Abstract
Quantum computers promise exponential scaling in computational power by leveraging quantum superposition. A critical milestone in this field is "quantum advantage," the point at which a quantum system can solve a problem that is classically intractable. Boson sampling—the study of multi-photon evolution in photonic networks—is a leading candidate for reaching this milestone. However, traditional photonic platforms are limited by the physical scaling of photon numbers and circuit modes. In this work, we present "timestamp membosonsampling," a novel variant that utilizes the timestamp information of single photons as a computational resource. By implementing a self-looped photonic chip, we experimentally achieved multi-photon registrations up to 56-fold across 750,000 modes, accessing a Hilbert space of $10^{254}$. This approach provides a cost-efficient, scalable path toward quantum advantage in photonic systems.
Introduction
Demonstrating quantum advantage requires a system to operate in a regime where classical supercomputers can no longer simulate its behavior. Boson sampling has long been proposed as a path to this goal. In a standard boson sampling experiment, multiple photons are sent through a complex network of beam splitters (modes), and their output distribution is measured.
The primary obstacle to scaling these experiments is physical. Constructing chips with thousands of waveguides is an immense engineering challenge, and generating large numbers of simultaneous, identical photons is equally difficult. Most current platforms remain at scales that can still be simulated by classical clusters. To overcome these limitations, we propose shifting the complexity from the spatial domain to the temporal domain.
Methodology: The "Time-Loop" Architecture
Our approach, termed timestamp membosonsampling, is inspired by the concept of a memristor—a component whose state depends on its history. We utilize a 16-mode photonic chip integrated with a "self-loop" mechanism, where an output channel is fed back into an input channel.
Temporal Layering
By matching the delay of the loop to the repetition rate of the laser source (80 MHz), photons from one pulse can be held back to interfere with photons from subsequent pulses. This effectively transforms a single physical chip into a multi-layered temporal network.
High-Resolution Timestamping
To manage this temporal complexity, we employed a time-of-flight module with 64-picosecond resolution. This allows us to record the exact arrival time of every photon, treating the "time of arrival" as a discrete computational mode. Because photons can "jump" between temporal layers via the loop, the system creates a global quantum interference pattern across the entire timeline, rather than just repeating independent experiments.
Experimental Results and Scaling
The experiment was conducted using the "Zhiyuan" machine, which ran continuously for 74 hours, generating 9.45 Terabytes of raw timestamp data.
Massive Hilbert Space
The system successfully extracted 56-fold photon coincidence events. When accounting for trigger photons, this corresponds to 112-fold events. This experimental scale is equivalent to a scattershot boson sampling experiment involving 200,000 photons in a 750,000-mode interferometer. The resulting state space, or Hilbert space, is approximately $10^{254}$—a figure that vastly exceeds the number of atoms in the observable universe and is far beyond the reach of any classical simulation.
Validation
To ensure the integrity of the quantum interference, we validated the system using smaller, verifiable configurations. For 2-layer configurations, we achieved a distribution fidelity of 86.7%, confirming that the multi-photon interference was high-quality and consistent with theoretical models.
Discussion and Future Outlook
While this method provides a massive leap in scaling, it introduces new challenges in verification. As quantum systems reach these unprecedented scales, verifying that the output is not merely random noise becomes a classically intractable problem. Future research must focus on developing more efficient validation tests for large-scale coincidence events.
Additionally, physical losses remain a factor; each loop iteration increases the probability of photon loss. Future iterations could benefit from active loops or storage-based methods to mitigate these losses. Finally, while the current implementation uses a random unitary matrix, the introduction of on-chip modulators could make the system fully programmable, potentially enabling applications in molecular vibronic spectra simulation and graph theory.
Conclusion
Timestamp membosonsampling represents a significant shift in photonic quantum computing. By utilizing time as a free resource, we have demonstrated a scalable and cost-effective shortcut to the quantum advantage regime, providing a robust platform for future quantum information processing.
Summary updated