Note: This post can also be downloaded at the bottom of this page.
Please cite as:
M.F.J. Vos, A.J.M. Mackus, W.M.M. Kessels. Atomic Layer Deposition Process Development – 10 steps to successfully develop, optimize and characterize ALD recipes. 2019, 3. AtomicLimits.
Development of atomic layer deposition (ALD) processes, either to enable fabrication of new materials by ALD, as well as to improve existing ALD materials, is an ongoing effort within the ALD-field (as was also illustrated by our recent blog post with an overview of binary materials prepared by ALD). This also holds for the research performed by our group; a large part of our publications concern novel ALD processes. Speaking from personal perspective, I have set up ALD recipes for deposition of Cobalt (Co), Ruthenium (Ru), Aluminum Fluoride (AlF3) and Molybdenum Oxide (MoO3) during my time in the PMP group (which started back in 2012). Along the way I think I gained some knowledge on the subject of process development, which I would like to (or at least try to) share in this blog. I hope it can serve as a starting point for people who are new to ALD process development, or provide some additional perspectives for those we are more experienced.
Atomic layer deposition (ALD) is a surface-controlled thin film deposition technique that can enable ultimate control over the film thickness, uniformity on large-area substrates and conformality on 3D (nano)structures.1–5 Each ALD cycle consists at least two half-cycles (but can be more complex), containing a precursor dose step and a co-reactant exposure step, separated by purge or pump steps (see Figure 1). Ideally the same amount of material is deposited in each cycle, due to the self-limiting nature of the reactions of the precursor and co-reactant with the surface groups on the substrate. By carrying out a certain number of ALD cycles, the targeted film thickness can be obtained.
Figure 1. Schematic illustration of a typical ALD cycle consisting of two half-cycles. Sequential precursor and co-reactant doses are separated by purge or pump steps, leading to self-limiting film growth. ‘M’ indicates the metal atom, which can for instance be bound to oxygen or nitrogen atoms (in blue), to form a metal oxide or metal nitride, respectively. Precursor ligands are colored green and are eliminated by reaction with the co-reactant before being purged away.
Although the principle of ALD appears to be relatively simple, developing an ALD process is not necessarily a trivial task and it requires careful attention.6 Here, steps are suggested and described to successfully set up an ALD process. Although the steps are roughly ordered chronologically, repetition of some of the steps might be needed at a later stage. In summary, it is advised that the following steps are taken when developing an ALD process:
- Reactant selection: Which precursor and co-reactant will be used for the process?
- Composition: Is the deposited film of the expected material composition?
- Thickness control: Does the film thickness proceed linearly with cycles?
- Saturation: Are the precursor, co-reactant and purge steps in saturation?
- Properties: Does the material have the desired material properties?
- Temperature: Is ALD behavior observed for a range of deposition temperatures?
- Uniformity: Does the film have the same thickness everywhere on the substrate table?
- Conformality: Is the film thickness the same everywhere along 3D-structures?
- Nucleation: Does the initial growth on the substrate differ from the steady-growth regime?
- Other aspects: Are aspects such as safety, stability, reproducibility, etc. met?
1. Precursor and co-reactant selection
Before setting up an ALD process, a suitable combination of precursor and co-reactant has to be decided on. Most importantly, the precursor and co-reactant molecules should consist of the appropriate elements, to obtain a material of the desired composition. In addition, they need to be reactive towards the surface groups present after the preceding subcycle, and in turn result in reactive surface groups after dosing.7 Moreover, volatility, thermal stability and reactivity need to be sufficiently high.8 Other requirements include availability and safety of the chemicals. Furthermore, reactor constraints and the application of the ALD film have to be considered, since they can limit the choice of possible chemicals. In addition to selecting the precursor it has to be determined how the precursor will be delivered to the chamber: vapor drawn, carrier gas assisted (i.e. carrier gas flowing over the precursor), bubbling (i.e. carrier gas flowing through the precursor), etc.
