The data reuploading trick, originally proposed for universal quantum computing, enables the universal approximation property. We recently extended this concept to achieve universal non-quantum photonic computing using practical photonic integrated circuits (PICs) composed solely of 50:50 beam splitters and phase shifters, eliminating the need for nonlinear photonic devices. In this approach, input data are repeatedly embedded as rotation angles. In this presentation, we explore strategies to enhance the performance of this method by increasing the number of layers and optical channels (lanes) in various configurations. For a classical two-dimensional, four-class classification problem of wavy lines, we evaluated a two-mode class-embedding circuit with output, a four-mode configuration, and four stacking methods of two-mode circuits with average pooling, alongside a baseline configuration using projection in a complex domain. The first three configurations demonstrated excellent accuracy. Additionally, we investigated the effect of shifting the order of input data layer by layer, which significantly improved performance in certain applications. These findings highlight a novel architectural approach to photonic neural networks enabled by data reuploading in PICs. This approach offers unique features that set it apart from traditional photonic
The data reuploading trick, originally proposed for universal quantum computing, enables the universal approximation property. We recently extended this concept to achieve universal non-quantum photonic computing using practical photonic integrated circuits (PICs) composed solely of 50:50 beam splitters and phase shifters, eliminating the need for nonlinear photonic devices. In this approach, input data are repeatedly embedded as rotation angles. In this presentation, we explore strategies to enhance the performance of this method by increasing the number of layers and optical channels (lanes) in various configurations. For a classical two-dimensional, four-class classification problem of wavy lines, we evaluated a two-mode class-embedding circuit with output, a four-mode configuration, and four stacking methods of two-mode circuits with average pooling, alongside a baseline configuration using projection in a complex domain. The first three configurations demonstrated excellent accuracy. Additionally, we investigated the effect of shifting the order of input data layer by layer, which significantly improved performance in certain applications. These findings highlight a novel architectural approach to photonic neural networks enabled by data reuploading in PICs. This approach offers unique features that set it apart from traditional photonic
The data reuploading trick was originally proposed for quantum computing to achieve the universal approximation property. In this paper, we introduce data reuploading to realize universal non-quantum photonic computing with practical photonic integrated circuits (PICs). We aim to comprehensively discuss the various advantages and implementation considerations of this approach. Our framework can eliminate the need of quantum squeezed lights, photon counters, and nonlinear photonics, which have been essential for enabling photonic neural networks in conventional configurations. Additionally, we explore ways to minimize the optical components by combining multiple functionalities into a single phase shifter, showing competitive performance when compared to using the same number of phase shifters, all without employing any nonlinear photonic devices. Considering these characteristics, our investigation into the use of PICs for data reuploading presents a novel architectural approach to realize photonic neural networks. This approach embodies unique features that distinctly set it apart from traditional photonic neural networks.
The growing computational demands of classical neural networks have intensified the search for energy-efficient and powerful computational alternatives. Quantum neural networks (QNNs) implemented on integrated photonic platforms offer a compelling avenue, offering exceptional computational power enhancements, with inherent programmability and scalability of integrated architectures. A critical challenge, however, is implementing the fundamental non-unitary and nonlinear activation function of QNNs within a linear quantum photonic system. Existing strategies, such as the adding ancillary qubits and measurement-based feedback or forward are constrained by high qubit resource costs, overhead devices, and poor cascadability. Here, we propose a novel deep photonic QNN with an expanded computational Hilbert space via input replication and mode expansion, which enables the realization of effective non-unitary and nonlinear activation on a linear programmable quantum photonic chip. This approach eliminates the need for physical ancillary qubits, measurement-induced qubit consumption and the measurement device burden, thereby significantly reduce resource costs. The fabricated chip integrates four high-quality entanglement sources and a programmable high-dimensional interferometric network, enabling a two-hidden-layer QNN that exhibits dimension-enhanced expressivity over the existing QNN architectures. We demonstrate its capabilities across diverse tasks, including nonlinear classification, image generation, and quantum Gibbs state preparation. This work establishes a scalable and efficient architecture toward practical quantum deep learning systems capable of tackling problems beyond the reach of classical computation.
The growing maturity of photonic integrated circuit (PIC) fabrication technology enables the high integration of an increasing number of optical components onto a single chip. With the incremental circuit complexity, the calibration of active phase shifters in a large-scale PIC becomes a crucial issue. The traditional one-by-one calibration techniques encounter significant hurdles with the propagation of calibration errors, and achieving the decoupling of all phase shifters for independent calibration is not straightforward. To address this issue, we propose a global calibration approach for large-scale PIC. Our method utilizes a custom network to simultaneously learn the nonlinear phase-current relations for all thermo-optic phase shifters on the PIC by minimizing the negative likelihood of the measurement datasets. Moreover, the reflectivities of all static beam splitter components can also be synchronizedly extracted using this calibration method. As an example, a quantum walk PIC with a circuit depth of 12 is calibrated, and a programmable discrete-time quantum walk is experimentally demonstrated. These results will greatly benefit the applications of large-scale PICs in photonic quantum information processing.
