In this paper we (i) describe a model for the stress distribution across a wafer during chemical‐mechanical polishing, which is solved using I‐DEAS (a commercial software package) and (ii) summarize the predicted effects of carrier film and pad compressibility on polishing nonuniformity. Results indicate that (i) the Von Mises stress correlates with polishing nonuniformity, while the normal stress does not correlate with the nonuniformity and (ii) CMP uniformity improves with decreasing polishing pad and carrier film compressibility.

The integra-differential equations which describe free molecular flow in long rectangular trenches in the absence of deposition and to both low pressure chemical vapor deposition (LPCVD) and physical vapor deposition (PVD) are derived. A pseudosteady state assumption is implicit in the formulation, i.e., the feature dimensions change slowly relative to the time required for the flux to redistribute in response to the changes. Numerical solution of the governing equations provides film and deposition rate profiles as a function of deposition time until the trench is completely filled. Solutions are discussed for selected values of the sticking coefficient from zero to unity. The calculated film profiles are consistent with empirical results which typically show poor uniformity in PVD and step coverage increasing with decreasing sticking coefficient in LPCVD. Film profiles compare well with Monte Carlo based simulations of deposition processes.

No abstract

Wafer-level three-dimensional integration ͑3D͒ is an emerging technology to increase the performance and functionality of integrated circuits ͑ICs͒, with adhesive wafer bonding a key step in one of the attractive technology platforms. In such an application, the dielectric adhesive layer needs to be very uniform, and precise wafer-to-wafer alignment accuracy ͑ϳ1 m͒ of the bonded wafers is required. In this paper we present a new adhesive wafer bonding process that involves partially curing ͑cross-linking͒ of the benzocyclobutene ͑BCB͒ coatings prior to bonding. The partially cured BCB layer essentially does not reflow during bonding, minimizing the impact of inhomogeneities in BCB reflow under compression and/or any shear forces at the bonding interface. The resultant nonuniformity of the BCB layer thickness after wafer bonding is less than 1% of the average layer thickness, and the wafers shift relative to each other during the wafer bonding process less than 1 m ͑average͒ for 200 mm diameter wafers. When bonding two silicon wafers using partially cured BCB, the critical adhesion energy is sufficiently high ͑ജ14 J/m 2 ͒ for subsequent IC processing.

This paper reports a molecular statics study of Cu surface diffusion barriers, particularly the facet–facet and step–facet barriers. The study focuses on two high-symmetry surfaces or facets, Cu{111} and Cu{100}. Our results show that these two barriers are distinct from conventional step barriers and are independent of facet size once it is beyond three atomic layers. Usually, the facet–facet barrier is substantially larger than diffusion barriers on flat surfaces or down monolayer steps, and the step–facet barrier is substantially larger than diffusion barriers along or across monolayer steps. Exceptions do exist. When two Cu{100} facets are involved, the two barriers decrease as the size of the ending facet increases from one layer to two layers, and then increase from two to three (or more) layers. As a result of the large facet–facet and step–facet barriers, surfaces of Cu thin films are of the order of 100 nm. The small facet–facet and step–facet barriers between two Cu{100} facets, when the ending facet is two to three layers, make it difficult to form another Cu{100} facet near one Cu{100} facet. For the same reason, nanowires along ⟨100⟩/{100} on the Cu{100} are unlikely, while nanowires along ⟨110⟩/{111} are feasible.

No abstract

Using classical molecular statics simulations, we show that nanoplate elasticity strongly depends on surface reconstruction and alignment of bond chains. Because of its well-established surface reconstructions and the readily available interatomic potential, diamond-cubic silicon is the prototype of this study. We focus on silicon nanoplates of high-symmetry surfaces, {111} and {100}; with 7×7 and 2×1 reconstructions. Nanoplates with unreconstructed {111} surfaces are elastically stiffer than bulk. In contrast, the same nanoplates with 7×7 reconstructed {111} surfaces are elastically softer than bulk. On {100} surfaces, the 2×1 surface reconstruction has little impact. The bond chains are along one of the two ⟨110⟩ directions, making the two ⟨110⟩ directions nonequivalent. The alignment of the bond chains on the opposite surfaces of a nanoplate dictates its elastic anisotropy. The sensitivity of nanoplate elasticity on details of surface atomic arrangements may impact the application of nanoplates (or nanocantilevers) as sensors.

The effects of thermal cycling on critical adhesion energy and residual stress at the interface between benzocyclobutene ͑BCB͒ and silicon dioxide ͑SiO 2 ͒ coated silicon wafers were evaluated by four-point bending and wafer curvature techniques. Wafers were bonded using BCB in an established ͑baseline͒ process, and the SiO 2 films were deposited by plasma-enhanced chemical vapor deposition ͑PECVD͒. Thermal cycling was done between room temperature and a peak temperature. In thermal cycling performed with 350 and 400°C peak temperatures, the critical adhesion energy increased significantly during the first thermal cycle. The increase in critical adhesion energy is attributed to relaxation of residual stress in PECVD SiO 2 , which in turn is attributed to condensation reactions in those films. Thermal cycling also cures the BCB beyond the ϳ88% achieved in the baseline process, and the residual stress in the BCB is reset at a glass transition temperature corresponding to the increased BCB cure conversion. As more thermal cycles are performed, stress hysteresis in the BCB decreases as the cure stabilizes at 94-95%.