We propose the photonic architecture based on microring resonator (MRR) arrays for binary convolutional neural networks (BCNNs) accelerated computing. The MRR crossbar unit is used for computing weight {−1, 1} and the single MRR is for input {0, 1}. The computing parallelism is improved through wavelength division multiplexing. The photonic BCNN achieves 97.29% classification accuracy on the MNIST test set which is only 1.94% lower than the accuracy of the 32-bit neural network, and saves $32\times $ at memory usage. We analyze effects of input and weight encoding errors on the photonic BCNN. When the input or weight error rate is less than 0.01%, the test accuracy remains unchanged. We evaluate the performance of the photonic BCNN architecture considering optical loss, inter-channel crosstalk, operation frequency and device power consumption. The energy efficiency of the designed photonic BCNN architecture is 1.72 pJ/MAC, which is $4.80\times $ and $61.32\times $ better than the 8-bit and 16-bit architecture respectively. The photonic BCNN is promising to be used for edge computing.

A chip-level optical beam steerer is an inevitable choice for next-generation light detection and ranging (LiDAR). The research on optical phased array (OPA) is the most intriguing. However, the complexity of control and calibration speed limit the full potential as the number of channels increases. In this paper, an improved stochastic parallel gradient-descent algorithm combined with the Nesterov accelerated gradient method (NSPGD) is presented and applied in a 512-channel OPA. This algorithm can reduce the phase calibration time of large-scale OPA and demonstrates a better convergence performance than traditional SPGD. Compared with the traditional SPGD and hill-climbing (HC) algorithm, optimized convergence performance of NSPGD is shown. The side mode suppression ratio (SMSR) of over 10dB for 512-channel OPA is obtained with the NSPGD algorithm, and the convergence speed is twice that of traditional SPGD. In addition, a temperature-controlled OPA is also studied to stabilize the whole calibration system.

We have proposed a 128-channel silicon-based optical phased array (OPA) chip with an overall antenna, and a large two-dimensional scanning range of 104.0° × 17.6° was achieved.

The focal plane array (FPA) is a promising solution for light detection and ranging (LiDAR) that offers advantages such as low optical loss, low power consumption, and a simple system architecture. To enhance the ranging performance of the FPA, we propose a novel structure featuring a single antenna for transmitting the laser beam and an antenna array for receiving the return signal. The theoretical directionality of the antennas is up to 91.6%. Utilizing the above structure, we fabricated a 6-channel FPA chip and successfully demonstrated scanning of object shapes. This chip is scalable and offers technical support for the development of larger-scale solid-state LiDAR.

Due to the advantages of fast scanning speed, good stability, flexible pointing, high resolution, small size and low cost, optical phased array has become a potential key component of beam scanners. The effective aperture of optical phased array determines properties such as divergence angles and ranging distance, and how to achieve it has been of interest to researchers. In this letter, we have designed and fabricated a 1024-channel passive optical phased array chip. A small divergence angle (0.05 $^{\circ }\,\,\times 0.04^{\circ }$ ) has been achieved by adopting a Si ₃ N ₄ perturbation grating structure on Si waveguide. Based on this chip, we have built a coherent detection demonstration system and carried out relevant experiments with it.

Optical neural networks (ONNs) are getting more and more attention due to their advantages such as high-speed and low power consumption. However, in a non-ideal environment, the noise and low-bits control may heavily lead to a decrease in the accuracy of ONNs. Since there is AD/DA conversion in a simulated neural network, it needs to be quantified in the model. In this paper, we propose a quantitative method to adapt ONN to a non-ideal environment with fixed-point transmission, based on the new chip structure we designed previously. An MNIST hand-written data set was used to test and simulate the model we established. The experimental results showed that the quantization-noise model we established has a good performance, for which the accuracy was up to about 96%. Compared with the electrical method, the proposed quantization method can effectively solve the non-ideal ONN problem.