2. Chemical composition
Soon after deposition of the first ALD film, it is valuable to check if the grown material consists of the intended elements. Common ways to investigate the chemical composition are X-ray photoelectron spectroscopy (XPS) and Rutherford backscattering spectroscopy (RBS), although XPS is typically more readily available. In case the material should be conductive, a simple four-point probe conductivity measurement can tell if the material is of high purity. Furthermore, a quick assessment of the refractive index can also indicate whether the desired material is obtained. If the deposited material is very different from the expectation, it might be good to reconsider Step 1, since proceeding can be a waste of time. However, in many cases, optimization of the deposition temperature or dosing and purge times can lead to improvement of the material composition. It is important to realize that the chemical composition and stoichiometry will determine the final material properties.
3. Thickness control
An important characteristic of ALD is the deposition of the same amount of material in each cycle, allowing for ultimate thickness control.9 To confirm this, the thickness or material increase is to be determined per cycle, which is called the growth per cycle (GPC). Determining the GPC can be done both in situ, by following the material increase during deposition, as well as ex situ, by depositing multiple samples with varying number of cycles. Typically, the film thickness is measured (e.g. by spectroscopic ellipsometry), although alternative ways to check for linear growth are by determining the number of deposited atoms (e.g. by Rutherford backscattering spectroscopy) or deposited mass (e.g. by a quartz-crystal microbalance). Figure 2 shows a typical example of the film thickness increasing linearly with the number of ALD cycles. Note that the initial growth on the substrate might occur differently than at a later stage, as will be discussed in Step 9. For this reason the focus should lie on relatively thick films, with thicknesses over ~15 nm.
Figure 2. Typical example showing the film thickness as a function of number of ALD cycles, illustrating the linear growth behavior which is typical for ALD. The growth per cycle (GPC) can be defined as the slope of the thickness as a function of cycles. In case the deposited atoms or mass is measured as function of cycles, the GPC is given in terms of atoms per unit area or grams per unit area deposited per cycle.
To confirm self-limiting growth, one of the key characteristics of ALD, the growth per cycle (GPC) has to be determined as a function of dosing and purge times. In the case of a standard AB-type (i.e. two-step) process, the precursor dosing time, precursor purge time, the co-reactant exposure time and co-reactant purge time need to be optimized. This is done by choosing a relatively long time for three of the four times and keeping those constant, while varying the fourth. This needs to be done for every step and, logically, the first step would be to confirm saturation of the precursor dosing time (See Figure 3a). After this the saturation of the other dosing times can be investigated, and perhaps the whole procedure has to be reiterated with slightly different dosing and purge times based on the findings.
Ideally, a clear plateau is observed when investigating saturation, meaning that the GPC does not increase or decrease when more precursor/co-reactant is dosed, or the purge time is extended.9 Certain phenomena can however cause deviation, such as precursor condensation and precursor decomposition in the case of the precursor saturation curve. In addition, too short co-reactant dosing times can cause impurity incorporation, while too short purge times (See Figure 3b) can lead to a parasitic CVD component (i.e. reaction of the precursor and co-reactant molecules in the gas phase or at the surface), which can impact conformal and uniform deposition. Too long dosing and purge times, on the other hand, can significantly slow down the experiment, or in the case of industrial application, fabrication time. Note that it can also be insightful to perform experiments where only the precursor or co-reactant is dosed, to see if this leads to film deposition or modification of the substrate. Furthermore, saturation should also be confirmed at other process conditions, such as different table temperatures, as will be addressed later.
Figure 3. Idealized growth per cycle (GPC) as a function of (a) reactant (precursor or co-reactant) dosing time and (b) purge time, illustrating the self-limiting ALD behavior. After saturation is reached, dosing additional precursor, or prolonging the purge time does not lead to a change in GPC. Note that it is important to include a measurement for the data point corresponding to 0 s precursor (or co-reactant) dosing time. When purge times are too short, parasitic CVD can occur, leading to a higher GPC.
5. Material properties
Apart from the desired chemical composition, several other material properties are highly important. Depending on the application of the ALD films, it can be of interest to check the following: optical properties (refractive index n, absorption coefficient α), electrical properties (resistivity ρ, carrier density Ne and mobility µ), film and surface morphology (roughness, crystallinity Xc), et cetera. Note that the material properties are strongly linked to the chemical composition of the films, which should be investigated in detail first, and that tuning of the chemical composition can enable different material properties.
