Laterally and depth-resolved cathodoluminescence spectroscopy (DRCLS) provided direct, nanoscale measurements of oxygen vacancy and oxygen vacancy complex distributions in undoped and Fe-doped SrTiO3 with high temperature electric field stress associated with dielectric resistance degradation. DRCLS provided direct and spatially resolved observation of oxygen vacancy migration driven by external electric fields from the anode to the cathode in undoped SrTiO3 between laterally separated electrodes, resulting in increased current leakage and lower thermal breakdown strength. DRCLS measurements through planar Pt electrodes after high temperature electric field cycling reveal pronounced oxygen vacancy depletion within the surface space region of the Pt/SrTiO3 Schottky barrier as predicted theoretically. These results provide a direct insight into the transient states impacting the conduction during the electromigration of the oxygen vacancies. The deconvolution of different peaks and their intensity variations relative to the direct bandgap provide methods to gauge the relative defect energetics of these gap states. These data are discussed in relation to providing a tool to further understand conduction in mixed ionic conductors.

SrTiO3 is widely used in many device applications including high energy density, high voltage capacitors,1–3 low voltage varistors,4 and tunable microwave circuit applications.5,6 Key to the performance of many of these applications is their stability at elevated temperature and under high electric field stress, which is limited by resistance degradation. Under high applied electric field and temperature, resistance degradation vs time can cause failure within circuits and electronic systems that contain them. The mechanism of such electrical as well as optical degradation has been investigated extensively in the past few decades.7–12 For capacitor applications, the breakdown of SrTiO3 as a high dielectric strength and tunable material for capacitor applications involves a number of physical mechanisms. Electromigration of oxygen vacancies that build ionic space charge regions within the dielectric is a major concern. Defects, especially oxygen vacancies, are commonly believed to play a key role in the degradation process. First-principles calculations show that formation energies of vacancy defects in SrTiO3 vary according to equilibrium conditions and different charge states but that charged oxygen vacancies are relatively more stable within most of the Fermi level position range.13,14 Furthermore, oxygen vacancies are mobile and well known to redistribute spatially under external electric fields.15,16 By investigating the degradation behavior of ceramic and single crystal SrTiO3 with different dopant concentrations at different temperatures and under different DC voltages, Waser et al. proposed that the degradation process is due to electromigration of oxygen vacancies between the anode and the cathode.8 Studies using thermally activated diffusion models of this migration and spatial redistribution17–20 showed this diffusion to be a function of many variables including temperature, oxygen pressure, electric fields, and light and can exhibit nonuniformities between crystal orientations21 as well as between the surface and the bulk.22 Under applied electric fields, oxygen vacancies are known to diffuse and change their complex nature.17 

With impedance spectroscopy and analytical transmission electron microscopy, Yang et al.23 and Randall et al.24 showed that the degradation in BaTiO3-based commercial capacitors is caused by oxygen vacancy concentrations that increased across the dielectric layers toward the cathode and decreased toward the anode region. Moballegh and Dickey investigated defect accumulation at the cathode electrode of TiO2 − x during degradation.25 Recently, Bayer et al. investigated the local conductivity and defect distributions of degraded Fe-doped and undoped SrTiO3 by in situ impedance studies. Activation energies measured before and after degradation were correlated with a defect model based on an additional acceptor-type trap.26 It is generally accepted that during the degradation process for Fe-doped SrTiO3, oxygen vacancies migrate from the positively charged anode toward the negatively charged cathode, resulting in an anode region depleted of oxygen vacancies and an accumulation of oxygen vacancies in the cathode region.8,17,27,28 The oxygen vacancy movement away from the anode and the subsequent local charge compensation result in an increased hole and Fe4+ concentration with increasing p-type conductivity, and often a electrocoloration change near the anode.29,30 whereas oxygen vacancies increasing toward the cathode introduce more donors and higher electron concentration. These changes are the dominant contributions to the orders-of-magnitude difference in electrical conductivity between the anode and the cathode since electron and hole mobilities in SrTiO3 are much higher than the ion mobility associated with the oxygen vacancy movement. Recently, using the established literature, with the impedance methods using electrical modulus, with other experimental input, data were modeled to account for the transitional changes in both spatial conduction and activation energies. The modeling was found to not correlate with specific conduction mechanisms but revealed the changes to be the result of a cumulative response embracing various non-equilibrium conductive processes that are even sensitive to majority and minority impurities.31 

The direct observation of these defect distributions spatially on a macroscopic scale, together with the defects' changes in physical nature and electronic energies on a nanoscale with applied electric fields at elevated temperatures has until now not been fully reported. Complementing our electrical measurements, spatially- and depth-resolved cathodoluminescence spectroscopy (DRCLS) provides direct microscopic evidence for the mechanism of resistance degradation in both Fe-doped and undoped SrTiO3. Cathodoluminescence is one of several processes that occur with electron irradiation of solids.32 In general, these incident electrons produce a cascade of secondary electrons that lose energy by (i) Auger electron excitation whose unscattered Auger electrons escaping from the outer few angstroms of the free surface form the basis for Auger electron spectroscopy,33,34 (ii) secondary electrons resulting from collisions as well as ionization of atoms at depths that depend on the incident beam energy, (iii) backscattered electrons at greater depths resulting from random collisions of electrons, (iv) x-rays characteristic of specific atomic transitions, (v) a continuum of x-rays that result from secondary x-ray excitation, (vi) fluorescent x-rays due to electrons initially excited by x-rays, and (vii) the generation of electron–hole pairs due to impact ionization, i.e., electron collision with the solid's atoms and detachment of outer shell electrons at the final stage of the electron cascade.35 X-ray generation dominates electron losses above a few tens of keV while plasmon generation is predominant for electron energies down to energies below 100 eV.36 At the final stage of the electron cascade, electrons have only enough energy for impact ionization of the solid's atoms, creating electron–hole pairs.

