In situ atomic layer deposition and electron tunneling characterization of monolayer Al 2 O 3 on Fe for magnetic tunnel junctions

Magnetic tunnel junctions (MTJs), formed through sandwiching an ultrathin insulating film (so-called tunnel barrier or TB), with ferromagnetic metal electrodes, are fundamental building blocks in magnetoresistive random access memory (MRAM), spintronics, etc. The current MTJ technology employs physical vapor deposition (PVD) to fabricate either amorphous AlOx or epitaxial MgO TBs of thickness around 1 nm or larger to avoid leakage caused by defects in TBs. Motivated by the fundamental limitation in PVD in, and the need for atomically thin and defect-free TBs in MTJs, this work explores atomic layer deposition (ALD) of 1-6 A thick Al2O3 TBs both directly on Fe films and with an ultrathin Al wetting layer. In situ characterization of the ALD Al2O3 TB was carried out using scanning tunneling spectroscopy (STS). Despite a moderate decrease in TB height Eb with reducing Al wetting layer thicknesses, a remarkable Eb of ∼1.25 eV was obtained on 1 A thick ALD Al2O3 TB grown directly on an Fe electrode, which is more than twice of that of thermal AlOx TB (∼0.6 eV). Achieving such an atomically thin low-defect TB represents a major step towards improving spin current tunneling in MTJs.


I. INTRODUCTION
Magnetic Tunnel Junction (MTJs) are the building blocks for non-volatile magnetoresistive random access memory (MRAM) and are created by sandwiching an ultrathin insulator between two ferromagnetic electrodes. 1-4 MRAM has significant performance and power consumption advantages over standard dynamic random access memory or flash memory with fast read, write times and a memory state which is retained without power draw. 1,2,5 MTJs operate with a differing spin-polarized electron tunneling resistance for parallel and antiparallel magnetization of the ferromagnetic layers. The tunneling current decreases exponentially with the tunnel barrier (TB) thickness and can be further affected due to scattering from defects in TBs. The current MTJ technology based on physical vapor deposition (PVD) employs either amorphous AlOx or epitaxial MgO TBs of thickness around 1 nm or larger to avoid leakage caused by defects in TBs. The TMR values are around 10-70% for AlO x TBs 6-8 and at least a factor of three higher for crystalline MgO TBs due to enhanced coherent spin current tunneling. 5,[9][10][11][12][13][14][15] It should be realized that TBs with a smaller thickness approaching atomic scale, that are pinhole-free with low defect concentration are the key to high TMR values desirable for achieving greater signal to noise ratio, lower power consumption, higher speed, and large design margin for device fabrication. Therefore, further reduction of the TB thickness to the sub-nanometer regime remains a major challenge in the research and development of MTJs.
Atomic layer deposition (ALD) provides a promising alternative to grow low-defect TBs at the atomic-thickness scale for MTJs. ALD is a chemical vapor deposition (CVD) process and has several unique advantages in growth of the TBs for tunnel junctions. 16 First, ALD can provide a highly conformal coating even on high aspect ratio surfaces, [16][17][18] which is important to minimization of the pinholes and other defects that can cause leakage in TBs. On the other hand, ALD allows atomic-scale accuracy in thickness control due to its self-limiting mechanism. Finally, ALD growth occurs via well-designed chemical ligand exchange between selected pulses of precursors on the sample surface, resulting in minimal un-reacted dangling bonds in the stoichiometric ALD films and hence a very low defect concentration. The low defect-density, high conformality, and excellent thickness control make ALD especially attractive for industrial-scale fabrication of MTJs on large wafers. This has motivated research in ALD growth of TBs for MTJ applications. [19][20][21][22] Using ex situ ALD deposition, 1.6 nm, or thicker, ALD TBs have been obtained for MTJs, while lower thicknesses produce TBs with pinholes. 20 The primary challenge in the ALD growth of TBs of sub-nanometer thickness is the formation of an interfacial layer between the metal electrode and the dielectric ALD TB, typically formed when the metal surface is exposed to ambient or low vacuum in ex situ processes. This defective interface has a profound effect on the ALD dielectric film quality. For example, the dielectric constant (ε r ) is significantly reduced from the bulk single crystals values when the ALD film thickness is on the order of tens of nanometers or lower, [23][24][25][26][27] indicative a high defect concentration in ALD dielectric films grown on defective metal-insulator interfaces. Even in the in situ ALD process, a prolonged exposure to vacuum (even at high vacuum of ∼10 -7 Torr) during heating of a metal electrode for ALD TB growth can cause formation of such an interface. 27 This interface is even more detrimental to ALD TBs for MTJs since the additional TB defects can impair spin current tunneling current in MTJs. 24,26,27 In a recent work, we reported a dynamic heating process for in situ ALD TB growth which is capable of growing high quality ALD Al 2 O 3 TBs on aluminum at the monolayer scale with an excellent control of the Al-Al 2 O 3 interface to a negligible level. 23,[28][29][30][31] A key step is in control of the hydroxylation of the metal surface during the first monolayer growth of ALD Al 2 O 3. 23,31 This control leads to an order of magnitude reduction in ALD Al 2 O 3 TB thickness, from 1 nm to about 1 Å, and the dielectric constant of ALD Al 2 O 3 films of 3 nm in thickness comparable to that of the bulk Al 2 O 3 single crystals.
