📌 Accepted at ICLAD 2025
This repository hosts the official implementation, dataset, and experimental results for our paper:
"TPU-Gen: LLM-Driven Custom Tensor Processing Unit Generator"
Authored by Kartik Pandit, Deepak Vungarala, Mohammed Elbtity, Shaahin Angizi, Ramtin Zand, and collaborators.
TPU-Gen is the first end-to-end framework that uses fine-tuned LLMs with RAG (Retrieval-Augmented Generation) to generate both exact and approximate TPU architectures.
It consists of:
- ✅ Prompt generation from user input
- ✅ LLMs enhanced with multi-shot learning
- ✅ RAG module for dependency fetching
- ✅ Automated design validation and PPA reporting