The DRCLS technique involves electron beams with incident energies EB controlled from a few tens of keV down to a few hundred eV or less. For EB in this range, plasmons dominate electron energy loss, followed by impact ionization. Electron–hole pair creation requires that electron energies must exceed a multiple of the semiconductor/insulator bandgap in order to satisfy both energy and momentum conservation.37 Recombination of these free carriers for optical emission occurs by conduction-to-valence band and gap state-band edge electronic transitions. The excitation depth of the incident beam depends on the distribution of electrons with energies in this low energy range, i.e., a few tens of eV or less. For semiconductors, plasmon energies are in the range of ∼15–20 eV. For electron kinetic energies in the range of 50–100 eV, secondary electrons have minimum inelastic mean free paths of only 2–10 Å38,39 so that optical excitation depth of incident electron beams in the low keV range will be in the range of tens to hundreds of nm. In order to compute the depth distribution of this cathodoluminescence excitation, we use Monte Carlo simulations of the electron cascade process that provide profiles of electron density vs depth for electron kinetic energies of 50 eV including backscattered electrons.40 Depths are accurate to within 1–2 monolayers since only 1–2 additional plasmon losses and scattering lengths are involved as electron kinetic energies decrease further.

In this study, we used lateral- and depth-resolved cathodoluminescence to measure both spatial redistribution of the dominant oxygen vacancy-related defect as well as changes in the physical nature of the defect itself, relative to pristine single crystal SrTiO3 in undoped and doped (Fe) cases. DRCLS allows us to monitor relative defect state changes in three dimensions with spatial resolution on a scale of tens of nanometers. DRCLS spectral features also revealed energy level shifts correlated to defect configuration changes that occurred with the defect migration under electrical bias. In turn, these changes at this time are qualitatively correlated with the conductivity changes commonly termed degradation in SrTiO3 resistance.

Nominally undoped SrTiO3 single crystal samples purchased from MTI corporation were prepared, processed, and electrically measured at the Pennsylvania State University as metal–SrTiO3–metal structures on 5 × 5× 0.5 mm3 thick squares with either planar (metal on square faces) or lateral (metal on 0.5 mm faces) geometries. These sample stacks were processed and contacted with Pt overlayers in multiple ways: Samples 1 and 2 consisted of two undoped SrTiO3 samples whose square planar surfaces were polished followed by annealing at 900 °C for 12 min in air and then coated with 80 nm thick Pt electrodes for electrical characterization. Sample 1 was degraded by applying a 6 V potential at 210 °C between the square face electrodes. Following this degradation, the 6 V voltage was maintained while decreasing the temperature from 210 to 100 °C (in 20 °C steps). In situ impedance spectroscopy vs temperature analysis using an Agilent Precision LCR Meter E4980A between 150 and 400 °C under this 6 V bias yielded temperature-dependent electrical properties. Samples 3 and 4 consisted of two undoped SrTiO3 samples both degraded in the same way as Sample 1 with planar (metal on square faces) (Sample 3) or lateral (metal on 0.5 mm faces) (Sample 4) orientations. Planar and lateral surfaces for both samples were polished followed annealing at 900 °C for 12 min in air and then coated with 20 nm thick Pt electrodes for electrical characterizations. Electrodes on facing 0.5 × 5 mm2 edges applied electric bias across the square planar crystal. Electrodes on facing 5 × 5 mm2 planar surfaces applied electric bias across the 0.5 mm crystal thickness. Pt electrode thicknesses were limited to 20 nm to permit keV electron beams to pass through the Pt and excite the SrTiO3 below as well as to permit luminescence from the semiconductor to pass back out through the metal.

Sample 5 was a SrTiO3 specimen, Fe-doped (0.01 wt %), (100) orientation, annealed in an oxygen partial pressure of 2 Pa at 900 °C, equilibrated for >15 h, quenched to room temperature, with dimensions 3 × 3 × 0.5 mm3 and electrodes on facing 0.5 × 3 mm2 edges. Electrodes were 100–150 nm thick. Electric field degradation for all postdegradation specimens followed the voltage/temperature schedule described above and as reported previously.26Table I summarizes the basic crystal characteristics, contact orientations, and incident electron beam geometries for each of these samples.

TABLE I.

Sample descriptions, contact orientations, and DRCLS geometries.

Sample No.DopingDegradationContact thickness (nm)Anode (A) cathode (C) geometriesDRCLS beam direction
None Post ∼80 Square planar Through A, C 
None Pre ∼80 Square planar Through A, C 
None Post ∼20 Square planar Through A, C 
None Post … Rectangular lateral Across A to C 
Fe Post … Rectangular lateral Across A to C 
Sample No.DopingDegradationContact thickness (nm)Anode (A) cathode (C) geometriesDRCLS beam direction
None Post ∼80 Square planar Through A, C 
None Pre ∼80 Square planar Through A, C 
None Post ∼20 Square planar Through A, C 
None Post … Rectangular lateral Across A to C 
Fe Post … Rectangular lateral Across A to C 

DRCLS measurements employed a 0.3 mm diameter glancing electron beam in ultrahigh vacuum (UHV) with energies EB ranging from 0.5 to 5 keV incident on the samples. In this range, electron beams generate electron cascades inside the solid that lose energy initially by plasmon loss and subsequently by impact ionization—the creation of electron–hole (e–h) pairs. Recombination of these free carriers occurs by conduction-to-valence band and gap state-to-band edge electronic transitions, resulting in cathodoluminescence spectra (CL) measured using monochromators and photodetectors.41 Monte Carlo simulations (supplementary material) provide rates of e–h pair creation and corresponding maximum excitation depths, i.e., Bohr–Bethe ranges RB ranging from 26 to 166 nm through a 20 nm Pt overlayer into the SrTiO3 for EB = 2–5 keV, respectively.42 Differential DRCLS (DDRCLS) analysis43,44 further refined the excitation depths probed for electron penetrations at different energies through the Pt overlayer (see the supplementary material).

