Surface Characterization for and by Semiconductor Wafer Direct Bonding
Over recent decades, the direct bonding of semiconductor wafers has become a well-established process in science and the semiconductor industry, for stacking and mechanically joining wafers without any interlayer, based on surface tension forces . In addition to the requirement of excellent cleaning and activation of the wafer surfaces to be bonded, the mechanical surface quality of both wafers – very low roughness, no scratches, no residues - is essential, but difficult to guarantee in production processes when structured wafers have to be bonded. Additionally, a full wafer investigation of the surface quality is quite difficult, since measurement methods such as Atomic Force Microscopy (AFM) can only analyse very small areas, while surface problems can be very local and lie outside the standard measurement positions. In this paper, a method for analysing the surface quality in relation to the bonding process itself is demonstrated, as well as its use for process improvement.
Surface Quality Analysis by Direct Wafer BondingIn semiconductor direct wafer bonding processes, the wafers are typically cleaned (wet processes) and their surface activated either by this cleaning itself or by plasma treatment, often followed by a DI water rinse. The purpose is to hydrophilize the wafer surfaces to allow good hydrophilic bonding. This is followed by mechanical contacting of the wafers (pre-bonding) and then thermal annealing to transform the OH bonds into oxygen bonds with much higher strength.
In the pre-bonding step, a so-called bonding front travels across the wafers after they are initially contacted at just one point. The bonding front is the actual position where the pre-bond is formed between the wafers, and through the surface forces of the bonding process, it travels across the wafer to finally form a full-area wafer bond. The behaviour of the bonding front can be evaluated in different ways in order to describe the bonding process and its influencing factors. If the activation process is kept constant, differences in the front speed and its behaviour at obstacles, such as particles or scratches in the bond interface, can be used to evaluate the bonding behaviour of the surfaces to be bonded. While the bonding front speed and direction can be easily observed by an infrared camera during the bonding of standard silicon wafers, the obstacles which are a measure of the actual surface energy in the bond interface need to be well defined.
Here, two blades are placed between the wafers before they are brought in to contact (Fig. 1). The bonding front stops before reaching the blades, since this is a surface separation, and the open area between the bonding front and blade is measured as an indication of the surface energy in the pre-bonding process. This is ultimately a modification of Maszara’s blade test  for pre-bonding evaluation. By using this method, the process of high temperature direct bonding of cavity wafers for absolute pressure sensors (Fig. 2) was improved.
After etching the cavities in the handle wafer, a silicon oxide hard mask has to be removed without damaging or roughening the surface for the subsequent wafer bonding. Investigations using the proposed method (Table 1) have shown that the bonding front travels faster if the oxide is removed using HF compared to the use of buffered oxide etchants (BOE). This indicates that the BOE is causing some slight roughening of the silicon, so that the use of the HF as etchant can ensure process reproducibility. The detailed process conditions can also have an influnce on the bonding speed, while the initial pre-bond strength of all versions is comparable (length of unbonded area).
Observation of the bonding front propagation in a special setup with pre-inserted blades can be used to evaluate the surfaces to be bonded, for the understanding and improvement of direct wafer bonding processes.
 Q.-Y. Tong, U. Gössele: Semicondutor Wafer Bonding Sience and Technology, John Wiley &Sons, Inc. 1999