CRC 1459 - C02: Opto-electronic neuromorphic architectures

Basic data for this project

Description

In this project, we will develop adaptive opto-electronic networks using linear photonic crossbar arrays combined with non-linear dopant network processing units (DNPUs) for machine learning in materia. The DNPUs will provide multi-terminal non-linear activations which are individually trainable and thus enable novel material-based learning algorithms to be implemented in hardware. We will further operate our envisaged architecture in reverse mode using DNPUs as nonlinear input modules to photonic crossbar arrays. Based on such hybrid opto-electronic networks, we will create nanoscale matter systems with optical and electrical feedback, enabling learning capability.