<<< Back | Print


ALD Breaks Materials, Conformality Barriers
At a Glance
ALD, widely considered the next step in thin-film deposition, promises unparalleled deposition control as well as capabilities impossible with current methods — all of it at the smallest possible dimensions. Three-dimensional films that coat around surfaces, air-gap structures, 3-D vertical structures without bottoms or with covered bottoms, and other new topographies are almost within reach.
Sidebars:
Doing Molecular ALD

Atomic layer deposition (ALD) is a thin-film deposition technique developed in Finland more than 20 years ago.1 Like CVD, chemical reactants in ALD are introduced into the deposition chamber as gases (Fig. 1). In CVD, all reactants required for film growth are simultaneously exposed to a wafer surface, where they continuously deposit a thin film. CVD deposition rates can be surface-limited at lower temperatures, or mass-flow-limited at higher temperatures where deposition rates are relatively higher. In ALD, reactants are supplied in pulses, separated from each other in the flow stream by a purge gas. Each reactant pulse chemically reacts with the wafer surface.2 It is the chemical reactions between the reactants and the surface that make ALD a self-limiting process inherently capable of achieving precise monolayer growth3 (Fig. 2).

ALD has been used to deposit a variety of materials, including II-VI and III-V compound semiconductors, elemental silicon, SiO2, and metal oxides and nitrides. When ALD is used to deposit a film containing a metal such as Al, Ta, Hf, etc., and a non-metal constituent such as O or N, the first reactant can contain the metal and the second the non-metal. The first pulse deposits a metal-containing layer, and the second one reacts with that layer to form the complete film of metal oxide or metal nitride. Both reactants react chemically with the surface on which they deposit, and each reaction is self-limiting. Thus, ALD is a self-limiting, wholly conformal process, enabling film thickness control to previously impossible accuracy levels. Film thickness can be controlled to within a monolayer solely by counting pulses. Depending on the process, films can be amorphous, epitaxial or polycrystalline, and extremely uniform and conformal. ALD typically has a low deposition rate, on the order of 1 Å/cycle, where each cycle lasts a few seconds. However, such rates are suitable for ultrathin films whose thicknesses range from 3 to 100 Å.

1. ALD, the next step in thin-film deposition technology, provides unparalleled deposition control at the smallest possible dimensions. (Source: Genus)

ALD reaction temperatures are typically in the 200-400°C range. When the deposition temperature is too high, chemical bonding cannot be sustained, or the density of chemically reactive sites is reduced, decreasing deposition rate. At low temperatures, the deposition rate increases because the chemisorption and film-formation reactions are thermally activated. Therefore, as ALD process temperatures are varied from low to high, the reaction rate first increases, reaches a maximum, then decreases. This creates a temperature window within which ALD can be carried out at reasonable deposition rates.

ALD reaction rate at low temperatures can be increased by using an extremely reactive element, such as a radical.4 For thermal- or radical-assisted ALD, the process's unique enabling conformality is related to surface-saturating and self-limiting reactions originating in complete chemical reactions in each half-reaction of the ALD process. Chemically reacted layers have self-limiting properties, providing ideal conformality and smooth high-density films. This is in contrast to processes involving physically adsorbed layers, in which thickness must be controlled by manipulating the reactant dose. These physisorption-based processes are not self-limiting, and thus lack the inherent monolayer thickness control of the ALD process.

"There's no doubt that ALD is the technology for deposition beyond CVD," said Moris Kori, vice president and general manager of the CMI Product Business Group at Applied Materials (Santa Clara, Calif.) "The only question is when, and this depends on the technology being available for the films of interest and on having systems that are production-worthy."

"The situation today favors ALD, or ALCVD as we call it," said Suvi Haukka, research manager at ASM Microchemistry (Espoo, Finland). "The semiconductor industry recognizes that ALCVD will play an important role as a future thin-film deposition technique."

"After almost a quarter century, ALD is a technique whose time has finally come," added Tom Seidel, CTO at Genus (Sunnyvale, Calif.).

Positioning ALD

For ICs, ALD fits into three sectors, according to Steven Rossnagel, research staff member of the IBM Research Division at the T.J. Watson Research Center (Yorktown Heights, N.Y.). "First, in the DRAM world for dielectrics — deep-trench dielectrics. We're looking at very high-aspect-ratio trenches, somewhat conventional dielectrics — things like aluminum oxide and simple oxides, not the perovskites like BST. ALD is compatible with this." IBM is actively working with ALD for DRAM development, which is one of the more mature technology applications for it. There are commercial tools, from both Genus and ASM, and there is considerable interest in the groundbreaking work done in the nanolaminant, dielectric area. There is also some interesting physics in these applications, and this may allow further shrinkage of DRAM trench dimensions and even higher aspect ratios. "Nanolaminants aren't really manufacturable yet, but it's intriguing to see these 20 Å films in alternating layers," Rossnagel said.

