On May 1st, we kicked off our new research project entitled “Spatial Atomic Layer Deposition – More Materials, More Demanding Applications” , granted by the NWO Domain Applied and Engineering Sciences. As the title suggests, the project is geared at making more materials by spatial atomic layer deposition (S-ALD) for more demanding applications, of which I will give some examples below. With the Eindhoven University of Technology (TU/e) as central coordinator, the project will bring together nine companies, most of them working on spatial ALD. As hopefully reflected by the cover image of this post, it is also great to see that this is very much a Dutch endeavor, centered around the Brainport region Eindhoven! The project aims to strengthen our knowledge position on spatial ALD technology further through fundamental research, which in turn will give the S-ALD-hub in the region a strong boost!
In this blog post, I will give a flavor of what this new project is about. How is spatial ALD different from “normal” ALD? Why is it interesting and what kind of aspects will be researched? At the end I will also briefly discuss some of the project partners and their contributions to the project, with as highlight a new spatial ALD tool from the company SALD that is coming to our labs!
Before continuing with the blog post, I’d like to thank especially Vincent Vandalon, Bas van de Loo and co-applicant Erwin Kessels for their contribution to the project proposal, it was really a team effort! Also Paul Poodt is highly thanked for his involvement, being both part-time professor in our group as well as being part of TNO, an institute that obviously has a longstanding history in S-ALD and with whom we’ve collaborated a lot over the years.
How is Spatial ALD different from conventional ALD?
In order to explain the difference between conventional (temporal) ALD and spatial ALD (S-ALD), we made the following animation:
ALD is a vapor phase deposition technique in which a film is grown on a substrate by alternating exposures to precursor and reactant species. Both the precursor and reactant species react with the surface in a self-limiting manner, leading to the controlled formation of a film. The precursor and reactant exposure steps are separated by purge steps to ensure only the self-limiting surface reactions contribute to film growth. As can be seen in the animation, the key difference between temporal and spatial ALD is the way the various ALD steps are separated: in temporal ALD they are separated in time (time-sequenced), in spatial ALD they are separated in space (spatially divided). In other words, in temporal ALD, the substrate is stationary in a (typically low-pressure) reactor into which the various gases are dosed one after the other. In spatial ALD, the substrate is introduced to different zones containing the precursor or reactant gas. This is often done at atmospheric pressure such that the zones can be separated by N2 gas curtains.
Although this project is geared towards spatial ALD, it is probably good to note that I don’t think either temporal or spatial is better. Both are already successful in industry, and whether one or the other is preferred will largely depend on the application:
Temporal ALD tools have become a cornerstone in for example semiconductor (chip) processing, e.g. for the deposition of the high-k metal gate (HKMG) stacks in finFETs, metal-insulator-metal (MIM) stacks in DRAM, or SiO2 spacers for self-aligned multiple patterning (SAxP). As you can probably imagine, ALD isn’t the fastest of deposition methods. Although throughput is always an important aspect, for high-end products such as computers chips relatively many €’s can be spent and throughput can be achieved by having many ALD reactors in parallel, all processing a single wafer containing many high-value chips. For memory devices, throughput is increased by using batch reactors while for SAxP the throughput can be enhanced by rotating semi-batch reactors which are a kind of spatial ALD (see blog). On the other hand, in markets that are even much more cost-competitive and where extremely large areas need to be covered by films, throughput is even much more important. For example, in the production of silicon solar cells, batch ALD systems are used to coat billions of 6” silicon wafers per year with Al2O3 passivation layers at a cost on the order of a few cents per wafer. Such batch ALD system can process close to 10,000 wafers per run!
Spatial ALD tools seem to be especially appealing for a range of upcoming large-area and low-cost commodity products such as solar cells, displays and batteries. For these markets, there are a few good reasons to opt for the spatial variant:Moving substrates through reaction zones makes it possible to do rapid inline (serial) processing. Doing this in a continuous R2R or S2S fashion is also quite a natural step for spatial ALD. Also, since spatial ALD often operates near atmospheric pressure, there is no need for vacuum equipment and fast-acting valves. Since precursor and reactant don’t meet in the tool, there is ideally no deposition on the reactor walls, only on the substrate. Because of this, precursor utilization can be relatively high. The thermal budget (temperature times time) can be lower due to the low cycle time, which can be advantageous with temperature-sensitive materials such as perovskite solar cells and OLEDs. Finally, S-ALD seems well-suited for making compound materials by co-dosing of multiple precursors.
