The adoption of optical computing in Artificial Intelligence (AI) systems offers a promising solution to overcome the limitations of conventional electronic hardware in space applications. Building on our previous theoretical design of a hybrid optical-electronic Vector-Matrix Multiplication (VMM) architecture [1], this work details the construction and evaluation of a Proof-of-Concept (POC) demonstrator. The proposed system utilizes a Mach-Zehnder-Interferometer Network (MZI-Network) for VMM and ring modulators or Variable Optical Attenuators (VOAs) for optical vector encoding. A custom payload Printed Circuit Board (PCB) was developed to interface conventional electronics with optical components, supporting future integration into nanosatellite platforms such as the TU-Berlin innovative neXt generation satellite bus (10 kg) (TUBiX10). To assess the feasibility of the POC, a Convolutional Neural Network (CNN) trained on the Modified National Institute of Standards and Technology (MNIST) dataset was partially executed on the optical hardware, demonstrating the successful implementation of optical convolution. Experiments on a limited subset of MNIST achieved a classification accuracy of 93.3% using the optical system, compared to 96.6% on conventional hardware. While preliminary, these results illustrate the potential of optical AI accelerators for space applications. Although the current laboratory setup is not yet optimized for space deployment, this work marks a significant step toward realizing compact, low-power, and radiation-resilient optical computing systems for future spacecraft.