Previous studies show that resistance degradation of Fe-doped SrTiO3 by applied electric fields at elevated temperature is accompanied by the formation of a color front with a transition from dark brown at the anode to transparent at the cathode. Electron paramagnetic resonance (EPR) and optical absorption measurements showed that Fe3+ ions in SrTiO3 change to Fe4+ in oxidized SrTiO3 with lower oxygen vacancy (VO) concentrations and the reduced local charge compensation that results. This change from Fe3+ to Fe4+ is responsible for the coloration in degraded Fe-doped SrTiO3 at the anode,12,28,45,46

To illustrate how DRCL spectra change with defect movement under an applied dc electric field, Fig. 1 shows the DRCLS spectra for Sample 5, the Fe-doped SrTiO3 sample, with laterally spaced electrodes at three discrete spots between the cathode (clear) and the anode (dark). The photo inset in Fig. 1(c) shows the color change due to defect migration and electrical degradation. Spot 1 spectra near cathode in Fig. 1(a) display peak features at energies attributed to 1.9 eV Sr vacancy (VSr),47,48 2.3 eV and 2.9 eV oxygen vacancy-related (VO-R) defect-related,47,49 as well as the 3.2 eV indirect gap, and 3.6 eV direct bandgap features.50 Our previous work shows that 1.8 eV and 2.3 eV features in (Ba,Sr)TiO3 are consistent with Ba vacancies based on close agreement with previous theoretical calculations51 and cathodoluminescence measurements of Koschek and Kubalek.52 Given the structural similarities between SrTiO3 and BaTiO3 and the close peak energy correspondence to the 1.9 eV shoulder reported in this paper, we assign this feature to a Sr vacancy-related defect. Sr vacancies (VSr) are immobile under the experimental conditions used and are distributed in accordance to the crystal growth conditions. However, as both VO and VSr defects have opposite charge, VSr may permit formation of an associated defect complex. Features above 3.6 eV correspond to higher lying conduction band to valence band transitions.50 The Spot 2 spectrum in Fig. 1(b) at the boundary between clear and dark shows a significant decrease in all three defect features relative to the higher energy bulk conduction band peaks, and these features are further reduced in the Spot 3 spectrum near the anode in Fig. 1(c). This trend indicates a systematic increase in defect intensities relative to their 3.6 eV bandgap intensities from the anode to the cathode. Total intensity variations at different EB are due to minor changes in data collection times.

FIG. 1.

DRCL spectra of degraded Fe doped SrTiO3 Sample 5 with laterally separated cathode (left) and anode (right). Photo inset illustrates positions of (a) spot 1 near clear cathode, (b) spot 2 at clear/dark interface, and (c) spot 3 near dark anode. Differences in visual appearance of SrTiO3 following electrical degradation correlate with changes in specific defect transitions as well as the impedance spectroscopy (see Figs. S4 and S5 in the supplementary material).

FIG. 1.

DRCL spectra of degraded Fe doped SrTiO3 Sample 5 with laterally separated cathode (left) and anode (right). Photo inset illustrates positions of (a) spot 1 near clear cathode, (b) spot 2 at clear/dark interface, and (c) spot 3 near dark anode. Differences in visual appearance of SrTiO3 following electrical degradation correlate with changes in specific defect transitions as well as the impedance spectroscopy (see Figs. S4 and S5 in the supplementary material).

Close modal

The systematic decrease in native point defects from the cathode to the anode provides direct, spatially resolved information on the density and nature of point defect complexes in Fe-doped SrTiO3 after electromigration of the oxygen vacancies. Based on a previous DRCLS (Ba,Sr)TiO3 calibration with positron annihilation spectroscopy (PAS),47 a 10:1 I(2.9 eV)/I(3.6 eV) peak height ratio corresponds to ∼1018 cm−3 so that the 1.2 ratio in Fig. 1(a) would correspond to ∼2 × 1017 cm−3, within an order of magnitude of free carrier densities extracted from modulus spectroscopy measurements, notwithstanding differences between (Ba,Sr)TiO3 and SrTiO3.26 These spectra illustrate how differences in visual appearance of degraded Fe-doped SrTiO3 following electrical stress correlate with orders of magnitude changes in specific defects that result from defect migration on a mm scale. These results directly complement previous studies of defect migration based on electrical measurements and modeling.8,25,26

Since the activation energy at the cathode is ∼1.0 eV in the Fe-doped case,26 then the dominant states with luminescence transitions in Fig. 1 at 2.31 and 2.88 eV correspond to states above midgap that add electron density, moving Fermi level EF closer to the conduction band EC as confirmed by earlier x-ray photoemission measurements of EF above midgap to be discussed below. At the anode, these transitions are less intense relative to bulk conduction band features as expected since the Fe dopants would be more oxidized and have lower oxygen vacancy density. Figure 1(c) inset shows the electrocoloration at the anode of these Fe-doped crystals. However, density function theory (DFT) calculations indicate Fe4+ states near the anode that are responsible for the acceptor levels and p-type behavior to be 0.48 eV above the valence band,31 well below the 1.2 eV cut-off energy of our Si-based CCD.

Figure 2 shows DRCL spectra for Sample 4—undoped, degraded SrTiO3 in the same geometry of electric field stress as in Fig. 1, obtained at EB = 5 keV at five spots between the anode and the cathode across a 5 × 5 mm2 area. These spectra exhibit similar 2.6 eV and 2.9–3.0 eV features associated with oxygen vacancy movements47,48 between the anode and the cathode after electrical stress at elevated temperature. Both features increase steadily from the anode to the cathode, indicating the direction of in-plane Vo-R migration during the degradation process. Deconvolved peak areas for each spectrum at each spot normalized by the bandgap bulk feature (supplementary material) distinguish and quantify the magnitude of each peak. With degradation, the near-cathode 2.6 eV (3.0 eV) peak area normalized by 3.6 eV near band edge (NBE) peak area increases by 6× (3×) from the anode to the cathode. These oxygen vacancy-related increases are consistent with the electron density increase near the cathode that contributes to the overall conductivity increase with electric field-induced vacancy-related diffusion.

FIG. 2.

Spatially resolved 5 keV DRCL spectra at distances shown between the cathode to the anode on the degraded undoped SrTiO3 sample (Sample 4) with laterally spaced contacts. Both 2.6 eV and 3.0 eV Vo-related defects exhibit systematic increases from the anode to the cathode.