6. Temperature dependence
Varying the deposition temperature can have a significant effect on both the growth behavior and the material properties. For this reason, it is advised to deposit films over a range of substrate temperatures, for example between 50 °C and 350 °C. ALD researchers often speak about an ALD window, i.e. a region where the GPC is (nearly) constant, allowing for reliable and repeatable results despite slight temperature variations (See Figure 4).10 However, this constant GPC is not a necessity for an ALD process, and it is mostly important that saturation behavior is found for all temperatures. Effects such as precursor condensation (at low temperature) and precursor decomposition (at high temperature) can namely become visible by verifying saturation at various temperatures. While a constant GPC is a desirable feature, numerous reports actually exist of ALD processes that demonstrate a temperature dependent GPC, but do show saturating ALD behavior over a wide temperature range.
Figure 4. Growth per cycle (GPC) as a function of deposition temperature, showing the idealized ALD window. In addition, precursor related phenomena which can occur at high or low temperature are indicated.
It should be noted that the actual temperature of the sample can be significantly lower than the set temperature of the substrate table, especially when working with a warm (instead of hot) wall reactor at high vacuum. This difference in temperature can be due to the poor thermal contact between the table and the sample, which is pressure dependent.
Another important merit of ALD is the uniformity of the film over large area substrates.11 Often, saturation is verified using coupon samples at the center of the substrate table, which doesn’t mean that precursor or co-reactant dosing is sufficient everywhere. Moreover, non-uniformity is generally a clearer signal of a CVD component than obtained through saturation curves. Hence, a good uniformity is evidence that the process is a well-behaving ALD process. Therefore, it is a recommended that a film is deposited on the largest substrate that fits in the ALD reactor. Most importantly, the thickness variation can easily be determined, either manually, or using an automated mapping stage. However, certain material properties such as composition and resistivity can also be important with respect to uniformity. It might for example be that the thickness uniformity of the deposited film is good, while the resistivity of the film varies significantly over the substrate.
Although often disregarded during ALD process development, the conformality of the ALD film should be taken into account. Conformality is the ability to conformally deposit a film on 3D structures, i.e. with (ideally) no variation in thickness along the structure.12 Instead of the conventional planar coupons, specific samples are required to assess the conformality, which contain trenches or vias. A way to quantify the conformality is to deposit in a vertical trench or via with a certain aspect ratio (AR), and, after preparation of a cross-section, calculate the ratio between the thickness at different locations (See Figure 5). Another approach is to use a specifically designed structure with a lateral via, such as Pillar Hall, allowing for assessment of the thickness profile without the need of a cross-section. In addition to conformality in terms of film thickness, variations in material properties along the 3D structure can also be significant.
Figure 5. Schematic explaining conformality, which can be defined as the film thicknesses at different positions in the trench relative to the thickness at the top of the trench. The trench has an aspect ratio (AR), defined as the height of the trench divided by the width.
9. Nucleation behavior
The film growth during an ALD process might initially show a different behavior than at a later stage of the deposition (See Figure 6). The reason for this is that reactions of the precursor with the material and the chemical groups present on the substrate will probably occur differently than reactions with surface groups on the deposited film. As is shown in Figure 6, it is generally possible to distinguish three cases that can manifest themselves during the initial cycles: linear growth, accelerated or enhanced growth, and delayed growth.13 The nucleation behavior can affect material properties, such as defect or pinhole density, crystalline structure, and film resistivity. Moreover, sometimes differences between growth on various substrates can be observed, which can be the starting point for an area-selective ALD process. This means that a delayed growth can be beneficial in certain cases, although usually fast nucleation is desired for regular applications
Figure 6. Example of the thickness as a function of number of ALD cycles, illustrating the difference between accelerated, linear, and delayed growth during the initial ALD cycles.
The nucleation behavior is strongly dependent on the growth mode of the ALD material. ALD of a metal on a metal oxide often starts with island growth, also Volmer-Weber-type growth, before a closed films is obtained, while certain other materials growth in a layer-by-layer fashion or Frank-Van Der Merwe mode.14 Note that the growth mode depends on the difference in surface energy between the substrate and the deposited film. After investigating the nucleation behavior, it is important to realize that specific thickness effects may start playing a role in the growth behavior after the initial nucleation phase. For instance, crystallization of the film after a certain film thickness is reached might lead to an increase of the GPC.
