products and services, such as advanced driver assistance systems and predictive maintenance applications, driven by analytics based on industrial data. Analytics modeling is the ability to describe and predict a system’s behavior from historical data using domain-specific techniques for data preparation, feature engineering and machine learning. Combining these capabilities with automatic code generation, targeting edge-to-cloud, enables reuse while automating actions and decisions. By leveraging the increased availability of ‘big industrial data’, compute power and scalable software tools, it becomes easier than ever to use machine learning in engineering applications. ML methods ‘learn’ information directly from the industrial data without relying on a predetermined equation as a model and are particularly suitable for today’s complex systems. However, two of the most common challenges faced by engineers and scientists who are modeling with machine learning relate to choosing a suitable ML model to classify their domain-specific data and eliminating data overfitting. Classification models assign items to a discrete category based on a 34

specific set of engineering features extracted from the historical data. Determining the best model often presents difficulties given the uniqueness of each data set and the desired outcome. Overfitting occurs when the model is too closely aligned with limited training data that may contain noise or errors. An overfitted model is unable to generalize well to data outside the training set, limiting its usefulness in a production system.

Choosing a classification model

Choosing a classification model type can be challenging because each has its own characteristics, which could be a strength or weakness depending on the problem. For starters, you must answer a few questions about the type and purpose of data: what’s the model meant to accomplish? How much data is there and of what type? How much detail is needed? Is storage a limiting factor? Answering these questions can help narrow the choices and select the correct classification model. You can use cross-validation to test how accurately a model will evaluate data. Afterward, you can select the best-fitting classification model. There are many types of classification models. A common type is logistic regression. Because of its

simplicity, this model is often used as a baseline. It’s applied to problems that have two possible classes into which data may be categorized. A logistic regression model returns probabilities for how likely a data point belongs to each class. Another common type is k-nearest neighbor (kNN). This simple yet effective way of classification categorizes data points based on their distance to other points in a training data set. It has a short training time, but it can confuse irrelevant attributes for important ones unless weights are applied to the data, especially as the number of data points grows. A third common classification model type is the decision tree. This model predicts responses visually and it’s relatively easy to follow the decision path taken from root to leaf. It’s especially useful when it’s important to show how the conclusion was reached. A fourth common type is the support vector machine (SVM). This model uses a hyperplane to separate data into two or more classes. It’s accurate, tends not to overfit and is relatively easy to interpret, but training time can be somewhat long, especially for larger data sets. A fifth common type is the artificial neural network (ANN). These networks can be configured and

trained to solve a variety of different problems, including classification and time series prediction. However, the trained models are known to be difficult to interpret. You can simplify the decisionmaking process by using scalable software tools to determine which model best fits a set of features, assess classifier performance, compare and improve model accuracy and, finally, export the best model. These tools also help users explore the data, select features, specify validation schemes and train multiple models.

Eliminating data overﬁtting

Overfitting occurs when a model fits a data set but doesn’t generalize well to new data. This is typically hard to avoid because it’s often the result of insufficient training data, especially when those responsible for the model didn’t gather the data themselves. The best way to avoid overfitting is by using enough training data to accurately reflect the model’s diversity and complexity. Data regularization and generalization are two additional methods you can apply to check for overfitting. Regularization is a technique that prevents the model from over-

relying on individual data points. Regularization algorithms introduce additional information into the model and handle multicollinearity and redundant predictors by making the model more parsimonious and accurate. These algorithms typically work by applying a penalty for complexity, such as adding the model’s coefficients into the minimization or including a roughness penalty. Generalization divides available data into three subsets. The first set is the training set, the second is the validation set. The error on the validation set is monitored during the training process and the model is fine-tuned until accurate. The third subset is the test set, used on the fully trained classifier after the training and cross-validation phases to test that the model hasn’t overfitted the training and validation data. There are six cross-validation methods that can help prevent overfitting. K-fold partitions data into k randomly chosen subsets (or folds) of roughly equal size, with one used to validate the model trained with the remaining subsets. This process is repeated k times, as each subset is used exactly once for validation. Holdout separates data into two

subsets of a specified ratio for training and validation. Leaveout partitions data using the k-fold approach, where k equals the total number of observations in the data. Repeated random subsampling performs Monte Carlo repetitions of randomly separating data and aggregates results over all the runs. Stratify partitions data so both training and test sets have roughly the same class proportions in the response or target. Resubstitution uses the training data for validation without separating it. This method often produces overly optimistic estimates for performance and must be avoided if there’s enough data. ML veterans and beginners alike run into trouble with classification and overfitting. While the challenges can seem daunting, leveraging the right tools and utilizing the validation methods will help apply machine learning more easily to real-world projects. Paola Jaramillo Garcia is a data scientist at Mathworks. Mohamed Anas is a regional engineering manager at Mathworks. Edited by Nieke Roos

1 35

11 MARCH 2020 VERKADEFABRIEK â&#x20AC;&#x2122;S-HERTOGENBOSCH

Silver sponsor

Sponsors

MLCON.NL

#BCML

PROGRAM Chair Aad Vredenbregt (Valoli) 09:30

Registration

Keynote 09:45

AI on the edge: interacting swiftly and surely in a dynamic world Nicolas Lehment (NXP)