AI SDK for Ceva-NeuPro NPUs

• Consumer IoT
• Automotive
• Industrial 

Ceva-NeuPro Studio provides the tools for a complete end-to-end flow transforming trained neural network models into executable code on Ceva-NeuPro NPUs. Users can select from a wide range of common neural-network models already trained and optimized by Ceva in the Ceva Model Zoo, or bring their own model. Models are quantized, optimized and compiled by an MLIR-based graph compiler that partitions the workload across the NPU cores..

n addition, the toolchain includes profiling tools to estimate throughput, power and accuracy, a runtime stack for CNN and LLM inference on the target, and simulation and emulation environments for pre-silicon development.

Ceva-NeuPro Studio is a single SDK that serves the complete Ceva-NeuPro NPU portfolio, from Ceva-NeuPro-Nano to Ceva-NeuPro-Ultra. Customers develop once and reuse their models, tools and expertise across every core, reducing development effort across their product portfolio and giving them the freedom to choose the right NPU for each product.

The same SDK supports the full development cycle, from IP evaluation and pre-silicon development through to production deployment. Teams can assess performance and accuracy before silicon is available, then move to on-device inference using the same models, tools and flows.

• Import from major training frameworks including PyTorch, ONNX, TensorFlow, or LiteRT
• Support for pre-trained Ceva AI models and Bring Your Own Model (BYOM) approaches
• Powerful quantization and compression exploit Ceva-NeuPro NPU features
• Graph compiler produces optimized code to implement networks
• Software or hardware simulation and debug environment
• Performance profiling using Arch Planner