Overcoming Cleaning Challenges for AI Devices: Optimizing a Cleaning Process for High Reliability

Every technological revolution has profoundly impacted people’s lifestyles and the knowledge acquired in each era. The advent of search engines in the 1990s emphasized the importance of utilizing available information on the Internet and transforming it into valuable outputs rather than attempting to memorize extensive and intricate information. The advancements in artificial intelligence (AI) applications are expected to bring substantial changes and opportunities across various industries, including manufacturing, healthcare, finance, and transportation. With the ability to analyze and process large amounts of data, AI applications enable people to work more efficiently and increase productivity. Companies like Nvidia, AMD, and Google utilize TSMC’s CoWoS packaging technology to incorporate multiple processors and High Bandwidth Memory (HBMs) within a single package. From a packaging perspective, this can be considered a form of large flip-chip packaging, with package sizes up to at least 60x60mm. Additionally, the use of μ-bumps or Cu pillar bumps creates small standoff gap heights typically ranging from 25-40um. This poses a significant challenge in achieving thorough cleanliness of AI devices, as existing wafer form cleaning tools require extensive time and effort. If any residue from the ionic flux used in the process is left behind, the reliability of the AI devices may be compromised. In this presentation, KYZEN will explore cleaning strategies tailored to overcome the challenges of CoWoS packaging technologies. The discussion will cover the cleaning obstacles unique to these packages, along with steps to optimize the cleaning process and achieve the desired level of cleanliness for AI devices. This will include cleaning tool selection, cleaning chemical selection, wash and rinse parameters, and inspection methods. The goal is to outline a successful cleaning process that guarantees the intricate devices’ high reliability and extended device life.