10. Other important aspects
Several other aspects are also important during ALD process development, although they are not necessarily specific to ALD.
- Safety: Apart from considering the safety of the used chemicals, it is a very good idea to inform yourself whether any poisonous or potentially harmful reaction products are produced during the ALD reactions, or upon contact with the
environment. Moreover, it should be checked if the film material itself is not creating any safety risk.
- Film stability: The stability of the deposited films over time and their sensitivity to the environment (e.g. the ambient) need to be taken into account. Especially in case the film is intended for use in a certain environment, it should be confirmed if it can withstand these conditions (e.g temperature, humidity, etc.).
- Reproducibility: It should be confirmed that the same film thickness and film properties can be obtained by repeating the same deposition recipe. Conditioning of the reactor wall can play a role in this context, although variations in substrate temperature or reactor pressure can also be responsible for unwanted effects.
- Precursor consumption: Efficient use of the precursor becomes more significant for expensive precursors and can for instance be achieved by minimizing the size of the reactor chamber and by avoiding overdosing. Note that the required precursor dosing typically depends on the total surface area of the substrate, which is larger when working on 3D substrates.
- Precursor stability: Prolonged heating of the precursor, which is usually needed to obtain a certain vapor pressure, can sometimes cause degradation of the precursor. For this reason it might be needed to turn off the precursor heating during periods that no deposition runs are being performed.
- Literature comparison: It is advised to verify whether the obtained results are in agreement with previous reports on the same (or very similar) ALD process, especially in terms of chemical composition and growth behavior. If this is not the case, an explanation for the distinctions should be found.
- Device performance: The real test for the developed ALD process will be the application in a working device, such as a transistor or solar cell. Certain phenomena, for instance the presence of defect states or film pinholes or incorporation of particles, might not be detected by basic characterization techniques and might only become visible when testing the film in a device.
Download PDF of this post
(1) Leskelä, M.; Ritala, M. Atomic Layer Deposition (ALD): From Precursors to Thin Film Structures. Thin Solid Films 2002, 409, 138–146.
(2) Puurunen, R. L. Surface Chemistry of Atomic Layer Deposition: A Case Study for the Trimethylaluminum/Water Process. J. Appl. Phys. 2005, 97, 121301.
(3) George, S. M. Atomic Layer Deposition: An Overview. Chem. Rev. 2010, 110, 111–131.
(4) Miikkulainen, V.; Leskelä, M.; Ritala, M.; Puurunen, R. L. Crystallinity of Inorganic Films Grown by Atomic Layer Deposition: Overview and General Trends. J. Appl. Phys. 2013, 113, 021301.
(5) Johnson, R. W.; Hultqvist, A.; Bent, S. F. A Brief Review of Atomic Layer Deposition: From Fundamentals to Applications. Mater. Today 2014, 17, 236–246.
(6) Ritala, M.; Leskela, M. Atomic Layer Deposition. In Handbook of Thin Film Materials Volume 1: Deposition and Processing of Thin Films; 2002; Vol. 1, pp 103–159.
(7) Knoops, H. C. M.; Potts, S. E.; Bol, A. A.; Kessels, W. M. M. Atomic Layer Deposition. In Handbook of Crystal Growth: Thin Films and Epitaxy: Second Edition; Elsevier B.V., 2014; Vol. 3, pp 1101–1134.
(8) Koponen, S. E.; Gordon, P. G.; Barry, S. T. Principles of Precursor Design for Vapour Deposition Methods. Polyhedron 2016, 108, 59–66.
(9) Puurunen, R. L. Growth per Cycle in Atomic Layer Deposition: Real Application Examples of a Theoretical Model. Chem. Vap. Depos. 2003, 9, 327–332.
(10) Leskelä, M.; Niinistö, J.; Ritala, M. 4.05 – Atomic Layer Deposition A2 – Hashmi, Saleem. In Comprehensive Materials Processing; 2014.
(11) Elers, K. E.; Blomberg, T.; Peussa, M.; Aitchison, B.; Haukka, S.; Marcus, S. Film Uniformity in Atomic Layer Deposition. Chem. Vap. Depos. 2006, 12, 13–24.
