In situ XPS study on atomic layer etching of Fe thin film using Cl₂ and acetylacetone

Magnetic random-access memory (MRAM) is a promising candidate for the next generation of semiconductor memory because of high read and write speeds, longer endurance, and lower energy dissipation.^1,2 The key unit in MRAM is the magnetic tunnel junction (MTJ),^3,4 which is a stack of magnetic metal layers, including Fe, Co, and Ni, with a metal oxide tunneling barrier in the middle. How to effectively etch these magnetic metals has become one of the most critical challenges in the fabrication of MRAM.^5,6

Ion milling was one of the first technologies used to etch the magnetic metals in MTJ fabrication.^6–8 In this process, redeposition of the metals on the sidewalls required subsequent processing steps.⁹ This problem was then overcome by using reactive ion etching (RIE). In the past few decades, halogen plasma RIE has also been used for metal etch.^10–12 It can react with the metal surfaces to quickly form metal halide complexes; however, due to their low volatility, a relatively high temperature is required for sublimation and a post-etch cleaning is generally needed. Studies also show that halide plasma may lead to degradation of magnetic properties¹³ or deformation of patterns.¹⁴ Recently, some organic etchants like CH₄,^15–18 CO/NH₃,^19–21 and CH₃OH (Refs. 22–24) were used for RIE, in the hope that metal films were etched by the formation of volatile complexes. An organic gas plasma improves the sidewall etching profile and the elemental selectivity compared with a halogen gas plasma. However, because of the lack of direct evidence of volatile coordination complex as an etching product, it is still uncertain if the etching is a result from chemical reaction or ion milling. Organic plasma also has another problem in that a carbon layer may form on the surface and passivate the sample surface, limiting further etching.

Given so many problems in the continual etching processes, recently multistep processes such as atomic layer etching (ALE) or quasi-ALE utilizing separate doses of different gases was developed. ALE is the time-reversed process of atomic layer deposition,^25,26 where multiple gas precursors are dosed sequentially and etch the film layer-by-layer in an atomic scale. Atomic layer etching of polymer,²⁷ Si,²⁸ or metal oxides^29,30 has been well studied in the past. Recently, cyclic etching for transition metals using Cl₂ and H₂ has been reported.^31,32 ALE for transition metal etching using O₂ and organic precursors has also been developed to avoid corrosive residues when using Cl₂.^33,34

In this work, we will investigate the etching condition and reaction mechanism of the ALE process with a halogen precursor Cl₂ and an organic precursor acetylacetone (acac) as shown in Fig. 1. We will discuss the ALE performance including the influence of temperature, pressure, and self-limiting behaviors. We will also study the mechanism of the chemical reaction in this process by using in situ x-ray photoelectron spectroscopy (XPS) and density functional theory (DFT) simulation. A better understanding of the mechanism is important for future precursor design and selection. Cl₂ is chosen to activate the Fe surface instead of O₂ as described in previous reports for two reasons. (1) The MO_x residues of O₂ etching will lead to degradation in the magnetization of the MTJ (Ref. 35) and need to be removed by sputtering.³⁶ (2) Halogen-based etching precursors are commonly used in industry. Acac is chosen to make volatile metal complexes. The classical acac reaction with the metal chlorides is usually based on an adsorbate substitution mechanism. However, the real surface reaction between acac and the Cl⁻ dosed Fe surface might be more complicated due to the decomposition of beta-diketone molecules which has been reported on FeO_x surfaces.³⁷ 

The structure of the in situ reactor is shown in Fig. 2. Samples were loaded from the XPS main chamber by a transfer arm and then placed on a stainless-steel heating stage with a cartridge heater inside. The temperature was monitored by a K-type thermocouple attached to the stage. Sample surface temperature versus thermocouple temperature was calibrated at the operating pressures of the reaction. Acac (99.5% purity) was purchased from Aldrich Sigma and stored in a Pyrex tube. Before use, acac went through three freeze, pump, and thaw cycles. Acac was delivered to the reaction chamber through a leak valve by evaporation. Cl₂ gas (99.9% purity) was purchased from Keen Gas. The pressure of Cl₂ and acac was controlled by leak valves. The background pressure is 10⁻⁹ Torr in the XPS main chamber and 10⁻⁶ Torr in the reaction chamber. In order to determine the contamination in the chamber background, we placed a clean Fe sample in the chamber and heated it to 140 °C for 2 h without dosing any gases. The atomic concentrations of adventitious C, O, and Cl found on the surface were 17%, 10%, and 8%, respectively. Therefore, the etching experiment is considered to be finished when the Cl atomic concentration approaches 8%.

