AI is here, and it needs glass
Corning’s glass innovations help artificial intelligence enhance the way we work, learn, and live.
Bring up artificial intelligence in conversation and you may be met with a raised eyebrow.
AI is the simulation of human intelligence by “trained” computers – an impressive feat of technology decades in the making.
Its future is still unfolding, but AI is already making our lives easier: The facial recognition in your security doorbell, customized social media experiences, tailored playlists, and personalized shopping recommendations.
Growing bigger daily, AI can help in ways yet to be discovered, too. The technology can help doctors diagnose and treat disease, help manufacturers quickly spot and correct defects, and even enable self-driving cars.
To do all these things, however, AI needs Corning’s crucial component: glass.
Corning invents, makes, and sells revolutionary glass that pushes AI technology forward – namely, Corning’s internet-enabling optical fiber as well as glass components for the semiconductor industry.
So why does AI require so much glass?
Corning optical fiber allow for GPUs to be networked more closely together, enabling denser server spaces in data centers.
An exponential desire for computing
Let’s first understand the sheer amount of computing needed for AI.
With large language models that power systems like ChatGPT increasing in popularity – the AI text-generating website saw 1.6 billion monthly visits in July 2023 – the strain is on data centers to keep up with demand.
Data centers, a.k.a. the “brains” of the internet that live in large warehouses with rows of servers, essentially relay all our data and compute our every request. Now, AI is turbocharging that data usage, and data centers are expanding to meet the demand.
For example, GPT-3 launched in 2020 with 175 billion parameters. GPT-4, introduced in 2023, is estimated to have over a trillion parameters.
“That’s over 10 times larger than three years ago,” says Xavier Lafosse, Advanced Optics commercial technology director at Corning. “The technology cannot cope with this, as is.”
Lafosse, who works closely with the semiconductor industry (the industry that creates microchips that power our everyday devices), thinks Corning can unlock the conundrum. And it all comes down to the graphics processing unit (GPU).
A graphics processing unit can take many shapes and sizes, but one thing stays the same: the need for fast connectivity.
The glass that makes the world smarter
“GPUs are the star of the show when it comes to accelerating AI,” Lafosse says.
Regular computing uses central processing units (CPUs) – a basic unit of electrical circuitry that performs computing instructions stored in its memory, like arithmetic and input/output operations. CPUs are the core components of computers and have been since the 1950s.
But AI requires additional graphics processing units (GPUs). These are specialized processors essential for gaming, video editing, machine learning, and other applications that require high-quality graphics and effects. The power of GPUs lies in their ability to process many pieces of data simultaneously.
“Without our ultralow expansion glass for extreme ultraviolet lithography, people could not make these GPUs,” Lafosse says.
Extreme ultraviolet lithography is the primary approach to fabricating more advanced chips within the GPU. Corning helps companies make chips smaller – down to nanometer-sized features. For scale, a human hair is approximately 80,000 to 100,000 nanometers wide.
Within the GPU, you’ll find high-bandwidth memory, which stores instructions for the GPU’s computational tasks. Corning supplies glass carrier wafers that support the silicon memory wafers during manufacturing. Once the memory wafer is thinned to the appropriate size – say, 40 micron – the glass wafer is detached. Corning’s glass carriers also help support the interposer wafer, which connects the GPU to the memory.
Glass wafers can strengthen the structure of GPUs and enable better electrical and mechanical reliability.
“Glass can also serve as the actual substrates for GPUs,” Lafosse says. “We believe glass will help strengthen the structure of GPUs as they get bigger and enable better electrical and mechanical reliability, therefore enhancing performance.”
Lafosse points out the hang-up in the AI trajectory.
“The bottleneck is the speed of the communication between the GPU and the memory,” he says. This leads to the next order of Corning’s expertise…
Optical fiber carries high-speed data in highly dense server environments.
Optical fiber and the increased flow of data
GPUs need to be networked closer together so they can do their job properly. Optical fiber, the revolutionary glass material invented by Corning over 50 years ago, has already been proven for carrying high-speed data from GPU to GPU. And Corning’s fiber is famous for its ability to fit into highly dense server environments, allowing for more GPUs and therefore more processing.
Corning fiber paves the way for AI.
Corning experts estimate that data centers running AI large language models will require five times more optical connectivity than they have today. That means those data centers need more of the glass optical fiber that transfers data from place to place in the form of light signals.
“We have heard from data center operators that speed to build is their #1 challenge,” says Mike O’Day, vice president, Corning Optical Communications. “AI is happening much faster than anyone expected, and the race is on.”
Listen to Corning experts discuss AI as a major opportunity in the telecommunications industry (36:36) on the Bloomberg Intelligence podcast below.
Data centers need quick solutions for scaling up.
Where installers once spent hours on the surgery of connecting fiber networks in the data center, Corning’s plug-and-play solutions shrink that time. The EDGE™ Distribution System, for example, simplifies the deployment of cabling within the data center by consolidating dozens of cords into a single assembly. This not only saves time and space, but it has the sustainability benefit of using less material.
Also key are new, smaller fibers such as the Corning® SMF-28® Contour Optical Fiber with superior bendability and low signal loss.
“These new fibers, in turn, enable a cable revolution,” O’Day says, “with the ability to create much higher cable density and fiber counts while reducing the cable diameters.”
As data centers continue to scale up, Corning glass remains at the forefront of our digitally expanding world.
“Artificial intelligence changes the game for data centers,” O’Day says. “At Corning, we’re excited to be working with the world’s largest hyperscale cloud operators to lay the fiber-rich foundation required for AI.”
Whether it’s enabling AI-focused data centers with fiber, helping etch the chips needed for advanced computing, or serving as a support for vital GPU components during manufacturing, glass is core to our future with AI.
And not even AI can predict what glass will enable next.
