POSTER: A Reliable Multi-FPGA RISC-V Based Cluster for Space AI Inference

The growing interest in space oriented AI inference is rapidly increasing the demand for computing platforms that are both high-performance and reliable. In this context, SRAM-based FPGAs are widely adopted thanks to their flexibility, reprogrammability, and the ability to integrate heterogeneous computing modules within a single device. In this work, we propose a versatile FPGA-based clustered architecture composed of multiple interconnected computing nodes. Each node integrates a lightweight RISC-V processor responsible for inter-board management and communication, running FreeRTOS, and a TPU-like accelerator to enable efficient on-board AI inference. The nodes are interconnected through high-speed optical links using the Aurora protocol, enabling a fully connected cluster that can distribute inference workloads across the system while supporting reliability-oriented recovery via partial reconfiguration.