The leakage current suppression mechanism in a RuO₂/SrTiO₃/Ru capacitor induced by introduction of an ultra-thin GeO₂ interfacial layer at the bottom interface

Abstract

This study examines the chemical and electrical properties of RuO₂/SrTiO₃ (STO)/Ru and RuO₂/STO/GeO₂/Ru capacitors to elucidate the effect that a 6 Å-thick GeO₂ interfacial layer has on current leakage. The insertion of GeO₂ at the STO/Ru interface effectively suppresses microstructural defect formation during STO deposition and post deposition annealing, which is a principal contributor to high leakage current. The Schottky barrier height increases from 0.32 eV (STO) to 0.74 eV (STO/GeO₂), and the internal bias is alleviated from 0.9 V to 0.3 V, attributable to the improved STO/Ru contact properties obtained through preservation of the RuO_2−x interfacial layer and by facilitating oxygen vacancy curing. Consequently, the STO/GeO₂ material achieves a minimum equivalent oxide thickness of 0.40 nm at a physical thickness of 11 nm, which is a significant improvement over STO (0.69 nm at 27 nm). The conduction mechanisms under applied bias and the measured temperature of STO and STO/GeO₂ were systematically analyzed, demonstrating that GeO₂ interfacial engineering markedly improves dielectric performance in dynamic random access memory capacitors.

---

Introduction

SrTiO₃ (STO) has attracted significant attention as an important high-k dielectric material for next-generation dynamic random access memory (DRAM) capacitors due to its ultrahigh dielectric constant (k ∼ 300), which is substantially higher than that of conventional ZrO₂-based dielectric materials (k ∼ 30–40).^1–7 Atomic layer deposition (ALD) has been employed to deposit conformal STO thin films with high aspect ratios for use as capacitors owing to the excellent thickness control and uniformity that can be achieved through the use of ALD.⁸ However, the narrow bandgap of STO (3.2 eV) and its intrinsic n-type defects, such as oxygen vacancies (V_O), significantly exacerbate leakage currents in metal–insulator–metal (MIM) capacitors.^9–11 Therefore, high work function electrodes such as Ru (4.7 eV) or RuO₂ (5.2 eV) have been employed to increase the charge-injection barrier.^8,9 Nevertheless, strong interfacial interactions between STO and these electrodes lead to non-ideal ALD overgrowth, resulting in low-density films. Nano-cracks and nano-voids form in these films due to shrinkage during high-temperature post-deposition annealing (PDA) for crystallization, significantly increasing the leakage current.^12,13

To address these challenges, a two-step deposition process was implemented, in which a crystalline STO layer was grown on a pre-annealed crystalline seed STO layer to prevent defect formation during PDA. In addition, acceptor-doping strategies using the incorporation of Al³⁺ into Ti⁴⁺ sites to shift the Fermi level toward the valence band edge and increase the barrier for electron injection, thus increasing the Schottky barrier height by ∼0.13 eV, have been explored.¹⁴ Although Al doping lowered the leakage current, the smaller ionic radius of Al³⁺ induced lattice distortion that degrades the capacitance. As a result, a relatively large minimum equivalent oxide thickness of 0.63 nm was achieved, which is undesirable for further device scaling.^14–18

Meanwhile, the insertion of an ultra-thin (∼6 Å) GeO₂ interfacial layer (IL) at the STO/Ru interface between the STO film and Ru bottom electrode (BE) via a single-step ALD process, without using the STO seed layer, effectively suppresses the interaction between Ru and STO.¹⁹ Unlike previously introduced ILs (e.g., Al₂O₃ or TiO₂), GeO₂ could achieve effective Ru passivation at sub-nanometer thicknesses, thereby minimizing capacitance degradation. The adoption of the ultrathin GeO₂ IL resulted in the formation of dense STO films with minimal film defects, decreasing leakage current densities by several orders of magnitude compared to STO without the GeO₂ IL. However, the specific electrical effects of the interposed GeO₂ IL have not yet been fully elucidated. Thus, in-depth studies are required to determine how the GeO₂ layer modifies leakage current mechanisms. Notably, the isovalent substitution of Ge⁴⁺ for Ti⁴⁺ would not be expected to introduce acceptor states, indicating that mechanisms beyond those relating to the doping effect must be operating.

In this work, the impact of the GeO₂ insertion layer on the electrical properties of STO was systematically investigated by comparing the leakage current of RuO₂/STO/Ru and RuO₂/STO/GeO₂/Ru capacitors. Comprehensive electrical characterization elucidated the influence of GeO₂-mediated defect curing on Schottky barrier height modulation, internal bias, and leakage suppression mechanisms. The findings presented are expected to aid the optimization of the integration of high-k dielectric materials in next-generation DRAM capacitors.

