At the Albany NanoTech Complex in upstate New York, where researchers from academia and industry carry out some of the world’s most advanced research into semiconductors, there is a surprising quiet.
From outside the complex, unassuming white walls and green tinted windows belie the urgency of the work happening inside. Walking down the halls, you’ll see people wheeling around what look like mailroom carts, but upon closer inspection, they’re carrying cutting-edge semiconductor wafers ready for testing in the complex’s various facilities. Sitting in a certain quiet transverse, you can watch the sun move across the sky, while robots on tracks in the ceilings silently whisk cases of wafers from room to room. Inside the labs, however, there is a fervor and din from researchers and machines working to change the way we compute.
For years, Albany has been one of the world’s epicenters for semiconductor design and fabrication. But more recently, computer chip production has shifted east — primarily into Asia. The pandemic rocked supply chains for just about everything you can think of, with people working at a reduced capacity due to social distancing. That included chip production, as well as shipping devices around the world. But spinning up a new chip factory closer to where chips are needed is not a quick solution. And the answers to tomorrow’s problems won’t be solved as they were in the past.
The rise of the internet fundamentally changed how we use semiconductors. They’ve moved from just powering our computers to being integral to almost every aspect of our lives. A regular sedan can have thousands of semiconductors in it. Entire businesses are run off smartphones. A pair of AirPods are more powerful than the computers we originally used to send astronauts to the Moon. Without access to increasingly powerful chips, our lives will grind to a halt.
At the same time, there’s a growing demand for entirely new types of computing resources. For years, chip producers have continued to chase Moore’s Law, which states that the number of transistors on a microchip will double roughly every two years. But it’s not just about packing more processing power into a chip anymore — there’s a need to build chips built for specific purposes, such as running complex AI models and algorithms on quantum computers.
Demand for traditional computer chips needs to be met, but we also need to recognize that chips are more than just traditional computer processors these days. The recent passage of the U.S. CHIPS and Science Act aims to help on both fronts, providing funds to boost chip production and research.
To come out stronger from the hardships the pandemic caused, we need to address the chip shortage now, but we also need new chips — and new ways to design them. And part of that solution might be found in the quiet halls of the Albany NanoTech Complex.
Some of the earliest semiconductors, the ones made in the late 1960s, were fabricated with processes that resemble modern techniques. These semis had transistors that measured 20 micrometers across — for comparison, the average human hair is about 75 micrometers thick. Over the decades, researchers have looked for ways to keep Moore’s Law alive by shrinking the size of each transistor. By 2005, at the dawn of the mobile internet age, transistors had been shrunk down to about 65 nanometers — over 1,000 times smaller than the ones from the Sixties.
In the last few years, researchers from IBM’s Albany lab have hit a string of breakthroughs. In 2015, they unveiled the world’s first 7 nanometer (nm) transistor, at a time when the state of the art was 14 nm. They also coined the term “nanosheet” and unveiled this new device architecture to help continue shrinking the size of transistors. The 5 nm transistor followed, and by 2021, IBM unveiled the world’s first 2nm transistor.
We’re getting to the point where transistors can’t get much smaller — a single silicon atom is only about 0.2 nm after all. At these sizes, the standard laws of physics start to go out the window. “We’re firmly in the realm of quantum physics,” Mukesh Khare, IBM Research VP of Hybrid Cloud and head of the Albany lab, said.
These breakthroughs end up in the sorts of chips we rely on every day, whether that’s through third-party chip producers who rely on IBM’s innovations to power their own products, or IBM’s own hardware. The newest IBM z16 system runs on an AI inferencing processor designed in Albany called Telum that gives businesses the power to run AI models while transactions are in progress. A bank can now detect whether a charge is likely fraudulent as the credit card is being swiped.
“The work we do here at Albany NanoTech is always focused towards product,” Khare said. “We want to make sure that the technology we develop ends up in our product, and our partners’ products.”
It’s not just research on the semiconductors themselves that makes Albany unique. Researchers are working to improve every aspect of the chipmaking process, from the tools needed to etch chips, to the processes needed to package all their parts together. When first unveiling 7 nm chips, IBM said that the only way to produce these at scale would be to master a complicated process called extreme ultraviolet (or EUV) lithography. Lithography is a common way of etching out areas of a material that sits on top of a silicon wafer, with the etched channels being where the circuitry sits in the chips.
