Nearfield Instruments ended 2020 on a high note. The Rotterdam-based TNO spinoff shipped its first metrology system to an undisclosed customer. Its scanning probe microscope – the Quadra – is capable of sensing challenging 3D semiconductor structures, at a sufficiently high speed to be used in chip mass production. How did a small startup manage to carve out a place for itself in that highly competitive market?
It took Nearfield Instruments (NFI) ten years of development, tinkering and engineering, but in December, the Rotterdam-based company finally announced that it had sold its first machine. Founder and CTO Hamed Sadeghian isn’t allowed to disclose who purchased the high-end metrology system, but chances are that it’s Samsung. Together with investment fund Innovation Industries, the Korean chip giant put ten million euros into NFI in 2017. In the second round of investment, in 2019, Samsung bought another stake. “It’s a big player in the semiconductor industry,” is all Sadeghian is prepared to divulge.
NFI’s Quadra metrology machine is intended for chip companies like Samsung. Such systems are often optical in nature, but for the most advanced chips, that has become inadequate. The chip structures are completely three-dimensional, the details too small and the channels too narrow and deep. There’s an urgent need for a solution because a few misplaced atoms can already have major consequences for the chip’s operation.
The Quadra is basically an atomic force microscope (AFM). Such a system produces a relief drawing of a surface by scanning it with a needle. Thanks to its atomic resolution, the technique is very interesting for imaging state-of-the-art ICs. The principle isn’t new, especially not in the research world, but for a long time, the AFM was considered too slow for industrial applications. Moreover, a standard AFM has great difficulty in measuring three-dimensional structures with a high aspect ratio. It therefore never came to a production system – until now. NFI has managed to ensure that its AFM solution does a good job of mapping those 3D structures and that the technology is also fast enough for the high throughput requirements of chip manufacturers.
To achieve the desired speed and capture complex nanostructures, NFI developed a completely new scanning mechanism in which Sadeghian took full advantage of the principle of advanced feedforward control. “With that control loop, we can very accurately plan the trajectory of the probe in the chip channel, with a very small tracking error. In fact, it’s nothing more than a solid mechatronic architecture,” says Sadeghian with a sense of understatement. “We’ve increased the measurement and control bandwidth and miniaturized the scan heads so that many heads can work in parallel. The system consists of many functions that all have to work at lightning speed and super-accuracy. Across a 300 mm wafer, we can navigate to a measurement point with a precision of 500 nanometers. The stage has a resolution of about 2 ångström. The alignment systems also had to become very intelligent; they have to scan the whole wafer, recognize the marking points, prepare the coordinates of the measurement points and transmit them to the scan heads.”
Does the probe touch the surface? “That’s almost a philosophical question,” Sadeghian finds. “The definition of ‘contact’ is very vague at the subnanometer level. It’s about controlling and minimizing the forces properly so that you don’t deform or damage the surface. That’s the core of our machine. We keep the interaction force at such a low level – in the order of piconewtons – that the probe doesn’t deform or otherwise damage the sample.”
The Quadra isn’t an inspection tool that allows chip manufacturers to probe their entire wafer, Sadeghian emphasizes. “It’s all about sampling, typically twenty to a hundred locations per wafer,” explains the Iranian-born. “The speed depends on the application. For example, do you measure the line roughness after a lithography step or the depth of a profile after an etching process? In the former case, we’re at a throughput of fifty to sixty wafers per hour.”
That’s fast enough for sampling, but of course, customers would like the measurement system to be even faster. “There’s a roadmap in place towards higher productivity where we can eventually measure all the wafers that come out of the scanner,” Sadeghian points out. To reach that final goal, NFI thinks it will need three more generations. The Quadra machine has four scan heads, but the underlying platform is ready for expansion to thirty heads. “We expect to have the second generation ready in 2024. The third generation will take five to six years.” The schedule is related to another development that Sadeghian and his team have started (see box).
NFI’s history goes back to 2011. At TNO, Sadeghian and Roland van Vliet – the current COO of the company – sketched the first version of the architecture on paper and listed all the market requirements. In 2015, the proof of concept was ready. With the prototype, the researchers could test wafers for potential customers, to show them the added value of the system.
