Experimental boson sampling

Universal quantum computers¹ promise a dramatic increase in speed over classical computers, but their full-size realization remains challenging². However, intermediate quantum computational models^{3, 4, 5} have been proposed that are not universal but can solve problems that are believed to be classically hard. Aaronson and Arkhipov⁶ have shown that interference of single photons in random optical networks can solve the hard problem of sampling the bosonic output distribution. Remarkably, this computation does not require measurement-based interactions^{7, 8} or adaptive feed-forward techniques⁹. Here, we demonstrate this model of computation using laser-written integrated quantum networks that were designed to implement unitary matrix transformations. We characterize the integrated devices using an in situ reconstruction method and observe three-photon interference^{10, 11, 12} that leads to the boson-sampling output distribution. Our results set a benchmark for a type of quantum computer with the potential to outperform a conventional computer through the use of only a few photons and linear-optical elements¹³.