Sign In

Communications of the ACM

ACM News

Simulating a Faster Path to Quantum

Fujitsu's A64FX processor is used in Fujitsu's Fugaku supercomputer, ranked second on the most recent TOP500 list of the most powerful supercomputers in the world; it also is used in the company's quantum simulator.


In a bid to accelerate the development of quantum computing applications, supercomputer manufacturer Fujitsu has come up with what it says is the world's fastest 36-qubit quantum simulator.

Quantum simulators attempt to emulate quantum operations on a classical computer. While their speeds are slower than quantum machines, they can serve as a bridge technology before wider implementation of quantum computers, which have limited computing range and can be prone to errors.

The new Fujitsu machine operates on a 64-node PRIMEHPC FX 700 hardware system with the same A64FX processors used in Fugaku, the Fujitsu supercomputer that was knocked off its leading position in late May in the new TOP500 ranking of the world's most powerful supercomputers.

The simulator has a theoretical peak performance of 3.072 teraflops in double-precision floating-point format calculations, and can run the open-source Qulacs quantum circuit simulator software at about twice the speed of other major simulators in 36-qubit operations, according to the company. The machine performs multiple calculations simultaneously using scalable sector extension (SVE) instructions. The SVE specification was incorporated in the A64FX, which Fujitsu developed with Britain's Arm.

"The technology that made this possible is based on the optimization technology of calculation and communication that we have cultivated through the development of supercomputers," said Hirotaka Oshima of Fujitsu Research's Quantum Laboratory. "We greatly reduced the amount of communication required for quantum computation by efficiently rearranging quantum information in a distributed memory system in accordance with successive quantum gate operations."

A Fujitsu arXiv paper related to the announcement describes the simulator, mpiQulacs, which runs on a new 64-node A64FX-based cluster system named Todoroki. Fujitsu researchers compared its performance to the Intel-QS, JUQCS-G and Qiskit Aer distributed simulators. "We show that mpiQulacs outperforms them for large-scale simulation on tens of nodes and achieves nearly ideal weak and strong scaling," the Fujitsu researchers concluded.

Francois Le Gall, a professor in the Graduate School of Mathematics at Nagoya University, said Fujitsu's claim that the simulator is twice as fast as rivals appears justified. Simulating large-scale quantum computing with classical computers including supercomputers is not feasible, he said, but for the next five to 10 years the scale of quantum computing will remain sufficiently limited that for practical applications, classical simulations of quantum computing will be likely possible, as well as cheaper than using a real quantum computer.

"Having such simulators is essential to start investigating immediately the potential and applications of medium-scale quantum algorithms that are not currently implementable with quantum computing technology," Le Gall said.

Simulators are critical to development of the quantum computing systems ecosystem, said Rod Van Meter, a professor in Keio University's Faculty of Environment and Information Studies; he is involved in a government project led by Qulacs developer Keisuke Fujii, one of the initiators of Qulacs, a Python/C++ library for fast simulation of large, noisy, or parametric quantum circuits.

"But the challenge with simulators is the same as the attractiveness of quantum computers," said Van Meter. "Absent both convenient symmetries in the quantum state and intelligent software, the storage and computation to represent the state of a quantum machine grows exponentially with the number of qubits. Ten qubits requires 16 kilobytes, 20 requires 16 megabytes, 30 requires 16 gigabytes, and 40 requires 16 terabytes. Tensor networks, which I think of as being spiritually similar to the concept of memoization developed in the 1960s for classical software, can serve as that intelligence."

Intel, which is developing a range of quantum applications including qubit (quantum-bit) devices and software architecture, said it would not be affected by the development due to its cloud-ready simulator. "Fujitsu's 36-qubit quantum simulator will not have an impact on Intel's quantum results and services," said Intel spokesperson Laura Stadler, because "Intel has its own open-source high-performance 45-qubit simulator that uses parallel distributed nodes. The Intel Quantum Simulator has been available for about five years."

Fujitsu is working with other companies on applications for its simulator. For instance, Japanese imaging and pharmaceutical company Fujifilm is using it to evaluate quantum algorithms that can simulate chemical reactions for the development of new materials.

Planning to improve the scale and speed of the simulator, Fujitsu aims to have a 40-qubit version that could be used for financial and drug-discovery applications available by September, according to Oshima.

Aside from discovering innovative new materials science, Fujitsu says it wants to use quantum computing technology to meet other kinds of social needs. For instance, the company claims its quantum-inspired algorithms have reduced traffic jams and lowered greenhouse gas emissions at the port of Hamburg.

Tim Hornyak is a Canadian journalist based in Tokyo, Japan, who writes extensively about technology, science, culture, and business in Japan.


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account