“Frontier’s record-breaking performance will ensure our country’s ability to lead the world in science that improves the lives and economic prosperity of all Americans and the entire world,” Secretary of Energy Rick Perry said in a statement. “Frontier will accelerate innovation in AI by giving American researchers world-class data and computing resources to ensure the next great inventions are made in the United States.”

Cray CEO Pete Ungaro said that at 1.5 exaflops, Frontier will have the combined power of the current top 160 supercomputers in the world. An exaflop is equivalent to a quintillion calculations per second. If everyone on Earth carried out one calculation per second, it would take us more than six years to perform as many calculations as Frontier will be able to in a single second.

Oak Ridge National Laboratory director Thomas Zacharia noted that when it’s activated, Frontier will be the “most powerful AI machine anyone has seen up until that point in time.” He also said it will be “by far the single most expensive machine that has ever been procured,” with the contract for Frontier’s system and technology worth more than $600 million. Oak Ridge also houses Summit, which is the current supercomputer record holder.

Frontier will use Cray’s Shasta supercomputer system and Slingshot interconnect, as well as AMD’s GPUs and custom EPYC processors. The companies are working together on programming tools to help developers harness the supercomputer’s power, while Cray, AMD and Oak Ridge jointly developed a custom environment for the hardware. Frontier will take up almost the same space as 2 basketball courts, include 90 miles of cabling and have enough network bandwidth to download 100,000 HD movies in a single second.

Scientists and engineers will be able to use Frontier for an array of projects, including climate change forecasting, predicting the path of hurricanes and developing new pharmaceuticals. They’ll also be modeling fusion energy sources, simulating combustion engine systems and conducting cosmology simulations for the distribution of billions of galaxies.

It remains to be seen whether Aurora or Frontier will be the first exascale supercomputer to be activated in the US. But there’s plenty of competition facing the country in the exascale sphere. China might pip the US to be the first with such a supercomputer, as the National Supercomputing Center plans to build one by late 2020 or in the first half of 2021. Japan aims to turn on an exascale machine in 2021, and Europe may have its own exascale systems up and running in 2022 or 2023.

“The idea is everything has its own heartbeat,” said Shalin Shah, the lead author of the paper who is a Ph.D. student in electrical and computer engineering at Duke University. The technique invented by Shah, along with Duke computer science professor John Reif and postdoctoral researcher Abhishek Dubey of Oak Ridge National Laboratory, would allow researchers to identify thousands of molecules at a fraction of the cost of other methods. Unlike dyes or stains, the temporal barcodes wouldn’t fade over time.

As researchers found molecules they wanted to study, they tweaked the length or the number of repeating sequences on one of their strands. When that strand collided with a free-floating strand, previously stained with a fluorescent dye, it would briefly light-up. Each molecule ended up with its own unique flashing pattern, which acted almost like a fingerprint or barcode.

In an email to Engadget, Shah compared molecules to old black and white films. “You want to give it color, that’s what our temporal barcodes can do,” said Shah.

Using computer simulations, the researchers found that they could distinguish as many as 56 different molecules with their own unique flashing barcode. But if scientists opted for different colors of fluorescent dyes, the researchers think that number could increase to thousands.

What’s next for the technology? Shah told Engadget that the researchers at Duke want to use artificial technology to scale up their technology even further. Rather than dealing with the daunting task of manually identifying thousands of temporal barcodes, the scientists want to figure out how they can use AI to automatically tell them apart. They also hope to apply their method using cellular or protein components.