FIG. 2.

Spatially resolved 5 keV DRCL spectra at distances shown between the cathode to the anode on the degraded undoped SrTiO3 sample (Sample 4) with laterally spaced contacts. Both 2.6 eV and 3.0 eV Vo-related defects exhibit systematic increases from the anode to the cathode.

Close modal

DRCLS measurements inside the undoped SrTiO3 under the electrodes in a planar bias configuration further confirm the increase in defects from the anode to the cathode after degradation. Monte Carlo simulations (supplementary material) show that incident beam energies EB of 2–5 keV can penetrate the 20 nm thick Pt planar electrodes with Bohr–Bethe maximum ranges RB from ∼25 to 166 nm below the interface, respectively. DRCLS spectra in Fig. 3 show much lower intensity NBE-normalized 2.6/3.0 eV Vo-related defect features at all EB into the anode in Fig. 3(a) vs into the cathode in Fig. 3(b), consistent with Figs. 1 and 2.

FIG. 3.

DRCL spectra of degraded undoped SrTiO3 (Sample 3) with 20 nm Pt contacts on top/bottom surfaces through positive anode (a) and negative charged cathode (b) contacts. 2.6/3.0 eV defect features beneath the anode are much lower than through the cathode.

FIG. 3.

DRCL spectra of degraded undoped SrTiO3 (Sample 3) with 20 nm Pt contacts on top/bottom surfaces through positive anode (a) and negative charged cathode (b) contacts. 2.6/3.0 eV defect features beneath the anode are much lower than through the cathode.

Close modal

Figure 4 shows depth profiles for both 2.6/3.0 eV Vo-R deconvolved peak areas normalized by NBE area features vs incident beam energies that illustrate the higher cathode vs anode defect intensities at maximum (∼166 nm) excitation depth. These depth profiles on a scale of tens of nanometers reveal an additional aspect of the Pt cathode–SrTiO3 interface, namely, dramatic decreases of both defects between ∼166 nm and the metal cathode interface but not at the metal anode interface, which is depleted of VO-R defects. Such large defect variations at metal–semiconductor junctions within depths comparable to space charge regions appear at other metal–semiconductor junctions and may be associated with local band bending,53 interface chemical interactions,54 or effects of crystal orientation.55 Significantly, the depth over which the NBE-normalized 2.6/3.0 eV Vo-R peak areas decreases from the bulk to the Pt interface is comparable to the depletion width

d=[2ε(V0V)/qN]1/2
(1)

at the Pt/SrTiO3 interface. Based on previously measured temperature-dependent dielectric permittivity ɛ ∼ 250 ɛ0 and carrier density n = 3–3.5 × 1018 cm−3 on similar undoped SrTiO3 samples,26 the depletion width d for a Pt/SrTiO3 Schottky barrier height V0 = 0.8 eV56 at V = 0 applied bias V using Eq. (1) equals 86 nm. Figure S3 in the supplementary material shows DDRCLS distributions at representative EB. With the Monte Carlo distributions of EB = 5.0 keV and 4.5 keV shown in Fig. S1 in the supplementary material and normalized at the Pt/SrTiO3 interface, their differential Monte Carlo distribution peaks at U0' = 77 nm, decreasing with increasing depth to a 1/e value at 135 nm. For all cathode data points shown in Fig. 4, U0' values for EB = 2.5, 3.0, 3.5, 4.0, 4.5, and 5.0 keV are 25, 30, 36, 56, 71, and 77 nm, respectively, with electron–hole pair excitation rates decreasing to 1/e at 38, 53, 68, 83, 132, and 135 nm depths, respectively. Monte Carlo distributions for incident electron beams on bare surfaces are narrower in depth since EB can be lower for the same depth range without a Pt overlayer. Electrical impedance measurements indicate pronounced depletion regions at acceptor-doped SrTiO3 grain boundaries57,58 with highly resistive layers of 84 nm for Ni-doped SrTiO3.57 

FIG. 4.

(a) 2.6 eV and (b) 3.0 eV peak areas vs depth through outer 20 nm Pt contacts and ∼166 nm layers of the cathode and the anode for degraded, undoped SrTiO3 (Sample 3) on top/bottom surfaces. Dashed arrows illustrate direction of VO migration during degradation, causing an overall increase toward the cathode side. Starting ∼146 nm away from cathode Pt/STO interface, DDRCL spectra (see Fig. S4 in the supplementary material) show VO-related defects decreasing dramatically toward the Pt cathode. (c) 3.0 eV peak areas vs depth expanded to show 1/e U0' Monte Carlo maximum depths.

FIG. 4.

(a) 2.6 eV and (b) 3.0 eV peak areas vs depth through outer 20 nm Pt contacts and ∼166 nm layers of the cathode and the anode for degraded, undoped SrTiO3 (Sample 3) on top/bottom surfaces. Dashed arrows illustrate direction of VO migration during degradation, causing an overall increase toward the cathode side. Starting ∼146 nm away from cathode Pt/STO interface, DDRCL spectra (see Fig. S4 in the supplementary material) show VO-related defects decreasing dramatically toward the Pt cathode. (c) 3.0 eV peak areas vs depth expanded to show 1/e U0' Monte Carlo maximum depths.

Close modal

The direct observation of oxygen vacancy-related defect depletion in the Pt–SrTiO3 surface space charge region reported here provides support for thermodynamic modeling predictions of De Souza18 They indicate that space-charge formation at the TiO2-terminated (100) surface is driven by the Gibbs formation energy of charged oxygen vacancies at the interface being lower than in the bulk.18,59,60 The spectra and depth profiles shown in Figs. 3 and 4, respectively, demonstrate the ability of spatially localized DRCLS to distinguish local defect variations on a tens of nanometer-scale from macroscopic defect variations on a scale of millimeters. Those systematic changes on a macro-scale provide direct physical evidence for the migration of oxygen vacancy defects from the anode to the cathode that confirm the proposed mechanism of resistance degradation based on electrical modulus measurements.