2. Chemical reactants in the ALD process are introduced into the deposition chamber as gases, and supplied in pulses delivered to the reactor at different times. Reactants are separated from one another in the flow stream by a purge gas. Each reactant pulse chemically reacts with the wafer surface, making ALD a self-limiting process capable of precise monolayer growth. (Source: ASM)

The second area is gate dielectrics. "This is to replace SiO2, which is running out of steam as it gets thinner. Much remains to be done, and the limitation seems to be that, although you do not want to make SiO2, you're making a little of it at the interface between the ALD material and the underlying silicon wafer," Rossnagel said. That bit of oxidized silicon can significantly degrade the junction operation. "There's a lot of work there — aluminum oxide, hafnium oxide — that's possibly the most visible work in the ALD realm," he added.

Thickness control and ALD go hand-in-glove. If exactly six layers are needed, where each atom counts, ALD is the solution. "ALD's advantage is that you make the thickness by counting," Rossnagel explained. "With virtually every other film deposition technology, you calibrate the rate by making a thicker sample, get a rate-per-unit time, and then take less time for a thinner sample. It's difficult to produce an accurate film at 10 Å using that technique; whereas with ALD, if you want 10 Å you just do 12 pulses or whatever number is needed for the specific thickness and stop. It's ideal for gate dielectrics." The situation is similar for gate metals. As gate dielectrics change from the current oxidized silicon or silicon oxynitride, it becomes necessary to change the gate metal.

Rossnagel's third area is the interconnect back end: diffusion barrier, seed layer, etc. "There, PVD or ionized PVD is used, as well as permutations of things like hollow cathode magnetron, SIP, etc. However, these are all basically sputtering, and it's generally assumed that sputtering won't work anymore. Over the years, we've been able to eke it out a little longer, figuring that CVD will come in for the next generation, but it never does. PVD was first extended by collimation, then by ionization, and there are still a couple of tricks left, but sooner or later it'll run out of steam because PVD atoms tend to stick where they hit, requiring considerable extra work to produce a conformal film"

ALD and copper

ALD seems well-matched for the interconnect back-end barrier world but less so for seeds, the copper seed that would be used for plating. The barrier is the tantalum system or perhaps the titanium system, and ALD work has been ongoing with those materials. "It isn't ready for production yet, and won't be needed for another two or three years," Rossnagel said. "Eventually you'll have a problem in the back end with the volume of copper that you have left in the line. In the line and vias you get <1000 Å dimensions — 800, 700 Å— so every atom removed from that via or line to put in a liner atom reduces conductivity and frequency response."

Future interconnect liners will have to be no more than five to 10 atoms thick, highly conformal and tailored, and ALD is perfect for this. "You do a couple of atoms of tantalum, an atom or two of nitrogen, one of titanium and you get a multifunction film stack only six or seven atoms thick, which doesn't displace too much copper," Rossnagel said. ALD works well in the Ta, Ti and TiN systems, but it is more difficult for TaN. Copper ALD has been demonstrated, but it is more chemically complicated.

3. Top-view SEM images of tungsten contact plugs after CMP. Figure a shows ALD tungsten nucleation followed by CVD tungsten bulk fill. The average dimple size is <50 Å and dimple density in the array is <5%. Figure b shows a conventional CVD tungsten fill, with an average dimple size of 400-500 Å. Array density is >60%. The diagram depicts a scenario where copper wiring on top of the tungsten plug diffuses through tungsten dimple holes. (Source: Applied Materials)

Selectivity is a hurdle for ALD. With ALD, precursor molecules uniformly sit on all the surfaces in the structure producing a monolayer. But it would be best if they did not land or remain on certain surfaces. It is a common problem; selective CVD has always been a headache. "I believe ALD is the obvious direction as we begin making 800 Å structures," Rossnagel added. "CVD will probably be skipped as a generation. Seed layers are still a mixed bag."

Platforms and ALD

Both Applied and Novellus have developed ALD into an attractive, unique process for tungsten plug applications. The underlying problem with the CVD WF6 chemistry used for plugs is film nucleation. Over the years, there have been various fixes — such as high-pressure silane exposure — to facilitate nucleation. However, using an ALD approach to the same process chemistry results in highly conformal nucleation of thin W films. Once the film is formed to a few tens of angstroms thickness using ALD steps, the process is changed back to conventional CVD to complete the filling of the via. This can be done in the same chamber, or sequentially in an adjacent integrated chamber. It appears that this ALD/CVD process sequence will be widely used in the 300 mm generation. Both companies have really come up with an elegant solution.

