Electrical Properties, Deep Levels and Luminescence Related to Fe in Bulk Semi-Insulating β-Ga₂O₃ Doped with Fe

Measurements between the top and back ohmic contacts showed that the contacts were reasonably ohmic and the temperature dependence of the current at fixed high voltage (50V) showed activation energy of 0.85 eV, which should be close to the depth of the trap level pinning the Fermi level. For Schottky diodes, rectification appeared starting at 350K- at lower temperatures the high resistance of the quasi-neutral region overpowered the impact of the Schottky barrier height. For high temperatures, the rectification pointed to the prevalence of n-type conductivity (Fig. 1a). The activation energy for the forward current of Schottky diodes measured at high forward voltage of 50V was 0.85 eV, close to the activation energy measured with ohmic contacts. Illumination with high-power 365 nm LED increased the reverse current. The open-circuit voltage was positive, indicating that the sample was high-resistivity n-type (Fig. 1b. The spectral dependence of reverse photocurrent at −10V is shown in Fig. 1c.

Zoom In Zoom Out Reset image size

At temperatures >350K, the series resistance of the Schottky diodes became such that it allowed measurements of capacitance dependence on voltage and of admittance spectra. Fig. 2a shows capacitance versus frequency C-f characteristics of two Schottky diodes measured at −2V at 460K. These characteristics show almost a plateau in capacitance for frequencies up to ∼1 kHz and at these low frequencies meaningful C-V profiling could be attempted. Fig. 2b presents the dark 1/C² versus voltage (V) plots measured in the dark and under illumination with 365 nm LED at 460K. The plots are linear at low reverse voltages, showing slopes corresponding to the concentrations of deep centers filled with electrons and pinning the Fermi level (the change of the space charge width of the diodes with changing applied bias is driven by the change in electron occupation of these centers). The respective concentrations calculated from the slopes at low reverse voltage were 1.4 × 10¹⁷ cm⁻³ for one diode in Fig. 2b and 1.8 × 10¹⁷ cm⁻³ for the other, with the voltage cutoff 1.3–1.5 V. The diodes showed photocapacitance when excited with high-power 365 nm LEDs. Fig. 2c shows the concentration profiles of the charged centers computed from the data in Fig. 2b. The dark profiles are flat whereas the profiles measured under illumination with 365 nm LED (photon energy 3.4 eV) show that the near-surface regions of the diodes were enriched with the deep centers excited by the below-bandgap illumination.

Zoom In Zoom Out Reset image size

The evolution of the temperature dependence of capacitance and AC conductance G measured at different low frequencies (f) display well defined steps in capacitance and corresponding peaks in conductance, as illustrated by Figs. 3a, 3b (the conductance in Fig. 3b is divided by the angular frequency ω = 2πf, as usual in presenting the conductance peaks in admittance spectra.¹⁸ The Arrhenius plots of Ln(1/(τT_M²)) versus 1000/T_M obtained from the shift of the G/ω peak/C step temperature T_M with changing the measurement frequency f are shown in Fig. 3c (ω = 1/(2πf)¹⁸). The activation energy E_a determined from the slope and the electron capture cross sections obtained from the pre-exponential factor¹⁸ were E_a = 0.75–0.82 eV, σ_n = (2–5) × 10⁻¹⁵ cm², close to the parameters of the Fe acceptors in Ga₂O₃.¹⁰ These traps pin the Fermi level, while the concentration measured by C-V profiling in the dark gives the number of electrons occupying the centers (the space charge region width at the given reverse voltage in high-resistivity semi-conductor is determined by the change in occupation of the deep trap that pins the Fermi level in the quasi-neutral region^12,17). As the Fe concentration must be close to 10¹⁸ cm⁻³,¹⁸ a considerable portion of these centers remains free of electrons and can effectively participate in charge capture producing current collapse in Ga₂O₃ based HEMTs.¹⁴ The data presented above indicate that the Fermi level in SI β-Ga₂O₃(Fe) samples is pinned by centers whose energy level is very close to the level of Fe determined in DLTS measurements¹⁰ that give the parameters of the Fe acceptor with the charge transfer level Fe³⁺/Fe²⁺. The concentration of these Fe states filled by electrons, Fe²⁺ states, is estimated to be only ∼15% of the total Fe density, the rest of Fe existing as neutral Fe³⁺ states. This should have profound effects for electrons trapping in β-Ga₂O₃-based HEMTs and FETs with Fe doped substrates. One would expect strong capture of electrons from FET channels by Fe³⁺ in the buffer and consequent current collapse, as demonstrated by actual trapping studies reported for (Al_xGa_1-x)₂O₃/Ga₂O₃(Fe) HEMTs.²⁰ This situation is also conducive to detection of intracenter transitions in Fe³⁺ by PL/MCL spectra measurements.