In addition to spatial measurements of defect densities between the anode and the cathode, DRCLS peak positions can provide energy levels associated with defect complexes that correspond directly to intraband and activation energies before and after electric field degradation, as well as the electromigration of oxygen vacancies and possible electron injection and trapping at the interfaces. This is because the spectral features correspond to optical transitions between defect levels in the bandgap and the SrTiO3 band edges. The energy separation of a defect level relative to the band edges represents an electronic activation energy for trapped charge to reach a band edge and conduct. For SrTiO3, the basic direct bandgap is 3.62 eV, and the indirect gap is 3.25 eV.

If we consider the most general case for a mixed conductor with both high mobilities and concentrations of ionic and electronic carriers, the overall activation energy is a complex function that depends on the mobility and concentration of each carrier as well as their temperature dependence. For example, the partial conductivity arising from a carrier i of charge qi with temperature-dependent concentration ci and mobility μi is σi = qiciμi. In a material exhibiting mixed conduction, the total conductivity is given by σtot = ∑σi.

The temperature dependence of electronic conductivity can be expressed as

σ=σ0exp(EA/kBT),
(2)

where μ0 is a pre-exponential factor, kB is the Boltzmann constant, T is the temperature, and EA is the activation energy determined from excitation from the trapped electronic states.26 In oxides with various stoichiometric and non-stoichiometric defects and unintentional background impurities, there are several other factors that can determine the positions of electronic states in the bandgap. Besides electrons and holes, the two forms of electronic carrier within the band theory, other forms of electronic conduction include polaron hopping and its variants. The ionic diffusion is also described by a thermally activated diffusion model where the ionic diffusion between the anode and the cathode can be described by a thermal activation energy according to20 

μ(VOR)(T)=μ0exp(EA/kBT),
(3)

where μ0 is a pre-exponential mobility factor in an otherwise identical expression to Eq. (2). For Fe-doped SrTiO3, this ionic activation energy has been of considerable research interest and numerous studies report values ranging from 0.59 to 0.78 eV.20 For oxygen vacancies, enthalpy of vacancy migration in SrTiO3 can be used to describe an activation barrier of ≈0.6 eV for an ion-hopping process monitored chemically via tracer diffusion.59,60 The conductivity determined by impedance spectroscopy in the pristine crystals indicates an activation energy of ∼0.6 eV in undoped and doped SrTiO3,26 consistent with the expected ionic dominated conduction, as shown in Figs. S5(b) and S6(a) in the supplementary material. Under degradation in the undoped SrTiO3, there is an activation energy of ∼0.64 eV, indicating that this is still ionically dominated, in contrast to the degraded doped SrTiO3 changes with the conduction then being controlled by two regions with electronically controlled conduction with ∼1.0 eV activation energies.

Figure 5 shows the comparison of 5 keV CL spectra through 80 nm of Pt planar contact of (a) pristine, undoped SrTiO3 (Sample 2) vs (b) degraded, undoped SrTiO3 (Sample 1). Optical features common to both spectra correlate with several SrTiO3 native point defects including 1.6 eV Ti3+46 (a missing oxygen atom in the 6 O2− octahedron surrounding a Ti4+ ion61) and 1.9 eV VSr46,47 defect features. Dominant in both spectra are broad 2.55 and 2.65 eV features before and after degradation, respectively, and, therefore, the clear 0.1 eV shift to higher energy with degradation correlates to a higher n-type contribution to the electronic conductivity and the overall mixed ionic conductivity.

FIG. 5.

EB = 5 keV cathodoluminescence spectra of (a) undoped pre-bias (Sample 2) and (b) undoped degraded (Sample 1) SrTiO3 through 80 nm contacts on top/bottom surfaces with EB = 5 keV. Colored lines indicate deconvolved optical features. The dominant 2.55 eV feature before degradation shifts from 2.55 eV to 2.65 eV after degradation. Activation energies of conductivity that is ionically controlled and shifts from 0.70 to 0.64 eV measured for these samples before (after) degradation due to mixed ionic diffusion and electronic thermal activation.

FIG. 5.

EB = 5 keV cathodoluminescence spectra of (a) undoped pre-bias (Sample 2) and (b) undoped degraded (Sample 1) SrTiO3 through 80 nm contacts on top/bottom surfaces with EB = 5 keV. Colored lines indicate deconvolved optical features. The dominant 2.55 eV feature before degradation shifts from 2.55 eV to 2.65 eV after degradation. Activation energies of conductivity that is ionically controlled and shifts from 0.70 to 0.64 eV measured for these samples before (after) degradation due to mixed ionic diffusion and electronic thermal activation.

Close modal

Optical and electron paramagnetic resonance (EPR) studies of Fe-doped SrTiO3 report 425 nm (2.91 eV) and 585 nm (2.12 eV) EPR absorption bands associated with Fe3+ as well as a 500 nm (2.48 eV) band associated with above bandgap photons.12 DRCL spectra commonly observe all three such bands, suggesting that all three involve nearest neighbor oxygen vacancy complexes with the Fe.49 

In Fig. 5, the shift in peak energy from 2.55 eV to 2.65 eV with electrical stress degradation can be interpreted as a change in complex formation of oxygen vacancies with foreign acceptor impurities vs oxygen vacancies dissociated from those impurities. Since oxygen vacancies are relatively mobile but locally are interacting with negative point defects within the lattice, they can form complexes after dissociating, enabling new VO complexes to form. These may not necessarily be at the nearest neighbor positions but rather next nearest neighbors as deduced and modeled with previous studies20,62 Our previous studies of very high purity SrTiO3 grown by molecular beam epitaxy (MBE) also exhibited VO-related DRCLS features at 2.65 eV and 2.9–3.0 eV with similar intensity variations with oxygen growth conditions and subsequent oxidation.49 Therefore, the shift to 2.65 eV suggests the formation of VO-related defects such as isolated VO and VO clusters after impurity dissociation and new VO complex formation.

