Vision Processor IP

• Fully programmable to efficiently execute Neural Networks, feature extraction, signal processing, audio and control code
• Scalable performance by design to meet wide range of use cases with MAC configurations with up to 64 int8 (native 128 of 4x8) MACs per cycle
• Future proof architecture that supports the most advanced ML data types and operators

• Ultra-fast Response Time
• Zero-latency Switching
• Low Power

• Support wide range of activations & weights data types, from 32-bit Floating Point down to 2-bit Binary Neural Networks (BNN)

• The SFA 250A has been designed to be easy to adapt to suit the support needs of the customer’s IP as it is scalable, both in terms of function and performance, as well as modular as multiple versions can be combined to form larger solutions.

• Can cut design cost and time to market by up to 30%.

Enables up to 256 streams decoding with robustness, high throughput single-core solution, or multi-core solution, supporting VVC, AV1, HEVC, H.264, AVS3, AVS2, VP9 video formats, JPEG, and legacy formats.

• Enables up to 256 streams encoding with high video quality single-core solution, or multi-core solution, supporting AV1, HEVC, H.264, VP9 video formats and JPEG.

• Unified architecture supports multiple formats
• VVC main profile 8/10bit
• AV1 main profile 8/10bit

• Up to 8K @ 60 FPS
• Up to 4 pixels per clock
• FLEXA PSI interface in streaming mode or AXI interface in memory-to-memory mode

• PC820 IP offers a rich array of foundational video and image processing capabilities, facilitating seamless integration with major video and graphics processing IPs, AI processors, and data pathways of the Host CPU.