Predictive Evaluation of Atomic Layer Deposition Characteristics for Synthesis of Al₂O₃ thin Films ^†

Abstract

The atomic layer deposition (ALD) synthesis process is being heavily researched for its conformality, high aspect ratio with thickness control, selective area deposition versatility, and variety of low-temperature oxide, nitride, and transition metal dichalcogenide (TMDC) precursors for a multitude of applications. Repeatability and reproducibility are essential, along with large-scale deposition with high throughput from the commercialization perspective. JMP and Design of Experiment (DoE) are industrially practiced tools to study and reduce process variations. This research paper demonstrates the application of DoE in JMP for the predictive evaluation of ALD for the synthesis of Al₂O₃.

1. Introduction

Statistical methods are the driving tools for industrial problem-solving. The mass production of products is possible due to tolerances, interchangeability of the products, and process capability of instruments based on their accuracy, precision, repeatability, and reproducibility. The basic tool for monitoring variations in the processes is statistical process control. Understanding the processes, their parameters, and responses is easily and accurately studied using JMP tools. JMP is the industry standard for data evaluation and modeling, like DoE. DoE is the cost-effective strategy used to design processes with a significant amount of predictive control. A research gap has been identified related to new synthesis processes being researched. Atomic layer deposition (ALD) lacks emphasis on the use of DoE like tools to establish the repeatability and reproducibility of synthesized 2D thin films. One factor at a time (OFAT) experimentation is still the highly relied on strategy for understanding the cause–effect relationship [1]. The serious drawback of the OFAT experimentation strategy lies in the fact that engineering processes involve process parameters which interact. At the nanoscale, parametric interactions have a significant effect and thus necessitate the usage of experimentation strategies like DoE. To gauge repeatability and reproducibility, Monte Carlo simulations, when integrated with DoE, provide a dependable means to evaluate and predict the performance of processes. Significant research efforts are being devoted to understanding the underlying phenomenological aspects by using density functional theory (DFT) and molecular dynamics (MD) simulations [2,3]. Simple but effective DoE analysis may aid the DFT/MD results to make new processes and product development more predictive and sustainable. Thus, this research attempts to acquaint the reader with the ALD synthesis of Al₂O₃ thin films. Further, we aim to study the effect of the input parameters of ALD, such as the pulse duration of the precursor, pulsing sequence in the ALD recipe on the output response of the thickness, and the stoichiometry of synthesized Al₂O₃ thin films. The effectiveness of DoE with JMP has been examined, with a notable point being made regarding the ease of using Monte Carlo simulations for predictive evaluation. The simultaneous pulsing of ozone and H₂O pulse for the first time is evaluated using DoE. A significantly high deposition rate of Al₂O₃ was observed in this study.

This paper evaluates parametric interactions of the Vecco Savannah thermal ALD system for the deposition of Al₂O₃ thin films. Trimethyl aluminum (TMA) and water (H₂O) were the precursors used for depositing high-K dielectric layers on Si/SiO₂ substrates. The high-quality pin hole free deposition of Al₂O₃ was carried out by varying the pulse time for TMA and H₂O from 0.01 s to 0.02 s. Temperature was changed from 100 to 200 °C. The effect of ozone pulse was also evaluated. The response characteristics of thickness were measured using spectroscopic ellipsometry. The number of cycles was kept constant at 100, as this is known to have a direct relationship with the thickness of Al₂O₃ thin film deposition. JMP was used for the DoE of ALD. It was identified that if the interaction of temperature and H₂O pulse is dominant than the two parameters when considered independently. The ozone pulse, too, has a significant impact. Understanding the parametric interactions is useful in the predictive configuration of the ALD process development for the deposition of Al₂O₃ thin films. ALD is extensively used in industry, and so is DoE, for the purpose of process evaluation or qualification for production. The novelty of this paper is the exceptionally high GPC achieved by pulsing H₂O and ozone together. Furthermore, from the industrial perspective, JMP for DoE is used to understand the relationship of factors with the response characteristics of interest. Various practical considerations for developing an optimized DoE model in JMP are being discussed. Thus, this article provides a simple and effective pathway to understanding the ALD synthesis of Al₂O₃.

2. Materials and Methods

Atomic layer deposition (ALD) is the variant of chemical vapor deposition where the reactants are supplied sequentially as a timed pulse. The pulse of the reactant precursor is usually separated by purging the step of inert gas pulse or having intermittent sufficient purge time. The precursors introduced to the ALD chamber chemisorb in a self-limiting manner. Chemisorption of the precursor is reactive-site-dependent and thus provides very precise thickness control over intricate surfaces with a high conformality and aspect ratio [4]. These aspects of atomically controlled thickness due to self-limiting reaction mechanisms with high conformality and ease of operation have made ALD the most prospective deposition technique for semiconductor and synthetic biological applications.

The ALD process is briefly introduced here to understand the role of factors and responses. Process input parameters are termed factors, and the performance characteristics are termed responses. The accompanying schematic of ALD in Figure 1 is for the Vecco Savannah thermal ALD system. It has four cannisters for ozone (O₃), trimethyl aluminum (TMA), cobalt (Co), and water (H₂O). Strem Chemicals, Air Liquide, or Sigma Aldrich are the vendors who provided ALD precursors with suitable vapor pressure. It is imperative to know the vapor pressure of the precursor at various temperatures or else the cannisters are heated to increase vapor pressure. Volatility due to the vapor pressure of the precursors is critical for chemisorption [5,6]. The Swagelok cylinders are connected to a manifold that provides an independent path for each precursor to the ALD chamber. Solenoid valves control the amount of precursor pulse based on the valve timing. The 6″ or 8″ silicon (100) wafer with a layer of 300 nm SiO₂ was procured from University Wafers. Before the deposition of Al₂O₃, the substrate was cleaned with IPA, rinsed with DI water, and dried with N₂. Figure 2 demonstrates the ALD recipe that details instructions and the channel with units for process parameters.

