铜微凸点
Search documents
HBM,太难了
半导体行业观察· 2025-11-12 01:20
Core Insights - High Bandwidth Memory (HBM) is a critical driver for artificial intelligence and is at the forefront of multiple technological developments, but it is also one of the most challenging modules to manufacture [2] - The shrinking dimensions and spacing of Through-Silicon Vias (TSV) and microbumps present significant challenges, impacting yield rates and defect detection [2][10] - The transition to HBM4 is expected to increase the number of stacked layers from 16 to potentially 20, while maintaining a total height limit of 775 micrometers [8] Manufacturing Challenges - The need to stack 16 chips on a single wafer requires significant thinning, down to 20 micrometers, and the use of backside inspection technology to ensure wafer flatness [4] - Major HBM manufacturers are considering a shift to hybrid bonding technology to achieve shorter interconnects and lower signal latency [4][10] - Inconsistencies in bump height can negatively affect yield, reliability, and performance, leading to issues such as contact failures and reduced signal integrity [4][10] Detection and Inspection Techniques - Manufacturers focus on identifying issues after the electroplating step and before reflow soldering, with confocal laser detection being preferred over white light detection for rough metal surfaces [5] - Advanced packaging requires strict control of coplanarity and flatness, with flexibility being crucial for adapting to different development stages and achieving mass production [7] - Various detection methods, including automated optical inspection (AOI) and X-ray tools, are being optimized to identify critical defects in microbumps [9] Transition to HBM4 - The transition to HBM4 involves challenges such as reducing copper microbump sizes to 10 micrometers and determining the optimal timing and method for migrating from microbumps to hybrid bonding [13] - Defects in microbumps include pad misalignment, solder necking, and localized cracking, necessitating rapid analysis of thousands of images to ensure high yield [13][14] - A robust framework for analysis and detection of defects is essential as HBM suppliers transition to HBM4, focusing on optimizing solder paste usage and ensuring proper alignment during assembly [14]