Zoom In Zoom Out Reset image size

PICTS spectra measured on Schottky diodes at reverse bias of −10V, with excitation by the 365 nm high-power LED were performed. Fig. 4a shows the ΔI versus T spectra, Fig. 4b presents the spectra normalized to I_ph measured simultaneously with PICTS spectra. Measurable photocurrent appeared above 330K. Both spectra showed one prominent peak. Arrhenius plots for both types of spectra are shown in Fig. 4c. The normalized ΔI/I_ph spectra yielded the apparent trap activation energy of Ea = 0.8 eV with the electron capture cross section of σ_n = 2.7 × 10⁻¹⁶ cm², close to the Fe level signature reported for DLTS measurements and observed in admittance spectra in Fig. 3. For the spectra not normalized by photocurrent, the apparent activation energy was 0.95 eV, yielding an activation energy for the μτ product of 0.15 eV coming predominantly from the temperature dependence of τ. PICTS results indicate that Fe acceptors are the most prominent deep centers observed in these semi-insulating β-Ga₂O₃ (Fe) crystals for the probed temperature range. The presence of additional deep traps could be detected by measurements of reverse photocurrent dependence on photon energy displayed in Fig. 1c. The spectrum of photocurrent is dominated by deep traps with optical ionization thresholds of 1.5 eV and 2.3 eV, both of which should be fully occupied in the dark, given the Fermi level pinning near E_c-0.8 eV (see above). The centers with optical ionization threshold near 2.3 eV are common for bulk β-Ga₂O₃ and for epitaxial films grown by HVPE or metallorganic chemical vapor deposition.^21,22

Zoom In Zoom Out Reset image size

Deeper electron traps with optical threshold near 1.5 eV and 2.3 eV are expected to produce their peaks in PICTS at much higher temperatures than used here. These traps should be fully occupied with electrons in the dark and should not be a concern for current collapse in HEMTs, but could present a problem under illumination, particularly in the case of very deep and relatively abundant traps with optical threshold of 2.3 eV (Fig. 1).

