I CAN’T BELIEVE IT’S NOT PLASMA!
Making Catalyzed Thermal ALD Viable for Manufacturing
As scientists and engineers, when we are faced with the limits of our technology, we work tirelessly to innovate around it. In this process, we try our best to maintain the core identity and benefits of the original technology to develop a new “flavor” that potentially unlocks new use cases.
One example for those working in Atomic Layer Deposition, or ALD, was the invention of plasma-enhanced atomic layer deposition, known commonly as PE-ALD.
Introduced in the 90s, PE-ALD sought to combat one limitation of thermal ALD – that some materials could not be deposited by thermal energy only. They needed a bit of a boost to activate the surface and there existed a need in the semiconductor industry for a couple of materials not available with thermal ALD – namely titanium nitride (TiN) and tantalum nitride (TaN)1.
PE-ALD gained a lot of traction since then. It has been hailed for its ability to unlock new materials, including many elemental films, impart flexibility in process development with additional knobs to turn, and even offer higher growth rates or enhanced film properties.
Limitations of plasma-enhanced ALD
However, while the technique addresses some of the limitations of thermal ALD, PE-ALD does not come without its drawbacks. The main downside of plasma is its effectiveness in addressing high-aspect ratio (HAR) structures.
Figure 1 compares trenches of different aspect ratios (AR). The AR of a trench is defined by its height divided by its length. The higher the aspect ratio, the more difficult it is to achieve a conformal coating to the bottom of the trench.
Fundamentally, PE-ALD works by moving highly energized ions and neutral species (radicals, excited vibrational states, etc.) to a substrate surface to give it extra energy for that half-reaction. But unlike thermal ALD, these reactive species cannot move around in perpetuity. Over time, and particularly over distance, recombination of radicals and ions with sidewalls and each other will begin to decrease the number of energetic species available for reaction.
Figure 1
Figure 1. A comparison of trenches with different aspect ratios. PE-ALD works well for structures with an AR of 10 or less. Thermal ALD can achieve conformal films in structures with an AR greater than 100.
In high aspect ratio structures, especially over ARs of 10, due this recombination effect, there will typically not be enough energetic species left to deposit a robust, conformal film2,3. Some plasma-based oxide films can deposit conformal films in HAR structures with enough plasma exposure, but this is particularly an issue with metals films. Thermal ALD, however, is only limited by the amount of time it takes to reach the bottom of an HAR trench.
For example, Figure 2 shows the thickness of films grown by thermal ALD and PEALD along the depth of trenches of varying aspect ratios. The films grown by thermal ALD are uniform along the entire depth of a trench with an aspect ratio of 22:1 (Figure 2a, green triangles). The thickness profile in the 50:1 aspect ratio structure (Figure 2a, purple triangles) was used to illustrate how too short a dose affects conformality in a thermal ALD process. In contrast, films grown by PEALD struggle to uniformly coat the entire trench already at an aspect ratio of 10:1 even with ample plasma doses (Figure 2b, red circles)3.
Figure 2
Figure 2. Film thickness versus depth inside trench structures at varying aspect ratios for films deposited by (a) thermal ALD and (b) PE-ALD3. The green triangles directly compare the ability of thermal ALD (a) and PEALD (b) to conformally coat a structure with an aspect ratio of 22. PEALD struggles with aspect ratios even as small as 10 (red circles). In this example, thermal ALD does not struggle with conformality unless the precursor is not given enough time to reach the bottom of the trench (purple triangles).
For many applications, especially in emerging semiconductor markets, this aspect ratio restriction may limit the ability to deposit conformal films, resulting in voids, high non-uniformity, and ultimately, decreased yields. Copper barrier and seed layers for 3D packaging applications are particularly susceptible to this as high-quality, conformal films are necessary for proper barrier performance against shorting or high-resistivity device failures.
Today, plasmas are commonplace on ALD reactors. We’ve readily adopted PE-ALD as the go-to deposition method for hard-to-deposit materials, yet a growing list of materials will remain difficult to scale down due to the inherent limitations of using plasma. Eventually, we will be unable to conformally coat high AR structures without the need for additional capabilities outside of plasma.
We like to challenge the status quo at Forge Nano so, we’ve asked, “Is it always necessary to use plasma to deposit difficult materials at low temperatures?”
Lowering the barrier
If we go back to the days of general chemistry and asked ourselves how to lower the activation barrier of a reaction, we would all immediately shout, “Catalysts!” And, in a way, we do use plasma as a catalyst for these difficult reactions, but we must change the entire chemistry and hardware to make it happen.
What if there was a way to catalyze reactions while doing thermal ALD without having to sacrifice the benefits of the technique?
It turns out, there is a way. Catalyzed thermal ALD reactions have existed for some time, being explored as early as 19973. They are based on the exact same tenants as all ALD processes – sequential, self-limiting surface reactions – but during each half-cycle, we use a small amount of a Lewis acid or Lewis base, an amine in many instances, to help push the reaction forward.
Let’s use the deposition of SiO2 as an example.
A typical ALD process for the deposition of silica is sequential doses of silicon tetrachloride (SiCl4) and water (H2O). It follows a typical mechanism where strong Si-O bonds are formed, and hydrochloric acid (HCl) is created as a byproduct. Importantly though, as an uncatalyzed thermal ALD process is quite challenging, this reaction proceeds at temperatures >325 °C and requires large precursor exposures (>1000 torr·s)4.
