Some quantum computers might need more power than supercomputers

Large quantum computers may be able to solve problems that are impossible for even the best traditional supercomputers – but to do so, some of them might need much more energy than these supercomputers.

Existing quantum computers are relatively small, with most having fewer than a thousand components called qubits. They are also susceptible to errors during operation due to the fragility of these qubits. This makes these computers incapable of solving the economic and industrial problems they are supposed to excel at, such as helping with drug discovery. Researchers largely agree that truly useful quantum computers must have radically larger numbers of qubits and an ability to correct errors, making them fault-tolerant quantum computers (FTQC). But getting there remains a formidable engineering challenge, in part because there are several competing designs.

Olivier Ezratty of the Quantum Energy Initiative (QEI), an international organization, says one of the overlooked concerns when building large-scale FTQCs is their potential energy consumption. At the Q2B Silicon Valley conference in Santa Clara, California, on December 9, he presented preliminary estimates. Amazingly, several FTQC designs have exceeded the energy footprint of the world’s largest supercomputers.

The world’s fastest supercomputer, El Capitan, installed at the Lawrence Livermore National Laboratory in California, requires about 20 megawatts of electrical power, about triple the energy consumption of the neighboring town of Livermore, population 88,000. By Ezratty’s estimate, two FTQC designs, scaled up to 4,000 logical qubits or error-corrected, would require even more. The most greedy among them could need up to 200 megawatts of energy.

Basing his estimates on publicly available data, proprietary information from quantum computing companies, and theoretical models, Ezratty identified a broad spectrum of possible energy footprints for future FTQCs, ranging from 100 kilowatts to 200 megawatts. Notably, by Ezratty’s estimate, three FTQC models currently under development would ultimately require less than 1 megawatt of electricity, which is comparable to typical supercomputers used by research facilities. According to him, this spectrum could influence the evolution of the industry, for example by expanding the market for quantum computing if less energy-intensive designs came to dominate.

The big difference in projected power consumption primarily reflects the diversity of competing ways in which quantum computing companies build qubits and use them. In some cases, power consumption is driven by the need to keep different parts of the device cold, for example for some light-based qubits, where light sources and detectors perform less well when hot. Ezratty says this can be particularly energy-intensive. In other cases, such as qubits made from superconducting circuits, entire chips must be placed in giant refrigerators, while quantum computers based on trapped ions or ultracold atoms require power for the lasers and microwaves that control the qubits.

Oliver Dial at IBM, which makes superconducting quantum computers, says he expects the company’s large-scale FTQC to require a little less than 2 or 3 megawatts to operate. Dial says this is only a fraction of what is expected to be needed for large-scale AI data centers, and could be even lower if FTQC were integrated into an existing supercomputer. The team at ultra-cold atom quantum computing company QuEra estimates that its FTQC would require around 100 kilowatts, which is at the lower end of Ezratty’s spectrum.

Xanadu, which builds light-based quantum computers, and Google Quantum AI, whose quantum computers are based on superconducting qubits, declined to comment. PsiQuantum, which also makes qubits from light, did not respond New scientistrequest for comment.

Ezratty says there are also many costs associated with traditional electronics used to direct and monitor qubits, especially when it comes to FTQC where qubits are given additional instructions to detect and correct their own errors. This further complicates the situation because it means that the details of the error correction algorithms also contribute to the energy footprint of the devices. And then there’s the question of how long a quantum computer needs to run to complete an operation, since the energy savings from using fewer qubits could be countered if they had to run for longer.

To untangle all of these factors—the basic energy cost of making qubits, the cost of cooling and controlling them, and the cost and runtime of quantum software—the industry should develop standards and benchmarks for determining and reporting the energy footprint of its machines, Ezratty says. This is part of QEI’s mission. He says related projects are underway in the United States and the European Union.

In the same way that the entire quantum computing industry is still developing, Ezratty says his work is in its early stages and should lead to efforts to better understand FTQC’s energy consumption and build on that understanding to reduce it. “There are many technical options that could help reduce the energy footprint. »