Data Repository for Chip-Chat: Challenges and Opportunities in Conversational Hardware Design

# Data Repository for Chip-Chat: Challenges and Opportunities in Conversational Hardware Design
This repository accompanies the manuscript accepted at MLCAD 2023, titled "Chip-Chat: Challenges and Opportunities in Conversational Hardware Design".
It contains the following:
- `free-chat-gpt4-tt03` - this contains the data used for the paper, which examines free-form process when exploring the potential applications for LLMs in hardware design. The task here was to generate the Verilog for a full (albeit small) processor design. Here, the chats are presented (and annotated) in the `/chats` subdirectory, which also includes a python script for extracting metadata (presented in table IV in the manuscript). Note that this directory also includes `/assembler` which provides a basic assembler (also written in Python) to make it easier to write demo programs (examples included) for the processor.
- `scripted-benchmarks` - this contains additional data not used in the paper, which examines a more rigid process when exploring the potential applications for LLMs in hardware design. Here, each model chats are separated by subdirectory.
- `scripted-benchmarks-gpt4-tt03` - this contains just the benchmarks not used in the paper, made by the first run of GPT-4, which were used for tapeout in Tiny Tapeout 3.
The two tt03 directories contain the GitHub action scripts required to invoke OpenLane and produce synthesis files, as well as used to perform simulation tests.