This paper addresses a 0.35µm CMOS process (X-FAB XH035 technology) based integrated sensors. The sensor structure is constructed by using standard features of the CMOS components. To reach the required performance on chip, ASIC including memories are needed. All these features are on one and the same wafer – MEMS and CMOS are side by side monolithically integrated in the same piece of silicon. To make the sensor functional, parts of the CMOS structures need to be undercut by silicon etching from the wafer front side. This requires the opening of the CMOS passivation to get access to the silicon and a very special release etch process. In a next step, these released very fragile structures need to be protected mechanically on wafer level by bonding a structured cap wafer. Additionally, in this step a hermetically sealed cavity with the MEMS structure is formed. By a dicing process the silicon over the wire bond pad area is finally removed to enable wafer probing (Fig. 1 and 2 show this configuration). By advanced packaging processes (die and wire bonding and special plastic moulding) the final device is realized in a small and thin package.
In this very complex technology wafer bonding is a key process. Glass frit wafer bonding was chosen as bonding process as it allows a quite simple process set up. The Glass paste can be printed on top of the structured cap wafer and after the multiple temperature step glass conditioning (firing) the wafer can be bonded on the processed MEMS/CMOS wafer. However, many side effects need to be considered in this process to fulfil the sensor performance spec and to get a stable production process. These aspects are here shortly named and will be discussed in detail in the related full paper:
The bonding has to be done on top of CMOS structures, actually on top of the CMOS passivation. The passivation stack itself should not be changed - so the bonding surface is silicon nitride, due to its hydrophobic nature it is rather difficult to bond. Additionally, any damages of passivation need to be avoided. Due to the fact that any glass frit is containing sharp filler particles to adapt the thermal coefficient of extension of the glass to the silicon, a paste material has to be used with low particle sizes to prevent a damaging of the passivation. By additionally using a flat passivation, variant local stress can be avoided - this is preventing from passivation cracks.
The bond interface has to be of high quality. The bond needs to be mechanically strong and hermetically sealed. The requirement for this is a very good screen printing process – fig. 3 shows such an optimal structure. By means of screen printing process improvements this could be stably achieved. The second important point here is the bonding temperature: it has to be high enough to let the glass flow wet the surface to allow a sealing and strong bonding. Here optimizations resulted in a nearly 100 % yield of sealed devices and prevented the removal of any caps in the later process steps like dicing.
This application of the glass fit bonding shows that this process can be used for very sophisticated applications. The investigation provided a new understanding of the material, the bonding process and boundary conditions.
Acknowledgements: The authors would like to thank Dr. Frank Seifert (Fraunhofer ENAS Chemnitz) for supporting the screen printing process optimization and Dr. Appo van der Wiel (Melexis) for advising about device integration aspects.