Synaptics pivots and embraces open-source for its edge AI processors (www.edn.com)

At Synaptics Tech Day (Oct 15, 2025) the company unveiled its AI-native Astra SL2610 edge SoC and the Torq platform, signaling a major pivot toward open-source hardware and software for edge AI. Synaptics confirmed integration of Google’s Coral open NPU (a RISC‑V–based front-end descended from Google’s mini‑TPU/Coral efforts) and open compilers/runtimes (IREE/MLIR) into its stack, building on a January 2025 agreement to pair Google’s ML core with Synaptics’ Astra software. The SL2610 nevertheless combines conventional IP—Cortex‑A55 application cores, Cortex‑M52 microcontroller cores with Helium, and a Mali GPU—while using a performance‑oriented RISC‑V neural core as a common front-end. Torq provides an open toolchain that can port models from any ML framework onto the Coral NPU and Synaptics’ accelerators, demonstrated in e‑bikes, EV charging, industrial AR glasses, PoS, speech recognition and smart‑home devices. The move matters because it targets the biggest friction in edge AI today: fragmented, proprietary toolchains that stall deployment. By adopting an open-stack approach (RISC‑V front-end + IREE/MLIR toolchain + open runtimes), Synaptics and Google aim to standardize model portability, reduce vendor lock‑in, and enable low‑power, multimodal, context‑aware inference at scale. For developers and silicon vendors this could mean faster integration of LLMs and multimodal models on resource‑constrained devices, easier cloud‑to‑edge workflows, and a more interoperable ecosystem that accelerates real‑world edge AI adoption.