Control systems for deep-tech tools share several common denominators, whether in a particle accelerator or a proton therapy linac: real-time operation, deterministic behavior and zero room for failure. Cosylab built this expertise over 25+ years. Now, it’s applying it in the semiconductor industry.
Every new precision system, from a chip machine to a metrology system to a proton therapy accelerator, has to do something its predecessor couldn’t. That may mean measuring a smaller feature, positioning a component more precisely or controlling a process differently. Each advance creates a challenge for both the mechanics and the control system, which has to coordinate the moving parts with extremely precise accuracy and timing. That means control system engineers have to work out what the new machine demands as the design takes shape.
Those are situations that Rok Vintar and his team at Cosylab discuss with their customers. The Slovenian company develops the control system layer beneath the equipment controller, covering everything from PLCs and FPGAs to motion control, data acquisition and safety interlocks. It builds on more than 25 years of experience in fields ranging from particle accelerators to cancer treatment. Five years ago, it entered the semiconductor industry.
Vintar, who began his career as an electrical engineer and now leads a multidisciplinary control systems team, has worked on multiple projects in physics research and fusion, as well as medical technology. “While the machines serve very different purposes, each depends on many components working together accurately and reliably,” he explains. “For Cosylab, that means understanding the machine as a whole before designing how it will be controlled.”
Integral parts
A control system connects sensors, moving parts and other subsystems, each developed by specialists with their own priorities and constraints. It can’t be designed in isolation. Cosylab’s teams bring together expertise in software, electrical engineering, systems engineering and systems integration. This range of knowledge helps them communicate with a machine’s designers early on, rather than waiting for a finished list of software requirements.
“The first task is to understand what the machine is intended to do,” Vintar says. “Which functions demand the greatest precision? How must its components interact? And where could a design choice create difficulties later?” These conversations help turn the machine’s overall purpose into concrete requirements for its control system. “Having engineers that understand the mechanical and physics side of the machine helps reveal questions that may be easy to overlook when only thinking in software.”
Those questions become especially important when the machine has to do something that hasn’t been done before. A decision in one discipline can affect what’s possible in another. By involving control engineers with mechanical knowledge in early discussions, the team can identify dependencies and technical risks while there’s still time to address them.
Cosylab has a rich history in particle accelerators and medical systems. In an accelerator, the control system has to coordinate the equipment that generates, steers and monitors a particle beam. In proton therapy, beam control has to work with the treatment system’s safety functions to deliver radiation exactly. These projects taught the Slovenian company to consider timing, measurement and safety as integral parts, rather than separate software tasks.
Timing
About five years ago, an existing customer brought Cosylab into semiconductor equipment development. While the applications were different, Vintar recognized the engineering challenge: “We had to understand the complete machine, establish what its control system must deliver and anticipate where a new design could run into trouble. We do this with our other projects as well; only in the semiconductor industry, this has to happen at nanoscale.”
A wafer passes through many processing and measurement steps before its chips are ready. At each step, the equipment must carry out its task consistently, often across different operating settings or “recipes.” The control system coordinates the machine’s movements, process settings and timing to carry out each recipe consistently. That consistency is essential to achieving a high yield of usable chips from each wafer.
In a metrology tool, the machine control software has to move the stage, trigger the source and read out the detectors in exact synchronization, then turn the result into trustworthy data, for every wafer, under every recipe. “Timing is everything,” Vintar points out. “If the stage, the trigger and the detector disagree by a fraction of a microsecond, you’re measuring the wrong thing. There’s no room for error.”
As precision requirements rise, every change to the optics or mechanics becomes a change to the machine control software, too. This is why Cosylab prefers to be in the room before the design is finalized rather than afterward.
Focus on the gaps
The Netherlands has a dense network of semiconductor equipment makers and specialized suppliers. Some build complete machines, others contribute a critical part of one. For smaller companies in particular, developing a new capability can stretch the engineering capacity available across disciplines. According to Vintar, Cosylab already works with several players in the semiconductor industry. It also works alongside Dutch high-tech companies on customer projects. Some customers require specialist control systems expertise, while others have the expertise in-house but need additional capacity. Cosylab can help shape a new machine’s control system alongside the customer’s engineers or take on a defined part of the development work.
In one project, the customer had a solid mechanical design for a new tool, but the machine control software was still at an early prototype stage. The company was under pressure to deliver within six months. “They had already done a great deal of work,” Vintar recalls. “We looked at how to get from what they had to a stable, working machine within the time available.”
Where other companies suggested a complete control redesign from scratch, Cosylab began by examining the existing system controls with the customer’s engineers. Together, they identified which parts could be reused, which needed modification and which had to be developed anew. “We first needed to understand what they had already done and what was critical for the machine to work,” Vintar says. From there, the team could define the requirements and focus its development effort on the gaps, making use of the work the customer had already completed.
Companies may initially bring Cosylab in for a specific task, while keeping their legacy in the hands of their own engineers. Vintar: “Once they understand our expertise and are willing to share their earlier work, we can help maintain or update those machines, freeing up the customer’s team for new development work.”
The same pains
Across the industries, Vintar sees the same pattern wherever machine complexity rises. The first trend is artificial intelligence becoming a serious tool, using machine data to identify problems, support maintenance and improve yield. “Machines are already becoming more efficient with the use of AI,” he says, “though in critical applications, there’s still a human in the loop.”
Cosylab is actively accelerating its development of AI for machine data and applying those methods across its industries. Here, too, the industries share more than they differ. “A proton therapy system and a metrology tool have different purposes, but many of the same pains. What we learn about machine data in one industry, we can apply in the next. This efficiency is especially welcome in an industry where there are already too few engineers for the number of machines made.”
This article was written in close collaboration with Cosylab.