Figure 6 shows schematic diagrams that illustrate (a) energy level positions in the SrTiO3 bandgap corresponding to the dominant 2.55 and 2.65 eV optical transitions in Fig. 5, (b) reported energy levels for Fe -doped SrTiO3 from ESR,63 and (c) EF positions of multiple (Ba,Sr)TiO3 specimen near the conduction band EC measured by x-ray photoemission (XPS).64 The shift in Fig. 6(a) optical transition energies from 2.55 eV to 2.65 eV CL under degradation is consistent with lower electron activity energy and with a higher EF − EV at the cathode as the oxygen vacancy concentration builds under accumulation with the electromigration. The energy levels in Fig. 6(a) do not coincide with any Fe complex in Fig. 6(b), the Fe3+ at 2.92 eV above EV being the closest. The VO-related energy levels positioned above midgap are consistent with EF always stabilizing above midgap. In Fig. 6(c), x-ray photoemission spectroscopy (XPS) measurements of EF at (Ba,Sr)TiO3 surfaces lie within a band of energies above midgap, regardless of donor vs acceptor doping and synthesis route.65 Furthermore, XPS measurements of SrTiO3 display EF shifts at the cathode from EF − EV = 2.2 eV up to 2.7 eV with degradation, close to the 2.65 eV level after degradation in Fig. 6(a).66 

FIG. 6.

Schematic illustrations of (a) dominant defect transitions measured by CL in Fig. 5 before and after electrical degradation, respectively, from energy levels in the SrTiO3 bandgap corresponding to the 2.55 eV–2.65 eV shift in energy and relative shift of the Vo-R complex under electrical degradation, (b) the dominant defect states for Fe complexes in SrTiO3 (after Morin and Oliver63), and (c) reproduction of the experimentally determined Fermi level positions in donor and acceptor-doped (Ba,Sr)TiO3 (after Klein64) and the SrTiO3 Fermi level movement with electrical degradation (after Giesecke et al.65).

FIG. 6.

Schematic illustrations of (a) dominant defect transitions measured by CL in Fig. 5 before and after electrical degradation, respectively, from energy levels in the SrTiO3 bandgap corresponding to the 2.55 eV–2.65 eV shift in energy and relative shift of the Vo-R complex under electrical degradation, (b) the dominant defect states for Fe complexes in SrTiO3 (after Morin and Oliver63), and (c) reproduction of the experimentally determined Fermi level positions in donor and acceptor-doped (Ba,Sr)TiO3 (after Klein64) and the SrTiO3 Fermi level movement with electrical degradation (after Giesecke et al.65).

Close modal

Further microscopic studies are needed to test these proposed correlations. DRCLS spatial data would be highly useful especially if relative activation energies, distributions, the relative concentrations, and the local atomic/electronic structure and nature of the emissions can be determined. As pointed out by Long et al,31 it is the interfacial regions where there is crossovers between ionic and electronic controlled conduction that have the highest activation energies. These results suggest that DRCLS could prove to be highly effective in exploring the science of important processes such as time-dependent breakdown of dielectric materials to further aid the identification and quantification of such effects, and to determine the electronic states and their complexes with oxygen vacancies and other point defects. Extensive DFT approaches would need to be developed to consider these defects complexes and their emission spectra under electron excitations,66,67 Also required would be detailed methodologies to systematically understand the various configurations and their excited states.

In summary, we used spatially and depth-resolved cathodoluminescence spectroscopy to study the behavior of defects before and after electric field-stressed degradation. This work focused on oxygen vacancy-related defect complexes in undoped and Fe-doped SrTiO3 with high temperature electric field stress in three dimensions on a nanometer scale, with respect to spatial defect distributions and energy levels associated with specific defect complexes. Consistent with impedance studies, DRCLS provided direct and spatially resolved observation of oxygen vacancy migration driven by external electric fields from the anode to the cathode in undoped SrTiO3 between laterally separated electrodes, which result in local carrier compensation and conductivity increase. DRCLS measurements through planar Pt electrodes after high temperature electric field cycling confirm oxygen vacancy movement from the anode to the cathode together with a strong decrease in defect densities within 100–150 nm of the Pt/STO interface. In addition, a 0.1 eV spectral shift in the dominant 2.55 eV energy level is observed, consistent with a change in oxygen vacancy defect configuration during migration, i.e., a 0.06 eV decrease in free carrier activation energy, and the increase of conductivity during the resistance degradation process. These spatially resolved measurements of defect complex provide direct evidence for the defect nature and movement responsible for the well-known resistance degradation in SrTiO3 and may be applicable to other oxide semiconductors as well.

See the supplementary material for Monte Carlo distributions of e–h pair creation vs depth and incident beam energy, and deconvolved, differential Monte Carlo distributions through a Pt cathode/SrTiO3 interface, NBE-normalized peak areas vs lateral position of undoped SrTiO3, and impedance spectroscopy discussion of basic conductivity in doped and undoped SrTiO3 with degradation.

The authors gratefully acknowledge financial support for this work provided by AFOSR under Grant Nos. FA9550-18-1-0066 (H.G. and L.J.B.) and FA9550-14-1-0067 (S.S. and C.A.R.) (Program Manager: Ali Sayir).

