Ara Software Development Kit

ARA-SDK

The Ara AI-software development kit (SDK) streamlines deployment of AI models onto Ara-1 and Ara240 discrete neural processing units (DNPUs) and modules. It features an extensible compiler that supports diverse model architectures, including convolutional neural network (CNN), large language models (LLMs) and vision language models (VLMs). The SDK offers efficient dataflow optimization, flexible quantization methods and support for multiple datatypes, ensuring high-performance and low-latency for edge-AI applications.

Diagram

Choose a diagram:

Machine learning deployment flow

ARA-1-DNPU-BD

NXP Applications Processor and Ara DNPU Connection

ARA-1-DNPU-BD-2

Integration of Ara DNPU and Applications Processors Host with Drivers

ARA-1-DNPU-BD-3
Get diagram PDF

Key Features

  • Optimal execution planning to automatically determine efficient data and compute flow for any AI graph
  • Extensible compiler that supports models from CNNs to complex vision transformers (ViTs) and LMMs
  • Programmable compute engines easily accommodate new model operators
  • Multiple datatype support - Handles 8-bit integer (INT8) (Ara-1, Ara-2), 8-bit integer (INT4) and Microsoft Floating Point 16 (MSFP16) (Ara-2 only)
  • Software-defined tensor partitioning and routing optimized for dataflow
  • Flexible quantization methods

Supported Devices

  • ARA240: Ara240 Discrete Neural Processing Unit (DNPU)
  • ARA-1-DNPU: Ara-1: Discrete NPU for Optimized Edge-AI

Design Resources

Hardware

Quick reference to our board types.
Filters

3 hardware offerings

Support

What do you need help with?

  • Narrow your search
Search