Experimental

STO and GeO₂ films were deposited on a sputtered Ru (40 nm)/Ta (3 nm)/SiO₂ (100 nm)/Si wafer at a substrate temperature of 350 °C using a traveling wave-type ALD reactor (CN-1, Atomic Classic). Sr(ⁱPr₃Cp)₂, Ti(Me₅Cp)(OMe)₃ (synthesized by Air Liquide), and Ge(NMePh)(NMe₂)₃ (synthesized by EGTM) precursors were used as the Sr, Ti, and Ge precursors (Pr = propyl, Cp = cyclopentadienyl, Me = methyl, and Ph = phenyl), respectively. The canisters of the Sr, Ti, and Ge precursors were heated to 110 °C, 70 °C, and 100 °C, respectively. O₃ (220 g cm⁻³) was used as an oxygen source for TiO₂ and GeO₂, and H₂O (5 °C) was used for SrO. The ALD sequence consisted of precursor injection, Ar purge, reactant injection, and Ar purge. The durations of each step were 3 s–10 s–2 s–10 s for SrO and TiO₂, and 5 s–10 s–2 s–10 s for GeO₂. The deposition of STO involved four TiO₂ sub-cycles and one SrO sub-cycle to achieve a stoichiometric Sr/Ti composition. Rapid thermal annealing (RTA) was conducted under an N₂ atmosphere for 2 min at 650 °C after STO deposition. Details of the ALD process are reported elsewhere.^20,21

The amounts of Sr, Ti and Ge deposited were measured via X-ray fluorescence (XRF; Thermo-Fisher, ARL Quant’X). The physical thicknesses and film densities were estimated by ellipsometer (SE; M-2000, J. A. Woollam) and X-ray reflectivity (XRR; PANalytical, X'pert Pro). The chemical bonding properties of the films were analyzed using X-ray photoelectron spectroscopy (XPS; Kratos, AXIS SUPRA), with a take-off angle of 54.7° for the X-ray photoelectrons. The work function of the electrode was analyzed by ultraviolet photoelectron spectroscopy (UPS; Kratos, AXIS SUPRA). The surface morphology of the films was analyzed using field-emission scanning electron microscopy (FE-SEM; Carl Zeiss, SUPRA 55VP). For electrical characterization, 20 nm-thick RuO₂ and 50 nm-thick Pt films were sequentially deposited via sputtering on top of the STO films to form the top electrodes (TE) of the metal–insulator–metal (MIM) capacitor structure, using a shadow mask with a 300 μm diameter hole pattern, as illustrated in Fig. S1. Leakage currents were measured using an HP4140 picoammeter at temperatures ranging from room temperature to 100 °C. Capacitances were measured using an HP4194A impedance analyzer at an AC oscillation voltage of 50 mV and a frequency of 10 kHz.

Results and discussion

Fig. 1a shows the equivalent oxide thickness (t_ox)–physical thickness (t_phy) plots for the STO films and the STO films with 6 Å-thick GeO₂ layer inserted at the BE interface (b-Ge-STO), which were annealed at 650 °C after STO deposition. The bulk dielectric constant (k_bulk) and interfacial equivalent oxide thickness (tⁱ_ox) values were estimated from the slope and y-intercept from the best linear-fitted graphs (t_ox = tⁱ_ox + 3.9 × t_phy/k_bulk). Considering the k_bulk values were 199–210, each STO layer feasibly formed a high-k perovskite structure regardless of the presence of the GeO₂ IL. The tⁱ_ox value of STO without the GeO₂ IL was 1.4 Å due to the presence of a low-k interfacial layer induced by abnormal overgrowth at the initial stage of the STO ALD process on the Ru BE.^21,22 Despite deposition of the 6 Å-thick GeO₂ IL at the BE interface, the tⁱ_ox value of the b-Ge-STO sample was only 1.7 Å. If this tⁱ_ox value originated from the 6 Å-thick GeO₂ IL, its k-value must be 13.8, which is unreasonably high when compared to the reported dielectric constant of crystalline GeO₂ (4–13).^23,24 If the presence of the original low-k IL of STO, which induced a tⁱ_ox value of 1.4 Å, was assumed, the tⁱ_ox value increase caused by the 6 Å-thick GeO₂ IL was only 0.3 Å, which is even more unreasonable. Therefore, it was concluded that the GeO₂ IL at the BE interface feasibly suppressed the adverse interfacial low-k layer formation during the ALD, and a substantial amount of Ge ions diffused into the STO film during the PDA at 650 °C rather than forming a distinctive IL. Fig. 1b shows the ToF-SIMS profile of b-Ge-STO after PDA, confirming that substantial diffusion of Ge into the STO film had occurred.

Fig. 1 (a) t_ox–t_phy plots for STO and b-Ge-STO. (b) ToF-SIMS profile of 12 nm-thick b-Ge-STO after PDA. (c) J–V curves of 10- and 20-nm thick STO and b-Ge-STO under positive and negative biases. (d) The Ge defect formation energy as a function of the chemical potential in STO under Sr-rich and Ti-rich conditions. (e) Electronic band structures of STO and b-Ge-STO. (f) J–t_ox curves for STO and b-Ge-STO.

Fig. 1c compares leakage current densities (J) of 10 and 20 nm-thick STO and b-Ge-STO films when positive (closed data points) and negative (open data points) biases were applied to the TE of each MIM capacitor. The leakage current of STO was excessively high, ∼10⁻³ A cm⁻² with a 10 nm-thick STO, and ∼10⁻⁵ A cm⁻² with a 20 nm-thick STO at +0.8 V. These values make STO an unsuitable DRAM capacitor dielectric material, which requires a leakage current below 10⁻⁷ A cm⁻² at ±0.8 V for stable data retention and device reliability. The leakage current of STO under positive bias was about one order of magnitude higher than under negative bias, which demonstrates that the TE RuO₂ has a higher work function of than the BE Ru. The high leakage current of STO was attributed to the inherent n-type band structure of the STO film (discussed later) and its rough morphology resulting from overgrowth on the Ru BE. The numerous nano-voids in the STO film formed after PDA seen in the planar SEM image in Fig. S2a support this hypothesis as they make STO vulnerable to increased leakage current.