While atomically thin ALD Al 2 O 3 TBs may be promising for MTJs, two fundamental issues are present in growth of the ALD Al 2 O 3 TBs on ferromagnetic electrodes. An Al wetting layer on the ferromagnetic metal electrode, such as Fe, is expected to reduce the TMR value by incoherently scattering the spin-polarized electron tunneling. 32 This means that the Al wetting layer should be as thin as possible since it is not a ferromagnetic material. This, however, may affect hydroxylation of the electrode surface and lead to defective nucleation of the ALD Al 2 O 3 on the Al/Fe or Fe film directly. An in-depth understanding of the nucleation and evolution of the ALD Al 2 O 3 TBs of sub-nanometer in thickness on Fe films with Al wetting layers of different thicknesses is therefore important to development of MTJs with ALD TBs. In this work, in situ Scanning Tunneling Spectroscopy (STS) was employed to characterize the electron tunneling properties of ALD Al2O3 TBs with thicknesses ranging from 1-6 Å in order to inspect TB quality, via barrier height measurement, as the Al layer was systematically reduced from 7 nm to zero. In tandem, Ab initio Molecular Dynamics (AIMD) simulations were carried out to examine the possibility of adapting the pre-ALD H 2 O pulse to produce a uniform layer of OH groups directly on the Fe surface.

II. METHODS
To create the metal-insulator structure for in situ STS, a trilayer of Nb (50 nm)/Fe (20 nm)/Al (0-7 nm) was DC magnetron sputtered using Argon in vacuum chambers with base pressures between 1.5 -5.5 × 10 -7 Torr onto an Au (50 nm) coated SiO 2 /Si substrate at deposition rates of 1.7 nm/s (330 W 14 mTorr), 1.0 nm/s (200W 4 mTorr), and 0.5 nm/s (90 W 14 mTorr) respectively, which were all previously empirically estimated. A washer was clamped to the Au surface of the substrate to create a connection between the sample and the STS ground. The niobium layer served to prevent diffusion between the Al and Au layers which may cause significant problems with ALD nucleation in thinner Al films. In order to understand the role of the Al wetting layer thickness, five different Al thicknesses of 7 nm, 4 nm, 2 nm, 1 nm and 0 nm were selected in this work. After metal sputtering, the samples were transferred in situ to the ALD chamber under high vacuum where a pre-ALD H 2 O pulse of 2s in duration to hydroxylate the metal surface, which was followed with a N 2 purge for 35s and a trimethylaluminum (TMA) pulse of 1s to form a monolayer of Al 2 O 3 using the optimized ALD conditions discussed in our previous papers. 30,31 With an additional N 2 purge, one cycle of ALD growth of an Al 2 O 3 monolayer was completed with the layer thickness of about 1.1-1.2 Å. 16,25,29,33 In this study, up to five cycles of ALD Al 2 O 3 TBs were grown on metal electrodes. After ALD, the samples were in situ transferred under high vacuum to the STS chamber which had a pressure of about 1x10 -9 Torr. dI/dV spectra were taken over 100-200 random locations on the sample's surface using a mechanically-cleaved Pt-Ir tip with a voltage modulation of 30 mV at 5 kHz and a set point of 2 V, 200 pA. This same tip was used to acquire all dI/dV spectra in this experiment except for a one of the final sets of data where the new tip was checked to assure it was comparable in quality to the previous tip. 34 The TB height was estimated by the intersection of two bisquare-method linear fits to ln(dI/dV ) and the ALD coverage was defined as the percentage of locations which had dI/dV with calculated E b values greater than 0.8 eV. 30,31 Ab-initio molecular dynamics (AIMD) simulations for H 2 O molecules on the Fe (100), Fe (110), and Fe (111) surface were conducted under a constant volume and temperature ensemble as implemented in VASP code. 35 The Fe slabs were constructed using experimentally determined lattice parameters 36 and were adjusted in size to accommodate thermal expansion. A "water pool" was added to the top site of each Fe surface to simulate the pre-ALD H 2 O pulse. Additional details for the simulation procedure can be found in our previous works. 