A 400 nm sputtered iron metal (Fe⁰) thin film was deposited on 150 mm diameter Si(100) wafers. The samples were cut into 1 cm by 1 cm coupons. For a fundamental understanding, Fe surfaces were sputter cleaned with 3 kV argon ion gun until the native oxide was removed as verified by XPS, and then Cl₂ gas was dosed in the reaction chamber for 3 min at room temperature at 30 mTorr. In situ XPS was used to analyze the chemical composition of the surface and near-surface regions during the etching reaction.

XPS studies were carried out using a PHI-5600 electron spectrometer. All XPS spectra were acquired using monochromatic Al Kα radiation (hν = 1486.6 eV) with 23 eV pass energy and 0.05 eV/step. Ar sputtering used a 3 kV ion gun and emission current at 25 mA. All of the binding energies were determined by reference to the Fe 2p_3/2 peak at 706.8 eV. The sputter rate was calibrated by sputtering a 100 nm SiO₂ thin film. Peak fitting was done with casa xps software. A Shirley-type background subtraction was subtracted from each C 1s, O 1s, and Cl 2p peak, and a mixture of Gaussian and Lorentzian functions was used for the fit. For determination of iron valence state, the Fe 2p_3/2 peak, Lorentzian asymmetric convoluted line-shapes were also used.^38,39

DFT (Refs. 40 and 41) calculations of total energies and structural optimizations of selected products were performed for Fe surfaces using the vienna ab initio simulation package (vasp).⁴² Spin polarized calculations using Perdew–Burke–Ernzerhof functional⁴³ and projector augmented plane wave potentials^44,45 were used. A five-layer Fe slab with (3 × 3) unit cells was used in the calculations in which the three top-most layers were relaxed while the bottom two layers were fixed at their bulk positions. A 20.0 Å vacuum layer was inserted along the z direction. Calculations were carried out using an energy cutoff of 400 eV, and integrations over the Brillouin zone were performed using a 6 × 6 × 1 mesh of special k-points.

Cl₂ adsorbed on Fe surfaces at room temperature was first studied by XPS. The Cl₂ exposure was fixed at 100 mTorr pressure for 3 min. Figure 3(a) displays Fe 2p_3/2 XPS spectra for Cl₂ dosed on the Fe surface. The Fe peaks are fitted with Gupta and Sen multiplets.^38,39 The peak at 706.8 eV is the Fe⁰ component, which is from the unreacted Fe layer under FeCl_x. The broad peak that is fitted into the 709.5, 710.1, and 711.2 eV peaks corresponds to the Fe²⁺ components split by spin–orbit coupling interactions. The peak at 712.1 eV corresponds to the small amount of Fe³⁺ surface components, likely resulting from defects. Finally, the peak at 715.1 eV corresponds to the Fe²⁺ satellite peak resulting from metal–adsorbate charge transfer. The ratio between atomic concentration of Fe²⁺ and Cl is close to 1:2, suggesting that the major product of Cl₂ adsorption on the surface is FeCl₂.

The effect of temperature on Cl₂ adsorption was then studied. The thickness of the FeCl_x layer after dosing Cl₂ at room temperature is around 0.5 nm as shown by the XPS depth profile in Fig. 3(b) and increases to 1 nm at 80 °C [Fig. 3(c)] and 4.5 nm at 140 °C [Fig. 3(d)], which is much thicker than one atomic layer. The interface between FeCl_x and Fe substrate also appears broader at 140 °C, which suggests an increasing roughness of the FeCl_x/Fe interface. Thus, both the thickness and the roughness of the FeCl_x layer are highly temperature dependent. As shown previously, Cl can diffuse into Cu films,^46,47 Cl₂ gas can also activate Fe surfaces by diffusing into the film when the temperature is high, and, as a result, has poor self-limiting performance.