(12) Dendooven, J.; Detavernier, C. Basics of Atomic Layer Deposition: Growth Characteristics and Conformality. In Atomic Layer Deposition in Energy Conversion Applications; 2017.
(13) Puurunen, R. L.; Vandervorst, W. Island Growth as a Growth Mode in Atomic Layer Deposition: A Phenomenological Model. J. Appl. Phys. 2004, 96, 7686.
(14) Puurunen, R. L. Random Deposition as a Growth Mode in Atomic Layer Deposition. Chem. Vap. Depos. 2004, 10, 159–170.
Dear Atomic Limits authors,
Thank you for a great post! I like the simplicity and clarity. Thank you also for the permission to re-use the materials (discussed in Twitter, https://twitter.com/rlpuu/status/1096129689758953474) – I will likely incorporate some e.g. in my teaching.
There is one term used, though, upon which I would like to raise discussion. This term is ”steady state”, used in the post to refer to the situation when the initial non-linearity of an ALD process has been passed.
I have learnt – having studied chemical engineering – that (quasi) steady state refers to conditions where reaction rates are constant with time. Here, the term is used to refer to something else. Reaction rates are not constant with time at the situation that is referred to as ”steady state”; the GPC is.
In (optimized) ALD processes, reactions rates (change of concentration over time – do not mix this with GPC) are almost never constant. They may be constant (i.e., zero) for a short time when the reactions have saturated (stopped) and the next reactant is not yet introduced. ALD processes are a transient state almost all the time – surface concentration change with time, gas phase composition changes with time, etc.
I have in purpose avoided to use the term ”steady state” for ALD in my earlier works – to which you nicely referred in the blog post, thank you for that. I chose to use ”steady regime” instead. Whether that term is much better, I am not sure; at least it avoided direct mix-up with the established term ”steady state”.
With this comment, I would like to (i) raise discussion in the field on the use of term “steady state” in relation with ALD and (ii) perhaps motivate you to update the terminology of the post in this respect. I am sure that may will follow your post and advice and terminology as well; your choices are important examples. Would you prefer to keep on using the term “steady state”, perhaps a brief justification could be worthwhile to include?
Greetings, Riikka Puurunen, Feb 19, 2019
There has been some discussion on this earlier, I refer to two initial tweets:
•https://twitter.com/rlpuu/status/794526833819664384?s=21 (refers to Chrijs Klein 2007 #EuroCVD proceedings Preface, https://doi.org/10.1016/j.surfcoat.2007.07.014 )
•https://twitter.com/rlpuu/status/794528423406407680?s=21 (refers to my JAP 2005 paper, https://doi.org/10.1063/1.1940727 )
Here are some relevant internet references that I quickly found:
•Wikipedia: steady state https://en.m.wikipedia.org/wiki/Steady_state
•Wikipedia: transient state https://en.m.wikipedia.org/wiki/Transient_state
•Also: quasi steady state (or pseudo steady state): https://www.quora.com/What-is-the-meaning-of-%E2%80%9Cquasi-steady-state-assumption%E2%80%9D
Thank you for the positive feedback and your suggestions. I’m happy to hear you can use parts of the blog. We are certainly supporting people to re-use material for their presentations etc. Moreover, input to this blog post is very much appreciated, and we like to see this as a living document. If things can be done better, we are open to make changes.
Regarding your comment on the terminology; A regime is typically observed in ALD where the GPC is not changing with the number of cycles, which in daily-life we sometimes refer to as ‘steady state growth’. Our group makes a lot of use of in situ spectroscopic ellipsometry, which is why we are tempted to count our time in cycles instead of minutes or seconds ;).
Anyway, I agree that ‘steady state growth’ is not correct from a chemical perspective, since indeed concentration and reaction rates are changing with time (in minutes). I believe that using the correct terminology is very important, and we have updated the text (including the downloadable PDF-file). We are now using the term ‘steady-growth regime’, to distinguish it from the initial nucleation regime.
Finally, I would like to add that the term ‘steady state growth’ is not rarely used in ALD literature. Maybe the community should adapt to the correct terminology, and indeed use ‘steady regime’ or ‘steady-growth regime’ (or something better if one of our colleagues has a good suggestion).
Martijn, February 20, 2019