Reservoir computing has emerged as a lightweight, high-speed machine learning paradigm. We introduce a new optoelectronic reservoir computer for image recognition, in which input data is first pre-processed offline using two convolutional neural network layers with randomly initialized weights, generating a series of random feature maps. These random feature maps are then multiplied by a random mask matrix to generate input nodes, which are then passed to the reservoir computer. Using the MNIST dataset in simulation, we achieve performance in line with state-of-the-art convolutional neural networks (1% error), while potentially offering order-of-magnitude improvement in training speeds.
Technology to apply 53 Gbaud 4-level pulse amplitude modulation (PAM4) to each single wavelength is essential for increasing a transceiver’s communication capacity. An electroabsorption-modulator-integrated laser (EML) having the several merits of high output, a high-speed response, and relatively good chirp characteristics is a promising device to provide such high baud-rate operation. We have demonstrated uncooled 53 Gbaud PAM4 operation at temperatures ranging from 20°C to 80°C using a newly developed TO-CAN EML with a novel lead-pin-less structure. We have also demonstrated 400 Gb/s PAM4 signal transmission over 10 km of single mode fiber (SMF) with no penalty using a compact integrated LAN-WDM EML TOSA that we have developed employing stripline with an aperture on the FPC enabling electrical crosstalk to be reduced. The uncooled TO-CAN EML exhibited a transmitter and dispersion eye closure (TDECQ) for PAM4 of better than 1.4 dB and an extinction ratio of over 4.5 dB at temperatures ranging from 20C to 80C thanks to the introduction of the lead-pin-less structure to the TO-CAN header in order to reduce the impedance mismatch and improve the 3-dB bandwidth. Meanwhile, the 400 Gb/s EML TOSA exhibited TDECQs of better than 2.5 dB and extinction ratios of over 4.8 dB for all lanes, as well as achieving a low crosstalk penalty of less than 0.4 dB. We believe that the uncooled TO-CAN EML and the EML TOSA will become key components for next generation 100 Gb/s and 400 Gb/s transceivers for data center applications.