Inline Bondwave Monitoring for Direct Bonding, Process Optimization and Impact on Post-Bond Distortion

Abstract

Direct wafer bonding, named also fusion wafer bonding, is based on the formation of molecular bonds between the two surfaces placed in contact. By the initiation of the contact between the two surfaces the bond is self-propagating: at the initiation point a bondwave is generated, which propagates free until the entire surfaces are in contact (bonded). The dynamic of the bondwave propagation is depending on various material (substrates) and process parameters. As shown by Rieutord et al. (1), and experimentally confirmed (2), for a given material and geometry, the speed is function of adhesion energy and fluid viscosity. For industrial applications, this means that investigating the bondwave speed by inline monitoring can be a way to monitor adhesion energy. Adhesion energy itself is a function of different parameters such as surface roughness, WET treatment or water presence (2,3).Inline IR monitoring can be also a practical solution for defect detection. Compared to post bond IR inspection, the live information allows for easier root cause analysis, e.g., particle defect, surface inhomogeneity or chuck effect. Inline early detection of defective bonds and process outliers enables bonded pairs to be reworked (separated and re-bonded) without any additional metrology steps such as post bonding IR inspection or Scanning Acoustic Microscopy (SAM). In case of fast bondwave, one has to monitor any unwanted effect such as edge voids or increased distortion (4).In an automated fusion bonding equipment, EVG®850LT, a bonding chamber has been equipped with an IR camera to record the bondwave propagation. The chamber has also connected different gas and vent lines and a humidifier to control the bonding atmosphere. The bond is initiated with a mobile pin equipped with a load cell, allowing to start the bondwave in a repeatable way at different locations (each bond must be initiated in a single point, but location might vary between center or edge proximity). As shown in figure 1 for a 300 mm Si-SiO2 bonded pair, spatial information can be extracted from the IR images, allowing for bondwave speed mapping (5). The equipment is also equipped with a plasma chamber for pre-processing. This allows to evaluate, using inline bondwave observation, various process conditions such as: plasma treatment, bond chamber gas type, pressure, humidity and bond pin location. Moreover, a metrology equipment was used for bonded pair shape measurement. This data is used to predict high resolution distortion maps to monitor the effect of bonding wave speed and bonding parameters.Different use cases of bondwave monitoring for process development will be presented, as well as the influence of different process parameters on bondwave propagation. A detailed analysis of the impact of the bond wave speed on bonded stack distortion will be presented. As shown in figure 1 (a) and (b), bonding wave velocity has no impact on high order distortion but will significantly increase the runout or scaling of the bonding pair.1. F. Rieutord, B. Bataillou, and H. Moriceau, Phys. Rev. Lett., 94, 236101 (2005).2. V. Larrey, G. Mauguen, F. Fournel, D. Radisson, F. Rieutord, C. Morales, C. Bridoux, and H. Moriceau, ECS Trans., 75, 145 (2016).3. D. Radisson, Phd thesis, Université Grenoble Alpes, (2014).4. M. Broekaart, A. Castex, K. Landry, R. Fontaniere, and C. Lagahe-Blanchard, ECS Trans., 50, 371 (2013).5. Fournel, F, B Rousset, V Larrey, C Morales, R Sachs, L Sudrie, and C Morvan, WaferBond’22 Proceeding, 2022. Figure 1