RISC-V and the X280: an AI Coprocessor for Mobile and Space

Published on May 18, 2026 | Translated from Spanish

SiFive introduces the Intelligence X280, a RISC-V core with vector extensions designed for AI tasks. Its versatility allows it to be used in both mobile devices and satellites. This processor aims to compete in a market dominated by other architectures, offering an open and scalable alternative for inference and data processing.

RISC-V X280 chip architecture cross-section, vector processing units actively routing data through AI tensor cores, mobile device motherboard on one side connecting via PCIe lanes, satellite radiation-hardened casing on the other side demonstrating dual-use capability, glowing orange data paths illustrating inference workload distribution, blue circuit traces branching into parallel compute clusters, engineering visualization style, metallic silicon wafer texture, microscopic transistor details visible, cool blue and warm amber lighting contrasting terrestrial and space environments, photorealistic technical illustration, clean industrial aesthetic, high-contrast shadows emphasizing 3D depth

Vectorized for the Edge: Performance Under Power Control 🚀

The X280 implements the RISC-V vector specification (v1.0) with support for lengths up to 256 bits. Its modular design allows for multi-cluster configurations, optimizing performance per watt. It includes accelerators for matrix operations (int8, fp16) and a local memory unit (LMU) interface that reduces latency in AI workloads. For space environments, it is hardened against radiation using logic redundancy techniques.

From Earth to the Moon: The X280 Doesn't Need a Spacesuit 🌙

The fact that a RISC-V chip can run AI on a satellite while you can barely get your phone to last a day without a charger has its charm. The X280 promises extreme efficiency, but one wonders if it will also come with a don't lose my signal mode for when the Mars rover decides to take a selfie. At least, if it fails, we'll know it was the fault of cosmic rays and not poor software optimization.