Although this is a great list of pros for spatial ALD, this does not mean there are no challenges! As I will elaborate below, these actually have a central role in this project.
Why “More Materials, More Demanding Applications?”
If you look at research papers, you can find many examples of demonstrators of how ALD layers can be an enabling technology in for example batteries, solar cells and displays. These successes are often demonstrated on lab-scale temporal ALD reactors, and make use of the merits of ALD: for example, the ability to make ultrathin, conformal films with good compositional control. It makes sense that these success stories spark the interest of industry, which in turn looks to e.g S-ALD companies for scale-up.
A logical line of thinking would then be: “It has been done by ALD, so you can do it by spatial ALD, right?” The answer to this question is quite often yes, but not always! If you think about it, the main difference between the growth process for temporal and spatial ALD is the time scale and the (partial) pressure regime. Temporal ALD typically operates at the seconds scale at millibars of pressure, whereas spatial ALD operates at the (tens of) milliseconds scale at atmospheric pressure. In other words, in spatial ALD the growth surface experiences a high pressure of precursor in a short time, while the total dose (in Langmuir: the product of pressure and time) can be similar to temporal ALD. In the case of “ideal” ALD, this should not matter: as long as your surface reactions are in saturation, the growth should be insensitive to the dose, the timescale and pressure regime.
It is often the cases of non-ideality where things get interesting! You can encounter non-idealities in the ALD process itself (e.g. soft saturation) or when the growth is not yet in steady-state: for example during nucleation of the film or when switching between ALD processes for the preparation of mixed or doped materials or nanolaminates. When the reactive surface chemistry deviates from that of steady-state growth, this can easily result in effects such as nucleation delays, etching, exchange reactions or even enhanced growth. If you want to make ultrathin, conformal films with good compositional control, detailed insights into these effects and their causes are a prerequisite. This of course holds for both temporal and spatial ALD! For temporal ALD, great strides have been made in getting a hold on these aspects over the years. Also for S-ALD there is good progress in this respect, although I think it is safe to say that the level of understanding of these phenomena for S-ALD has not reached the level of maturity as for temporal ALD. This project therefore strives to push the level of understanding and know-how of S-ALD further in this respect. This will allow us to chart and understand the opportunities but also the limits of S-ALD better. To give a few examples of aspects we will address: “If you grow a film by S-ALD, does the high precursor partial pressure lead to a change in nucleation and consequently a difference in film closure?“, “If you want to deposit in a high-aspect ratio or porous structure, what are the consequences of the typical short timescales used on the precursor supply and the purge steps?, “What are the opportunities and challenges when co-dosing precursors or using supercycles? For which materials or regimes is either of the methods preferred?”
Strengthening the S-ALD “hub” in Brainport region Eindhoven
As mentioned at the start of this blog, this project aims to strengthen the S-ALD “hub” , bringing together the list of the partners shown below.
Central to the project will be a new spatial ALD tool developed by the company SALD that will be installed in our research lab at the Eindhoven University of Technology. This will be invaluable to the project, especially since it is highly versatile in terms of:
- Substrates: various heights, shapes, types (wafers, foils, …)
- Multiple precursor lines
- Compound materials can be made by both co-dosing and supercycles
- Optional plasma unit
The focus of the project is on fundamental aspects of S-ALD, such that results obtained on the tool will be relevant to all S-ALD users. The research is moreover mostly pre-competitive which means that results can easily be shared among all partners.
Also worth mentioning are the non-Dutch partners Chipmetrics (Finland) and J.A. Woollam (USA). Our group has a longstanding history with J.A. Woollam, with (mainly in–situ) ellipsometry (see previous blog post) playing a key role in most of our ALD work. With Chipmetrics, we have more recently demonstrated how valuable their PillarHallTM structures can be for studying plasma ALD (see our previous blog post). We are eager to see if they can be equally valuable for S-ALD!
To conclude, I think we have a very exciting four year research project ahead of us! We look forward to sharing the results with you at a later stage.