How IBM Research built a lab for the future of computingAccelerated DiscoveryAIHybrid CloudQuantumSemiconductors

For almost the entirety of the Computer Age, the world has focused on making classical computer structures smaller, faster, and more efficient. At its simplest, the concept of how today's most powerful supercomputer can run computations in the blink of an eye isn't much different than how the first computers ran all those decades ago. Albeit on a far more modest scale.

In recent years, however, new ways of thinking have started to emerge. Quantum computing has moved from theoretical discussions to massive machines that are on the cusp of answering some of the world's toughest questions. AI can generate fantastical content and make pressing business decisions simpler, at scales previously impossible. And new ways of connecting and running classical supercomputers has put an imperceptibly massive amount of compute power at the hands of just about anyone who needs it.

Each of these computing paradigms - grounded in bits, neurons, and qubits - are groundbreaking on their own. But when they're used to complement each other, the results will be even more transformative. It's the concept of accelerated discovery that IBM Research has been exploring in recent years. There's a belief here that it's how we'll usher in the next era of computing.

It's why IBM spent the better part of 2023 building a new lab within the Yorktown Heights, New York headquarters of IBM Research. Its goal: to bring together the latest in quantum computing, AI processing, and hybrid cloud technologies. The new space is meant to be the physical manifestation of what's next in computing. It's a working lab, where scientists from different domains can work together to shape new ideas and partnerships that likely wouldn't have occurred by working alone in their own spaces.

Bringing the platforms together

There will be many new types of computations where having machines close together will be useful. Just from a practical perspective, being able to troubleshoot work on multiple systems in real time could make it easier for researchers to figure out how best to optimize their algorithms for the hardware they're working on. Similarly, trying to figure out how to optimize multi-cloud structures is a lot easier when everything you're working on is right next to each other. But there's also the issue of latency, according to Zaira Nazario, director of Science and Technology in the Office of the Director of IBM Research.

Many quantum computations don't need classical computers next to them to carry out their tasks - that can happen through the cloud. "There are some, however, where you need them very close to each other, because you need to be able to run things in a matter of hundreds of nanoseconds or microseconds," Nazario said. "The time scale and the classical computing requirements that need to be brought to quantum computing depend on the problem at hand, the needs of different types of developers, and the level of abstraction at which you're working."

"Those are the ones that we would like to discover - the bits, qubits, and neurons story - and for those, co-location of the systems becomes important," she said.

The space also serves to highlight how the nature of computing itself is changing. In the past, supercomputers and datacenters have primarily consisted of endless rows of server blades. Today, we're starting to see new chips and processors enter the fray, from GPUs and AI accelerators for handling the burgeoning AI workloads the world is using, and chiplets to tackle how we move beyond Moore's Law. We're seeing datacenter owners and national labs reconsider what high-performance computing (HPC) systems look like with the AI explosion - but this is just the start.

The datacenter of tomorrow

"What do data centers that support these big missions start to look like in the future? These facilities will have the three types of systems in order to support users," Nazario said. And these systems won't always be interconnected, either - if you're a large facility supporting many users, some will likely only be interested in access to quantum systems, some will want to run traditional HPC or AI workloads - and some will be after some combination of all three.

In the Think Lab, the team set out to build a space that could accommodate the modular new IBM Quantum System Two, along with a cluster of IBM's prototype AIU chips, and even future Z systems. The space isn't meant to look the same forever, but rather adapt to the needs of researchers and those who visit the lab, according to IBM Research Creative Director Mark Podlaseck. Tables and chairs can be wheeled away or pulled over to sketch something out, the AIU computing cluster is extensible, and there's even room for an entire second System Two to be installed if need be - even that system was installed under the expectation that it could be modified or expanded.

It's a vision that could be replicated at any institution that would see value in adopting these new forms of computation. "You could be a national lab, or a pharmaceutical company that has a large research division, and you would need different types of systems to support the simulations that you need to do to discover drugs, and you want that to be modular so that it grows with user need," Nazario said. "We also wanted to be modular so that is easier to upgrade as advances come in."

The lab even has space to adapt beyond the systems that are in there now. The AIU cluster that's been installed is working on real workloads for IBM right now, including hate, abuse, and profanity (or HAP) filtering for watsonx AI models, using less energy to do so than many similar AI computing systems would. The cluster is set up on standard PCIe cards, meaning new versions can be easily installed. And for all the systems installed today, they're meant to be easily worked on, Podlaseck said. The lab does not need to be shut down to do maintenance on the Quantum System Two, or the AIU cluster.

The AIU, first unveiled in 2022, is a prototype chip design where computation and memory are tightly intertwined on the chip to reduce the latency in for deep learning and AI computations. It came out of the research that went into the TelumCPU first built for IBM's z16 in 2021.

As AI gets embedded into ever more business processes, the team is working to take the insights from deploying Telum and the AIU for the next generation of Z systems, which will have a home at the Think Lab in the future. This includes the recently announced Spyre AI Accelerator.

The new lab is much about building the platforms for researchers to create what's next in computing as it is redefining the notion of what a datacenter looks like in the future. It's a combination of some of the most cutting-edge technology in the world, and a space that's meant to honor IBM's design history, including that of the building it's in.

Informed by the past to design the future

The IBM Research headquarters was designed by the storied architect Eero Saarinen and is filled with furniture designed by Ray and Charles Eames. Podlaseck said in designing the new lab, they wanted to "embrace as much as possible the infrastructure of the old building, its rawness and honesty."

When Saarinen was designing the office, he wanted to use the natural beauty in the area as inspiration for the researchers who would work in the office. Some 4,000 tons of local Westchester fieldstone were used inside and out in building the lab, and Saarinen put massive windows along the entire perimeter of the lab, to allow researchers constant refreshing nature breaks between their intensive work sessions. At the same time, he designed libraries, a large central cafeteria, and meeting nooks throughout the facility to ensure that people could, perhaps with a little serendipity, come across others who worked on completely different tasks and find new ways to be inspired. That's something that the new Think lab has tried to maintain within its design.

It was unclear whether the original concrete from 1961 would be stable enough to host these massive new systems, as any slight fluctuation in stability can throw off the extremely delicate quantum computations happening inside this System Two. Podlaseck said they did countless tests to ensure the space would be able to support everything that was envisioned for it, and in the end, the team was surprised how well the concrete and rock used in Saarinen's original designs have held up. The IBM Quantum System Two is clad with polished anodized aluminum and surrounded by 70/30 glass, creating the impression that the system exists within an infinity of reflections - and the vast multidimensional spaces that quantum computing operates within.

You get a similar sense of grandiosity when first walking into the lab today. Although the systems are monolithic and demanding of your gaze, Podlaseck and team have ensured that the space is welcoming - and befitting of the wider lab. The furniture that was added fits the Mid-Century era in which the lab was built. There's a record player off to the side that was spinning Coltrane when I came to speak to Podlaseck. You can make yourself an espresso, sit and listen to the quantum computer's rhythmic thrum between meetings. It's inviting, encouraging folks to drop in and see what's new, and perhaps make a new connection with a colleague in doing so. As former CEO Thomas Watson Jr. once said, "Good design is good business."

The lab is as much a testament to how IBM has approached building new technologies over its lifetime as it is a space to answer the latest burning questions in computing. And it will grow and evolve, much like IBM itself has over the decades, as we begin to unlock the answers to some of those questions.

"This is not a static lab - it was built as a lab, not as a showroom," Nazario said. "As our research evolves, the lab will change."