1.
C.
Hou
,
W.
Huang
,
W.
Zhao
,
D.
Zhang
,
Y.
Yin
, and
X.
Li
,
ACS Appl. Mater. Interfaces
9
,
20484
(
2017
).
2.
S.
Nishigaki
,
J.
Nakamura
,
K.
Murano
, and
K.
Tatsuki
,
Jpn. J. Appl. Phys.
22
,
87
(
1983
).
3.
T.
Correia
,
M.
Stewart
,
A.
Ellmore
, and
K.
Albertsen
,
Adv. Eng. Mater.
19
,
1700019
(
2017
).
4.
T. R. N.
Kutty
and
S.
Philip
,
Mat. Sci. Eng.
33
,
58
(
1995
).
5.
A. B.
Kozyrev
,
T. B.
Samoilova
,
A. A.
Golovkov
,
E. K.
Hollmann
,
D. A.
Kalinikos
,
V. E.
Loginov
,
A. M.
Pruden
,
I. O.
Soldatenkov
,
D.
Galt
,
C. H.
Mueller
,
T. V.
Rivkin
, and
G. A.
Koepf
,
J. Appl. Phys.
84
,
3326
(
1998
).
6.
S. W.
Kim
,
H. I.
Chai
,
M. H.
Lee
,
J. S.
Park
,
D. J.
Kim
,
D.
Do
,
M. H.
Kim
,
T. K.
Song
, and
W. J.
Kim
,
Ceram. Int.
39
,
S487
(
2013
).
7.
S.
Zhao
,
S. J.
Zhang
,
W.
Liu
,
N. J.
Donnelly
,
Z.
Xu
, and
C. A.
Randall
,
J. Appl. Phys.
105
,
053705
(
2009
).
8.
R.
Waser
,
T.
Baiatu
, and
K.-H.
Härdtl
,
J. Am. Ceram. Soc.
73
,
1654
(
1990
).
9.
Y. A.
Genenko
,
J.
Glaum
,
M. J.
Hoffmann
, and
K.
Albe
,
Mater. Sci. Eng. B
192
,
52
(
2015
).
10.
J.
Watson
and
G.
Castro
,
J. Mater. Sci. Mater. Electron.
26
,
9226
(
2015
).
11.
H.
Chazono
and
H.
Kishi
,
Jpn. J. Appl. Phys.
40
,
5624
(
2001
).
12.
B. W.
Faughnan
,
Phys. Rev. B
4
,
3623
(
1971
).
13.
C.
Mitra
,
C.
Lin
,
J.
Robertson
, and
A. A.
Demkov
,
Phys. Rev. B
86
,
155105
(
2012
).
14.
A.
Janotti
,
J. B.
Varley
,
M.
Choi
, and
C. G.
Van de Walle
,
Phys. Rev. B
90
,
085202
(
2014
).
15.
R.
Moos
and
K. H.
Härdtl
,
J. Am. Ceram. Soc.
80
,
2549
(
1997
).
16.
S.
Rodewald
,
J.
Fleig
, and
J.
Maier
,
J. Am. Ceram. Soc.
83
,
1969
(
2000
).
17.
Y.-Y.
Pai
,
A.
Tylan-Tyler
,
P.
Irvin
, and
J.
Levy
,
Rep. Prog. Phys.
81
,
036503
(
2018
).
18.
R. A.
De Souza
,
Adv. Funct. Mat.
25
,
6326
(
2015
).
19.
R.
Merkle
and
J.
Maier
,
Angew. Chem. Int. Ed.
47
,
3874
(
2008
).
20.
R. A.
Maier
and
C. A.
Randall
,
J. Am. Ceram. Soc.
99
,
3350
(
2016
).
21.
J.
Hanzig
,
M.
Zschornak
,
E.
Mehner
,
F.
Hanzig
,
W.
Münschgesang
,
T.
Leisegang
,
H.
Stöcker
, and
D. C.
Meyer
,
J. Phys. Condens. Matter
28
,
225001
(
2016
).
22.
Z. Q.
Liu
,
D. P.
Leusink
,
Z.
Wang
,
W. M.
,
K.
Gopinadhan
,
A.
Annadi
,
Y. I.
Zhao
,
X. H.
Huang
,
S. W.
Zeng
,
Z.
Huang
,
A.
Srivastava
,
S.
Dhar
,
T.
Venkatesan
, and
Ariando
,
Phys. Rev. Lett.
107
,
146802
(
2011
).
23.
G. Y.
Yang
,
E. C.
Dickey
,
C. A.
Randall
,
D. E.
Barber
,
P.
Pinceloup
,
M. A.
Henderson
,
R. A.
Hill
,
J. J.
Beeson
, and
D. J.
Skamser
,
J. Appl. Phys.
96
,
7492
(
2004
).
24.
C. A.
Randall
,
R.
Maier
,
W.
Qu
,
K.
Kobayashi
,
K.
Morita
,
Y.
Mizuno
,
N.
Inoue
, and
T.
Oguni
,
J. Appl. Phys.
113
,
014101
(
2013
).
25.
A.
Moballegh
and
E. C.
Dickey
,
Acta. Mater.
86
,
352
(
2015
).
26.
T. J. M.
Bayer
,
J. J.
Carter
,
J.-J.
Wang
,
A.
Klein
,
L.-Q.
Chen
, and
C. A.
Randall
,
J. Appl. Phys.
122
,
244101
(
2017
).
27.
J.-J.
Wang
,
H.-B.
Huang
,
T. J. M.
Bayer
,
A.
Moballegh
,
Y.
Cao
,
A.
Klein
,
E. C.
Dickey
,
D. L.
Irving
,
C. A.
Randall
, and
L.-Q.
Chen
,
Acta Mater.
108
,
229
(
2016
).
28.
T.
Baiatu
,
R.
Waser
, and
K.-H.
Härdtl
,
J. Am. Ceram. Soc.
73
,
1663
(
1990
).
29.
J.
Blanc
and
D. L.
Staebler
,
Phys. Rev. B
4
,
3548
(
1971
).
30.
R.
Waser
,
T.
Bieger
, and
J.
Maier
,
Sol. State Commun.
76
,
1077
(
1990
).
31.
D. M.
Long
,
B.
Cai
,
J. N.
Baker
,
P. C.
Bowes
,
T. J. M.
Bayer
,
J.-J.
Wang
,
R.
Wang
,
L.-Q.
Chen
,
C. A.
Randall
,
D. L.
Irving
, and
E. C.
Dickey
,
J. Am. Ceram. Soc.
102
,
3567
(
2019
).
32.
B. G.
Yacobi
and
D. B.
Holt
,
Cathodoluminescence Microscopy of Inorganic Solids
