A laser pulse lasting one and a half nanoseconds hits a nonlinear crystal, and a stream of squeezed light is born. That is how an experiment on the Jiuzhang 4.0 quantum processor in China begins. In 25.6 microseconds, this maze of mirrors and optical fibers produces a sample that would take the most powerful classical supercomputer, El Capitan, more than $10^{42}$ years to generate. The age of the Universe is only 1.4×$10^{10}$ years.
Gaussian boson sampling is a quantum experiment with entangled photons that turns a photonic processor into a random number generator with unimaginably complex statistics. Each photon, as it runs through a maze of 8176 modes, effectively chooses one of countless paths, and its choice depends on all the others. A classical computer has to enumerate these paths one by one, and their number grows exponentially. Jiuzhang 4.0 uses hybrid spatial-temporal encoding: photons are spread out both in spatial position and in arrival time, which yields a cubic growth in connectivity: 16×16×16 = 4096 possible combinations with a linear growth in physical resources.
Previously, the main bane of photonic quantum computing was loss: photons were lost in optical elements, giving classical algorithms a loophole: they could simulate not the ideal system but a 'leaky' one, which is much easier. In the new work, losses have been squeezed down to 49% across the entire circuit: the squeezed light sources have an efficiency of 92%, the detectors 93%, and the phase is stabilized to an accuracy of λ/200. This has allowed, for the first time, scaling Gaussian boson sampling to 1024 input squeezed states and 8176 output modes, registering up to 3050 photon coincidences, more than an order of magnitude above previous records.
Comparison with the best classical simulators shows that none of them can reproduce the device's statistics. Even the recent matrix product state (MPS) algorithm, a way to compactly describe a quantum system, requires a bond dimension χ greater than 8×$10^{21}$, making the computation impossible. The estimated simulation time on El Capitan exceeds $10^{42}$ years, whereas the quantum processor produces a sample in 25.6 microseconds. The speedup is more than $10^{54}$ times. This is no longer a race, but a chasm.
Until now, it was unclear whether Gaussian boson sampling could be scaled to thousands of modes without photon losses negating the quantum advantage. Jiuzhang 4.0 has answered: yes, it can. By creating a hybrid spatial-temporal scheme with cubic connectivity and highly efficient sources, physicists have shown that with a total efficiency of 51% and 1024 squeezed states, classical algorithms, including MPS, cannot reproduce the experiment's statistics even in theory. This platform can now serve as a testbed for generating cluster states, entangled networks needed for measurement-based fault-tolerant quantum computing.
The next step is to use the same architecture to create massive three-dimensional cluster states. Hybrid spatial-temporal encoding already allows control of thousands of highly entangled modes, and further improvements in sources and detectors could push efficiency and photon numbers even higher. New classical algorithms targeting this specific architecture may also appear, so the race will continue, but now the quantum side has a solid margin of safety. The maze in which photons find their way in microseconds, while a classical computer would not have enough time in the lifetime of the Universe, remains insurmountable for now.
The number of possible photon configurations in the L1024 experiment is estimated at $10^{2461}$, which is larger than the number of atoms in the observable universe (about $10^{80}$).
The idea of quantum supremacy echoes Greg Egan's story 'The Halo', where characters use quantum computing to model consciousness. Here, though, the task is more down-to-earth but no less impressive: proving that a quantum machine can be faster than any classical one.