Applied has combined a dedicated ALD tungsten nucleation chamber with a CVD tungsten bulk fill chamber. Separating the nucleation and bulk deposition chambers — in what is described as a "production-ready" system — enables a void-free, high-throughput plug fill for <100 nm devices.

"This works for front-end applications such as high-k materials, gates and capacitors — things like aluminum oxide, hafnium, etc.," said Applied's Kori. However, for ALD to become mainstream, it must find a place in device areas, especially on the interconnect side, where traditionally the growth, interest and most benefit are.

"We must look at ALD for thin metal films, specifically for barrier applications," he said. "It began when we introduced ALD tungsten, because it's the interconnect's first layer, and even under copper it is critical that we have a good tungsten plug past 100 nm. ALD's tungsten nucleation not only enables an almost perfect plug film, but also the seed layer itself, which is a barrier." Applied considers the concept as proven and has a chamber that is in pilot environments (Fig. 3).

"Next, we must determine whether we can move up into the device," Kori explained. "By this I mean copper. Today, this can be addressed at the 130 nm node. However, as we approach 100 nm, we must think beyond current, essentially PVD-based processes, and look at CVD or ALD — copper barrier, seed, and possibly other applications." The issue is one of technology: finding the right chemistry and precursor, which not only enables the right film to be deposited, but makes the process production-worthy. This has been copper CVD's biggest hurdle; nobody has yet figured out how to do it in a production-worthy way. ALD faces the same issues.

At 100 nm, conformality requirements make ALD necessary. "It depends on yield objectives," Kori said. "Some will say, 'We're getting the yields we want with PVD or CVD.' At some point they'll realize that, by switching to ALD, they attain another level. This happened with the TiN barrier. For years TiN meant PVD. It took time, but by going to MOCVD TiN there was a great yield improvement, and it became the industry's de facto standard. We now have an ALD TiN process that demonstrates virtually 100% conformality on 7.25 µm-deep trench structures with uniform composition along the substrate's entire topography."

From a hardware standpoint, ALD has fewer unknowns. "The switch from CVD to ALD — it's an incremental, although critical, hardware change," Kori said. "Our existing CVD process chambers can be upgraded to ALD chambers. ALD is also chemical deposition; you just do it differently. The transition will be easier."

Genus introduced its ALD products on already-developed CVD wafer-handling and process-module platform and architectures. Its ALD platforms are extensions of its CVD platforms, and many of the parts are common to a platform already used for high-volume CVD processes. Its Lynx platform has run marathon ALD processes without maintenance to the 2000-wafer level.

If ALD can be used to improve deposition, a good seed may enable better and easier plating by expanding plating's process window. However, there are still difficulties. The current barrier is PVD-based-tantalum, TaN. If ALD TaN is done, it becomes necessary to use a Ta metal organic precursor, which is expensive and difficult to handle. Using alternative materials is problematic, because manufacturers prefer to use familiar ones, and they can affect many things throughout a process.

A hardware problem lies in the chamber. "Like CVD, ALCVD coats the reaction chamber," said ASM's Haukka. "The film is not only deposited on the wafer, but on reaction chamber surfaces as well. For production, ways must be found to clean the chamber in situ. Much will depend on the new materials chosen."

ALD and HAR structures

ALD can provide conformality for very high-aspect-ratio structures. "There's no peer to this capability," said Genus' Seidel, adding he expects the first applications to take place where conformality needs cannot be met with any other standard method. "People are going to deeper, higher-aspect-ratio structures, demanding uniformity for the whole topology, even for aggressive feature sizes with aspect ratios greater than 30:1."

In this ideal implementation, ALD surface-saturates all the layers as they run their reactions to completion. "You won't get different compositions anywhere on the film such as, for instance, along the sidewall," Seidel said. "A major problem with some CVD and PVD processes is that composition can vary along the sidewall of these very high-aspect-ratio structures." He added that this problem is expected to occur along deep trench and high-aspect-ratio topologies. "Its solution should come with surface-saturation chemistry, along with conformality."

ALD's first and foremost advantage is conformality, and then uniformity across the wafer and from wafer to wafer. "We've demonstrated that you can get control to better than a couple of angstroms," Seidel explained. "Besides process control, the industry needs to support advances in metrology for precision and reproducibility. When working at the angstrom level, there are concerns about making measurements that verify angstrom-level control." Presently, measurement technology is limited to the level of an angstrom. As layers thin to the tens of angstroms level, and percentage control is needed for verifying corresponding good percentage control in device performance, a way must be found to measure or evaluate by interference at sub-angstrom levels. "This possibly might be done through some sort of a sophisticated closed-loop control optical ellipsometric method," he added.