To get down to the ridiculously narrow level of precision needed for a 7 nm, IBM showed that you need to use EUV machines, like the giant one that sits in NFX, or NanoFab Extension, the newest “clean room” lab at Albany NanoTech. It’s so large, it took three jumbo jets to bring the machine’s 100,000 pieces to the facility, and the 180-ton machine sits permanently attached to a crane in NFX to move it when it requires maintenance.
Clearly, having the space and resources to house such a machine is not something everyone can do, and it’s part of what makes Albany so special.
“When we announced 7 nm, we were the first team of companies who told the world that EUV technology is required for successful 7 nm implementation,” Khare said. “And here we are, EUV is part of 7 nm technology. We developed the core technology that the world is using.”
“The work that we do here, now, will allow us to have better-performing devices in a decade or so,” Luciana Meli, the manager of IBM’s lithography research at Albany, said.
The ubiquity of smartphones, other connected devices, and the march of cloud computing means that chips are likely only going to become more integral to our lives. But they’re also not going to be the only types of chips the world needs in the future.
We’ve seen an inkling of what’s likely going to be coming next. Over the last decade, as AI has moved out of academic institutions into real-world uses, graphics processors have moved to the fore. GPUs, chips designed to render graphics on computers, have been repurposed by AI practitioners to train and run models. CPUs are designed to handle myriad different tasks at once. GPUs, however, are optimized to take on complex tasks in parallel, making them great for carrying out the complicated mathematical equations needed to render graphics, or processing the massive datasets needed to train AI models.
But GPUs weren’t designed for running AI workloads and are inefficient as a result. This is one of the many challenges IBM is aiming to solve in Albany. One division, the AI Hardware Center, is experimenting with several different AI chips, from powerful digital devices to analog devices that could one day process information in much the same way as our brains do.
“There will be a time when we can't just throw more compute at the problem, you can't simply solve it in that way,” Nicole Saulnier, who runs AI hardware integration and materials research at the center, said. “The traditional ways are too computationally intensive; they consume a lot of energy.”
Several labs throughout the NanoTech Complex are focused on designing and testing the chips of the future. In the Heterogenous Integration lab, researchers work on the ways that future systems, composed of devices like CPUs, memory, and AI chips, interact and work together as efficiently as possible. This unassuming lab, dotted with powerful microscopes, drill presses, and industrial-size fridges, aims to speed up the flow of data through a chip to reduce latency in a computer chip.
Take the Telum processor, for example, which was designed in the AI Hardware Center. One reason it can run AI algorithms so quickly is that it’s integrated into the CPU of the IBM z16 system. Instead of the AI processor being installed near the CPU on the motherboard, having to pass information back and forth on miniscule circuitry, it’s integrated into the CPU architecture. This dramatically increases efficiency — much the same way that visiting someone downstairs from you takes less time than someone down the road.
Another lab, the Offline Wafer Testing lab, is where chip designs go to be tested to their limits. Many devices, from smartphones down to copper wiring, are stress-tested before they’re sold to consumers. You can’t realistically test multiple chip designs for defects over years and years. So in this lab, filled with machines that can heat up or chill chip wafers to extreme temperatures, researchers can quickly find and fix potential flaws.
Wafers head here from all over the NanoTech Complex. Off the corridor leading to NFX, cases of wafers to places like the testing lab, where researchers check whether the chips designs being created in the amber-hued clean rooms are viable. Albany is just one location where IBM researches the future of chips, however: Across IBM’s research facilities around the globe, researchers working on every aspect of the future of AI chips, from designing AI cores, to developing AI algorithms and mapping neural networks onto physical chips. A lot of the earliest work is done in smaller IBM Research facilities in Yorktown Heights, New York, and in Zurich, Switzerland, where their best ideas are sent to Albany for fuller testing and working on ideas at a rapid clip.
“We couldn't necessarily do what we're doing without the input from the labs, but in those facilities, it's very hard to do things at scale,” Saulnier said. “We can process wafers much faster and get a lot more cycles of learning a lot faster at this facility here.”
The ability to scale quickly allows researchers at Albany to think about what computing hardware will look like well into the future. “Some projects are focused towards what we need in the next two to three years, the digital part of our AI Hardware Center,” said Khare.
