AVS 71 Session EL2-TuA: Spectroscopic Ellipsometry Novel Methodologies

Thermal dry etching uses gas-phase reactants in a vacuum and physicochemical reactions based on thermal activation, providing isotropic material removal for lateral patterning without line of sight. Thermal dry etching covers atomic layer etching (ALE) and spontaneous etching. ALE is defined by self-limiting reactions, separated by purge steps. These half-reactions modify and sequentially volatilize a thin film surface, thereby removing material digitally one ultra-thin layer per cycle. Conversely, spontaneous etching is characterized by a sustained reaction of a thin film surface with one etchant only, thereby removing a targeted material with a continuous etch rate.

This invited talk reviews exemplary studies of thermal ALE and spontaneous etching, utilizing in situ spectroscopic ellipsometry (iSE) to reveal thickness changes, self-limiting behavior, synergy between half-reactions, and selectivity between different materials. An iSE instrument (J.A. Woollam Co.) acquired ellipsometric spectra for 5 s at the end of reactant purge steps. Interference enhancement enabled thickness precision of ±0.01 Å.

Al₂O₃ thermal ALE using sequential hydrogen fluoride (HF)/trimethylaluminum (TMA) exposures exhibited a linear etch per cycle (EPC) at 275℃. After initial fluorination, consecutive HF exposures gave virtually no Al₂O₃ thickness loss. This self-limiting behavior corresponded to ideal ALE synergy because all material removal resulted solely from a favorable interaction of the HF/TMA sequence and no etching occurred by either HF or TMA alone. SiO₂ thermal ALE using sequential TMA/HF exposures likewise exhibited a linear EPC at 275℃. Consecutive HF exposures displayed negligible SiO₂ thickness loss, especially after eliminating H₂O during the fluorination step. This self-limiting behavior revealed near-ideal synergy for SiO₂ ALE.

SiN_x thermal ALE using sequential TMA/HF exposures discovered no ALE synergy because consecutive exposures of HF alone caused predominant SiN_x spontaneous etching. This difference between near-ideal versus no ALE synergy obtained great inherent selectivity between major SiN_x versus minor SiO₂ spontaneous etching using anhydrous HF vapor at 275℃. Using anhydrous HF at temperatures >150℃ also discovered facile spontaneous etching of single-crystalline, poly-crystalline, and amorphous Si films with high selectivity compared to SiO₂ retention.

In contrast, co-adsorbing polar molecules with anhydrous HF had a drastic effect. Co-dosing NH₃+HF at 275℃ obtained exceptional selectivity for rapid SiO₂ versus negligible SiN_x spontaneous etching. Similarly, co-adsorbing dimethylamine with HF at 200℃ enabled substantial SiO₂ spontaneous etching.

Laser sources played an important role in the introduction of ellipsometry to the semiconductor industry. This method was largely displaced by spectroscopic methods. Over the past five years, there has been a resurgence of interest in laser ellipsometry particularly for high precision measurements of ultra-thin dielectric films. Advances in laser, phase modulation, optical-design, and electronic instrumentation enable significant improvement in precision and stability relative to previous generations. The industrial requirements and operating principle for an exemplar of this new class of systems will be reviewed. Examples of measurements on thin films including high-k gate dielectrics shall be provided.

The optical properties and thickness of a thin absorbing film depositing on a known substrate can be determined using ellipsometry in real time for any isotropic substrate. The desired film parameters are related to visible light reflection measurements through Maxwells equations, wavelength, and geometry. An emerging method for obtaining film properties from measurements is that of artificial intelligence (AI) in the form of artificial neural networks. One of the primary advantages of the AI method is speed. It can be a thousand times faster than the pre-existing curve fitting methods for example Levenberg- Marquardt. Prior work by the authors focused on ITO on silicon and chromium on BK-7 glass. The work here describes further the developments in the use of such Artificial Intelligence methods to enable real-time, in situ monitoring of thin film growth in a broader range of applications for any absorbing film on any homogeneous, isotropic substrate. Examples are given using a single angle of incidence (55°) and three angles of incidence (55°, 65° and, 75°) for comparison. Thin absorbing films (up to 40 nm) are developed using a multilayer perceptron configuration of the artificial neural network.The networks consist of either 4 or 12 input neurons and 4 output neurons with two hidden layers of 80 neurons each. Other configurations are possible. A separate network is developed independently across the spectrum from which particular wavelengths are selected. Solutionsdo not rely on fitting functions.

Overall predictions depend upon two steps. The first step is the training step in which a large training data set is presented to the artificial neural network (ANN) and an error backpropagation algorithm is employed to incrementally adjust the ANN weights. This step is computationally intensive but only needs to be performed once. The second step is prediction in which ellipsometry measurements are presented to the trained network. This step is fast and may be carried out on a small edge device. Once trained the ANN normally does not need additional training