This guide is for a system designer, possibly with access to the Arm Flexible Access program. This guide will help you to develop a System on Chip (SoC) that can perform machine learning at the edge. The SoC that is presented in this guide can handle machine learning tasks related to image recognition. Image recognition is a reasonably complex machine learning task, which usually requires more performance than, for example, keyword recognition.

The guide is also relevant to system designers who want to create an SoC for a high-end smart device, for example a smartphone.

Specifically, the guide explains:

  • Why specific pieces of IP were chosen for this SoC
  • How to configure the pieces of IP to use them in this SoC
  • How to connect the pieces of IP together

The aim is to provide a broad view of how the pieces of IP work together. Use the SoC presented in this guide as an example. It is expected that you customize the SoC to suit your exact requirements. This involves further understanding and configuration of each piece of IP, the possible removal of some IP, or the addition of extra IP.  

This guide uses IP from the Arm Flexible Access program. The Arm Flexible Access program provides low-cost access to a wide range of Arm IP, so that you can experiment and design with a complete IP portfolio. If you have a license for the Arm Flexible Access program, or hold licenses for the individual pieces of IP, use this guide as a starting point. You can then begin to take practical steps with the IP and tweak the design to suit your individual requirements.

If you do not have any licenses, this guide still provides you with a unique overview. You can then go deeper into the specifics of any individual piece of IP included in the SoC.