Palimpsest: Bill of Materials Prediction - A Case Study with Solid State Drives

Accurately quantifying product carbon footprints (PCFs) is critical for organizations to measure environmental impacts and develop decarbonization strategies. However, traditional methods require Bills of Materials (BOMs) data as a key input for PCF estimation, which is time-intensive and limits scalability. We present Palimpsest, an automated BOM generation algorithm given product specification as input using Large Language Models (LLMs) and a reference dataset. Palimpsest extracts data from teardown reports to build a BOM repository, retrieves reference products based on an their attribute list, generates BOMs by systematically modifying reference BOMs based on attribute differences, and standardizes the output to enable automated PCF estimation. We also introduce a novel impact-based evaluation framework that compares predicted BOMs with ground truth, focusing on the accuracy in carbon impact. We benchmark our model against a naive LLM solution and a traditional PCF estimation approach for solid state drives and find it outperforms these methods with a weighted F1 of 99.5%. By streamlining and automating BOM prediction, our method reduces the manual effort required for PCF estimation, driving progress toward net-zero emissions targets across industries.