But then, Khare added, there are longer-term projects that are attempting to mimic the way our own brains compute. “Our brain is not digital — if you want to learn from how we make decisions, we have to learn how we create analog systems,” Khare said. “So we have both a near-term roadmap as well as longer-term roadmap addressing the type of workload needs that we understand today, and the workload that we will hopefully see in the future as the technology evolves.”
What’s happening in Albany these days isn’t entirely a new phenomenon, even if the technology is so cutting-edge.
By the early 1800s, Albany was one of the largest cities in the country, thanks to its easy access to New York City via the Hudson River, and the construction of the Erie Canal connecting the city to the Great Lakes to the west. Albany helped supply raw materials and finished goods to much of the growing United States. These waterways meant factories could spring up to produce everything from gloves to books and beer.
Albany also became a center of scientific breakthroughs. One of the first electric motors, using electromagnetic self-induction (a phenomenon that eventually became the DC motor), was invented in the city by Joseph Henry. The precursors to major technical education institutions, Rensselaer Polytechnic Institute (RPI) and the State University of New York, Albany, were both founded in the mid-1800s to meet the demand for technical talent. (Another, the State University of New York Polytechnic Institute, sprung up about a century later.)
But after World War II, much of the manufacturing and industry that put Albany on the map moved elsewhere. It wasn’t until the 1990s that tech started to return to the area. The Hudson River corridor north of New York City came to be known as “Tech Valley,” with IBM Research’s headquarters in Yorktown Heights to the south, and new startups forming in and around Albany to the north.
RPI and other nearby universities now offer technical training in chip fabrication and validation. The infrastructure that made Albany a production powerhouse in the past could lure talent from Boston and New York City, where the cost of living is much higher. The NanoTech Complex is now just one of several cutting-edge chip facilities in Albany’s Capital Region.
When the Albany NanoTech complex opened in 1997, there was just one building with 4,000 sq. ft. of clean room space. Now, there’s over 130,000 sq. ft. of clean room space across the six buildings on campus. Alongside SUNY Polytechnic Institute researchers and students are employees of companies from around the world, including GlobalFoundries, Samsung, Applied Materials, Tokyo Electron (TEL), ASML, Lam Research, and of course, IBM.
The environment fosters a near-instant feedback loop between researchers designing and testing chips, those making the machines needed to do all that, and the companies that will eventually mass-produce those chips. A great example is the recent breakthrough that IBM Research and Tokyo Electron made in developing a new process that simplifies 3D chip stacking technology.by stacking chip parts on top of each other, rather than side by side, potentially saving space and boosting the processing power of traditional devices.
“The future has never been brighter for IBM and for TEL, for the work we do here together in this ecosystem,” Alex Oscilowski, the president for TEL’s US research, said. “For the industry, we're really in unprecedented times, the technical challenges are at a level that they've never been at before.”
It’s a measure of their success that researchers are beginning to run out of space a. Plans are underway to build a mirror image of the newest NanoFab Extension building, coincidentally called NanoFab Reflection, that will bring another 50,000 sq. ft. of clean room space for IBM, Tokyo Electron, Applied Materials, and other tenants.
The New York Center for Research, Economic Advancement, Technology, Engineering and Science (CREATES), a partnership between the state and SUNY that operates the NanoTech Complex, put out a request for proposals earlier this year for the new building. The passage of the CHIPS and Science Act may lead to further expansion soon.
Beyond AI workloads, other computing demands lie on the horizon. In the last decade, quantum computers have moved rapidly from research prototypes to solving a variety of real-world problems. It won’t be long before new types of computing resources will be needed in an instant, delivered by the cloud. New use cases require their own types of chips, and being able to research what’s next, and what’s next after that — along with being able to produce the end results in the U.S. — has the power to be transformative.
“Semiconductors have moved to the point where they're not only ubiquitous, but they're essential for the things we all do every day and for how we get work done in the future,” TEL’s Oscilowski said. “When you look at the potential that's out there, and you look at the capability that we have here, with the CHIPS and Science Act, has the possibility to take what we do here to a whole new level.”
Within the walls of the Albany NanoTech Complex, there’s a sense that whatever lies ahead can be greeted with open arms. Inside those white walls and green tinted windows, away from the din of mobile phones, cars, ATMs, kitchen fridges, and airplanes, as the wafers whisk by, researchers are quietly inventing what’s next.