Barrier-breaking materials

ALD provides the means to create entire classes of materials. Although semiconductor manufacturers resist changes in materials, preferring instead to extend their use and capabilities to the maximum, it is clear that existing materials are about to hit barriers. "We do demos and provide some support for several metal oxides and metal nitrides," Seidel said. "People know what they are, but in the case of metal nitrides they would like lower resistivity, while keeping the benefits of conformality, low-trace impurities, etc." ALD will meet the need for new dielectric materials for capacitors, gates and metallic materials for capacitor electrodes and barriers.

4. ALD provides hereto impossible materials engineering possibilities. Shown are SEMs of nanolaminates, alternating Al2O3 and Ta2O5. There are five layers of Al2O5, each 14 Å, and five layers of Ta2O5, 27 Å each. (Source: Genus)

ALD enables new materials to be engineered in their layering. It is possible to put down one material followed by another, even in the same tool, preserving the qualities of the interface within the same ALD chamber. "You can deposit one dielectric followed by another with very little hit on throughput," Seidel noted. "We call it 'no-throughput penalty' for doing alloys or nanolaminates." (Fig. 4)

ALD technology's sweet spot — for now — is in the 10-50 Å area. At 100 Å or beyond, ALD's slowness leads to throughput problems. There are several approaches to improve throughput. The most obvious one is parallel processing — using parallel cluster tools, or exposing several wafers at the same time. Also, new chemistries are being explored (see "Doing Molecular ALD") to produce precursors that permit the ALD of two elements in one self-saturating chemical reaction. "If you're trying to make, for example, a silicate such as HfSixOy, and you have a precursor that'll deposit the silicon and the metal together directly, then you don't need a two-step, just a one-step, process," Seidel said.

Today, ALD's primary driver force is atomic layer conformal low-temperature films. The industry thinks differently about wafers than it does about materials. It tends to look either to the old world of reactive etching, planar films and gap fill, or damascene technology. If it becomes possible to make 3-D films that can coat around surfaces, phenomenal new structures are in the offing — things like the so-called air-gap structures, 3-D vertical structures without bottoms or with covered bottoms. It opens up a world of new topographies of unguessed possibilities, all of it at the tiniest possible dimensions.


References
  1. U.S. Patent 4,058,420.
  2. Atomic Layer Epitaxy, T. Suntola and M. Simpson, Eds., Blackie and Sons, 1990.
  3. Id., p. 4.
  4. S. Imai, et al, "Atomic Layer Epitaxy of Si Using Atomic H," Thin Solid Films 225, 1993, p. 168; S.M. Bedair, "Atomic Layer Epitaxy Deposition Processes," J. Vac. Sci. Technol. B, January/February 1994.
 

Doing Molecular ALD

Roy Gordon, Harvard University, Cambridge, Mass.

Layers of clusters of atoms may be deposited by ALD (Figure). For example, a layer of SiO4 clusters results from the reaction of (C4H9O)3 SiOH with surfaces previously reacted with a metal amide such as Hf(NMe2)4. An amorphous hafnium silicate is deposited at a higher rate — 3-4 Å/cycle — than normal ALD reactions yielding single atoms (typically <1 Å/cycle). A recent paper by me and my co-workers ("Vapor Deposition of Metal Oxides and Silicates: Possible Gate Insulators for Future Microelectronics," Chem. of Mat., Aug. 20, 2001), reports this ALD process.


Cross-sectional SEM of a HfO2coating inside a hole with an aspect ratio of 35:1. (Source: Joseph Shepard, IBM DRAM Development)

Another advantage of these reagents is that the oxygen is already bound to silicon, so there is no tendency to oxidize a silicon substrate. This feature is particularly important for potential applications of this material as a high-k gate oxide. After a couple of layers of hafnium silicate have been deposited to protect the silicon from oxidation, pulses of water vapor can be used in place of the silanol to deposit pure hafnium oxide, a higher-k material than the silicate. The films are smooth (<0.5% roughness by AFM), and highly uniform in thickness over a 300 mm deposition zone. As shown in the Figure, the step coverage is outstanding, showing a coating inside a hole 0.17 µm wide by 6 µm deep. No chlorine is present in these reactants, so there is no chloride contamination of the films or corrosion of substrates or equipment. These characteristics make this ALD process a good candidate for producing trench capacitors as well as gate oxides.

Analogous ALD reactions could be used to make silicates and oxides of most metals. Amides provide a wider selection of metals than are available from metal chloride sources traditionally used for ALD, because some of the metal chlorides are not sufficiently volatile for use in ALD.

Applied Materials http://www.appliedmaterials.com/
ASM Microchemistry http://www.asm.com/
IBM http://www.chips.ibm.com/
Genus http://www.genus.com/

 


<<< Back | Print