For the catalyzed process, the amine is dosed simultaneously with the precursors, helping to weaken bonds at the surface and allowing for the reactions to proceed at lower temperatures. In the case of the SiCl4 half reaction, strong hydrogen bonds form between the amine and surface hydroxyl groups, greatly weakening the SiO-H bond, increasing the nucleophilicity of the oxygen, and encouraging the bond with the electron-deficient Si of the SiCl4 precursor (Figure 3a)4. Step one done.
The catalyst is also used in the H2O half reaction. It works in a similar fashion via hydrogen bonding but this time with the incoming water (Figure 3b)4. The strong hydrogen bonds make the water’s oxygen a much stronger nucleophile, again allowing oxygen to “attack” the electron-deficient Si on the surface4,5,6,7.
And voila! A catalyzed ALD cycle is complete.
Figure 3
Figure 3. Mechanism of pyridine-catalyzed ALD during the (A) SiCl4 half reaction and (B) H2O half reaction4. In the SiCl4 half reaction, hydrogen bonding of the hydroxyl group with the amine allows the surface-bonded oxygen to more easily bond with the incoming silicon. For the water half reaction, the same principle applies, but now the incoming oxygen can bond easier with the surface-bonded silicon.
And here’s the kicker. The catalyzed reaction proceeds at temperatures as low as 27 °C and uses four orders of magnitude less precursor exposure. It turns out that you, in fact, do not need to use plasma to get low temperature depositions of hard-to-deposit materials, like SiO2. Table 1 summarizes the key process differences between the uncatalyzed and catalyzed reactions.
Table 1
Table 1. Comparison of deposition properties of amine-catalyzed and uncatalyzed SiO2 ALD. The catalyzed ALD process runs at a lower temperature, uses less precursor and has a higher growth rate than the uncatalyzed process.
CRISP: Where chemistry meets engineering
Okay, you might now ask yourself, “If these catalyzed reactions are so great, why are we not using them in industry?” To which we would answer, “Excellent question!”
Clever chemistry can only get you so far and when it comes to catalyzed thermal ALD processes, it has proven difficult to crack as a manufacturing process. Most of the issues come from purging needs.
During a deposition like SiO2, the HCl byproduct can react with the amine catalyst creating an ammonium chloride salt. This salt does have a vapor pressure and can be purged away, but it tends to take much longer than feasible for a manufacturing process. Any shortcuts are bound to leave behind detrimental impurities in the film.
Others have tried to integrate the catalyst into the ALD precursor. One process uses a silane precursor with an ethoxyamino functionality (APTES) to deposit the SiO2 at 150 °C8. While the deposition is successful, ozone is needed as an extra step to remove the surface-bound ligands, and the APTES precursor is extremely difficult to purge. Again, extra steps and extra time are death knells for a manufacturing process.
So, let us introduce you to CRISP, or Catalyzing Reactions for Induced Surface Process.
CRISP is Forge Nano’s solution for making catalyzed thermal ALD possible in manufacturing tools. We have engineered a line of ALD tools, known as ALDx, with a reactor design and hardware that allows us to carry out catalyzed reactions at manufacturing throughputs.
The two features of ALDx equipment that enable CRISP are our proprietary Fast Pneumatic Valves (FPVs) and modulated pressure control in the reaction chamber.
Long story short, the incredible actuating speeds of our FPVs allow us to dose vanishingly small amounts of catalyst into the chamber, while the ability to control the in situ pressure regimes lets us purge the catalyst in record time.
The result: Some of the best dielectric films you could desire, at low temperature, without a single piece of extra hardware.
So how does Forge Nano’s CRISP SiO2 compare with bleeding edge PE-ALD SiO2? Table 2 compares key process characteristics and film properties of SiO2 grown with the ALDx CRISP process and with PE-ALD.
The Forge Nano CRISP process offers many improvements over the PE-ALD process. The deposition time is three times faster with a precursor usage nearly two orders of magnitude lower than achieved with a PE-ALD process. This exemplifies the power of the ALDx equipment to make processes previously dying on the lab shelf into a truly viable manufacturing technique.
Throughput is not the full story, however. Just as notable are the film properties. The Forge Nano CRISP process deposits SiO2 films with higher density and higher breakdown voltage, making them even more suitable for dielectric and barrier applications than PE-ALD SiO2.
Table 2
Table 2. Comparison of SiO2 process and film properties as deposited by the Forge Nano CRISP process in an ALDx reactor and as deposited by a traditional PE-ALD process. The Forge Nano CRISP process shows numerous improvements for the deposition process and as-deposited material.
Forge Nano continues to work on new CRISP depositions, with numerous oxides, nitrides and elemental metal films in development and validation, including HfO2, TiN, TaN and Ru.
Table 3
Table 3. Summary of CRISP chemistries in development by Forge Nano’s Research & Development department.
To put a fine point on it, plasma is not an absolute necessity when it comes to hard-to-deposit materials. With a bit of clever chemistry and a bit of clever engineering, we can preserve the core benefits of thermal ALD processes at lower temperatures and transfer them to manufacturing tools.
If you’re interested in how a CRISP deposition could be implemented in your workflow, contact us for a consultation with our experts.
Sources:
1 J. Vac. Sci. Technol. B 2003, 21, 2231–2261
2Appl. Phys. Rev. 2019, 6, 021302
3J. Electrochem. Soc.2010, 157, G111
4Science 1997, 278, 1934−1936
5Surf. Rev. Lett. 1999, 06, 435−448
6Thin Solid Films 2005, 491, 43−53
7J. Phys. Chem. C 2007, 111, 219−226
8Chem. Mater. 2011, 23, 9, 2312–2316