HotspotFusion: A Generative AI Approach to Predicting CMP Hotspot in Semiconductor Manufacturing

Abstract

The semiconductor industry thrives on rapid technological advancements, crucial for superior product performance and cost efficiency. Chip design houses and consumer electronics companies must continuously pursue New Tape Out (NTO) to maintain technological leadership. Timely NTO completion expedites product launches, crucial in the competitive semiconductor market. This paper addresses Chemical Mechanical Polishing (CMP) hotspot, critical in NTO quality and cycle time, affecting wafer surface topology. Hotspot defects can degrade wafer performance, demanding swift detection and resolution. Traditional methods can only identify CMP hotspot after manufacturing, necessitating repeated adjustments to IC design. We propose HotspotFusion, leveraging pattern density data from Graphic Design System (GDS) to predict CMP hotspot early in the design phase. Utilizing a generative AI model, HotspotFusion significantly reduces NTO cycle time by enabling proactive hotspot detection and process optimization, fostering efficiency and competitiveness in semiconductor manufacturing.