Intel Announces Starfire SoC for Space Use, Targeting AI Inference in Orbit
Intel aims to transition space computing from control and communication to AI inference with Starfire, though radiation certification is not yet complete.
Reporting from 1 source: GIGAZINE.
Intel announced Starfire, a space-grade SoC for the U.S. government, combining CPU, NPU, and GPU chiplets via Foveros packaging. Two variants offer up to 45 or 75 TOPS of AI performance, with sample shipments starting Q3 2026. The chip is designed for AI inference in orbit, but radiation hardness certification is still ongoing.
Intel's Starfire SoC is built on two process nodes: the CPU and NPU use Intel 18A, while the GPU uses Intel 3. The Low Power variant runs at 10W TDP and delivers up to 45 TOPS, while the Performance variant reaches 35W and 75 TOPS. Both include 4 P-cores, 4 low-power E-cores, a 3-tile NPU, and a 64 EU Xe GPU. Intel compares Starfire to the RAD750, a radiation-hardened processor used in over 150 spacecraft that operates at 110-200 MHz. Starfire is intended for AI inference on sensor data in orbit, reducing the need to send all data to ground. However, Intel notes that evaluations for radiation tolerance are ongoing, and specifications may change.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.