Ceva’s Next-Generation NeuPro Brings a Unified AI Platform from Sensors to Infrastructure

What happened

Ceva announced a next-generation NeuPro NPU portfolio introducing four NPU families sharing a common architecture and unified software stack: NeuPro Nano targets always-on sensing and entry-level perception, NeuPro Prime supports mainstream CNN and transformer workloads, NeuPro Max extends into advanced edge applications including high-resolution perception, multimodal models, robotics and on-device assistants, and NeuPro Ultra targets AI infrastructure, supporting complex language model inference and integration into intelligent memory, connectivity and data-processing silicon. Ceva reports up to twice the AI throughput per square millimeter and a 50% improvement in energy efficiency compared with the first generation. The architecture supports FP4, FP8 and mixed precision, higher memory bandwidth, dynamic shapes, paged key-value caching, and Mixture-of-Experts and linear attention. A single SDK spans all four families, and designers can combine multiple NeuPro engines within one SoC. Ceva says lessons from more than 15 first-generation NeuPro licensees helped shape the new portfolio. Selected customers are already receiving early access, with general availability planned for the second quarter of 2027.

Impact and signal

Affected components and suppliers

Source and time

Source: SemiWiki · 10/28/2026 · View original source