Abstract
Here we report on the fabrication and characterization of dense arrays of up to 900 microlasers for the implementation of a photonic hardware platform for reservoir computing [1, 2], a powerful concept of efficient neuromorphic data processing in which the input data gets transferred into a random and fixed recurrent network of coupled neural nodes. The network projects the input information into the high-dimensional reservoir space, which then can be processed simply by a trained readout system. This leads to interesting applications in the field of ultra-fast pattern recognition, or the prediction of chaotic temporal sequences [2]. To realize this interesting concept with nanophotonic nodes, we present a hardware platform based on a dense array of semiconductor micropillar lasers, which can be optically coupled and injected within a fully parallel concept. Our scheme is implemented by a diffractively multiplied external resonator, which redirects the emission of individual laser nodes in the array to their neighbouring nodes [3]. To enable this scheme the pitch of the laser array must be precisely adjusted, and most importantly their spectral homogeneity needs to be on the order of the emission linewidth of the cavity mode.
© 2019 IEEE
PDF ArticleMore Like This
Afeeqah Amalina Haya, Chihiro Sugano, Kazutaka Kanno, and Atsushi Uchida
JW3A.95 Frontiers in Optics (FiO) 2019
Lorenz Butschek, Akram Akrout, Evangelia Dimitriadou, Marc Haelterman, and Serge Massar
JW3A.122 Frontiers in Optics (FiO) 2019
Shi Li, Sourav Dev, Simon Ohlendorf, Kambiz Jamshidi, and Stephan Pachnicke
M4C.4 Asia Communications and Photonics Conference (ACP) 2019