Meanwhile, the leakage currents of b-Ge-STO were significantly suppressed compared to STO, regardless of the bias polarity, where values of ∼10⁻⁶ A cm⁻² for a 10 nm-thick b-Ge-STO film and 10⁻⁸ A cm⁻² for a 20 nm-thick b-Ge-STO film were observed. The SEM image in Fig. S2b reveals that the b-Ge-STO films have improved morphology without nano-voids due to the Ru passivation effect of GeO₂ during the STO ALD process.¹⁹ Since the electrical field tends to be locally concentrated at voids and surface protrusions, this morphological improvement significantly decreased leakage current, particularly by suppressing nano-void formation. Despite maintaining the same asymmetric RuO₂/b-Ge-STO/Ru electrode configuration, b-Ge-STO exhibited near-symmetric leakage current characteristics under both bias polarities, which indicates that the suppression of leakage current under positive bias was more pronounced in the b-Ge-STO film than in the STO film. These observations suggest that the GeO₂ interfacial layer not only influenced the morphology of the film but also altered the electrical properties at the BE interface, particularly with regard to oxygen vacancy (V_O) formation. As discussed in detail below, the formation of point defects, such as V_O, played a more crucial role in governing the leakage current than morphological variations.

Ab initio density-functional theory (DFT) calculations were performed to investigate the impact that the diffused Ge ions have on the electrical properties of STO. The details of the calculations are included in the SI. Fig. 1d compares the formation energies of the possible Ge defect sites – substitutional Ge_Ti and Ge_Sr sites, and interstitial Ge_i,tetra and Ge_i,octa sites. The range of the Fermi level is defined by the band offset with the adjacent electrodes, Ru and RuO₂. Among the investigated Ge defects, the neutral species is the most stable defect species throughout the Fermi level range. Since the Ge⁴⁺ ions substitute with Ti⁴⁺ sites, no net charge change is induced by such a substitution. The other defect species are not energetically favoured. Fig. 1e compares the electronic band diagrams of Ge-STO and STO, showing that the species does not introduce defect levels within the bandgap. The estimated bandgaps are almost identical (1.83 eV and 1.81 eV, respectively), and are underestimated by the functional selected for the DFT calculations. Fig. 1d and e confirm that Ge substitution alone did not directly influence the leakage behavior in b-Ge-STO. Instead, the concurrent V_O curing induced by the diffusive GeO₂ interlayer drives the observed improvement in electrical performance.

Fig. 1f compares the J–t_ox curves of STO and the b-Ge-STO films. Although STO exhibited low t_ox values, its J–t_ox performance was inferior to that of b-Ge-STO due to the high leakage current. The minimum t_ox value (t_ox,min) and corresponding t_phy value that meet the DRAM leakage criterion are 0.69 and 27 nm for the STO film, and 0.40 and 11 nm for the b-Ge-STO film, respectively. This result is noteworthy, as it demonstrates superior electrical properties without needing a pre-crystallized seed STO layer, which was previously required.^20,21,25 Consequently, b-Ge-STO was identified as a promising dielectric material for DRAM capacitors, and its conduction mechanism was systematically investigated in this study.

Fig. 2 displays the deconvoluted XPS Ru 3d spectra of the Ru substrates in the STO and b-Ge-STO films at different ALD stages. Each spectrum was deconvoluted into Ru metal (280.0 eV), RuO_x (280.1–280.6 eV), and satellites, and the insets in each figure show the schematic diagram of the Ru and RuO_x surfaces at each deposition stage. Fig. 2a–c display the Ru 3d spectra of a bare Ru substrate, a Ru substrate covered with four Ti–O ALD cycles (the first four sub-cycles of the STO super-cycle), and a Ru substrate covered with 5 nm-thick STO film, respectively. As shown in Fig. 2a, the bare Ru surface exhibits a peak at 280.6 eV, attributed to native surface oxidation. The STO ALD proceeded with four TiO₂ ALD subcycles followed by one SrO ALD subcycle. When the Ti precursors were introduced to the process, the RuO_x initially present or formed later during the subsequent O₃ supply steps was reduced to metallic Ru (Fig. 2b), and the produced oxygen atoms were supplied to the incoming Ti-precursors to form the thick and relatively irregular TiO₂ films, as shown in the inset schematic diagram. The reduction of RuO_x by the Ti precursors was attributed to Ti–O having a higher oxygen binding energy (ΔG_f,600K: −834 kJ mol⁻¹) than Ru–O (ΔG_f,600K: −201 kJ mol⁻¹). Because the four TiO₂ ALD subcycles end with the final supply of O₃ (and purge), the Ru 3d spectrum returns to its initial configuration. However, when the better oxygen scavenger Sr precursor (ΔG_f,600K: −857 kJ mol⁻¹) and the worse oxidant H₂O were injected, the interfacial RuO_x mostly decomposed, as shown in Fig. 2c, with excessive growth of SrO observed after the SrO ALD subcycle. Fig. 2d shows Ru 3d spectrum of a Ru substrate covered with two Ge–O ALD cycles (corresponding to 6 Å-thick layers). Since Ge–O binds oxygen much more weakly (ΔG_f,600K: −251 kJ mol⁻¹) than Ti–O, less reduction of RuO_x occurred during the deposition of GeO₂, thus resulting in a minor RuO_x peak shift to 280.4 eV. Furthermore, GeO₂ uniformly covered the Ru substrate without creating any pin-holes, thus minimizing the exposure of RuO_x during subsequent STO deposition. Fig. 2e shows the Ru 3d XP spectrum after deposition of a 5 nm-thick b-Ge-STO film. Compared with the Ru 3d peak variation between Fig. 2b and c, the Ru 3d peak in Fig. 2e shows a similar configuration to Fig. 2d, suggesting that the intervening GeO₂ film suppressed the adverse chemical interaction between the interfacial RuO_x and growing STO film. The RuO_x peak remained at 280.4 eV, which indicates no further reduction of the RuO_x surface. The electronic properties of the O₃-treated Ru surface were examined using UPS, as they play a crucial role in determining the electrical performance of the STO films grown on top. Fig. S3 displays the UPS valence band spectra of the O₃-treated Ru substrate. The O₃ treatment involved injecting O₃ for 3 seconds and purging with Ar for 10 seconds and repeating for four cycles. The extracted work function (Φ = hν − (E_cut-off − E_F)) of the surface-oxidized Ru substrate was 4.9 eV, higher than that of Ru metal (4.7 eV). The work function of RuO₂ is 5.2 eV, suggesting that the oxidized layer (RuO_x, x < 2) at the Ru surface could enhance the effective work function. The b-Ge-STO film exhibits a stronger RuO_x peak in the Ru 3d XPS spectrum than the STO film which indicates that there is a higher barrier at the BE interface that suppress leakage current under positive bias.

