Supervised learning of soliton X-junctions in lithium niobate films on insulator.

Abstract

In this Letter, the first implementation, to our knowledge, of X-junctions between photorefractive soliton waveguides in lithium niobate-on-insulator (LNOI) films is reported. The experiments were performed on 8 µm thick films of congruent undoped LiNbO3. Compared with bulk crystals, the use of films reduces the soliton formation time, allows more control over the interaction between the injected soliton beams, and opens a route to integration with silicon optoelectronics functions. The created X-junction structures show effective supervised learning, directing the signals propagated inside the soliton waveguides into the output channels highlighted by the control assigned by the external supervisor. Thus, the obtained X-junctions have behaviors analogous to biological neurons.