Qualcomm AI Stack

Qualcomm® AI Stack is an end-to-end AI software offering that combines Qualcomm AI software capabilities within one unified software stack to support multiple product lines.

The Qualcomm AI Stack helps to unify our existing AI software offerings, making them easier to use and more accessible to developers and manufacturers. Use a single stack to repurpose the AI applications you build, without having to port across hardware silos.

Hardware

  • Smartphones
  • XR (VR/AR) devices
  • IoT devices and industrial IoT (IIoT) machines
  • Robots
  • Snapdragon Digital Chassis
  • Windows on Snapdragon
  • Cloud computing

The Qualcomm AI Stack is designed to support a wide variety of use cases including:

Functions

  • General artificial intelligence and machine learning
  • Model optimization and quantization
  • AI-based camera, voice/audio features
  • Augmented reality and virtual reality
  • Advanced driver assistance systems (ADAS) and assisted driving (AD)
  • Machine vision
  • Natural language processing

As part of the stack, the Qualcomm® AI Engine Direct scales across every AI accelerator inside a broad range of products. OEMs and developers can develop and optimize AI models using the Qualcomm AI Stack once, then move the same model across different Qualcomm products and tiers. It’s a true “develop once, deploy anywhere” model that helps OEMs and developers save time and resources.

If you are developing a solution that uses the following SDKs, then you are already using the Qualcomm AI Stack. You can also choose to use Qualcomm AI Stack directly if you are developing on other platforms including smartphones, Windows on Snapdragon® PCs and IoT devices.

Domain-specific SDKs

  • Qualcomm Neural Processing SDK — For optimizing the performance of trained neural networks on devices with Snapdragon
  • Snapdragon Ride SDK — Scalable, fully customizable ADAS platform
  • Intelligent Multimedia SDK — Brings together multimedia, AI and ML, computer vision (CV), and networking building blocks to support end-to-end deployment of robotic applications
  • Snapdragon Spaces™ XR Developer Platform — For creating XR applications around the Snapdragon platform, Unity, Unreal and other technologies

Tools

  • Qualcomm AI Model Efficiency Toolkit (AIMET) — Open-source library for optimizing (compressing and quantizing) trained neural network models.
  • AIMET Model Zoo – For direct access to highly accurate, quantized models that boost performance and reduce energy consumption, latency and memory.
  • NAS - NAS was designed to use artificial intelligence to help create and optimize networks automatically by creating the best neural topology for your specific hardware or performance targets. Working with Google, we will be integrating Vertex AI NAS into the Qualcomm Neural Processing SDK to enable OEMs/Ecosystem to create, test, and deploy unique and efficient experiences on the edge — in just weeks — while using significantly fewer resources.
Click Image to Enlarge

At the highest level, the Qualcomm AI Stack works with popular AI frameworks and runtimes such as TensorFlow, PyTorch and ONNX, and inferencing SDKs like the Qualcomm Neural Processing SDK.

At the level of developer libraries and services, you will find support for current programming languages and for virtual platforms like the open-sourced QEMU and for compilers like LLVM and TVM.

The system software level contains the real-time operating system, system interfaces and drivers.

The OS level includes support for operating systems like Android, Chrome OS, Windows and Linux, plus infrastructure OSes like Prometheus, Kubernetes and Docker.

The Qualcomm AI Stack helps give developers and manufacturers a competitive edge in shortening their time-to-market.


Snapdragon, Qualcomm AI Stack, Snapdragon Ride, Snapdragon Digital Chassis, Snapdragon Spaces, Qualcomm Neural Processing SDK and Qualcomm AI Engine Direct are products of Qualcomm Technologies, Inc. and/or its subsidiaries. AIMET is a product of Qualcomm Innovation Center, Inc.