Surface Plasmon Polariton neuromorphic circuit with sigmoid activation function

In the last decades the availability of large amounts of data and the necessity of processing it efficiently has led to the rapid development of machine-learning techniques. But, unlike real neural tissue, traditional computing architectures physically separate the core computing functions of memory and processing, making fast, efficient and low-energy computing difficult to achieve. In order to overcome these limitations, the investigation toward the design of new fundamental building blocks of brain tissue has been triggered [1]. In this way the analysis and the processing of data can be made directly on small units (chips). Based on the results of this research, we have successfully represented the excitation of Surface Plasmon Polaritons (SPP) via a metallic grating. In order to minimize the losses related to the backscattering light, we have reduced the number of grating grooves to only one grove per grating. The other effective parameters for the coupling of SPP such as the angle of incidence, the thickness of the metallic layer, width, and depth of grooves has been optimized and as a result, the coupling efficiency enhanced around 3 times compared to the primary condition. The effect of the different dielectric materials on the properties of SPP also studied and all of the parameters including the SPP wavelength, propagation length, penetration depth for metal and the dielectric layer, and required groove width for optimum coupling has been successfully characterized for different dielectric materials. Here, we demonstrate a fully functioning all-optical nonlinear activation function, based on the saturable absorber property of graphene. The synaptic behavior of this saturable absorber material provides a passive saturable optical response for the SPP signal, whose trend gets a sigmoid dependence. All of the optical properties of the graphene have been adopted from the recent experimental works and the results of this simulation are closely consonant with those results. The demonstrated SA scheme can be used to construct various neuromorphic architectures with intrinsic and passive optical computation.