Robust Architecture for Programmable Universal Unitaries

Abstract

Experimental implementation of a quantum computing algorithm strongly relies on the ability to construct required unitary transformations applied to the input quantum states. In particular, near-term linear optical computing requires universal programmable interferometers, capable of implementing an arbitrary transformation of input optical modes. So far these devices were composed as a circuit with well defined building blocks, such as balanced beamsplitters. This approach is vulnerable to manufacturing imperfections inevitable in any realistic experimental implementation, and the larger the circuit size grows, the more strict the tolerances become. In this work we demonstrate a new methodology for the design of the high-dimensional mode transformations, which overcomes this problem, and carefully investigate its features. The circuit in our architecture is composed of interchanging mode mixing layers, which may be almost arbitrary, and layers of variable phaseshifters, allowing to program the device to approximate any desired unitary transformation. arXiv:1906.06748v1 [quant-ph]

On-chip Reconfigurable Optical Neural Networks

“…Our proposed DONN is first trained on a conventional computer (see Supplementary Information Section 4 for details). To investigate the impact of fabrication errors on the performance of the DONN, the transfer matrix of MMI W used in training is replaced with a parametrized matrix, = (1 − α) + α , where R is a Haar-random perturbation, and α stands for the degree of error (0 < α < 1) 19 . The test set's recognition rate as a function of α is calculated using this matrix T and shown in Fig.…”

“…In this work, we propose an integrated, coherent diffractive optical neural network (DONN) based on a series of cascaded multi-mode interference (MMI) 18 structures. Our architecture can allow for an almost arbitrary transfer matrix out of the continuous unitary space 19 , which is crucial for in-situ training (See Supplementary Information Section 1 for details). This DONN encodes the input data in an optical phase domain.…”

Optical Implementation of 2 × 2 Universal Unitary Matrix Transformations

“…[30] b) BB architecture reported in refs. [16,[19][20][21][22][23]25,[32][33][34] and resulting SU(2) processor. [25,34] The rotations of the Bloch sphere that cannot be generated by each SU (2) processor are inset at the right of the figures.…”

Photonic Matrix Computing: From Fundamentals to Applications

“…After that, Saygin et al made an in-depth analysis of this integrated MPLC architecture, in which the flexibility and robustness of this architecture have been thoroughly discussed. The transfer matrices of the static MMI blocks can be randomly chosen from a continuous class of unitary matrices without sacrificing the quality of approximation for the target unitary transformation, making this scheme insensitive to errors [ 25 ]. The integrated MPLC matrix core provides an alternative viable solution to decompose large unitary matrices into small ones with high flexibility and robustness, and thus optical diffractive neural networks are possible to build on a chip based on this method.…”