A limitation to the use of direct wafer bonding methods for micromachining and thin film device manufacturing has been the necessity for high temperature anneals to strengthen the bonded interface. Obviously, strong interface strength is needed to withstand backthinning processes and the rigors of device fabrication. Unfortunately, the elevated temperature exposure has a detrimental effect on implanted or diffused etch stop layers via diffusive broadening. Additionally, for many micromachined applications wafer bonding could be used as a final assembly step, replacing epoxies. However, the sensitive components of the device must be protected from thermal effects. This paper describes the use of oxygen plasmas to develop chemical free, room temperature, wafer to wafer bonding methods. The bond developed between plasma-activated silicon wafers is virtually at full strength upon contact bonding and does not require further thermal strengthening. The results for silicon dioxide bonding show that full strength material is achieved with anneals below 300~ This process has been applied to a number of wafer materials including sapphire, silicon dioxide, silicon nitride, and gallium arsenide. The data presented are the results of strength tests, interracial defect etching, transmission electron microscopy analysis, initial interface reaction kinetics, and mechanisms studies. We also show preliminary results from a suggested model to explain the observed increases in kinetics compared to conventional aqueous solution processing of samples.

No abstract

No abstract

Atomic force microscopy has been used to quantitatively determine the surface roughness of silicon substrates as a function of processing and limitations to direct wafer bonding ability. This data is conveniently converted into a power spectrum creating a description of the topography which contains information about the amplitude and frequency of the surface undulations. Following initial characterization, the wafers were subjected to typical device manufacturing processes resulting in various degrees of increased roughness. An empirical correlation was developed between the roughness spectrum and bondability of (100) silicon wafers. Data on the roughening of wafers due to various standard integrated circuit processing steps were obtained and used to identify processes which promote wafer-to-wafer direct bonding. The fractal dimensions of the surfaces have been calculated and are discussed.in units of A2 p.m, where L is the scan length and Z(x) is the line profile. Although it is possible to analyze the entire wafer surface using multiple scans, the focus of this work was to identify a roughness size range that had the most impact on bondability. Fortunately, wafer warping and waviness are fairly limited in prime grade, uncoated silicon wafers due to the current state of the art in polishing. However, microroughness with spatial wavelengths much less than a millimeter and down to atomic dimensions is very difficult to assess quantitatively and often limits direct bonding. The atomic force microscope is well suited for studying roughness in this regime and using a power spectrum description, quicker and more comprehensive data can be obtained than from multiple profilometry scans for shorter and shorter wavelengths.' The information obtained can be used to develop an empirical relationship between the roughness spectrum and the bondability of the wafers. This allows for better control of the processing of wafers intended for direct bond applications.This study looked at the surface of silicon wafers exposed to various processing steps using an atomic force microscope (AFM). The surfaces are then described by power spectra. Finally the roughness data is correlated to the suitability of the surface to bond directly. ExperimentalMonsanto® prime grade (100) silicon wafers, n-type (3 to 5 fl-cm) were subjected to various common IC processes, specifically, RCA cleaning, wet thermal oxidation, and plasma-enhanced chemical vapor deposited (PECVD) nitride depositions. Each wafer was cleaved in quarter sections creating four samples from equivalent as-received surfaces. Each of the quarter sections underwent different processing steps and the roughness changes resulting from the process were monitored. These individual sections were examined with the AFM and then bonded to an identically processed section from another wafer for a total of four bonded samples for each process. Figures 1 and 2 give the descriptions of the processes performed on the individual sections and show the convention used for wafer labeling.The AFM w...

No abstract