“In 2016, we already did a lot of wafer tests for a large number of well-known chip manufacturers,” says Sadeghian. Samsung sent a heavy technical delegation and conducted extensive due diligence. NFI passed the test, as a year later it received a large investment from Samsung Venture Investment and others. This enabled the company to make really big strides.
NFI now has 81 employees on its payroll. About half have only joined the company in the past year and a half. “That was a challenge,” Sadeghian readily admits. “With a demanding customer who wanted the product as soon as possible, and then growing the company, hiring the right people with the right startup mentality, building cleanroom facilities: it was tough, but we succeeded. We benefited greatly from our collaboration with several high-tech suppliers.” Altran, Technolution and Aalberts division Integrated Dynamics Engineering (IDE), among others, were closely involved in the development and industrialization.
The latter was crucial. “We knew early on that we couldn’t reuse the architecture of the prototype for the production machine,” clarifies Sadeghian. The design had to meet very different requirements, which called for some drastic modifications. “To make the system ready for production environments, it had to be a lot more reliable, to begin with. The uptime also had to be increased and we had to be able to maintain and service the machine more quickly and easily.”
A production-ready machine isn’t just about performance, there are many other KPIs at play. “We’ve appointed architects for each KPI: cost of ownership, throughput, reliability, and so on,” explains Sadeghian. “This way all features get a boost without the system getting out of balance. Each module also has its own architect who keeps an eye on the overall performance of that component.” Sadeghian himself is responsible for the overall architecture.
Minimal viable product
With its first paying customer, NFI has entered a new phase. What lessons has Sadeghian learned on his path from research at TNO to a production-ready machine? “For starters, if you have to fail, fail fast. Don’t wait too long with your ideas and developments, but test them in practice as soon as possible. The later you discover that something doesn’t work or a concept doesn’t work well, the more expensive it is to fix.” He readily admits that he’s walked into that trap more than once himself.
Sadeghian’s second tip: “If you’re developing a new product, it’s vital to be on the same page with your customers. You have to know exactly what their wishes are and how you should translate them into system requirements. In doing so, you have to make sure that the system doesn’t become too expensive and that you’re left with a good margin. Define a minimal viable product. I’m totally against constantly improving the tool, as a result of which it doesn’t make it to the market at all. Instead, make sure you can deliver something as soon as possible so that customers can experience the added value and users can tell you what could be better. We’ve agreed on joint evaluation moments with our first customer. These will give rise to problems that we can then solve.”
NFI has benefited greatly from its outsourcing model. The wafer handler, positioning stage, optical alignment and active vibration isolation were all designed and built by certified partners. Sadeghian’s advice: “Outsource as much as possible. Just make sure to keep the core. We kept two modules completely in-house because they involve the machine core. You also really have to do the architecture yourself. I don’t believe you can outsource that. Not with this complexity, not in this market. That would be a mistake. Some companies have tried, but in the end, the result turned out to be a one-off solution, a one-off system that doesn’t take into account the roadmap.”
NFI also does most of the software in-house, with support from Altran. “The algorithms and data processing are so specific that it’s hard to hand that over,” Sadeghian has experienced. The partners create the modules and qualify them based on agreed test specifications. In Rotterdam, NFI takes care of the integration, calibration and factory acceptance testing.
Outsourcing costs money, and it’s precisely the financing that proves to be the biggest stumbling block for many startups. How did NFI manage in that respect? Sadeghian: “That was an enormous challenge, especially in the early years. Developments in the semiconductor market are expensive and risky. Few investors are willing to take that risk. You then have to have a mitigation plan. Limit the feeling that investing in your idea is super risky. You can put up all kinds of great stories and promises about how realistic and great your idea is, but only when a customer says they want to buy it too, do you have proof. It helps to focus on a minimal viable product, on a first system that you can actually sell.”
And it helps even more if you also have a paying customer. “You then suddenly have a completely different message to send to investors,” Sadeghian realizes. “We have a plan ready how we can scale up quickly now.” Apart from the first buyer, NFI is in contact with virtually all the major chip manufacturers. “They’ve all shown serious interest, although of course, we’re not at the same stage of discussions with everyone. Most first wanted to see that our solution really works.” Now that the first Quadra has been shipped, that requirement has been met and all systems are go for NFI to move forward.