Fig. 2 Ru 3d XP spectra of (a) the bare Ru substrate, and the Ru substrate covered by (b) four ALD TiO₂ cycles, (c) 5 nm-thick STO, (d) two ALD GeO₂ cycles, and (e) 5 nm-thick b-Ge-STO. The STO and b-Ge-STO films (c) and (e) were annealed at 650 °C.

Furthermore, intrinsic V_O defects were reported to be a dominant factor in inducing the n-type characteristics of STO.^12,26,27 V¹⁺_O and V²⁺_O defects were reported to create trap states at 0.57 eV and 0.28 eV below the conduction band edge (E_c) of STO, respectively.²⁸ These defect states contribute to the shift of the Fermi level (E_F) from the mid-gap towards the E_c. Consequently, a high V_O density would lower electrical carrier injection barrier (mostly electrons) of the MIM structure, degrading leakage performance. Angle-resolved XPS (ARXPS) analysis, in which the X-ray penetration depth was adjusted to reveal the chemical distribution, was conducted to confirm the V_O states in the STO and b-Ge-STO films. Fig. 3a displays the O 1s spectra of the 5 nm-thick STO and b-Ge-STO films obtained from the ARXPS analysis, where the angles corresponded to the tilting angle of the specimen; lower angles were associated with deeper, bulk-sensitive measurements, whereas higher angles were associated with shallower, surface-sensitive measurements.²⁹ An increase of the SrCO₃ peak (∼531.5 eV) relative to the STO (∼529.7 eV) peak was observed near the surface due to air exposure of the samples.

Fig. 3 O 1s XP spectra of the 5 nm-thick STO and b-Ge-STO films. The specimen tilting angle was varied over a range of 0–60° to elucidate the chemical distribution in the STO layer. (b) The area ratios as a function of the measurement angle of the oxygen vacancies in STO and b-Ge-STO and Ge-O in b-Ge-STO.

Each O 1s spectrum was deconvoluted to determine the relative concentrations of V_O and GeO₂ within the STO films based on the area ratios relative to the STO lattice oxygen, as shown in Fig. 3b. The relative V_O concentration in STO decreased from 15.3% in the bulk region (0°) to 9.5% in the surface region (60°), indicating that STO contained a high density of V_O defects, which were predominantly concentrated in the bottom interface region. In contrast, b-Ge-STO exhibited a uniform and low V_O concentration (∼5%) across the bulk and surface regions. As shown in Fig. 1b, substantial Ge and O ions dissociated from the GeO₂ layer and diffused into the STO film during PDA, thereby curing the V_O that existed in the STO layer. Since the diffusion profile of Ge–O gradually decreased nearer the STO surface, the diffused oxygen preferentially cures the V_O at the bottom interface region of the STO film. Consequently, the density of the V_O in the bottom region was significantly decreased, producing a uniformly low V_O concentration in the b-Ge-STO layer. Fig. S4 compares the deconvoluted Ti 2p XP spectra of STO and b-Ge-STO according to the ARXPS tilting angle. In the STO sample, the TiO_2−x component (∼528.0 eV) accounted for a larger fraction of the Ti 2p spectrum than in b-Ge-STO, and its relative area decreased systematically with increasing tilting angle. In contrast, the Ti 2p spectra of b-Ge-STO are dominated by the TiO₂ component (∼528.4 eV), with the TiO_2−x contribution remaining low and essentially invariant when the tilting angle is changed. Therefore, these results confirm that the STO film contains a higher V_O concentration near the bottom interface region. In contrast, the diffusion of the GeO₂ interlayer in b-Ge-STO effectively cured the inherent V_O in the STO throughout the entire thickness of the film. As confirmed by XPS analysis, such a decrease of n-type V_O defects in b-Ge-STO would shift the E_F back to the mid-gap. When considered in conjunction with Fig. 2, the pronounced suppression of leakage current observed in b-Ge-STO under positive bias in Fig. 1c could be ascribed to the increased charge injection barrier at the STO/Ru BE interface caused by such a E_F shift.