The second step in the ALE process is the reaction of acac with Cl-modified Fe surfaces. The Fe sample was first dosed with 100 mTorr Cl₂ at room temperature for 3 min, and then acac vapor was dosed under different conditions. In order to determine the optimum reaction temperature, we dosed 1 Torr of acac at different temperatures for 3 min. Figure 4 shows the change of Fe 2p and Cl 2p photoemission peaks after dosing acac at different temperatures. When increasing the reaction temperature from 23 to 110 °C, the intensity and components of Fe 2p and Cl 2p peaks showed little change compared to the spectra acquired before acac exposure. Changes in the XPS are observed when the reaction temperature reached 135 °C. The Cl 2p peak, Fe²⁺ components, and the satellite peak all diminish after reaction, while the intensity of Fe⁰ component increases significantly. Thus, most of the Cl⁻ and Fe²⁺ are removed from the surface with acac exposure as temperatures approach 135 °C.

Additionally, the influence of acac pressure was then studied at 140 °C reaction temperature. Figure 5 shows the change of surface atomic concentration of all elements with acac exposure time under different pressures. The reaction follows similar trends at 100, 200, and 400 mTorr, but the remaining organic groups on the surface after reaction are slightly different as shown in Figs. 5(a)–5(c), respectively. The Cl removal rate is not constant during the process, which increases at first with acac exposure time and then slows when the Cl fraction is less than 20%. At 100 mTorr acac pressure, the remaining carbon fraction is greater than 50% after 40 min exposure. But when using 400 mTorr acac pressure, only 40% carbon remains after 20 min. The removal of Cl can be the result of several processes involving the acac reaction on the surface including (1) the desorption of HCl, (2) volatilization of an organic chloride product, or (3) a volatile metal complex. Removal of Cl is a necessary step for the further reactions between Fe²⁺ and organic groups, including either the formation of a volatile complex or reducing Fe²⁺ to Fe⁰ by the organic molecule. If we assume that the etching rate follows the rate of Cl removal, we can estimate the reaction rate. The average Cl removal rate is taken to be the time required for the Cl fraction to decrease from 40 to 20 at. %. In this part of the reaction cycle, the reaction appears to be a first order reaction in acac pressure, since the average removal rate of Cl ratio is proportional to the acac pressure [see Fig. 5(d)]. Thus, a higher acac pressure is preferred for this process, yielding a shorter reaction time and less organic residue.

To prove that an Fe film can be etched by the two successive reactions, we dosed 100 mTorr Cl₂ for 3 min and 1 Torr acac for 3 min at 140 °C sequentially for 20 cycles. The thickness of the Fe film before and after etching was measured by XPS depth profiling. The thickness of the Fe film decreases from about 450 nm to about 350 nm (sputter depths are determined by sputter rates in a known thickness of SiO₂ films) as shown in Fig. 6. This etching rate is 5 nm/cycle, which is close to the thickness of the FeCl_x layer formed in each cycle [∼4.5 nm in Fig. 3(d)]. Presumably, the extent of Cl₂ reaction determined the etching rate, and acac then only reacted with FeCl_x. Based on this result, a self-limiting ALE process could be developed by using optimized conditions.

To study the reaction mechanism of the acac reaction on a Cl-dosed Fe surface, the chemical shifts were determined in the C 1s photoemission spectra acquired at high resolution. Figure 7 presents the data for Cl-dosed samples dosed for different acac exposure times, showing the chemical states of C 1s after the reaction. Based on the surface composition with acac exposure at 140 °C, 100 mTorr [Fig. 5(a)], three time points, 0, 15, and 30 min, were selected for conditions corresponding to before acac reaction, during acac reaction, and after acac reaction conditions.

