Through-silicon vias (TSVs) for 3D integration are superficially similar to damascene copper interconnects for integrated circuits. Both etch the via, into either silicon or a dielectric, line it with a barrier against copper diffusion, then deposit a seed layer prior to filling the via with copper using some form of aqueous deposition. In both processes, the integrity of the diffusion barrier and the uniformity of the seed layer are critical to overall reliability and yield.
There are important differences between the two processes, though. TSVs are not just up-scaled interconnects. In particular, assembly and packaging are among the most cost-sensitive aspects of device manufacturing. In almost all cases, a less expensive “good enough” process will be chosen over a more expensive “ideal” process. Cost is perhaps the key factor in deciding whether to use a 3D integration scheme at all.
This emphasis on cost has a profound impact on etch and fill processes for TSVs. These vias are very large by the standards of integrated circuit processes. Depending on the integration scheme, TSV via diameters range from 5 microns to several tens of microns, and TSV etch systems can be asked to remove a dozen microns or more per minute. While the standard Bosch etch process’ alternating etch and deposition steps are able to achieve the required removal rates, scalloped sidewall profiles typically result.
High aspect ratios and scalloped profiles are especially challenging for PVD barrier and seed deposition processes. PVD is a line of sight technique; it is difficult for the sputtering target to “see” all parts of the vias’ inner surfaces, even with the much lower aspect ratios typically seen in IC interconnects. Atomic layer deposition (ALD), proposed as a solution for the analogous problem in integrated circuit interconnects, is far too slow for the amount of material that TSV liners require. On the other hand, the larger dimensions of TSVs mean that the barrier layer can be as much as 10nm to 20nm thick without appreciably increasing total resistance. In IC interconnects, barriers can only be a few nanometers thick, and maintaining the integrity of such thin layers is a significant process challenge.
According to Frederic Raynal, chief technical officer of Alchimer, these differences help make Alchimer’s electrografting process a natural fit for TSVs. Like electrochemical deposition, electrografting applies an electrical current to an aqueous solution containing chemical precursors. In electrografting, however, the deposition reaction is electrically initiated but then proceeds chemically. Thus, electrografting requires much lower current densities. Variations of the process can be used to deposit both metals and polymers, onto both conducting and non-conducting surfaces.
Alchimer has demonstrated several different electrografting processes for use in TSVs. They begin by polymerizing 4-Vinyl pyridine monomers present in an aqueous solution. The resulting polymer forms a covalent bond, giving a conformal coating with excellent adhesion. Preliminary tests suggest that the polymer itself is a promising diffusion barrier. Alternatively, the polymer layer can be followed by deposition of a separate diffusion barrier. Next, an electrografted copper seed layer builds the foundation for copper fill using either electrografting or conventional electrochemical deposition. The bath composition and process parameters can be optimized depending on the desired deposition characteristics.
As an aqueous process, electrografting is inherently conformal. Alchimer has demonstrated successful filling of TSVs with aspect ratios as high as 30:1. The process is also inherently less expensive than dry PVD-based integration schemes: the company estimates a 43% cost of ownership savings for a fully electro/chemical-grafting process flow relative to an equivalent PVD-based flow. The savings are largely due to reduced equipment costs, which more than offset the higher consumable cost of the Alchimer process.
Electrografting, first reported in 1999, is not a new process. But the cost of dry processes and the larger, more forgiving dimensions of TSVs may allow it to come into its own as 3D integration schemes move from the laboratory to the mainstream.