30,31 Figure 1a illustrates schematically the samples fabricated using the procedures described earlier. Figure 1b shows the sequential ligand exchanges between pulses of H 2 O and TMA in one ALD cycle which leads to one atomic layer of Al 2 O 3 of thickness ∼1.1-1.2 Å. 37,38 To achieve 6 Å ALD Al 2 O 3 TB thickness, five ALD growth cycles were carried out. This fabrication was all done in situ using a specialized STS sample stage with the bias voltage applied between the STS tip and the Au bottom contact electrode, which is shown in Figure 1c. Figures 2a and 2b show representative dI/dV spectra taken on the Nb (50 nm)/Fe (20 nm)/Al (7 nm)/ALD Al 2 O 3 (1 Å) and Nb (50 nm)/Fe (20 nm)/ALD Al 2 O 3 (1 Å) respectively. Qualitatively, a similar nonlinear trend can be observed on both spectra, illustrating tunneling through the ALD Al 2 O 3 TB of 1 Å thickness with or without the Al wetting layer. The fitting (red lines) to the band gap region and the conduction band region allows extraction of TB height E b at the intersection of the fitting lines. The E b for the sample with a 7 nm Al wetting layer (Figure 2a) is ∼1.66 eV, which is consistent with the range we measured on Nb (20 nm)/Al (7 nm)/ALD Al 2 O 3 (1 Å) samples earlier without an Fe layer in the electrode stack. 30,31 This indicates a thick 7 nm Al wetting layer would thoroughly cover the surface of the underlying metal, either Nb or Fe. A lower E b ∼1.25 eV was observed on Nb (50 nm)/Fe (20 nm)/ALD Al 2 O 3 (1 Å) (Figure 2b), which reveals that the difference on the surface of Al and Fe sensitively affects the ALD Al 2 O 3 nucleation and will be discussed in detail later. When the Al wetting layer was varied within the range of 0-7 nm, the E b of the ALD Al 2 O 3 TB (1 Å) was found to remain constant in the Al thickness range of 4-7 nm, followed by a decrease when the Al thickness was further reduced below 4 nm, together with a drop in the ALD coverage from 96% to 93.9% (Figure 2c). The reduced E b values indicate an increased defect concentration in the ALD Al 2 O 3 TB (1 Å), possibly due to the difference in the nucleation of the ALD Al 2 O 3 on Fe as compared to Al assuming the surface of the Fe may not be fully covered at smaller Al wetting layer thicknesses due possibly to intermetallic diffusion or surface roughness (1.2 -1.4 nm) causing gaps in Al coverage. However, these effects are not significant. With 1 nm thick Al wetting layer, the E b ∼1.44 eV is still in the range of high quality ALD Al 2 O 3 TB (1 Å) and significantly better than the E b ∼1.0 eV observed on ALD Al 2 O 3 TB (1 Å) with a defective Al-Al 2 O 3 interfacial layer. 30 Comparatively, the E b values are in the range of 0.5-0.7 eV and 0.4-0.9 eV for thermal AlOx TBs and epitaxial MgO TBs respectively, 5,39,40 due to the presence of defects such as oxygen vacancies. Therefore, the observation of higher E b values in 1 Å thick ALD Al 2 O 3 TB, as opposed to their PVD counterparts with at least 5-10 times larger thickness, illustrates the promising potential of the ALD approach for synthesis of high-quality, sub-nm thick TBs for MTJ applications. Figure 2d depicts the E b and coverage of ALD Al 2 O 3 TBs as function of the ALD cycle number in the range of 1-5 (or TB thickness in the range of 1-6 Å) on the Nb (50 nm)/Fe (20 nm)/Al (1 nm)/ALD Al 2 O 3 samples. A constant E b and ALD coverage were observed in the Al 2 O 3 TB thickness range of 1-6 Å, which indicates that the ALD Al 2 O 3 TB is fairly uniform. This means that the reduced E b value of the ALD Al 2 O 3 TBs on Fe or on Fe with a very thin Al wetting layer is unlikely to be caused by the formation of an FeO x interface since a defective FeO x IL would reduce E b more significantly at a smaller TB thickness, resulting in a thickness-dependent E b . 30 This observation is important since a defective interfacial layer is highly detrimental to the TB quality, but can be minimized in the in situ ALD process with well controlled growth conditions. Therefore, the observation of the reduced E b value of the ALD Al 2 O 3 TBs grown on Fe without or with a very thin Al wetting layer raises a question on the difference in nucleation of the ALD Al 2 O 3 on Fe with respect to Al. To shed light on the differences of Fe and Al surfaces for ALD oxide growth, AIMD simulations were carried out to examine the growth kinetics of ALD Al 2 O 3 on Fe in comparison with on Al as we reported previously. 