(Invited) Physical Analysis and Modeling of Wafer-to-Wafer Copper Hybrid Bonding at Sub-Micron Pitches

Abstract

The bonding process has four sequential steps: plasma surface treatment, rinsing, bonding, and annealing. It is important to understand the details of all these steps to enable high performance bonding at sub-micron pitches. A multi-level back-end-of-line test vehicles at 1um and 0.5um pitches have been produced to evaluate the execution of the Tokyo Electron wafer-to-wafer (W2W) bonding tool platform. Subsequently several simulation and modeling approaches have been developed to better comprehend the impacts of different process conditions on the physical hybrid bonding results.Figure 1 shows the scanning electron microscope (SEM) image of the 1um bonding pad pitch Kelvin-via test structure that was successfully bonded and electrically tested. A median resistance of 0.44Ω has been measured for well-aligned 1um pitch kelvin-via measurements. Multiple variations of the test structures purposefully shift the top and bottom wafer patterning up to 100% mis-aligned to study the impact on electrical results from the bonding alignment. The corresponding electrical resistance, leakage and yield results provide statistically relevant feedback to improve the design and process integration of the test vehicle and W2W bonding platform.Maintaining a proper composition of the copper bond pads during plasma activation is necessary, especially if a lower anneal temperature post bond is desired. At low thermal budget bonding the oxides are not able to diffuse through any thick layer of modified copper formed between the pads during bonding which can significantly impede the electrical properties of the contact. The impact of plasma gas composition on the formation of surface copper species was determined using X-Ray Photoelectron Spectroscopy. For low temperature bonding, a simultaneous O2/H2 plasma treatment produces the thinnest copper oxide with the lowest theoretical resistivity. The activation gas chosen is based on the dielectric being used, but it has a significant impact on copper oxidation.If the bonding dielectric material is SiCN, then an O2 or O2/H2 mixed gas is best used for the activation process. The Cu2O content is less than the diffusion length for low temperature bonding for both gases, but the concentration of both CuO and Cu(OH)2 is high for the O2 treatment. At these concentrations, a low temperature O2 activation would create a copper contact with significant resistance. These results will help develop a plasma treatment to achieve good electrical contacts between hybrid bonded wafers with low thermal budget.A quantum chemistry simulation was developed to study how O/N-radicals modify the surface of SiO2 to increase the bond density. O atoms binding with H, N and C in the SiO2 films to create OH, CH2O, CO, HNO and nitic oxides (NO) groups. NO-groups are hydrolyzed to form additional surface OH during the rinsing step, increasing the density of Si-O-Si formation and total bonding energy. Atomistic molecular dynamic simulation also confirmed how nitrogen atoms introduce silanol and siloxane group to the surface during the surface activation treatment and the hydration reaction. The population effects of N atoms on bond formation were also studied, which indicated when more N atoms exist it will lead to improved bonding.Several plasma conditions were tested for the surface activation process to bond thermal silicon dioxide (TOX) to each other, and the bonding strength was calculated by the Maszara-Method after the annealing process. Comparing the bonding strength between TOX/TOX bond with N2 and O2 plasma under the same plasma conditions, the bonding strength using N2 plasma treatment is 1.7 times stronger than the bonding strength using O2 plasma treatment, indicating that the N2 plasma treatment is superior to the O2 plasma treatment.Finally, a multi-physics finite element model to simulate the post-bond distortion resulting from the wafer-to-wafer bonding process in room conditions was evaluated. The axisymmetric simulation model includes both the solid mechanics of the silicon wafers and the fluid mechanics of the air gap to accurately simulate the bond propagation dynamics. The two wafers, which may be warped by stressed films in preceding process steps, are initially separated at a small, fixed gap. As the center of upper wafer is pushed down by a striker, an adhesion model like Lennard-Jones potential simulates the contact propagation behavior between the two wafers. The simulation also includes other bonder process parameters and hardware designs to provide accurate distortion results. The outputs of the simulation are validated against a numerical model. The simulation model is also validated against direct wafer bonding experiment results. Figure 1