Quadrupole Mass Spectrometers (QMS) have been used widely in atomic layer deposition (ALD) for process optimization and troubleshooting. Unfortunately, QMS has had little application for in-situ process monitoring owing to its short lifetime and less persuasive value of process monitoring. This study is to improve the QMS sampling system design for ALD processes, which includes the ALD gas sampling and delivery, inlet valve design, and the range and control of the sampling system temperature. In addition, the QMS hardware and operating methods are also upgraded to improve the sensitivity and its lifetime for ALD processes. This QMS system has been applied to ALD processes for in-situ monitoring of chamber base vacuum, ALD precursors, process reaction by-products, and chamber contamination.

Pyroelectric calorimetry presents a new approach for the in situ study of the thermodynamics and kinetics of ALD reactions with unmatched thermal sensitivity and temporal resolution. Complementary techniques such as quartz crystal microgravimetry, spectroscopic ellipsometry, and mass spectrometry are limited by sampling times that are typically on the order of 50ms – 1s. We demonstrate response times down to 50ns and sensitivities of < 1μJ/cm², which corresponds to less than 1% of the heat generated by some ALD half reactions.

Large temperature changes, >10°C, as a result of ALD surface reactions have been observed on high surface area substrates. However, attempts to quantitatively measure reaction heats on planar substrates, ≪1°C, using thermocouples or pyroelectric detectors have previously been limited by hardware with insufficient thermal and temporal resolution, difficulties accounting for heat lost to cooling, and challenges with external calibration. To overcome these issues we designed and built pyroelectric calorimeters compatible with ALD systems, modeled the transient thermal and electric response, and quantitatively calibrated the system using a laser to provide precise heat generation.

This work was inspired by the techniques used in pyroelectric adsorption calorimetry, but the approaches used in this field resulted in unacceptable tradeoffs in responsivity in order to achieve the temporal resolution needed for ALD. A new calorimeter design was implemented to suspend the pyroelectric crystal with minimal physical contact. This reduced conductive cooling, enabling improved temperature sensitivity and faster responsivity.

Previously, ALD reaction energies have been computationally predicted, but not experimentally measured. ALD calorimetry contributes a new experimental technique with a wide range of applications including the study of fundamental reaction thermodynamics, the impact of non-ideal surfaces on nucleation, selective defect passivation, and growth inhibition.

This material is based upon work supported by Laboratory Directed Research and Development (LDRD) funding from Argonne National Laboratory, provided by the Director, Office of Science, of the U.S. Department of Energy under Contract No. DE-AC02-06CH11357

Despite the numerous applications of atomic layer deposition (ALD), used e.g. in the semiconductors industry to produce thin films with thickness control at the Ångström level and tunable composition, only limited knowledge exists about the kinetics of the surface chemical reaction, the reaction mechanism, and the role of the substrate. ALD-grown thin high-k oxide films, such as HfO₂ or Al₂O₃, help to significantly improve device performance of III-V semiconductor-based MOSFETs. Those materials, such as InAs, come along with an electron mobility 20 times larger than that of silicon, but unfortunately with a high defect density at the interface with its oxide. It has been demonstrated that so-called "self-cleaning" occurs during the early stage of the high-k ALD, in which the growing high-κ oxide substitutes the unwanted native oxide. However, in spite of all progress, the III-V/oxide interface still suffers from incomplete self-cleaning and too high defect levels.

In this work, we present a time-resolved study of the self-cleaning effect during the first ALD half cycle. We performed ALD of HfO₂ on InAs in the reaction cell of a synchrotron-based ambient pressure X-ray photoemission spectroscopy (APXPS) setup. Tetrakis(dimethylamido)-hafnium (TDMA-Hf) is used as the metallorganic precursor, and we followed in real time all relevant core levels to monitor the surface species during the deposition. Our findings challenge the established view of the ALD process based on the ligand exchange model and the surface's passive role. In fact, we establish a two-step reaction based on molecular adsorption followed by ligand exchange and the immediate formation of HfO_x, in which the native oxide acts as the oxygen source. In more detail, we find evidence for an As-In-O-Hf bonding configuration at the semiconductor/high-k interface.

Hence, time-resolved APXPS enables us to further understand the surface chemical process in ALD and how the III-V semiconductor/high-k interface can further be improved by optimizing ALD, promising superior device performance.

Process development in atomic layer deposition (ALD) is a time-consuming process requiring optimization of pulse / purge times and temperature for controlling growth rates.For example, a new precursor for ceria ALD process using Ce(iPrCp)₂(N-iPr-amd) takes tens of experiments for optimization [Golalikhani et al. J. Vac. Sci. Technol 36.5 (2018): 051502]. Against this backdrop, we report a powerful methodology of sequential pulsing aided by in situ ellipsometry, which is used to arrive at an optimal ceria ALD process in 10 experimental runs, requiring less than a day. We have performed saturation pulse experiments of precursor and oxidizer on a silicon surface monitored using in situ ellipsometry. This approach generates a time-temperature-thickness 3D map and makes it convenient to visualize the optimal pulse/purge times and a temperature ‘valley’ for the process. In situ ellipsometry data from saturation pulse experiments were correlated with adsorption isotherms. The film's growth rate deposited in the ALD temperature window was ~ 2 Å/cycle; the deposited film was further characterized using XPS, XRD, UV-Vis, and AFM. Furthermore, we develop a Machine Learning (ML)-based modeling to predict the growth rate for various process parameters. The transparency of our ML model is enhanced through explainability.