VeriSilicon and Google Unveil Open-Source Coral NPU IP to Power Ultra-Low-Energy Edge AI and LLM Applications
VeriSilicon, a leading silicon platform-as-a-service (SiPaaS®) company, today announced the joint launch of the Coral Neural Processing Unit (NPU) intellectual property (IP) with Google, marking a major milestone in the open-source AI hardware ecosystem. The new Coral NPU IP is specifically engineered for always-on, ultra-low-energy edge Large Language Model (LLM) applications and represents a critical step toward democratizing edge AI innovation through open collaboration.
The partnership brings together Google’s deep expertise in open machine learning compilers and VeriSilicon’s proven capabilities in custom silicon design, verification, and commercialization. Together, the companies aim to empower developers and enterprises with a unified, open-source AI hardware and software framework that lowers the barrier to entry for next-generation wearable, IoT, and ambient intelligence applications.
Building the Foundation for Open and Secure Edge AI
The Coral NPU IP is based on the open RISC-V instruction set architecture (ISA) and integrates native tensor processing capabilities optimized for modern AI workloads. Leveraging open standards, the architecture supports mainstream machine learning frameworks such as JAX, PyTorch, and TensorFlow Lite (TFLite), ensuring a seamless development experience for both research and commercial use cases.
At the compiler level, the design employs Multi-Level Intermediate Representation (MLIR) technology—an open compiler infrastructure under the LLVM project—to enable flexible, efficient translation from high-level ML models to hardware execution. This approach not only enhances compatibility but also ensures developers can easily adapt their AI workloads for Coral-based devices.
By grounding the IP in open standards and transparent architectures, VeriSilicon and Google are jointly fostering a new model of trustworthy, secure, and community-driven AI acceleration, critical for the rapid evolution of edge computing and AI ecosystems.
AI Security and Sustainability: Always-On, Ultra-Low-Power Design
A central design goal of the Coral NPU IP is to enable always-on intelligence at the extreme edge while maintaining minimal power consumption. With modern applications such as wearables, smart glasses, smart home devices, and ambient sensing systems requiring continuous environmental awareness and contextual responsiveness, power efficiency is paramount.
The Coral NPU incorporates AI security enhancements developed in collaboration with Google’s open research initiatives, ensuring safe and responsible AI deployment even in highly distributed, resource-constrained environments. Through these enhancements, developers can confidently deploy privacy-sensitive or proprietary LLM applications on devices without compromising user trust or data confidentiality.
This blend of security, sustainability, and openness makes the Coral NPU an ideal foundation for the next generation of edge AI—capable of supporting lightweight LLM inference, local voice and vision understanding, contextual awareness, and multimodal sensor fusion directly at the device level.
Open-Soure Access and Global Developer Availability
The Coral NPU IP has been officially open-sourced on the Google Developers platform, making it freely available to the global developer and research community. Through this release, developers can explore, customize, and optimize the NPU’s architecture for their own edge AI applications, promoting widespread experimentation and innovation.
The open-source version provides not only the hardware IP but also reference software stacks, documentation, and example implementations for integration with popular ML frameworks. This accessibility supports academic research, prototyping, and early-stage commercial exploration, helping organizations accelerate their AI edge computing roadmaps.
In parallel, VeriSilicon will offer a commercial-grade, enterprise-ready version of the Coral NPU IP. This enhanced version will be optimized for high-volume production, including rigorous design-for-manufacturability (DFM), verification, and performance tuning, meeting the quality and reliability requirements of industrial and consumer-grade AI devices.
This collaboration builds upon our prior experience with the Open Se Cura open-source project and represents a deeper integration between Google’s open-source innovation and VeriSilicon’s expertise in silicon design and commercialization,” said Wiseway Wang, Executive Vice President and General Manager of the Custom Silicon Platform Division at VeriSilicon. “We will continue to leverage our strengths in chip design, verification, and system-level optimization to advance the edge AI ecosystem, enable real-world deployment of open-source technologies, and accelerate the adoption of edge LLM applications that drive industry transformation.
Advancing Edge AI Ecosystems through Custom Silicon Innovation
As AI computation continues to shift from centralized cloud infrastructures toward distributed, intelligent edge devices, the need for customizable, energy-efficient, and secure AI accelerators has grown exponentially. VeriSilicon, with its SiPaaS® (Silicon Platform as a Service) business model, is uniquely positioned to deliver such solutions—bridging the gap between open-source IP innovation and commercial silicon deployment.
The Coral NPU IP serves as a flexible foundation for VeriSilicon’s custom silicon services, enabling customers to integrate advanced AI capabilities into their own chips and systems. By combining Coral’s open-source architecture with VeriSilicon’s extensive IP portfolio—spanning CPU, GPU, DSP, ISP, NPU, and analog modules— the company offers a true one-stop solution for edge AI system-on-chip (SoC) development.
VeriSilicon’s comprehensive chip design and verification flow, supported by AI-assisted EDA (Electronic Design Automation) tools and hardware-software co-optimization, ensures that customers can rapidly transform architectural innovations like Coral into production-ready, high-performance silicon.
Accelerating LLM Deployment at the Edge
Large Language Models (LLMs) have traditionally been confined to cloud environments due to their immense computational requirements. However, the industry trend toward on-device and hybrid AI is rapidly changing that dynamic. The Coral NPU IP is designed to bridge this gap, enabling lightweight, fine-tuned LLMs to run locally on edge devices for real-time intelligence.
VeriSilicon is already developing a validation chip based on the Coral NPU IP, aimed at AI/AR glasses, smart home controllers, and other low-power consumer electronics. This validation chip will demonstrate Coral’s potential for running advanced natural language, vision, and multimodal models in resource-constrained environments—unlocking new levels of interactivity and autonomy in edge devices.
By enabling real-time inference directly at the device, Coral minimizes dependency on cloud infrastructure, enhances data privacy, and ensures low-latency responses. This capability is particularly valuable for contextual AI assistants, environmental sensing, health monitoring, and wearable analytics, where connectivity and energy efficiency are critical.
Driving Open Collaboration Between Hardware and Software Communities
The Coral NPU initiative represents a broader industry trend toward open collaboration between hardware and software ecosystems. By grounding the IP in the open RISC-V ISA and MLIR compiler infrastructure, VeriSilicon and Google are providing a transparent, extensible foundation for future AI acceleration research and commercial development.
This approach fosters cross-community synergy among semiconductor engineers, machine learning researchers, and software developers—each able to contribute optimizations and enhancements that advance the collective ecosystem. Open-source access ensures that innovations developed for Coral can propagate across industries, encouraging shared standards and interoperability in edge AI.
By making the Coral NPU IP open-source, Google and VeriSilicon are helping to democratize access to advanced AI compute capabilities,” said a senior AI systems architect at a leading semiconductor firm. “This move empowers startups, research institutions, and established manufacturers alike to innovate at the edge without being locked into proprietary ecosystems.
A Vision for the Future: Democratizing Edge Intelligence
The joint launch of the Coral NPU IP represents more than a technological milestone—it symbolizes a shared vision for accessible, secure, and sustainable edge intelligence. As the AI industry transitions into the era of distributed computation and local learning, initiatives like Coral redefine how innovation happens—collaboratively, transparently, and efficiently.
VeriSilicon and Google’s partnership demonstrates the potential of combining open-source innovation with commercial-grade engineering excellence, enabling a new generation of AI devices that are both intelligent and energy-conscious. From LLM-driven wearables to adaptive smart home systems, the Coral NPU is poised to become a cornerstone of the future edge AI landscape.
Source Link:https://www.businesswire.com/