(
Plenum
,
New York
,
1990
).
33.
D.
Briggs
and
M. P.
Seah
, in
1990 Practical Surface Analysis, Vol 1, Auger and X-ray Photoelectron Spectroscopy
, edited by
D.
Briggs
and
M. P.
Seah
(
Wiley
,
New York
,
1990
), p.
5
.
34.
L. J.
Brillson
,
Surfaces and Interfaces of Electronic Materials
(
Wiley-VCH
,
Weinheim
,
2010
), Chap. 9.
35.
L. J.
Brillson
,
J. Phys. D Appl. Phys.
45
,
183001
(
2012
).
36.
A.
Rose
,
RCA Rev.
27
,
600
(
1966
).
37.
C. A.
Klein
,
J. Appl. Phys.
39
,
2029
(
1968
).
38.
M. P.
Seah
and
W. P.
Densch
,
Surf. Interface Anal.
1
,
2
(
1979
).
39.
C. J.
Powell
and
A.
Jablonski
,
Nucl. Instrum. Methods Phys. Rev. A
601
,
54
(
2009
).
40.
D.
Drouin
,
A.
Real
,
D.
Joly
,
X.
Tastet
,
V.
Aimez
, and
R.
Gauvin
,
Scanning
29
,
92
(
2007
).
41.
L. J.
Brillson
,
R. E.
Viturro
,
J. L.
Shaw
, and
H. W.
Richter
,
J. Vac. Sci. Technol. A
6
,
1437
(
1988
).
42.
P.
Hovington
,
D.
Drouin
, and
R.
Gauvin
,
Scanning
19
,
1
(
1997
).
43.
T. J.
Asel
,
H.
Gao
,
T. J.
Heinl
,
D.
Adkins
,
P. M.
Woodward
,
J.
Hoffman
,
A.
Bhattacharya
, and
L. J.
Brillson
,
J. Vac. Sci. Technol. B
33
,
04E103
(
2015
).
44.
S.
Shen
,
Y.
Liu
,
R.
Gordon
, and
L. J.
Brillson
,
Appl. Phys. Lett.
98
,
172902
(
2011
).
45.
M.
Fujimoto
and
W. D.
Kingery
,
J. Am. Ceram. Soc.
68
,
169
(
1985
).
46.
C.
Lenser
,
A.
Kalinko
,
A.
Kuzmin
,
D.
Berzins
,
J.
Purans
,
K.
Szot
,
R.
Waser
, and
R.
Dittmann
,
Phys. Chem. Chem. Phys.
13
,
20779
(
2011
).
47.
M. M.
Rutkowski
,
K.
McNicholas
,
Z.
Zeng
,
F.
Tuomisto
, and
L. J.
Brillson
,
J. Phys. D Appl. Phys.
47
,
255303
(
2014
).
48.
D.
Lee
,
H.
Wang
,
B. A.
Noesges
,
T.
Asel
,
J.
Pan
,
J.-W.
Lee
,
Q.
Yan
,
L.
Brillson
,
X.
Wu
, and
C.-B.
Eom
,
Phys. Rev. Mater.
2
,
060403(R)
(
2018
).
49.
J.
Zhang
,
S.
Walsh
,
C.
Brooks
,
D. G.
Schlom
, and
L. J.
Brillson
,
J. Vac. Sci. Technol. B
26
,
1466
(
2008
).
50.
K.
van Benthem
,
C.
Elsässer
, and
R. H.
French
,
J. Appl. Phys.
90
,
6156
(
2001
).
51.
J.
Daniels
and
K. H.
Härdtl
,
Philips Res. Rep.
31
,
489
(
1976
).
52.
G.
Koschek
and
E.
Kubalek
,
Ferroelectrics
68
,
293
(
1986
).
53.
G. M.
Foster
,
H.
Gao
,
G.
Mackessy
,
A. M.
Hyland
,
M. W.
Allen
,
B.
Wang
,
D. C.
Look
, and
L. J.
Brillson
,
Appl. Phys. Lett.
111
,
101604
(
2017
).
54.
L. J.
Brillson
,
J. Vac. Sci. Technol.
16
,
1137
(
1979
).
55.
S.
Balaz
,
J.
Singh
, and
L. J.
Brillson
, “Orientation-dependent defect segregation and dipole formation at SrTiO3 surfaces” (unpublished).
56.
N.
Ohashi
,
H.
Yoshikawa
,
Y.
Yamashita
,
S.
Ueda
,
J.
Li
,
H.
Okushi
,
K.
Kobayashi
, and
H.
Haneda
,
Appl. Phys. Lett.
101
,
251911
(
2012
).
57.
R.
Waser
,
Solid State Ionics
75
,
89
(
1995
).
58.
I.
Denk
,
J.
Claus
and
J.
Maier
,
J. Electrochem. Soc.
144
,
3526
(
1997
).
59.
R. A.
De Souza
,
V.
Metlenko
,
D.
Park
, and
T. E.
Weirich
,
Phys. Rev. B
85
,
174109
(
2012
).
60.
R. A.
De Souza
and
M.
Martin
,
Phys. Chem. Chem. Phys.
10
,
2356
(
2008
).
61.
R.
Plugaru
,
A.
Cremades
, and
J.
Piqueras
,
J. Phys. Condens. Matter
16
,
S261
(
2004
).
62.
M.
Schie
,
R.
Waser
, and
R. A.
De Souza
,
J. Phys. Chem. C,
118
,
15185
(
2014
).
63.
F. J.
Morin
and
J. R.
Oliver
,
Phys. Rev. B
8
,
5847
(
1973
).
64.
A.
Klein
,
J. Am. Ceram. Soc.
99
,
369
(
2016
).
65.
R.
Giesecke
,
R.
Hertwig
,
T. H. M.
Bayer
,
C. A.
Randall
, and
A.
Klein
,
J. Am. Ceram. Soc.
100
,
4590
(
2017
).
66.
F.
Furche
and
R.
Ahlrichs
,
J. Chem. Phys.
117
,
7433
(
2002
).
67.
P.
Rinke
,
A.
Janoffi
,
M.
Scheffler
, and
C. G.
van de Walle
,
Phys. Rev. Lett.
102
,
026402
(
2009
).

Supplementary Material