Highly scalable performance for classic and generative on-device and edge AI solutions

Scalable and Power-Efficient Neural Processing Units

The Neo NPUs offer energy-efficient hardware-based AI engines that can be paired with any host processor for offloading artificial intelligence and machine learning (AI/ML) processing. The Neo NPUs target a wide variety of applications, including sensor, audio, voice/speech, vision, radar, and more. The comprehensive performance range makes the Neo NPUs well-suited for ultra-power-sensitive applications such as IoT, hearables/wearables, high-performance systems in AR/VR, automotive, and more.

The product architecture natively supports the processing required for many network topologies and operators, allowing for a complete or near-complete offload from the host processor. Depending on the application’s needs, the host processor can be an application processor, a general-purpose MCU, or a DSP for pre-/post-processing and associated signal processing, with the inferencing managed by the NPU.

The Neo NPUs provide performance scalability from 256 up to 32k 8x8-bit MAC per cycle with a single core, suiting an extensive range of processing needs. Capacity configurations are available in power-2 increments, allowing for the right sizing in an SoC for the target applications. Int4, Int8, Int16, and FP16 are all natively supported data types, with mixed precision supported by the hardware and associated software tools, allowing for the best performance and accuracy tradeoffs.

Additional features of the Neo NPUs include compression/decompression to minimize system memory space and bandwidth consumption for a network and energy-optimized compute hardware to leverage network sparsity tradeoffs.

The Neo NPUs support typical clock frequencies of up to 1.25GHz in 7nm, and customers can target lower clock frequencies for specific product needs.

• Single-core performance up to 80 TOPS
  • Configurable range of 256 MACs per cycle to 32k MACs per cycle
  • Upward-scalable with multi-core topologies for 100s of TOPS
• Efficient offload and execution of neural network processing from any application host processor
• Built-in support for many networks, including CNN, RNN, Transformer, and more
• Built-in support for multiple data types, including int4, int8, int16, and fp16
• Application targets varied across many domains (sensor, audio, vision, radar) and markets (IoT, hearables/wearables, AR/VR, automotive)

• Single-core performance up to 80 TOPS
• Configurable range of 256 MACs per cycle to 32k MACs per cycle
• Upward-scalable with multi-core topologies for 100s of TOPS
• Efficient offload and execution of neural network processing from any application host processor
• Built-in support for many networks, including CNN, RNN, Transformer, and more
• Built-in support for multiple data types, including int4, int8, int16, and fp16
• Application targets varied across many domains (sensor, audio, vision, radar) and markets (IoT, hearables/wearables, AR/VR, automotive)

• Free Software Evaluation
• Try our SDK Software Development Toolkit for 15 days absolutely free. We want to show you how easy it is to use our Eclipse-based IDE.
• Online Support
• The Cadence Online Support (COS) system fields our entire library of accessible materials for self-study and step-by-step instruction.
• Xtensa Processor Generator (XPG)
• The Xtensa Processor Generator (XPG) is the heart of our technology - the patented cloud-based system that creates your correct-by-construction processor and all associated software, models, etc. (Login Required)
• Technical Forums
• Find community on the technical forums to discuss and elaborate on your design ideas.