When a Cl-dosed Fe film was placed in the reaction chamber and heated to 140 °C without dosing any acac, C 1s peaks were observed at 284.6, 283.5, 286.2, and 287.5 eV which correspond to hydrocarbon bonded only to hydrogen and carbon, C–Fe bonding, C–O and C=O groups, respectively. Only a weak O 1s peak at 531.9 eV was observed, which corresponds to C–O single bonds (e.g., alcohol or ether oxygen). These organic groups can be attributed to the chamber background contamination. Following 15 min exposure of acac, the C 1s and O 1s showed different carbon and oxygen peak distributions compared to those observed without acac treatment. In the C 1s region, the intensities of the C 1s peaks at 284.6, 286.2, and 288.7 eV have increased, while the intensities of the carbidelike peak at 283.5 eV and the C=O peak at 287.5 eV remained similar to those observed before dosing acac. In the O 1s region, peaks with binding energies of 530.3, 532.6, and 534.6 eV were observed, which correspond to O²⁻, C=O, and carboxyl groups, respectively. The intensity of Cl peak decreased during the process. The remaining Cl peak shifted from 199.1 to 199.5 eV, suggesting that either organic chloride compound was forming, or other electronegative groups coordinated with Fe and decreased the charge transfer from Fe to Cl. After dosing acac for 30 min, the binding energy of remaining Cl on the surface shifted back to 199.1 eV due to the FeCl_x residues. The atomic concentration of Cl is close to the Cl baseline 8% from the chamber background, indicating the ending of acac reactions.

Comparing the spectra of samples after 15 and 30 min acac exposure in Fig. 7, several things are obvious. First, Cl has largely been removed from the surface. The intensity of C 1s and O 1s peaks increased only slightly. Second, the intensity of C–Fe bonding peak, which increased after 15 min acac exposure, diminished after 30 min acac exposure. Third, comparing with the literature,⁴⁸ acac should only have two C 1s contributions: methyl and C=O group with a ratio 3:2. However, the intensity of the hydrocarbon peak increased much faster than the C=O group after 15 and 30 min acac exposure in our experiment.

The decomposition of acac occurs on chlorinated Fe surfaces just as on FeO_x surfaces³⁷ and increases the relative concentrations of O²⁻, hydrocarbon, and C–Fe bonding.⁴⁹ C–Fe bonding can result from either background contamination or acac decomposition. The O²⁻ or OH⁻ groups already present on the surface can also react with acac and lead to beta C–C bond scission.⁵⁰ When most of the FeCl_x was removed, the ratio of C–Fe bonding decreased. The C–Fe peak originally is probably a result of reaction of the chlorinated surface with adventitious carbon before exposure with acac. During exposure to acac, the chlorinated surface may react further with acac leading to the formation of C–Fe bonding. Once Cl is removed from the surface, acac (or adventitious carbon) no longer forms C–Fe bonds. This mechanism implies that the C–Fe species is an intermediate in the formation of a volatile etching product.

Due to the decomposition of acac, the etching mechanism does not appear to simply follow the proposed substitution reaction. The acac decomposition products, including C–Fe bonding, O²⁻ and C–O groups, could be either important intermediates in the process or ligands in the volatile organometallic complex instead of acac. However, a precise assignment of the intermediate and etching product is not possible at present.

From the in situ XPS studies described above, we found that the interface between FeCl_x and Fe was rough, and the reaction may not follow acac substitution for Cl. The thermodynamic energies in each proposed step can be used to explain these results. Given that sputter deposited Fe films are polycrystalline films, different crystal orientations will exist at the surface. The reactivity of Fe(110), the lowest energy surface of the bcc lattice, will be compared with Fe(100), which is a higher energy surface that may be present locally. The total energies of different structures were calculated using vasp. The change of enthalpy, E₁, in Cl₂ adsorption is calculated according to

where n is the number of adsorbed chlorine atoms adsorbed on the simulated portion of the slab. E_Cl/slab, E_slab, and $ECl2$ are the total energies of the Cl-dosed Fe surface, clean Fe surface, and an isolated chlorine molecule, respectively.

In the first step, the calculated Cl₂ adsorption energy on Fe(100) and Fe(110) surfaces is quite different, indicating different exothermicity on Fe(100) and Fe(110) surfaces. For the Fe(100) surface, the Cl coverage can reach 100% with a mixture of hollow site and bridge site bonding.⁵¹ The average adsorption energy E₁ for each Cl atom is −2.05 eV. However, for the close-packed Fe(110) surface, the adsorption of Cl in the simulation is unstable for a coverage of one Cl for each surface Fe. The maximum stable coverage is found to be 0.67 with adsorption energy E₁ −2.23 eV for each Cl atom. Due to this reactivity difference and the relatively low surface energy of Fe(110), presumably, Cl₂ will react mostly at grain boundaries or defects on the polycrystalline surface.