Fig. 4a displays the normalized C–V curves of the 13 nm-thick STO and b-Ge-STO films. The intrinsic voltage-dependent capacitance originated from soft-mode hardening of the STO: as the DC bias increased, anharmonic vibrations within the Ti–O lattice stiffen the soft optical phonon modes, thus decreasing differential polarizability (dP/dE) and consequently the capacitance.^30,31 The peak capacitance (C_max) value was observed at +0.9 V and +0.3 V in STO and b-Ge-STO, respectively, indicating the presence of the internal bias (V_int). The inset of Fig. 4a presents the C–V curves of STO and b-Ge-STO after shifting the voltage axis by V_int to superimpose their respective C_max values. Whereas both films exhibit nearly symmetric curves, the slope of the capacitance decrease differed, with STO showing a steeper decline than b-Ge-STO. Liu et al. have attributed this behavior to the formation of polarization by localized V_O–Ti³⁺ dipoles in STO, which becomes saturated under applied bias, leading to a more pronounced decrease in capacitance relative to C_max.^32,33 Since the STO film contained a higher density of V_O than the b-Ge-STO film, its capacitance decayed more steeply with bias. Fig. 4b illustrates the schematic band diagrams of the STO and b-Ge-STO MIM capacitors. In the case of STO, the relatively low V_O level at the upper region (as shown in Fig. 3) allowed the TE contact barrier (ϕ_b,TE) to be determined by the difference in the work function of the TE RuO₂ (Φ_TE, ∼5.2 eV) and the electron affinity of STO (χ_STO, 3.9–4.1 eV), which resulted in a ϕ_b,TE of ∼1.1–1.3 eV.^34,35 If the V_int contained in STO was 0.9 V, as shown in Fig. 4a, the BE contact barrier (ϕ_b,BE) would be 0.2–0.4 eV. It is smaller than the ΔΦ(Φ_TE − Φ_BE) of 0.6–0.8 eV between the BE Ru (Φ_TE, ∼4.7 eV) and χ_STO, suggesting that the high concentration of n-type V_O in the bottom interface causes E_F to shift toward E_c. In contrast, b-Ge-STO maintained a lower V_O level than STO across the film thickness, so assuming no E_F pinning by the V_O, the V_int would arise solely from the ΔΦ. Given the RuO₂ TE, the presence of RuO_x at the BE surface increased the ΔΦ_BE to ∼4.9 eV (as shown in Fig. 2 and Fig. S3). Consequently, the Φ between the two electrodes would be ∼0.3 eV, which matches the V_int observed in the C–V curve of the b-Ge-STO capacitor. The changes in the band structure of STO and b-Ge-STO were further examined by analyzing the electrical conduction mechanisms across different voltage regions to verify the leakage suppression.

Fig. 4 (a) Normalized C–V curves of 13 nm-thick STO and the b-Ge-STO film. (b) Schematic band diagrams of STO and b-Ge-STO containing internal biases.

Fig. 5a and b compare the J–V curves of the capacitors with the 20 nm-thick STO and b-Ge-STO films measured at 303–373 K, where a positive bias was applied to the TE to elucidate the band modulation at the BE interface. The applied voltage step was 0.02 V, and a delay time of 5 s was used to avoid the involvement of the possible local polarization and dielectric relaxation currents. As confirmed in Fig. 4, the positive V_int of 0.3 V for STO and 0.9 V for b-Ge-STO exist due to the asymmetric band structure, rendering it challenging to analyze the forward current injected from the BE if the applied bias (V_app) is smaller than V_int. Consequently, the V_app was adjusted by subtracting V_int to analyze the net current injected from the BE, and the effective bias (V_eff = V_app − V_int) was used in deriving the current equations presented below. Fig. 5c and d present the replotted J–E_eff curves of the STO and b-Ge-STO films, respectively.

Fig. 5 Leakage current density–applied voltage curves measured between 303 and 373 K for the 20 nm-thick (a) STO, and (b) b-Ge-STO films. Modified leakage current density-effective field curves for the (c) STO, and (d) b-Ge-STO films.

Fig. 6a shows the log J–log V_eff plots of the STO film under E_eff < 0.35 MV cm⁻¹, where the slope allows both the determination of the voltage dependence of the current and estimation of the leakage mechanisms across different field regions. In the low-field region, J linearly increased with V_eff, indicating ohmic conduction. The low ϕ_b,BE value (0.2–0.4 eV, estimated in Fig. 4) facilitated the maintenance of the equilibrium carrier concentration under such low bias conditions. Meanwhile, beyond a specific transition voltage (V_tr), where J increases with V_eff², a transition occurred suggesting a transition of the conduction mechanism to trap-limited space charge limited conduction (SCLC). The SCLC behavior can be described by the following equation:³⁶

where ε is the static dielectric constant, θ is the ratio of the free carrier density to total carrier density, and d is the film thickness. In the SCLC regime, as the voltage increased, charge carriers easily filled the traps, leading to a more rapid increase in current compared to the ohmic region. The presence of V²⁺_O traps (0.28 eV below E_c) near the E_F facilitated electron trapping. This carrier injection contributed to the rapid filling of traps and the transition to SCLC at lower voltages. As the temperature increased, carriers were more easily trapped, leading to a shift in the V_tr to lower values.

