This project explores the generation of custom ml hardware from ml software.
The goal is to create a custom ASIC that classifies 5x5 images of digits.
This is achieved by:
- Training a sklearn model on the dataset.
- Parsing the decision tree structure.
- Generating HDL from amaranth.
This project is made viable by the tinytapeout open ASIC initiative.
The dataset is composed of binary 5x5 images as shown in the following picture:
The model has flattened view of the image and has the following structure:
