Leveraging Pre-trained Models for Failure Analysis Triplets Generation

Abstract: Pre-trained LLMs recently gained traction in the NLP domain for text summarization, generation and question-answering tasks. This stems from the innovation introduced in Transformer models and their overwhelming performance compared with Recurrent Neural Network Models (Long Short Term Memory (LSTM)). In this paper, we leverage the attention mechanism of pre-trained causal LLMs such as Transformer model for the downstream task of generating Failure Analysis Triplets (FATs) - a sequence of steps for analyzing defected components in the semiconductor industry. We compare different transformer models for this generative task and observe that Generative Pre-trained Transformer 2 (GPT2) outperformed other transformer model for the failure analysis triplet generation (FATG) task. In particular, we observe that GPT2 (trained on 1.5B parameters) outperforms pre-trained BERT, BART and GPT3 by a large margin on ROUGE. Furthermore, we introduce Levenshstein Sequential Evaluation metric (LESE) for better evaluation of the structured FAT data and show that it compares exactly with human judgment than existing metrics.