Ramon.Space’s RC64 is a radiation-hardened, high-performance co-processor designed for Space applications, offering low power consumption and high processing efficiency. RC64 is employed in DSP computations and in NVM storage control of NuStream. However, AI testing on RC64 has been limited due to the absence of a generic AI framework. The Klepsydra AI inference engine, combined with its offline optimization tool, the Streaming Distribution Optimizer (SDO), addresses this gap by applying Convex Optimization techniques. This approach optimally allocates computing and co-processor resources, effectively minimizing latency and power consumption while maximizing data throughput, resulting in a more efficient and performant solution for spaceborne AI applications.

The ENORMITY industrial research project focuses on expanding Klepsydra AI and SDO for AI execution to RC64, providing a fully integrated hardware/software solution for Space-qualified AI inference. Its key objectives include: (1) integrating Klepsydra Streaming, AI, and SDO with RC64 for enhanced AI execution and storage, (2) enabling dynamic AI model loading for real-time execution without manual coding, and (3) demonstrating the European Space Agency’s OBPMark-ML AI benchmark on this optimized platform.

Additionally, Klepsydra has extended its AI software to support the GR740 processor, enabling a powerful combination where GR740 serves as the main onboard processor and RC64 as an AI accelerator. This fully radiation-hardened, high-performance AI-capable processing subsystem will support a wide range of critical Space applications, advancing the future of intelligent satellite operations.