Intel launches Starfire, a space-grade Panther Lake SoC with Intel 3 GPU for US government
Intel has unveiled Starfire, a space-grade SoC utilizing a Foveros-packaged Panther Lake architecture with 18A compute tiles and an Intel 3 graphics tile, designed for on-orbit AI inference.
Intel has introduced Starfire, a space-grade system-on-chip (SoC) designed for the U.S. government that employs a heterogeneous process strategy. According to Tom's Hardware, the chip utilizes the Panther Lake architecture and Foveros packaging to integrate 18A compute tiles with an Intel 3 graphics tile. This node division mirrors the approach used for the 288-core Xeon Clearwater Forest processor, which stacks 18A compute tiles on Intel 3 base tiles. By committing the 18A process to orbit, Intel is relying on RibbonFET architecture and design-level hardening to manage radiation sensitivity, rather than relying solely on the inherent tolerance of more mature, larger-feature nodes.
The Starfire SoC is available in two distinct SKUs, both rated for a 10-plus year operational lifetime and thermal operation between -55 and 125 degrees Celsius. As reported by Tom's Hardware, both versions feature four 18A P-cores, four low-power efficiency cores, a three-tile NPU on 18A, and a four-core Xe GPU with 64 execution units built on the Intel 3 process. The 10W variant achieves 45 TOPS, while the 35W performance SKU reaches 75 TOPS. Clock speeds vary significantly: the low-power part runs P-cores at 1.0 GHz and the GPU at 800 MHz to 1.0 GHz, while the performance part pushes P-cores to 3.1 GHz and the GPU to 2.0 GHz.
Both Starfire configurations include 12 PCIe Gen4 lanes and support for LPDDR5 or DDR5 memory. This integration is intended to facilitate advanced data processing and AI inference directly on orbiting platforms, a departure from the telemetry-focused tasks handled by legacy space-grade silicon. According to Tom's Hardware, the inclusion of a dedicated NPU on the 18A node positions Starfire for high-throughput AI workloads, moving beyond the capabilities of traditional radiation-hardened processors like the BAE Systems RAD750, which has been a standard for two decades but relies on 150nm or 250nm lithography.
While Starfire is designed for harsh environments, it is not yet fully radiation-qualified. Intel lists radiation data—specifically regarding total ionizing dose, single-event latch-up, and single-event effects—as being in the characterization process. The company has noted that these specifications are subject to change. The chip is being managed by Intel Government Technologies, which is leveraging Intel Foundry’s Trusted Foundry status and its participation in Pentagon programs such as RAMP-C and SHIP. This development highlights Intel's attempt to provide domestic manufacturing for leading-edge logic in aerospace, though industry-standard 18A yields are not anticipated until 2027.
The introduction of Starfire represents a shift in the aerospace semiconductor sector, where traditional reliance on mature, radiation-hardened processes is being challenged by the demand for on-orbit computational density. By utilizing 18A for compute-heavy tasks, Intel aims to provide a path for modern node technology in space hardware. With samples slated for Q3 2026, the project aims to offer market-competitive pricing and domestic supply chain security for government contracts. The success of this implementation will depend on how effectively the design-level hardening compensates for the inherent radiation sensitivity of smaller transistors compared to the legacy chips currently flying on missions like the Mars rovers.