30,31 Considering hydroxylation of the metal surface during the first H 2 O pulse is the most critical step in nucleation of the monolayer ALD Al 2 O 3 , Figure 3 illustrates this process on Fe surfaces of (110), (100) and (111) orientations, respectively. The first two are energetically preferred in polycrystalline Fe films and have smaller inter-Fe atom distance on the surface than in the (111) Fe case. The AIMD simulation suggests that a hydroxylated Fe surface is possible within the first H 2 O pulse. However, the effectiveness in generating adsorbed hydroxyl groups (OH ads ) without dissociation of H 2 O into oxygen and hydrogen is dependent on the Fe surface orientation. The Fe (110) and Fe (100) surfaces both result in OH ads creation without much oxygen formation whereas the Fe (111) surface leads to a high amount of oxygen. As we have shown earlier on the Al surface, a primary mechanism dictating the H 2 O dissociation into OH ads is the inter-molecular interaction between H 2 O molecules attached to the metal surface during the first H 2 O pulse. At an optimal pre-ALD H 2 O pulse duration to enable an adequate concentration of H 2 O molecules on a metal surface, the inter-atomic distance of metal atoms on the surface plays an important role in affecting the inter-molecular interaction between H 2 O molecules attached to the metal atoms and their dissociation paths.

III. RESULTS AND DISCUSSION
The FCC structured Al (111) facet is one of the favorable surfaces with a high density of surface atoms (0.143 atoms/Å 2 ), which leads to strong inter-molecular interaction between H 2 O molecules and facilitates their dissociation into OH ads on the Al surface. In contrast, Fe has a BCC structure. 41,42 Although the surface densities of 0.122 and 0.155 atoms/Å 2 on (100) and (110) facets respectively are excellent for the dissociation of H 2 O molecules into OH ads , the much reduced surface density of 0.063 atoms/Å 2 on (111) Fe may lead to reduced efficiency for such a dissociation. Consequently, the ALD Al 2 O 3 on Fe or on Fe with very thin Al may have lower density on average. In addition, the FeO x IL from the pre-ALD H 2 O pulse may form on the Fe (111) if H 2 O molecules cannot efficiently dissociate to OH ads , and rather to hydrogen and oxygen that may react with Fe to form an FeO x IL. This explains the reduced ALD Al 2 O 3 TB coverage on Fe surface without an Al wetting layer or with a very thin Al wetting layer. Figure 4a shows that the non-ALD dI/dV spectra. Overall, these spectra share similar characteristics of flatness and a rounded conduction band onset for samples with 7, 1, and 0 nm (on Fe) thick Al wetting layers while the E b values are significantly lower than that of the ALD TBs. This indicates that though the substrate makeups are different, many of these lower quality spots in the film are similar in composition. The low E b values in the range of 0.4-0.6 eV, which are comparable to the E b values measured on thermal AlOx TBs, 30 suggested that the non-ALD TBs formed on these spots are most probably defective AlOx TBs on FeOx IL/Fe (111) based on the simulation result in Figure 3. Nevertheless, after viewing the distributions in Figure 4b, and the coverage rates from Figure 2c, the percentage of the surface that is defective/lower quality increases with decreasing Al wetting layer thickness. Additionally, some of these defective surface spectra have increased noise and do not appear similar to those from Figure 4a possibly indicating different types of defects in the samples with very thin or no wetting layer. Intuitively, one would think that any exposure to heat and H 2 O molecules would cause oxidation in an Fe film based on how we expect Fe films to behave ex situ. Indeed, even in situ, the heating of the Fe film, its exposure to H 2 O during the first ALD pulse, and trace O 2 all have the potential to create FeOx. 43 However, our result indicates this does not appear to have happened to any significant level with a well-controlled in situ ALD process, especially the initial hydroxylation of the Fe surface, to minimize FeOx IL formation. In addition, a further improved ALD TB may be achieved on epitaxial Fe films with desired facets of (100) and (110).