Fig. 6 Re-plotted J–E curves of STO: (a) ln J–ln V_eff graphs at low field (E_eff < 0.35 MV cm⁻¹, ohmic conduction and SCLC), and (b) ln(J/E_eff²)–1/E_eff graphs at high field (E_eff > 0.35 MV cm⁻¹, F–N tunneling).

In the high-field region (E_eff > 0.35 MV cm⁻¹), the large electric field bent the band downward, significantly increasing the Fowler–Nordheim (F–N) tunneling current as the electron wave function directly penetrated the triangular potential barrier. Considering the relatively small barrier height at BE, F–N tunneling could make substantial contributions to the leakage current under a higher electric field, which is expressed by eqn (2):³⁷

where q is the elementary charge, h is the Planck constant, E_eff is the effective electric field, is the effective tunneling mass, and ϕ_b is the tunneling barrier. Fig. 6b shows the STO ln(J/E²)–1/E_eff curves replotted according to eqn (2). From the linear fitted plots of Fig. 6b, the ϕ_b could be calculated at each measurement temperature. The value used in the calculation was 0.15m₀, a value commonly used in previous studies for the tunneling mass of high-k oxides.^38,39 The average ϕ_b value was calculated to be ∼0.32 eV, which closely matches the value estimated from the internal bias in Fig. 4.

Fig. 7 analyzes the leakage current mechanism of the b-Ge-STO film. Fig. 7a presents the ln(J/T²)–E^1/2 graphs, which follow the Schottky emission equation given by eqn (3):⁴⁰

where A* is the effective Richardson constant, qϕ_b is the zero-bias Schottky barrier height (ϕ_b,BE), ε₀ is the vacuum permittivity, ε_i is the optical dielectric constant, T is the temperature, and k is the Boltzmann constant.¹² If the leakage current was governed by Schottky emission, a linear relationship between ln(J/T²) and E^1/2 should be obtained, and the ε_i calculated from the slope should be identical to n² (3.80), where n is the refractive index of STO (1.95, measured by an ellipsometer at a wavelength of 632.8 nm).⁴¹ The Schottky equation provided a feasible linear fit when the E_eff was below 0.5 MV cm⁻¹, as indicated by the solid lines in Fig. 7a. The average calculated ε_i value was 3.15, reasonably coincident with 3.80, implying that the charge injection at the BE was governed by Schottky emission. Fig. 7b shows the Arrhenius (ln(J/T²)–1/T) plots at each given field, and the barrier heights extracted from the slope of the best linear-fitted Arrhenius plots are presented in Fig. 7c. The ϕ_b value extracted from the extrapolation of the barrier heights at different biases to zero field was ∼0.74 eV, representing a ∼0.42 eV increase compared to STO (0.32 eV). Such an increase in the ϕ_b value was attributed to the presence of RuO_x on the Ru surface (Fig. 2) and V_O curing (Fig. 3), which together resulted in a higher ϕ_b value at the Ru BE contact point.

Fig. 7 (a)–(c) Re-plotted J–E curves of b-Ge-STO at low field (E_eff < 0.5 MV cm⁻¹, Schottky emission): (a) ln J/T²–E_eff^1/2 plots, (b) Arrhenius plots (ln J/T²–1/T), and (c) extracted barrier heights for Schottky emission as a function of E_eff^1/2. (d)–(f) Re-plotted J–E curves of b-Ge-STO at high field (E_eff > 0.5 MV cm⁻¹, TAT): (d) ln J–ln E_eff plots, (e) Arrhenius plots (ln J–1/T), and (f) extracted barrier heights for TAT as a function of E_eff.

In the high-field region (E_eff > 0.5 MV cm⁻¹), Schottky emission could no longer explain the leakage current. The following is the trap-assisted tunneling (TAT) equation that better captures the temperature-dependent conduction in b-Ge-STO film:³⁶

where a is the hopping distance, n is the charge density, μ is the mobility, and E_a is the activation energy. Fig. 7d and e show the ln J–E plots at different temperatures and the ln J–1/T plots at different electric fields of the b-Ge-STO film. From the linear fitting of each plot in Fig. 7d, a trap distance of 3.38 nm was extracted, representing the effective tunneling distance involved in the conduction process. In Fig. 7e, the barrier height (qaE − E_a) at each electric field was obtained from the Arrhenius plots' slopes, and extrapolating these values yielded an E_a of ∼0.56 eV. This value was in close agreement with the trap level of ∼0.57 eV previously reported in previous research, indicating that V⁺_O remained the dominant defect state contributing to the leakage current in the b-Ge-STO film under high field, as the V²⁺_O defects were preferentially cured by Ge–O diffusion.²⁸ Considering the film thickness (20 nm), the hopping distance of 3.38 nm indicates a relatively sparse V⁺_O trap distribution. This finding supports the interpretation that TAT occurred over a limited number of trap sites, decreasing the overall tunneling probability and contributing to the observed suppression of the leakage current in the b-Ge-STO film.

