How to Reduce BRAM Usage

[Ba1tu3han]

Hello All,

I trained a model by Brevitas QAT with 4 different bit-width then compiled by FINN with a custom configuration file. I would like to run them in Pynq-Z2.

While I was compiling them in estimation step of FINN, I realized that some of them are not fit to the HW.

Could you help me to reduce BRAM usage? How to reduce specifically BRAM in custom configuration file?

The situation is summarized in the table below

I am open to any feedback!