The dominant feature of the MCL spectra measured at different temperatures was a broad band peaked near 386 nm. Fig. 5a shows the MCL spectra in the region of this broad band for measurements at 295K and 80K. This feature has been often reported for MCL and PL spectra measurements of variously grown β-Ga₂O₃ crystals^23–28 and is close to the so called UV band of.²³ This UV band could in our case be accurately deconvoluted into two broad Gaussian bands peaked at 3.09 eV and 3.36 eV whose positions were not measurably affected by temperature change from 80K to 295K (Fig. 5a). The slight temperature dependence of the peak positions of both bands implies that they could be due to donor-acceptor transitions. There have been multiple attempts to associate various constituents of the broad UV band with transitions involving deep oxygen donors and deep Ga acceptors,^23–28 but the proposed identifications do not seem unambiguous and need further verification. In addition to the UV band discussed above our MCL spectra also showed a broad band peaked near 1.75 eV (Fig. 5b) that is most likely the second order reflection of the UV band at 386 nm. It is interesting, however, that on the high energy slope of this "parasitic" band one can observe two sharp peaks near 1.78 eV and 1.8 eV that have not been reported before and have to belong to some intracenter transitions judging by their extremely narrow width. Our MCL system does not allow to filter out the parasitic second order reflection and to study in detail these new peaks. To do that we measured PL spectra excited at various temperatures by 325 nm-wavelength (3.8 eV) laser excitation. As shown in Ref. 23 with such excitation the UV band is not excited and consequently there is no second order reflection near 750 nm. The 10 K PL spectrum measured with our He-Cd laser 3.8 eV excitation is shown in Fig. 6a. One can clearly see a very sharp line near 696 nm (1.78 eV) with two sidebands near 713 nm (1.74 eV) and 731 nm (1.7 eV), and a broad band peaked near 845 nm (1.46 eV). As far as the authors are aware neither of these features has been reported previously. The sharp line at 1.78 eV cannot be anything else then some sort of intracenter transition, most likely involving some transition metal impurity. Since it has only been detected in a heavily Fe doped sample and since Fe is by all means the most dominant deep level defect in the studied β-Ga₂O₃(Fe) it stands to perfect reason to associate the 1.78 eV line with intracenter transition in either the Fe²⁺ or Fe³⁺ states. From what we know about the pattern of levels in Fe ions in different III-V and II-VI compounds only transition between excited states in the Fe³⁺ ion produce near-IR PL bands, the Fe²⁺ transitions all lying in the far IR.¹³ The most obvious explanation of the observed 10K PL pattern is that we are dealing with the ⁴T₁→⁶A₁ and its phonon sidebands,¹³ the phonon energies being near 40 meV, close to the energy of optical phonons held responsible for the observed temperature dependence of mobility in n-Ga₂O₃.²⁹ Similar PL bands in GaN, AlN lie near 1.29 eV.¹³ In ZnO with increased ionic contribution to bonding the energy is increased to 1.78 eV in agreement with theoretical considerations that predict the energy to increase with increased ionicity.^13,30,31 In that sense Ga₂O₃ certainly follows suit because its bond is more ionic than in III-Nitrides.² If the attribution above is right one expects that, with increasing temperature the sharp line at 1.78 eV should show at higher temperatures an additional high energy line due to growing occupation of the excited state of the ⁴T_1. The relative intensity of this high energy line of the doublet should increase with temperature.¹³ As demonstrated by Fig. 6b this is indeed happening in our PL spectra. As the temperature increases from 10K to 50K, 100K, 150K an additional high energy line at 1.8 eV emerges and its intensity increases. We have to point out that the possibility of the second sharp line to be related to Fe occupying Ga sites in two different positions, the octahedral and the tetrahedral¹⁰ has been discarded above, firstly, because theory suggests the octahedral site to be energetically more favorable and secondly, because it is hard to understand why, with these two states of Fe, we should observe a doublet rather than a quartet. Of course, much more work including measurements of the observed levels splitting with applied uniaxial stress and magnetic field is necessary to establish the full pattern of excited states in Fe ions in β-Ga₂O₃. But locating the lines that are to be monitored in such experiments is the first necessary step. These studies could also have not only pure scientific interest. In GaN it has been quite convincingly demonstrated that Fe is a very efficient recombination center³² despite the fact that its charge transfer Fe³⁺/Fe²⁺ level near E_c-0.5 eV³³ is far away from midgap and thus should not be particularly suitable for fast capture of both electrons and holes via multiphonon emission model.³⁴ To explain the contradiction the energy recycling via excited states of Fe²⁺and Fe³⁺ has been invoked in Ref. 34 to achieve good agreement with experiment.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The PL band peaked near 850 nm (1.46 eV) is too broad to be related to intracenter transitions. Its origin has yet to be understood. The width of the band suggests either a strong lattice relaxation of the center involved or some sort of donor-acceptor recombination, but more work is needed here.