A major U.S. software and AI platform company builds custom AI silicon for vision

• Next-generation intelligent computing devices.
• Power-efficient AI acceleration for advanced on-device inference workloads including high end vision application.
• Image-quality enhancement, High-resolution, high-frame-rate vision processing

The customer needed to integrate Advanced AI inference workload into high end image-quality pipeline and implement advanced neural network optimizations tailored to its target AI workloads, under tight constraints of area and power efficiency.  

• Flexibility for custom AI silicon: Ceva provides an adaptable architectural foundation rather than a fixed, one-size-fits-all solution.

• A configurable NPU: NeuPro Max can be optimized around the customer’s specific vision workloads and custom silicon requirements.

• Performance, power and area efficiency: The architecture supports optimization across the key PPA requirements of next-generation intelligent computing devices.

• Optimization from software to silicon: Ceva enables the customer to connect software-platform requirements with workload-specific neural-network and silicon optimizations.

• NeuPro Max NPU IP serves as the foundation of the custom AI silicon program 

• Workload-specific optimizations are being built with Ceva in close collaboration with the customer.  

• The program supports OS-to-silicon optimization 

• Ceva-NeuPro Max (16K)