Given that the surface reaction between acac and Cl-dosed Fe surface is uncertain, we have assumed that etching results from acac substitution with Cl. One hypothetical etching product is shown in Fig. 8(a)—an Fe metal center with one Cl and one acac ligand. This structure is simple but relatively stable after structure optimization and might be an intermediate in the formation of another Fe–acac complex such as Fe(acac)₂ or Fe(acac)₃. In order to form this product, an acac molecule will react with one Cl atom on the surface and produce an HCl molecule. The energy of replacing one Cl⁻ with acac⁻ in this reaction E₂ is calculated according to

where E_acac/Slab, E_cl/slab, E_acac, and E_HCl represent the total energies of Cl-dosed Fe surface after an acac⁻ replaced a Cl⁻, a Cl-dosed Fe surface, an isolated acac molecule, and an isolated HCl molecule, respectively. E₂ was first calculated based on the optimized Cl-dosed smooth Fe surfaces. On Fe(110) surface, the energy needed for this reaction is 1.98 eV, while on Fe(100) surface the energy is 0.31 eV. Thus, acac substitution is not thermodynamically preferred on either Fe(110) or Fe(100) smooth surfaces.

The influence of defects on surface reactions was also considered, where we used an isolated Fe on a smooth Fe surface to simulate a defect site as shown in Fig. 8(b). With the surface defect atom, E₂ increases to 2.07 eV on Fe(110) surface because the defect site made the Cl-dosed surface even more stable. However, the defect atom reduced E₂ to −1.51 eV on Fe(100) as shown in Fig. 8(b). Desorption of this complex required 1.26 eV, which is close to the sublimation energy of some metal acac complexes.⁵² Given that the main surface of Fe crystalline is likely to be Fe(110), direct acac substitution for Cl is unlikely the etching reaction on Fe surfaces, even though the defect can assist the adsorbate substitution on Fe(100) surface.

We successfully etched Fe films by successively dosing Cl₂ and acac and investigated the influence of several conditions. Fe surfaces were activated by Cl₂ gas. XPS of the Fe 2p_3/2 peak and the ratio between Fe and Cl components indicate that the major reactant on the surface is FeCl₂. The thickness of FeCl_x layer and roughness of FeCl_x/Fe interface is highly temperature dependent, indicating that the Cl₂ step has poor self-limiting performance at high temperatures. The adsorption energies of Cl₂ were calculated by DFT simulation. Cl₂ is found to react more exothermally on defect sites or less energetically favorable surface orientations. In this step, the extent of Cl₂ reaction determined the etching rate, and the resulting nonuniformity issue is likely in this reaction. Based on this result, Cl₂ gas can be replaced by different gases or plasmas to improve the etching performance in the future. To achieve a more precise control, a less reactive gas or plasma such as NF₃, ICl, or IBr (Ref. 53) might reduce the adsorbate diffusion for better self-limiting performance. To increase the anisotropy of ALE for patterned samples, a plasma such as SiCl₄/Cl₂/Ar,⁵⁴ which could provide side wall protection, can also be used to replace Cl₂ gas.

Acac reactions were then carried out at various temperatures and pressures to identify the etching conditions. Acac was able to remove FeCl_x layer with a temperature higher than 135 °C, and the removal rate is proportional to the acac pressure. The mechanism of the acac reaction on the Cl-dosed Fe surface was studied by in situ XPS and DFT simulation, both suggesting that the reaction does not follow a simple adsorbate substitution pathway. During the reaction, acac was not the major organic species on the surface. Thus, it is likely that the decomposition of acac plays an important role in this reaction as it does on FeO_x surfaces. Species containing C–O, carboxyl, or carbidelike groups, which resulted from the acac decomposition, likely participate in the formation of a volatile metal complex instead of a pure acac substitution. DFT calculations also indicated that the acac substitution for Cl is not thermodynamically preferred on Fe surfaces. It is important to determine what volatile complex forms after acac decomposition.

The authors thank the Teplyalov group in the University of Delaware and Air Liquide for their assistant and consultation. They also acknowledge the financial support from the National Science Foundation GOALI: fundamental approach to atomic layer etching (NSF Grant No. 1609973).