ALD has become established as a way to create high-quality, thickness-controlled layers over large areas. However, fabricating intricate structures on a smaller scale usually requires complex additional stages like etching or imprinting, which can compromise sensitive surfaces.

These complicated extra processes can be avoided by using area-selective ALD (AS-ALD), in which the substrate is patterned first using focused electron beam-induced deposition (FEBID). Shapes inscribed directly in this way act as seed layers for subsequent cycles of ALD, causing deposition to occur preferentially upon the predefined pattern, and leaving the larger substrate pristine.

"The particular strength of AS-ALD using FEBID templates is its ability to directly address nanostructured patterns," says Michael Huth, whose research is reported in Nano Futures. However, the size of the structures now achievable means that the monitoring techniques for conventional ALD – like ellipsometry or X-ray photoelectron spectroscopy – are no longer useful. "All techniques for in situ monitoring that we could find in the literature with regard to ALD rely on large-area growth. For nanostructures, electrical conductance monitoring is probably the only technique possible."

To apply this technique to the ALD process, Huth and colleagues at Goethe University in Frankfurt used FEBID to deposit rectangular layers of platinum and carbon between electrodes that had been pre-fabricated on the substrate. When the researchers heated and exposed the sample to pulses of oxygen, the catalytic activity of the platinum grains caused the co-deposited carbon within the seed layers to be removed, resulting in a porous, roughly surfaced structure of pure platinum.

Measuring the conductance of the current pathways between the electrodes allowed the FEBID and subsequent ALD cycles to be monitored closely in real time. The researchers identified several trends in the signal that corresponded to different growth regimes. First, a slow increase in conductance reflected the nucleation process and infilling of the porous seed structure without significant connectivity enhancement. A steeper linear trend resulted from the thickening of the film after the seed islands had coalesced. Finally, a runaway rise in conductance was observed when deposition became unselective, and new current pathways were established on the substrate.

To investigate the dynamics of the ALD process, the researchers varied the flux of oxygen and precursor during the middle, linear regime, inducing periods of faster or slower growth. These were visible as greater or lesser increases in the conductance per ALD cycle. Huth and his team then fed this signal into a genetic algorithm that was allowed to modify the process parameters. When the algorithm was set the goal of maximizing the rate of conductance increase, the result was a system that automatically adjusted the variables to optimize the growth rate and decrease the overall cycling time.

Although in situ conductance monitoring was applied in this case because of specific needs related to FEBID/ALD, the technique can be used much more generally. The same conductance measurements could be fed into a similar genetic algorithm to optimize other ALD processes using different metals.

Next, Huth and his colleagues plan to move their work out of the scanning electron microscope chamber used in their recent experiments. "We are currently setting up a dedicated ALD reactor designed for use with FEBID-templated substrates and allowing for in situ conductance monitoring," says Huth. "This will help us to fully optimize the AS-ALD process to be used in our future research on mesoscopic metallic structures."