Ceva and embedUR systems to launch ModelNova for NeuPro NPU family

As the industry accelerates toward “physical AI” -- bringing artificial intelligence out of the cloud and into the real world, Ceva Inc., the leading licensor of silicon and software IP for the Smart Edge, has announced a partnership with embedUR systems to launch a dedicated instance of the ModelNova platform for Ceva’s NeuPro NPU family. 

Purpose-built for edge AI inference across a wide range of devices, this collaboration expands Ceva’s Edge AI ecosystem by delivering a rich library of pre-trained, quantized AI models optimized for ultra-low-power NeuPro-Nano NPUs and high-performance NeuPro-M NPUs.

ModelNova, developed and maintained by embedUR, is a developer-centric destination for Edge AI building blocks, including models, datasets, and use-case blueprints. The Ceva-specific instance of ModelNova will host a curated selection of models from both companies, enabling rapid prototyping and deployment of AI features in wearables, hearables, smart home devices, and other AIoT applications powered by NeuPro-Nano NPU IP.

It will also support larger model use cases for NeuPro-M such as industrial automation, robotics, smart cameras, and AR/VR, while providing a roadmap for future workloads that require transformer-based architectures, LLMs, and generative AI models.

Accelerating AI innovation at the edge
At launch, the platform will include multiple embedUR-supplied models and a growing set of Ceva-contributed models, all optimized using TensorFlow Lite Micro and integrated with Ceva’s NeuPro Studio SDK. These models span categories such as object detection, face recognition, pose estimation, audio denoising, keyword spotting, and anomaly detection. Each model is accompanied by performance metrics, execution scripts, and dataset descriptions to streamline evaluation and integration.

Physical AI: Intelligence where it matters
Physical AI is redefining how devices interact with the world. Instead of relying on cloud connectivity, intelligence is embedded directly into products, allowing them to sense, decide, and act autonomously. This shift is critical for applications like robotics, smart appliances, and health monitoring wearables, where latency, privacy, and reliability matter.

Ceva’s NeuPro NPU family delivers scalable AI acceleration, from NeuPro-Nano for ultra-low-power inference to NeuPro-M for compute-intensive edge AI, while embedUR’s ModelNova platform provides the ready-to-use models that make development faster and easier. Together, they enable the next generation of real-world AI systems.

“ModelNova is a powerful extension of our NeuPro ecosystem, giving developers access to a diverse and growing library of optimized AI models,” said Elia Shenberger, Senior Director of Product Marketing, AI Division at Ceva. “Our partnership with embedUR enables customers to build smarter, more responsive edge devices with minimal development overhead.”

“embedUR is excited to collaborate with Ceva to bring ModelNova to life for NeuPro NPU family,” said Eric Smiley, VP of Business Development, embedUR. “This platform provides developers with the tools they need to innovate quickly and efficiently, leveraging proven models tailored for real-world edge AI applications across consumer, industrial, and AIoT markets.”

The dedicated ModelNova instance for Ceva’s NeuPro NPU family is available immediately, with ongoing expansion to include new models and use-case blueprints for emerging edge AI applications.