IBM has announced its largest quantum processor to date, as the company seeks to show it is on track to create a commercially-useful quantum computer by the end of 2023.
The new quantum hardware, which IBM is calling Eagle, has 127 qubits, which are the information-processing units of a quantum computer. This is a large enough cluster to perform calculations that cannot be made by traditional computers in a reasonable timeframe, the company said.
But the company said it had not yet done a benchmark demonstration to prove that the new processor can perform tasks beyond the grasp of conventional computers, saying only that the new machine is powerful enough that it should be able to do so.
Quantum computers are machines that use phenomena from quantum physics to process information. In a traditional computer, information is represented in a binary form, known as a bit. A bit can be either a zero or one. In a quantum computer, information is represented by a quantum bit, or qubit for short, that can be placed into a quantum state in which it can represent both zero and one at the same time.
Also, in a classical computer, all the bits in a computer chip function independently. In a quantum computer, the qubits are “entangled” with others in the quantum processor, enabling them all to work together to reach a solution. Those two properties give quantum computers, in theory, exponentially more power than a traditional computer.
But to date, quantum computers have been too underpowered—meaning they have too few qubits and those qubits cannot remain in a quantum state long enough—to pose a major challenge to traditional computers. In 2019, Google achieved a milestone called “quantum supremacy” in which it performed a simulation of a quantum physics problem that could not be carried out on a traditional computer. But, as important as that achievement was in the annals of computer science, it did not have any immediate business applications.
There are two main problems holding back today’s quantum computers: they don’t have enough qubits in most cases to perform calculations that would give them an edge on standard computers. What’s more, those qubits can only remain in a quantum state for very short periods of time (often just a few hundred microseconds.) And when the qubits fall out of a quantum state, errors creep into their calculations. These errors need to be corrected, either by using more qubits, or by using software, but exactly how to do so efficiently remains an unsolved problem.
IBM last year unveiled a roadmap for the emerging technology that would see the company producing a quantum processor with more than 400 qubits by the end of next year and one with at least 1,000 qubits by 2023. A quantum computer of that size ought to be able to perform many useful business applications, the company has said.
The company is one of dozens around the world racing to commercialize quantum technology. Other leading contenders including tech titan Google, and industrial giant Honeywell, which recently spun off its quantum computing division into a separate public company, as well as D-Wave Systems, Rigetti Computing, and IonQ. Microsoft also has a quantum computing effort, although it has suffered setbacks.
IBM’s Eagle processor has almost two times the number of qubits as the company’s previous largest quantum processor, the 65-qubit Hummingbird, that it debuted last year.
Jerry Chow, the manager of IBM’s experimental quantum group, said that the company was still working to benchmark the performance of the new Eagle processor. He said the company still was not ready to say how long the Eagle’s qubits can remain in a quantum state or the degree to which the qubits are entangled.
He also said that IBM was debuting a new metric for measuring quantum performance called circuit layer operations per second, or CLOPS for short. This stat matters because a quantum computer does not produce a single, accurate result for a calculation, as a classical computer does. Instead, the answer can vary each time the calculation is run. As a result, to reach an accurate solution, the same calculation needs to be run through the quantum processor hundreds or even thousands of times, with the distribution of results converging over time on an accurate solution. In other words, if you ran the same calculation 100 times, and 85 times it produced answer A, then A is the accurate solution, even though 15 times the quantum computer spat out answer B.
But Chow said IBM was not yet ready to release a CLOPS figure for the new Eagle processor. “This is again an area where we are in process of measuring,” he told Fortune.
Chow also said that IBM is making progress in increasing the coherence times of its ealier 27-qubit Falcon processor. It said the qubits in this processor could now remain in a quantum state for as long as 300 microseconds, about three times the median rate for most other qubits built using superconducting materials like IBM’s. (Other companies are pursuing different methods of creating qubits, including using lasers to trap ions, and using silicon-based processors, similar to the materials used in standard computer chips, to create qubits.)
The new Eagle processor will be accessible through a cloud-based connection to companies that are part of IBM’s Q Network of early quantum adopters by the end of the year. Most of these companies, which include the likes of Toyota, Wells Fargo and Delta Airlines, have been experimenting with quantum computers and running small proof of concept projects, but have not deployed quantum computers into any real business units.