A new milestone has been reached for the NimbleAI.eu project! We’re proud to be a part of it!
We’re excited to share the latest progress on NimbleAI, a European project focused on advancing AI-driven solutions with remarkable agility and precision!
From NimbleAI.eu
Our computing engine integrating Menta SAS’s eFPGA IP, Codasip‘s RISC-V CPU, and CEA-List‘s C-SRAM, is now fully deployed on the NimbleAI prototyping MPSoC – ready to process data structures generated by IKERLAN‘s BEGI from DVS inputs.
By leveraging the flexibility and reprogrammability of Menta’s eFPGA, we can now implement custom instructions tailored to the most demanding operations in computer vision, including floating-point operations. This unlocks game-changing hardware acceleration, enabling our computing engine to keep pace with the ever-evolving needs of AI algorithms and use cases.
CEA’s C-SRAM (Computational-SRAM) addresses the memory wall problem by enabling near-memory computing, performing vectorized integer operations directly in the memory. This reduces data movement, improving performance and cutting power consumption in data-intensive applications like AI.
Our comprehensive toolchain leverages a custom Clang compiler to generate RISC-V executable code while automatically routing custom instructions to the eFPGA accelerator and offloading vectorized operations to the C-SRAM, thus enabling efficient heterogeneous computation.
IKERLAN’s BEGI is a digital processor optimized for integration into SoCs to bridge AI processing IP, such as our NimbleAI computing engine, with emerging event-based vision technology. It pre-processes DVS events to provide ultra-low-latency filtered event frames, time-surfaces, optical flow maps, and, when paired with light-field DVS, depth maps.
