Continuing with our look at the 2013 ECTC.
Siliconware detailed the “Integration Challenges of TSV Backside Via Reveal Processing.”
After via formation, finished CMOS wafers or interposers are temporarily bonded to glass carriers. The TSV are ‘ revealed ‘ by Si back grinding and plasma etch steps, passivated with PECVD nitride, and CMP’ed to open the Cu pillar area. The via reveal processes must maintain acceptably low
TTV to allow subsequent bonding/stacking steps. Also the process temperature must be lower than the carrier bonding adhesives which is a particular challenge for the dielectric deposition step. The SPIL backside reveal process is shown below.
A major challenge of the via-reveal process is control the exposed copper TSV height, because incoming wafers to via reveal can have significant compounded variation, such as TSV depth uniformity, glass thickness uniformity, adhesive thickness uniformity and Silicon thickness uniformity after grind.
SEMATECH and RPI reported on their studies on “Backside TSV protrusion induced by thermal shock and Thermal Cycling.”
The TSVs used for this study were fabricated on 300mm wafers with a TSV height of ~50Î¼m
and an aspect ratio of ~10:1. The front side of the TSV wafer is bonded face down on a handle wafer and backside-thinned to reveal the TSV and metallized to form testing lines and pads. The cross section schematic below shows the structure.
Various combinations of thermal loads (RT - 200 C - 400 C) and ramp-up/cool-down rates (0.167 C/s to 25 C/s) are used for thermal shock and thermal cycling tests. No TSV protrusion is visible at 200 C or below, while larger TSV protrusions are observed at higher peak temperatures.The avg TSV protrusion height, collected from 108 single TSVs under 3 testing pads over each die, increases from 0.1Î¼m at 250 C to about 0.5Î¼m at 400 C.
The TSV protrusion varies significantly from TSV to TSV, resulting in big error bars. This is reportedly due to the grain boundaries in each TSV (particularly near the Cu testing pads) being very different from TSV to TSV, indicating that the key mechanism for the protrusion could be related to the Cu grain boundary diffusion.
SEM images of the TSVs reveal delamination is observed at the interface between the Cu TSV and the Cu testing pad on top of the TSV-1, while delamination between Cu TSV sidewall and oxide liner is found in TSV-2.
IMEC also reported on protrusion issues in their paper “Impact of Post Plating Anneal and TSV Dimensions on Cu Pumping.”
When Cu-filled TSVs are exposed to high temperatures during BEOL processing, compressive stresses arise in the Cu TSV due to the large difference in coefficient of thermal expansion with the surrounding Si. These stresses are partly relaxed by irreversible extrusion of the Cu, a phenomenon known as ‘Cu pumping,’ which may damage the BEOL layers on top of the TSV. In order to reduce the amount of Cu pumping during BEOL processing, a high temperature anneal step can be applied after TSV plating and before Cu CMP.
IMEC, who is generally given credit for offering an anneal solution protrusion problem in 2011 has now used optical profilometry to study residual Cu pumping in TSVs with different post-plating anneals and different TSV dimensions ( 5 x 50um vs 10 x 100um ). In total ~ 4000 TSVs were inspected. Within one sample the Cu pumping values show an intrinsic large spread, therefore the distribution tail rather than the median is determining the impact on BEOL reliability. Lower pumping was found in TSVs annealed at higher temperatures and for longer times. The sinter conditions of 20 min at 420 °C were confirmed as optimal post-plating anneal conditions.. However, in order to effectively control the impact on BEOL reliability, development efforts should also be aimed at reducing the Cu pumping distribution width.
SUNY Binghamton and SEMATECH presented their work on the “Mechanism of Low Temp Cu-Cu Direct Bonding for 3D TSV Package Interconnect.”
While the solder-based approach for connecting chips to packages or chips to chips has become the industry standard for at least the first generation of 2.5/3D products, but the potential to significantly drive this approach to finer pitch interconnects is limited. The leading method for fine pitch chip-to-chip interconnects (pitch of 10 microns or less) is generally believed to be Cu-Cu direct bonding. In the direct bonding of Cu to Cu, the flatness of the surface on a small scale (~1 micron) or a large scale (wafer or die scale) and the chemical condition of the surface play important roles in the quality of the bond. Other factors such as the Cu grain size and grain orientation may also impact the quality of the Cu-Cu bond.
Therefore, it is necessary to use a reducing gas to decompose the oxides. Forming gas, which is a mixture of H2 and N2, can provide such a reducing environment to decompose copper oxides effectively. Prior attempts to surface passivate / clean including self-assembled monolayer passivation, plasma cleaning and chemical mechanical polishing with a formic acid clean all result in improved bond quality as a result of the passivation and cleaning approaches used. The Bingham / Sematech group CMP’ed the copper in the presence of benztriazole which protects the copper surface during CMP, but does not prevent oxidation once the CMP is complete. The cleaned surfaces were then exposed t atmospheric conditions for varying times and reexamined by XPS (Xray photoelectron spectroscopy)
Cu2O and CuO can observed on the clean Cu surface after a short atmosphere exposure (1 minute), while Cu(OH)2 and/or CuCO3 can be observed on the surface after longer exposures longer exposures (>30 minutes).
Wafers were cleaned by Ar sputter cleaned (NA) or annealed at 200ºC in forming gas (FGA). Both wafer pairs were bonded within minutes after cleaning by thermo-compression bonding with a force of 80 kN at 195°C for 5 minutes. Samples of NA and FGA wafers exposed to the atmosphere for 30 min and then examined by XPS showed both CuO and Cu2O but no Cu(OH)2.
The NA and FGA Bonded Wafers were characterized by CSAM looking for voiding. The image of the bond interface for the FGA wafers indicates an absence of voids for almost the entire interface, whereas the image of the bond interface for the NA indicates voids throughout the interface. They attribute the better bonding for the FGA wafers to more effective Cu oxide removal by the forming gas anneal.
For all the latest in 3DIC and advanced packaging, stay linked to IFTLE.