FPGA AI Suite: SoC Design Example User Guide

ID 768979
Date 7/31/2024
Public
Document Table of Contents

6.2.2. RAM considerations

An FPGA-based external memory interface is used to store all machine learning input, output, and intermediate data.

The FPGA AI Suite IP uses the DDR memory extensively in its operations.

Typically, you dedicate a memory to the FPGA AI Suite IP and avoid sharing it with the host CPU DDR memory. Although a design can use the host memory, other services that use the DDR memory impact the FPGA AI Suite IP performance and increase non-determinism in inference durations. Consider this impact when you choose to use a shared DDR resource.

The FPGA AI Suite IP requires an extensive depth of memory that prohibits the use of onboard RAM such as M20Ks. Consider the RAM/DDR memory footprint when you design with the FPGA AI Suite IP.