Reliable power is central to modern living and economic competitiveness, but concerns have grown that the increasing power demand from data centers may put serious strain on America’s electrical infrastructure. Dr. David Estrada made the case that, rather than simply seeking to add more power to the grid, we can solve the energy problem by rethinking the way semiconductors are made.
“All the nuclear reactors in the world won’t be able to supply the data centers by 2030. This is a fundamental materials problem at the transistor level.”
Mr. Roger Brown
Professor and Associate Director, Center for Advanced Energy Studies,
Boise State University
Demand for electricity is exploding at home in the United States, thanks in large part to increased usage by data centers. This process began when the economic potential of big data became apparent and was supercharged by the sudden rise of artificial intelligence in the past three years. As data centers of incredible size came online, and as the reach of AI stretches further, the strain they will place on the country’s grid will increase. Given that an AI query uses 10 times more energy than a Google search, increased reliance on AI is placing incredible strain on our energy infrastructure.
The solution to the massive problem of energy usage by data centers may lie in the microscopic problem of better materials for semiconductors. While it is easy to see the strain placed on the energy system as one of supply, Estrada proposed that it may be more effective to consider it as a demand problem. The reason that data centers require so much energy is because of the fundamental limitations of the way modern computers are constructed. The base material for modern semiconductors is silicon, cut into wafers, on which computing components can be built. But silicon as an element has fundamental material limitations; while billions of dollars of investment and decades of research has optimized silicon-based computing components, we are approaching the limit of how much energy we can put through a silicon semiconductor before it melts. These limitations mean that data centers are required to use more computing components at the same time to meet their needs, driving up their energy usage.
Estrada suggested that the limitations of silicon may soon be surpassed by a new class of computing components built out of new materials. Dubbed “2D” materials, these new creations, only a handful of atoms thick, are far more energy efficient. While these materials are in development, if commercialized the power savings may forestall the need for a large-scale buildout of additional grid capacity.
Much of the work to develop these new materials is being done in the Mountain West, specifically at Boise State University. The school has invested in a new machine – the first of its kind at a university – that will allow it to test new materials far beyond what was previously possible. Traditional methods of testing allow the creation of materials made up of two or three elements; the new facility at Boise State will allow Estrada and his team to use up to eighteen, allowing researchers at Boise State explore a host of new possibilities for 2D materials, including designing components that survive in harsh conditions like inside of nuclear reactors.
