Technical Insight
The attractive attributes of atomic layer deposition
Ammonia-free atomic layer deposition can yield tremendously smooth layers of GaN with incredibly high levels of uniformity.
BY RICHARD STEVENSON
The use of MOCVD to grow GaN films lies at the heart of the production of numerous commercial devices. This growth technology is used to form the LEDs that backlight countless screens and illuminate homes, offices and public spaces; it is used to manufacture blue and green lasers employed for reading discs and projecting images; and it is used to produce HEMTs for RF and power electronics that are deployed in radar, wireless communication and power supplies.
Given this wide-ranging, tremendous success, GaN growth by MOCVD clearly has its merits. But there are also undesirable aspects of this process that creates a film of GaN through the interaction of a gallium-based precursor with ammonia at highly elevated temperatures.
Some of the drawbacks are associated with the hardware that is used. MOCVD reactors designed for GaN growth must be capable of growth temperatures in excess of 1000°C, which makes them expensive to build and run. These tools are also challenging to scale, so significant investment is required when developing reactors to accommodate larger wafers.
Another major downside is the use of ammonia as the nitrogen source. This caustic, extremely hazardous gas has to be handled with great care, with abatement systems needed downstream to deal with any ammonia that has not been consumed in the growth process.
Precursors for the Nanomaster NLD-4000 are housed in a dedicated glove box.
There are also weaknesses associated with GaN films grown by MOCVD. The stoichiometry of the compound is imperfect, the films are not flat on an atomic scale, and the material is plagued with hydrogen. This latter weakness can lead to variations in etch rates, and it can hamper device performance. For example, if hydrogen diffuses to the gate of a GaN transistor, it can shift the threshold voltage.
It is possible to address all of these weaknesses associated with the growth of films of GaN by a novel, ammonia-free form of atomic layer deposition (ALD) pioneered by Nano-Master of Austin, Texas.
The NLD-4000 tool made by this company retails for around $250,000, compared to $1 million or more for an MOCVD reactor, and it is straightforward to scale growth from 2-inch substrates to those of 450 mm in diameter.
While the conventional form of ALD involves the repeated cycling of one precursor and then another to create a film one atomic layer at a time, the process pioneered by Nano-Master has a key difference: in this case, the plasma-formed nitrogen source is applied continuously, cutting cycle times in half.
Tracing the tool
The approach employed in the ammonia-free ALD tool produced by Nano-Master has its roots in the work undertaken by company president and CEO Birol Kuyel. When working at AT&T in the 1980s, he investigated the growth of SiN by plasma-enhanced CVD.
"I found out that there is no way to get the uniformity and stoichiometry simultaneously using ammonia chemistry," says Kuyel. But when he turned to pre-activated nitrogen, they could independently control the stoichiometry and physical characteristics of SiN. "We were successful, had very low levels of hydrogen, and we filed a patent on that," explains Kuyel.
The next milestone came about ten years ago, when a professor at the University of Arkansas approached Kuyel, who by now was CEO at Nano-Master. The academic wanted a conventional MOCVD tool for GaN growth, but could not afford one, so Kuyel suggested a plasma-enhanced variant with N2 rather than NH3. This switch lowered the growth temperature to around 600 °C, which allowed the use of lower-cost, simpler hardware, and eliminated NH3 abatement.
Use of a planar inductively couple plasma leads to very fast pump-down times, thanks to a very small gap between the substrate and the pump chamber.
Plasma-enhanced MOCVD proved a great success. "They were saying that they were getting the world's best gallium nitride," explains Kuyel. So pleasing were these results to these academics that they were unwilling to share the details of their success with Nano-Master. What Kuyel does know, however, is that the growth involved plasma-activated N2, and the GaN films had low levels of hydrogen.
The success led Kuyel and his colleagues to wonder what might be possible by applying a nitrogen-plasma process to ALD. But they did not have the finances to pursue this idea until early last year when they won an order for this novel growth tool. Thanks to the accumulated knowledge from manufacturing other growth systems, they were successful at the first attempt, getting the reactor out of the door before the end of 2015.
To prevent the plasma from damaging the GaN film, the tool features separate chambers for the plasma and for ALD. The nitrogen plasma is formed above the chamber and introduced through a showerhead that "kills" the plasma to inject activated nitrogen. "To do nitrogen chemistry, you don't need nitrogen ions "“ you need natural nitrogen, but in an activated state," says Kuyel.
To prevent oxidation, wafers can be placed in load-lock after growth, where they are flushed with nitrogen and cooled.
Using pulses of gallium precursors in bursts with a duration of typically 20 ms to 60 ms and growth temperatures of between 200°C and 400°C, the tool is capable of producing films with a thickness independent of temperature, pulse width and cycle time. "We were able to grow a number of [atomic] layers based only on the number of cycles we used," says Kuyel.
The growth rate is governed by the number of cycles, and has little dependence on temperature.
Another strength of the Nano-Master ALD tool is its very fast pump-down time to a typical growth pressure of 0.2-0.3 Torr. This stems from the use of a planar inductively coupled plasma system, rather than one based on a cylinder or a coil. By selecting this system architecture, the gap between the substrate and the pump chamber can be just a few centimetres.
Fantastically flat films
The tool is capable of producing some incredibly impressive results. For example, film thickness over a 6-inch substrate varies by less than 1 à…. "All those points [that are measured] have the same number of atoms," argues Kuyel. "MOCVD will depend on the flow patterns, reactivity concentrations and so on. This doesn't."
When films of GaN are grown on silicon they form a very strong bond with the substrate. "We did not even have to do special preparation of the silicon surface," remarks Kuyel.
The Achilles' heel of GaN growth with the Nano-Master ALD tool is the slow growth rate. Although it is faster than that of many ALD processes, thanks to pulsing of just the gallium source, growth rates are at best a few Angström per second. This effectively rules out the growth of thick films in a single-wafer processing chamber for chip production.
As the take-up of this tool is still in it infancy, it is not yet clear where it will find its greatest use. One option, says Kuyel, is to build a composite system with an ALD chamber and an MOCVD chamber and a transfer between them under high vacuum. He believes that such an approach could be used for making blue lasers, which would benefit from the high-quality foundation that results from ALD.
Another possible use of the Nano-Master ALD system is the production of mirrors. Using a gallium-tri-chloride source, it is possible to grow films with a surface roughness on the pico-metre scale. "This is better than mirror finishes that you obtain for optical elements," claims Kuyel. "That's why I'm thinking that making mirrors from multi-layer dielectrics would be possible."
The plans for the future involve raising awareness of the tool and driving up its sales. Kuyel wants to build dual-chamber systems, which could either be: a pair of ALD chambers; a combined ALD and MOCVD multi-chamber cluster tool; or multi-wafer batch processing tools. There is also the opportunity of using the tools to deposit oxides, such as HfO2, that could aid development of next-generation logic. So there is clearly much promise for this novel ALD tool.
Films of GaN that are deposited by ammonia-free ALD have a tremendous degree of flatness.