To quantify to what degree the FeOx IL formation occurs when in situ ALD Al 2 O 3 is grown on Fe, ALD Al 2 O 3 TBs were grown on Fe with varying cycle numbers. STS reveals, in Figure 5a, a nearly constant coverage of the ALD TB on the Fe surface with a 5% increase of E b from 1.25 eV to 1.31 eV with the increasing ALD cycle number from 1 to 5. Our previous work 30 and the data from Figure 2b show a constant barrier height with increasing cycle number can be obtained on Al surface with a negligible IL. Considering the 5% increases in the E b value in the ALD cycle number range of 1-5 is small, we argue the FeOx IL is unlikely significant in the ALD Al 2 O 3 TBs grown directly on Fe. This observation is therefore important to the potential implementation of ALD TBs with sub-nm thickness into tunnel junction devices. Furthermore, despite the almost constant E b values at different TB thicknesses, an E b value of 1.25-1.31 eV for ALD Al 2 O 3 TBs on Fe is lower than the same TB on Al 30,31 by about 25-30%. Considering the IL effect is unlikely significant, we hypothesize the lower E b value of the ALD Al 2 O 3 TBs on Fe may relate to the lower density of the ALD Al 2 O 3 nucleate on BCC Fe (100) and (111) surfaces with smaller densities of surface atoms, as compared to that on the FCC Al (111) surface. Confirmation of this hypothesis is certainly important to development of atomically thin and low-defect ALD dielectric on metal and semiconductor surfaces.
To rule out the possibility that this lower E b value of the ALD Al 2 O 3 TBs on Fe was due to un-optimized ALD growth condition originally optimized for ALD Al 2 O 3 TBs on the Al surface, three different dynamic heating times of 13,15, and 17 minutes were tested, which primarily reduced the pre-ALD heating time of the Fe film and also the initial ALD temperatures to 175, 190, and 200 • C, respectively, at the same fixed heating power. 31 From Figure 5b, the 15 minutes heating gave the optimal result with an E b of 1.25 ± 0.03 eV and ALD coverage of 93.9 ± 2.4%. At 13 minutes of heating with 175 • C as the starting temperature, less OH groups are expected to form on the Fe surface, frustrating the ALD process. When the heating time was 17 minutes with an initial temperature of 200 • C, it is possible that there is more FeOx formation due to the dissociation of OH and H 2 O molecules into O atoms. 30,31 This result shows that the ALD conditions ALD Al 2 O 3 TBs are nearly identical although the Fe and Al may have different properties, such as adsorption energies and surface atom density. In other words, the cause of the reduced E b value on an Fe surface is most probably the lower density of the ALD Al 2 O 3 TBs nucleated on BCC Fe surface due to the lower density of the surface atoms as compared to its counterpart on FCC Al surface.

IV. CONCLUSIONS
In conclusion, we report the first success in growth of atomically thin (1-6 Å in thickness), lowdefect ALD Al 2 O 3 TBs through atomic layer deposition directly on ferromagnetic Fe electrode. In situ STS studies have confirmed that high E b up to 1.25 eV can be obtained on the 1 Å thick ALD Al 2 O 3 TBs grown directly on Fe, which is more than twice of that of the thermal AlOx TBs. This indicates a significantly lower defect concentration can be obtained in the former, in addition to an order of magnitude reduction in the TB thickness. While Al wetting layer may provide a slightly better nucleation surface for the ALD Al 2 O 3 TBs as reflected in up to ∼30% improvement in the E b , we have found that this wetting layer is not necessary for high quality ALD Al 2 O 3 TBs to deposited directly on Fe electrode with a coverage of 94%, a slight decrease from 96% with an Al wetting layer. AIMD simulations of H 2 O molecules on the Fe (100), Fe (110), and Fe (111) surfaces during the pre-ALD H 2 O pulse reveal that the hydroxylation of the first two Fe surfaces, which are energetically more preferred orientations in polycrystalline Fe films, can be facilitated by a stronger inter-H 2 O molecule interaction as H 2 O molecules attached to the Fe atoms of shorter inter-atom distances. This is in contrast to the weakened inter-H 2 O molecule interaction on Fe (111) surface at a larger interatom distance, which may lead to detrimental dissociation of the H 2 O molecules into oxygen and hydrogen and hence formation of FeO x interfacial layer. Considering a possible small volume portion of Fe (111) may exist in the polycrystalline Fe films used in this study, a higher coverage of the ALD Al 2 O 3 TBs on epitaxial (100) and (110) Fe films is expected. Therefore, this work demonstrates that the fundamental roadblock towards achieving MTJs with atomically thin, defect-free dielectric TBs can be removed using the ALD TB approach.