For the determination of parameters and responses to be studied during experimentation by designing, it is very important to list and identify several parameters and responses (factors) associated with the ALD process in an Ishikawa diagram, as shown in Figure 3. The steps of the recipe in Figure 2 are self-explanatory; however, the deposition characteristics are modified by the order and time of the precursor supply, temperature, purging time, and number of cycles. The Vecco Savannah ALD system maintains a constant vacuum pressure of 0.5 Torr. From the practical perspective, the effectiveness of the DoE model depends on the sensitivity analysis of the process input factors. The sensitivity analysis of the input parameters means whether the limits chosen to vary cause a significant impact on the response characteristics of interest. Considering the linear relationship between the number of pulses and the thickness of the deposition, during this exercise, the number of cycles was kept constant. DoE is effective only if carefully designed with significant engineering insight and the experience of the parameters. Once effectively designed, there are numerous major benefits of DoE from a manufacturing perspective.

3. Results and Discussion

While new ALD variants and material chemistries are developed, consistency and reproducibility need to be addressed [7,8]. The JMP tool is an industry-standard for DoE. The “Custom Design” utility in the JMP-DoE menu equips users with intuitive DoE model development. First, the response characteristics are to be defined. The definition of response characteristics includes the name as the identifier and the targeted response to minimize, maximize, or match the target. For example, the Al₂O₃ layer thickness is to be minimized optimally without compromising the refractive index (IR) or density. If the IR or density are chosen as response characteristics, it is the property of the material so that it has to be matched as close to the expected value as possible. For the DoE of Al₂O₃, the stoichiometric response is to be matched in the ratio of 2:3. Once the response characteristics are amicably defined, the input parameters identified as factors are defined by naming the identifier and lower (−1) and upper (+1) limits of the parameter. The factors identified for v Al₂O₃ synthesis are the TMA pulse of 0.01 to 0.02 s, H₂O pulse of 0.01 to 0.02 s, ozone pulse on/off, and 100 to 200 °C temperature. When ON, the ozone pulse is 0.01 s. Out of the several types of model design available in JMP, response surface methodology (RSM) is chosen. Based on the number and types of responses and factors, the DoE table is derived (Figure 4).

The effects summary table of Figure 5 conclusively identifies that ozone is the most dominant factor for the chosen responses. Temperature is the second most influential factor. Subsequently, all other individual factors are not dominant, but the interaction of the factors is statistically significant. One important question arises here: in what manner are the dominant factors influencing the response? This requires an insight into the setting up of the DoE. As there are two response parameters chosen for analysis, these factors can be dominant with any one or both. Furthermore, the response characteristics for thickness were to minimize and for stoichiometry were “match target”; thus, the effects table points to this desirability defined during the construction of DoE.

The Figure 6 is normal quantile plots for actual vs. predicted values of response characteristics to be interpreted depend upon the expectations set during the construction of DoE. The blue line in the plot is for the mean value, and the red line depicts actual vs. predicted values for thickness and stoichiometry. The narrow plot for thickness indicates that there is good agreement in the actual vs. predicted value and the design significantly aligns with the targeted response. This validates the design and construction of the DoE. It can be inferred that that DoE model is developed to predict thickness more accurately. But conversely, there is large scatter in the stoichiometric ratios of the Al₂O₃ deposited in the model. Thus, the assumptions made during DoE construction with reference to the stoichiometry needed to be more reasonable, or the XPS analysis of the ALD Al₂O₃ needs to be refined. If the DoE model constructed is significant, the factors can be determined from the model to obtain the targeted output. Thus, Figure 7 is the optimized targeted response, where stoichiometry of 0.65 was achieved.

The most efficient feature of the DoE with the JMP is this desirability plot for achieving the targeted response. From the predictor profiler of Figure 8, engineers and scientists can choose the values of the factors to achieve the best combination of desirability. The red line varies highly for thickness at 300 A and at around 0.65 for stoichiometry. To achieve these response conditions, the factors required are TMA pulse of 0.0125, H₂O pulse of 0.015, ozone off, and temperature of 150 °C. If the limits of random variations in the ALD process for TMA and H₂O pulse are known, Monte Carlo simulation has the capability to provide a distribution of response, which can be further used to determine the defective parts per million as well for the specified variability in the ALD process.

Researchers can improve understanding, evaluate the process design strategy, and build repeatability using utilities in JMP [9].

4. Conclusions

The synthesis of Al₂O₃ with ALD has successfully resulted in the development of a predictive DoE model. The influence of the ozone is based on reaction kinetics. Ozone has suitable energetics for ligand removal as well as exchange to form Al₂O₃ from TMA. The removal of ozone during the purging cycle is also responsible for the conformal Al₂O₃ deposition. The dual pulse of H₂O and ozone has resulted in a combination reaction, causing an abruptly high GPC of about 3 A/cycle. The DoE predictor profiler and model was useful to determine optimized input factors to achieve a targeted response. Thus, in conclusion, it can be inferred that the development of the ALD recipe along with the use of JMP tools are effective for developing a predictive DoE model.