Concept of an in-orbit AI-system based on optical computing

Abstract Machine Learning (ML), or Artificial Intelligence (AI) in general, is among today’s fastest-growing methods to handle complex or computationally intensive tasks. ML is commonly implemented with Artificial Neural Networks (ANNs) on conventional computer systems that can limit their full potential. Even with access to specialized hardware such as graphics cards or Tensor Processing Units (TPUs), the demand for more computing power constantly increases. Although these hardware requirements can be met for terrestrial applications, an extraterrestrial or in-orbit application is considerably more challenging. Additional requirements for energy budget, thermal control, and radiation resistance can usually not be met, especially for small spacecraft. The benefits of an AI system for fast onboard data processing would, however, be remarkable. An optical approach to this problem can potentially be the solution. Optical computers promise to be much more energy efficient and better suitable for the mentioned space requirements. An implementation of an optical computing device on a spacecraft has not been done and can be considered as a technological leap. This work, along with the project Optical Computing for Machine Learning in Orbit (OMLO) of the Technical University of Berlin (TU-Berlin), aims to specify and conceptualize such a system.