What does Merck KGaA, Darmstadt, Germany contribute to the global chipmaking endeavour?
We provide end-to-end materials, equipment, instruments and services for wafer processing, including R&D and fab services. Merck KGaA, Darmstadt, Germany, as we refer to our company in the U.S./North America (and as Merck outside the U.S.)—comprises three sectors: life sciences, healthcare, and electronics. We have more than 60,000 people globally, with 8,200 in electronics. While it is the smallest of the three sectors at our company, it serves some of the largest players in semiconductors and has one of the broadest portfolios in the field.
We once did the math: there is over a 99% chance that any electronic device you buy has been produced using our materials or solutions – and we are in today’s AI chips as well. I always find that number impressive. Whether it is phones, laptops or even cars, the probability is extremely high that we were involved somewhere in enabling that device.
Can you outline in more detail the specialist and emerging technologies that Merck KGaA, Darmstadt, Germany works with?
Our sweet spot is thin films materials – those that are laid on the chip at the end to make the conducting and the insulating layers. Those materials are very differentiating due to the extreme quality and scale requirements to reach the smallest scales on a semiconductor. We also work with patterning materials, which help create chip structures, and planarization materials, used to flatten layers during chip building. On top of that, we offer specialty gases for deposition and processing, and delivery systems to get those materials to the right tools.
In terms of emerging technologies, a new add-on in that area is our metrology and inspection technology for heterogeneous integration, where wafers are cut into dice and then glued together in, for example, the NVIDIA AI systems. This is still done based on electrons today. The next frontier is integrating optoelectronics and photonic components to these chip systems, so light can be used for transmitting data. This is called co-packaged optics or silicon photonics, where the advantage of photons is that they create a lot less heat on the chip, so you consume less energy and create higher efficiency.
How mature is the rollout of heterogeneous integration technology and what is driving it?
Heterogeneous integration is moving very fast, propelled by the economic value of AI systems—especially for large language model training and cloud data centres. Those use cases require massive performance and efficiency, and this advanced packaging approach helps meet those demands.
Progress in the industry is defined by Moore’s Law, but soon, we will reach the limits of shrinkage and device improvement will have to come from somewhere else. More broadly, the entire linear data processing model of a separate CPU and memory unit that John von Neumann pioneered in 1945 is showing to be inefficient for today’s exponential growth in data. Given that, what does the future of semiconductor technology look like?
We have been hearing about the "end of Moore’s Law" for decades—I did my Master’s thesis in semiconductor design in 1989, and even back then people said we had reached the limits of light for lithography. I have joked before that there is a derived Moore’s Law: the number of people declaring its end doubles every two years.
That said, we are approaching economic and physical boundaries. There are different ways to deal with that, like furthering advanced lithography systems like EUV (Extreme Ultraviolet). However, at some point its advancement is prohibitively expensive. The other option is new architectures. You already referred to von Neumann's famous bottleneck. Here, neuromorphic or quantum computing could open a door. Each requires extremely advanced materials, which makes it a natural space for us to contribute.
How is Merck KGaA, Darmstadt, Germany investing in furthering neuromorphic and quantum computing?
Most of our R&D still focuses on shrinking structures on a standard semiconductor and improving CMOS technology. The second part goes into improving efficiency of these current systems, and this is where photonics and heterogeneous integration comes into play.
The next step, where we have some long-range R&D projects, is on additional phenomenon-based architectures on the chip, like neuromorphic computing systems—meaning where the compute is done right in the memory itself, so there is no more data transfer needed. On quantum computing, we are working in a consortium on advancing the usage of quantum computing to improve our R&D effectiveness.
What are some of the future applications on the table when it comes to quantum, neuromorphic computing?
The applications of quantum computing are very much in areas where classical compute is very complex and consumes a lot of time. For example, selecting the right molecule from billions of options to solve a chemical challenge on a chip. Simulating that fully is an enormous computational problem.
We are already using AI for this. Advanced algorithms help us avoid 99% of experiments, narrowing down the real-world trials to just 1%. It does not mean that our researchers are afraid of losing their jobs; the opposite is the case—they are more excited. They remove 99% of unsuccessful experiments to make the successful ones, which speeds up the process of R&D quite significantly. Other application fields for quantum are in logistics and financial planning—whenever you have to manage very complicated scenarios, this is where quantum computing comes in quite handy.
There is a term going around called bio-convergence. Can you explain it?
Take a look at supercomputers used for AI model training—they consume massive amounts of power. Now compare that to the incredible human brain, which runs on around only 20 watts. That’s a huge difference.
This is where biology comes in. Using DNA to store data—which is currently still expensive and slow to write and read— could open a whole new area of storage. I do not think we will see full-scale DNA-based storage within five years, but it is on the horizon. That is bio-convergence: integrating biological systems with data and computing architectures.
What is the mood like in the lab at Merck Darmstadt, Germany?
Historically, innovation always accelerates. So what we believe is unimaginable right now will be made possible at some point. That is why the teams are extremely excited about what we can already see.
In the R&D projects, we look probably two nodes down the road and see what our customers will introduce in the next five to 10 years. That is quite exciting to see. We are really in the Ångström scale of things already, which I believe has enormous potential.
Merck KGaA, Darmstadt, Germany is 357 years old, originally starting life as a pharmacy. Now it has pivoted to bleeding-edge semiconductor technologies. What kind of directions will the company pursue in the next five years to stay at the forefront of chip technology?
We clearly did not start with semiconductors in 1668. So how did pharma lead to chips? One moment stands out: in the mid-19th century, one of our founders made a quality promise: “I compensate for any impurity of the formulations that I sell”. That obsession with quality is in the genes of our company. And that is what ties us to semiconductors. At the atomic level, this industry demands extreme precision and purity. In fact, we can pinpoint purity down to parts per billion -- that is the equivalent of being able to find one football on a surface as large as Antarctica. We also use advanced data analytics and long term predictive modelling to address R&D, Quality and Supply Chain challenges in days and weeks vs. months or years with traditional approaches.
Second, when it comes to bio-convergence, a company like ours that is expert in biology, chemistry, physics, and microelectronics under one roof has a natural opportunity to contribute. It is a really exciting space.
