Google and VeriSilicon have unveiled the open source Coral NPU IP to accelerate edge AI and LLM deployment.
VeriSilicon (688521.SH) and Google have jointly launched the open source Coral NPU IP, marking a major leap in open hardware for edge Large Language Model (LLM) applications. Built to deliver always-on, ultra-low-energy performance, the Coral NPU aims to accelerate the global edge AI ecosystem by fusing Google’s open compiler research with VeriSilicon’s chip design expertise.
The Coral NPU IP is based on the open RISC-V instruction set architecture, featuring native tensor processing capabilities and AI security enhancements. It supports leading machine learning frameworks, including JAX, PyTorch, and TensorFlow Lite (TFLite), and leverages open-standard compiler tools such as Multi-Level Intermediate Representation (MLIR) from the LLVM project. This architecture is optimised for wearables, AI/AR glasses, ambient sensing systems, and smart home devices, all requiring ultra-efficient edge intelligence.
Now fully open-sourced and available on the Google Developers website, the Coral NPU IP allows developers worldwide to innovate, customise, and deploy AI solutions built on open standards. The initiative underscores Google’s commitment to open-source machine learning and VeriSilicon’s leadership in chip customisation, combining transparency with reliability.
VeriSilicon plans to release a commercial-ready version of the IP and is developing a validation chip for AI/AR and smart home use cases to speed up LLM adoption at the edge.
“This builds on our prior experience on the Open Se Cura open source project and represents the continued deep integration of Google’s open-source technology with VeriSilicon’s chip design and commercialisation capabilities,” said Wiseway Wang, Executive Vice President and General Manager of Custom Silicon Platform Division at VeriSilicon.