In recent years, quantum information has rapidly evolved as a field with transformative potential in science and technology. It provides secure encryption for communication, accurate emulation of practical physical systems, streamlined and accelerated machine learning models, and advanced computational powers that outperform classical supercomputers in solving complex problems. These advancements are all reliant on the superposition and entanglement properties of qubits. While various candidates such as trapped ions or superconducting qubits have been proposed to realize quantum technologies, photons have demonstrated a prominent advantage due to their robustness against decoherence and noise, high clock cycle rate, fast and easy measurement, and room temperature working environment. In addition, photonic integrated chip (PIC) technology has developed rapidly since the beginning of the 21st century. The number of devices integrated on a single chip has increased rapidly, and its complexity has also increased. Quantum computing combined with integrated photonics is expected to solve bottleneck problems in practical situations. In this thesis, based on the silicon-on-insulator (SOI) platform, an optical computing chip and a large-scale quantum computing chip with various practical applications are specifically studied. In the first part of the thesis, a novel integrated diffractive optical network is studied, and a scheme is designed to realize parallel Fourier transform, convolution operations, and application-specific optical computing. Due to the utilization of on-chip compact diffractive cells (slab waveguides), the footprint and power consumption of the proposed architecture is reduced from quadratic scaling in the input data dimensions required for Mach-Zehnder interferometer (MZI)-based optical neural network (ONN) architectures to linear scaling for the improved deep neural networks. A 10-fold reduction in both footprint and energy consumption and equal high accuracy with previous MZI-based ONNs was experimentally achieved for computations performed on the Modified National Institute of Standards and Technology (MNIST) and Fashion-MNIST datasets. This reduction in the resource scaling from quadratic to linear has a profound impact on the realization of large-scale silicon-photonics computing circuits with current fabrication technologies. A quantum photonic microprocessor based on Gaussian Boson sampling (GBS) that offers dynamic programmability to solve various graph-related NP-complete problems is then demonstrated. The system with optical, electrical, and thermal packaging implements a GBS with 16 modes of single-mode squeezed vacuum states, a universal programmable 16-mode interferometer, and a single photon readout on all outputs with high accuracy, generality, and controllability. The developed system is applied to demonstrate applications in solving NP-complete problems, manifesting the ability of photonic quantum computing to realize practical applications for conventionally intractable computations. The GBS-based quantum photonic microprocessor is applied to solve task assignment, Boolean satisfiability, graph clique, max cut, and vertex cover. These demonstrations suggest an excellent benchmarking platform, paving the way toward large-scale combinatorial optimization. Finally, this thesis explores the application possibility of large-scale PICs in molecular simulation. A new algorithm utilizing states with zero displacements (vacuum-squeezed states) coupling to a linear network is proposed to calculate molecular vibronic spectra. With a large-scale PIC, molecular vibronic spectra are accurately simulated for formic acid and thymine with high reconstructed fidelity (>91%) and for naphthalene, phenanthrene, and benzene under the non-Condon regime. Our experimental demonstration will open pathways toward conducting large-scale molecular quantum simulations, a task surpassing classical computers’ capabilities.
We experimentally investigated a single-qubit quantum classifier implemented on a silicon photonic integrated circuit, focusing on its performance under photon-limited conditions. Using the Data Reuploading method with layer-wise optimization via Sequential Minimal Optimization (SMO), input data were encoded into the photonic circuit, and classification was performed based on output detection probabilities. Heralded single photons, generated via spontaneous four-wave mixing in a silicon waveguide, served as the input states. Even when the average number of photon samples per input was reduced to approximately two, the classifier achieved nearly 90\% accuracy, provided that the training dataset was sufficiently large. The experimental results were consistent with numerical simulations, which also indicated that performance at low sample sizes can be improved by increasing the size of the training dataset. These findings demonstrate that photonic quantum classifiers can operate effectively with very few photons, supporting their practical feasibility for resource-efficient quantum machine learning on integrated photonic platforms.
In a single qubit system, a universal quantum classifier can be realized using the data reuploading technique. In this study, we propose a new quantum classifier applying this technique to bosonic systems and successfully demonstrate it using a silicon-based photonic integrated circuit. We established a theory of quantum machine learning algorithm applicable to bosonic systems and implemented a programmable optical circuit combined with an interferometer. Learning and classification using part of the implemented optical quantum circuit with uncorrelated two photons resulted in a classification with a success probability of 94±0.8% in the proof of principle experiment. As this method can be applied to an arbitrary two-mode N-photon system, further development of optical quantum classifiers, such as extensions to quantum entangled and multiphoton states, is expected in the future.