Duration: 6 months
Space missions increasingly require autonomous and efficient on-board data processing to overcome challenges such as limited communication bandwidth, high latency, and the need for real-time decision-making. While traditional artificial intelligence (AI) algorithms have demonstrated remarkable performance in these tasks, their high power consumption and computational demands make them less suitable for resource-constrained environments like space [6]. This proposal aims to advance satellite edge computing by developing a neuromorphic processing module that interfaces a RISC-V core with Spiking Neural Networks (SNNs). Inspired by the human brain's efficiency and adaptability, SNNs process information through intermittent spikes, activating neurons only when inputs exceed a threshold. This event-driven approach drastically reduces power consumption, specifically for continuous data streams, while maintaining high computational efficiency, making SNNs suitable for space applications [1][6]. Unlike conventional deep learning models, such as Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs), SNNs encode information in a time-dependent manner, enhancing robustness against radiation-induced errors—a critical requirement for space systems [2]. This activity will focus on adapting an existing RISC-V core to interface with a custom-designed SNN. Thanks to the high degree of customization offered by the RISC-V architecture and its open-source nature, I aim to develop an integrated module that combines a traditional processor with an SNN-based accelerator. To enable efficient control and communication between the processor and the SNN, the project will also explore the development of custom instructions for specific SNN tasks. The module will be validated through a practical use case to assess its performance in terms of computational efficiency and radiation resistance, ensuring its suitability for deployment in the challenging conditions of space.