Fig. 8 illustrates a schematic band diagram representing the dominant conduction mechanisms in the STO and b-Ge-STO films under various electric fields. In the RuO₂/STO/Ru MIM structure, a V_int was induced due to the Φ_TE − Φ_BE value. Additionally, the high concentration of V_O near the bottom electrode produced a low ϕ_b value at the STO/Ru BE interface, further intensifying the V_int. V_eff was used to elucidate the net current injected from the BE, which corresponds to the voltage required to achieve a flat band state (zero V_eff). STO exhibits a low ϕ_b (0.32 eV), with V²⁺_O concentrated near the BE interface, facilitating trap filling and leading to trap-limited SCLC at low V_eff. The low ϕ_b also facilitated the formation of a triangular potential barrier, making it easier for current injection via F–N tunneling at high V_eff, thereby contributing to high leakage current in STO. On the other hand, b-Ge-STO, with improved BE interface properties, had a relatively high ϕ_b value of 0.74 eV. Additionally, the V_O curing effect feasibly diminished the number of shallow V²⁺_O traps, minimizing the involvement of mechanisms such as ohmic conduction and SCLC. Consequently, at lowV_eff, Schottky emission dominates in the b-Ge-STO. As the V_eff increased, tunneling occurred along the bulk traps, with the low-concentration V⁺_O traps remaining after curing. The improved interface properties and defect curing caused by the GeO₂ IL lead to a more controlled conduction behavior, resulting in a notable suppression of leakage current compared to STO.

Fig. 8 Schematic band diagrams of MIM capacitors made from (a) STO, and (b) b-Ge-STO in each V_eff region.

Conclusions

In this study, inserting a thin GeO₂ interlayer at the STO/Ru bottom electrode interface proved to be an effective strategy for enhancing the electrical performance of STO films used in DRAM capacitors. The GeO₂ interlayer passivated the Ru surface during STO ALD and released oxygen during PDA, which cured the V_O throughout the STO while suppressing the formation of a low-k interfacial layer, yielding dense, defect-free films. The enhanced electrode symmetry and diminished E_F pinning caused by the V_O decreased (increased) the internal bias (Schottky barrier height at the BE interface) from +0.9 V (0.32 eV) in STO to +0.3 V (0.74 eV) in b-Ge-STO. Consequently, the leakage current decreased by orders of magnitude while achieving a minimum t_ox value of 0.40 nm that satisfies the DRAM leakage criterion without a pre-crystallized seed layer. Temperature- and field-dependent measurements revealed that STO exhibited ohmic conduction and SCLC at low fields and F–N tunneling at high fields. In contrast, b-Ge-STO was governed by Schottky emission at low fields and sparse-trap TAT at high fields. C–V measurements corroborated the decreased internal bias and the mitigation of V_O-induced non-linearity. Therefore, it was concluded that interfacial engineering using GeO₂ simultaneously optimized the interfacial chemistry, defect distribution, and band alignment through a single-step ALD modification compatible with Ru/RuO₂ electrodes, reinforcing the suitability of this approach for creating high-k dielectric materials for use in next-generation DRAM.

Author contributions

H. Paik designed and performed the experiments and wrote the manuscript. D. Kim and J.-H. Choi performed the DFT calculation. D. S. Kwon, J. Lim, H. Seo, T. K. Kim, J. H. Shin, H. Song, H. Yoon and D. G. Kim contributed to the discussion and assisted with the data interpretation. C. S. Hwang supervised the whole research project and manuscript preparation.

Conflicts of interest

There are no conflicts to declare.

Data availability

All data supporting the findings of this study are included in the article and its supplementary information. Additional data materials are available from the corresponding author upon reasonable request.

Supplementary information available: Supplementary analysis data and details of DFT calculations. See DOI: https://doi.org/10.1039/d5tc02736e.

Acknowledgements

This work was supported by the National Research Foundation of Korea (Grant No. 2020R1A3B2079882).

