Stigmergic Electronic Gates and Networks
Software implementations of neuronal systems have demonstrated their great effectiveness in managing and analyzing large amounts of data. While performing various signal processing tasks such as image processing, artificial intelligence and deep learning, neural software has limitations that derive from the characteristic structural division between processing and memory. To overcome such limitations, computing research has moved towards the realization of neuromorphic hardware models. A complete neural elementary unit should be able at the same time to perform signal recognition and to memorize information. Both functions can be implemented in photonic gates based on soliton waveguides: using X-junction geometries, soliton-based devices can be addressed according to specific signal feedbacks and can maintain the achieved status by exploiting the refractive index plasticity of the host material. Solitonic X junctions can implement either supervised or unsupervised learning as well as memory using extremely compact and simple geometries.