VeriSilicon and Google Open-Source Coral NPU IP for Edge LLM and Ultra-Low-Power AI Applications

VeriSilicon and Google have jointly launched the Coral NPU IP as an open-source neural processing unit intellectual property targeting always-on, ultra-low-energy edge Large Language Model (LLM) applications.

The Coral NPU IP is based on Google’s research in open machine learning compilers and includes AI security features. It provides a unified open-source platform intended to support development of an edge AI ecosystem.

Built on the open RISC-V instruction set architecture, the Coral NPU features native tensor processing capabilities and supports JAX, PyTorch, and TensorFlow Lite (TFLite) machine learning frameworks. It uses open-standard tools, including Multi-Level Intermediate Representation (MLIR) from the LLVM project, as its compiler infrastructure.

The design focuses on always-on, ultra-low-energy edge AI applications, including wearable devices and ambient sensing systems.

The Coral NPU IP has been open-sourced and is now available on the Google Developers website for developers worldwide.

VeriSilicon will provide a commercial-ready, enterprise-grade version of the IP. The company will also offer one-stop custom silicon services using its IP portfolio and chip design and verification capabilities.

VeriSilicon is currently developing a validation chip based on the Coral NPU IP, targeted at AI/AR glasses and smart home ...