References

1. A. Kosola, M. Putkonen, L.-S. Johansson and L. Niinistö, Appl. Surf. Sci., 2003, 211, 102–112.
2. S. W. Lee, O. S. Kwon and C. S. Hwang, Microelectron. Eng., 2005, 80, 158–161.
3. M. Pawlak, B. Kaczer, M.-S. Kim, M. Popovici, K. Tomida, J. Swerts, K. Opsomer, W. Polspoel, P. Favia and C. Vrancken, Appl. Phys. Lett., 2010, 97, 162906.
4. W. Weinreich, T. Tauchnitz, P. Polakowski, M. Drescher, S. Riedel, J. Sundqvist, K. Seidel, M. Shirazi, S. D. Elliott and S. Ohsiek, J. Vac. Sci. Technol., A, 2013, 31, 01A123.
5. J.-H. Kim, V. A. Ignatova, J. Heitmann and L. Oberbeck, J. Phys. D: Appl. Phys., 2008, 41, 172005.
6. H. Kashyap, P.-C. Lee, K. Chae, M. Passlack, A. K. Yadav, K. Wong, S. Nemani, E. Yieh, J. Spiegelman, K. Cho, A. I. Khan and A. C. Kummel, J. Vac. Sci. Technol., A, 2025, 43, 022401.
7. Y. N. Thakare, U. A. Wankhade and S. N. Kale, Int. J. Electron., 2025, 112, 2291–2310.
8. R. W. Johnson, A. Hultqvist and S. F. Bent, Mater. Today, 2014, 17, 236–246.
9. S. Saha, T. Sinha and A. Mookerjee, J. Phys.: Condens. Matter, 2000, 12, 3325.
10. K. Van Benthem, C. Elsässer and R. French, J. Appl. Phys., 2001, 90, 6156–6164.
11. D. A. Muller, N. Nakagawa, A. Ohtomo, J. L. Grazul and H. Y. Hwang, Nature, 2004, 430, 657–661.
12. S. W. Lee, J. H. Han and C. S. Hwang, Electrochem. Solid-State Lett., 2009, 12, G69.
13. W. Lee, S. Yoo, W. Jeon, Y. W. Yoo, C. H. An, M. J. Chung, H. J. Kim, S. W. Lee and C. S. Hwang, Thin Solid Films, 2015, 589, 723–729.
14. S. H. Kim, W. Lee, C. H. An, Y. Kim, D. S. Kwon, D.-G. Kim, S. H. Cha, S. T. Cho, J. Lim and C. S. Hwang, Phys. Status Solidi RRL, 2019, 13, 1900373.
15. S. K. Kim, G. J. Choi, S. Y. Lee, M. Seo, S. W. Lee, J. H. Han, H. S. Ahn, S. Han and C. S. Hwang, Adv. Mater., 2008, 20, 1429–1435.
16. S. K. Kim, G. J. Choi, J. H. Kim and C. S. Hwang, Chem. Mater., 2008, 20, 3723–3727.
17. B. Kim, Y. Choi, D. Lee, Y. Byun, C. Jung and H. Jeon, ECS J. Solid State Sci. Technol., 2021, 10, 083006.
18. J.-Y. Baek, S. Y. Lee and H. Seo, J. Mater. Sci. Technol., 2020, 42, 28–37.
19. H. Paik, J. Lim, H. Seo, T. K. Kim, J. Shin, H. Song, D. G. Kim, W. Lee, D. S. Kwon and C. S. Hwang, Mater. Horiz., 2025, 12, 7305–7317.
20. S. H. Kim, W. Lee, C. H. An, D. G. Kim, D. S. Kwon, S. T. Cho, S. H. Cha, J. I. Lim and C. S. Hwang, Phys. Status Solidi RRL, 2019, 13, 1800557.
21. W. Lee, J. H. Han, W. Jeon, Y. W. Yoo, S. W. Lee, S. K. Kim, C.-H. Ko, C. Lansalot-Matras and C. S. Hwang, Chem. Mater., 2013, 25, 953–961.
22. S. W. Lee, J. H. Han, S. Han, W. Lee, J. H. Jang, M. Seo, S. K. Kim, C. Dussarrat, J. Gatineau, Y.-S. Min and C. S. Hwang, Chem. Mater., 2011, 23, 2227–2236.
23. S. N. A. Murad, P. T. Baine, D. W. McNeill, S. J. N. Mitchell, B. M. Armstrong, M. Modreanu, G. Hughes and R. K. Chellappan, Solid-State Electron., 2012, 78, 136–140.
24. Y. Kamata, Mater. Today, 2008, 11, 30–38.
25. S. W. Lee, J. H. Han, O. S. Kwon and C. S. Hwang, J. Electrochem. Soc., 2008, 155, G253.
26. J. Wang, T. Wang, Z. Zhao, R. Wang, C. Wang, F. Zhou, S. Li, L. Zhao and M. Feng, J. Alloys Compd., 2022, 902, 163865.
27. R. Moos, W. Menesklou and K. H. Härdtl, Appl. Phys. A, 1995, 61, 389–395.
28. C. Mitra, C. Lin, J. Robertson and A. A. Demkov, Phys. Rev. B: Condens. Matter Mater. Phys., 2012, 86, 155105.
29. C. S. Fadley, Prog. Surf. Sci., 1984, 16, 275–388.
30. K. A. Müller and H. Burkard, Phys. Rev. B: Condens. Matter Mater. Phys., 1979, 19, 3593.
31. A. K. Tagantsev, V. O. Sherman, K. F. Astafiev, J. Venkatesh and N. Setter, J. Electroceram., 2003, 11, 5–66.
32. X. Z. Liu, B. W. Tao and Y. R. Li, J. Mater. Sci., 2007, 42, 389–392.
33. H. Trabelsi, M. Bejar, E. Dhahri, M. A. Valente and M. P. F. Graça, Phys. E, 2019, 108, 317–325.
34. J. Robertson and C. W. Chen, Appl. Phys. Lett., 1999, 74, 1168–1170.
35. A. Lübcke, F. Zamponi, R. Loetzsch, T. Kämpfer, I. Uschmann, V. Große, F. Schmidl, T. Köttig, M. Thürk, H. Schwoerer, E. Förster, P. Seidel and R. Sauerbrey, New J. Phys., 2010, 12, 083043.
36. F.-C. Chiu, Adv. Mater. Sci. Eng., 2014, 2014, 578168.
37. R. G. Forbes and J. H. Deane, Proc. R. Soc. London, Ser. A, 2007, 463, 2907–2927.
38. W. Jeon, S. H. Rha, W. Lee, Y. W. Yoo, C. H. An, K. H. Jung, S. K. Kim and C. S. Hwang, ACS Appl. Mater. Interfaces, 2014, 6, 7910–7917.
39. Y.-C. Yeo, T.-J. King and C. Hu, Appl. Phys. Lett., 2002, 81, 2091–2093.
40. J. Simmons, Phys. Rev. Lett., 1965, 15, 967.
41. M. Wöhlecke, V. Marrello and A. Onton, J. Appl. Phys., 1977, 48, 1748–1750.

---