Introduction—impact of defects on semiconductor interfaces

The nature of semiconductor interfaces with other materials and the impact of their defects on electronic properties has been of major research importance for over 75 years. In particular, the materials research of semiconductor interfaces with metals, insulators and other semiconductors has a rich history in which the Materials Research Society has played a major role. Since its inception fifty years ago, MRS has helped lead the understanding of semiconductor interfaces, driven first by industry’s need to make useful, predictable, and controllable contacts to metals and insulators in computer chips. Even before the invention of the transistor, the physical nature of the semiconductor–metal interface, the mechanisms controlling its charge rectification, and its applications have been of wide interest, generating hundreds of thousands of publications to date. Much early work centered on the materials science of the metal–silicon and silicon dioxide interfaces, particularly the interdiffusion and chemical reactions that could take place at their interfaces [1]. That branch of research addressed the materials interactions that determined the electrical resistivity and thermodynamic stability of those semiconductor interfaces. Nevertheless, technologists were also aware of the presence of electronic traps in and at the interfaces of the dielectrics needed for gate biasing of transistors—a realization described by John Bardeen to explain the reduced effect of transistor biasing [2]. Reflecting the rapidly expanding interest in semiconductor interfaces, MRS provided a platform for sharing research on defects in semiconductors. In the first MRS Proceedings issue published in 1980, defects in semiconductors comprised a major symposium with topics ranging across structural and electronic properties, techniques to measure them, and the effects of processing including ion implantation and radiation damage [3]. Indeed, since then, there have been more than 67 published MRS symposia on interfaces including the measurement and electronic impact of defects. By the late 1950’s, the drive for higher speed and optoelectronics had expanded research interests in semiconductor interface defects from Si to GaAs and other compound semiconductors. This review will first address how native point defects and impurities depend strongly on interface chemical structure, then how defects at nanoscale semiconductor–metal interfaces impact electronic structure, then how combining optical, electronic, and microscopic techniques provides a new direction for defect research at the nano/atomic scale, and finally the challenges and opportunities for new defect research at electronic material interfaces at a deeper level.

Defects and interface chemical structure

With the advent of ultrahigh vacuum (UHV) hardware, growth and analytic techniques to explore semiconductor surfaces and interfaces, researchers worldwide began to focus on the fundamental electronic properties of ideal metal–semiconductor interfaces, free from any intervening impurities [4,5,6]. Using ultraviolet photoemission spectroscopy (UPS) and X-ray photoemission spectroscopy (XPS), researchers worldwide were now able to monitor the Fermi level movements and band bending of semiconductors as metal atoms were added, sub-monolayer by monolayer, to form rectifying, i.e., Schottky barrier, and ohmic contacts [7,8,9]. With electron loss spectroscopy (ELS), Auger electron spectroscopy, and electron dispersive X-ray spectroscopy (EDX), it became possible to show how metal atoms can react with semiconductor surfaces to produce new interfacial species,even near room temperature [10, 13]! Furthermore, as shown in Fig. 1(a), these interactions appeared to follow thermodynamics so that metals that could form more stable metal-anion products exhibited a threshold of electronic behavior, i.e., Schottky barrier formation, between reactive vs unreactive metal—semiconductor interfaces.10.

Figure 1
figure 1

(a) Transition in Schottky barrier formation with interface chemical reactivity. [10] (b) Stoichiometry of metal—InP outdiffusion vs reactivity and Schottky barrier height [11, 12].

The development of soft x-ray photoemission spectroscopy (SXPS) further expanded the range of interfacial processes due to metal deposition on semiconductors. These included metal indiffusion and semiconductor outdiffusion, materials processes that until then seemed to be limited to high temperature thermal processing. As shown in Fig. 1(b), room temperature semiconductor outdiffusion for metals deposited on compound semiconductors such as InP in particular could be non-stoichiometric with anion outdiffusion sensitive to even single monolayers of reactive vs unreactive species [14, 15]. This unbalanced outdiffusion could result in missing atoms in the outermost semiconductor layers, which could result in charge sites near the metal interface, which in turn could control the Fermi level movement and resultant band bending. In this sense, there was now a “chemical basis” for Schottky barrier formation.

SXPS provided further evidence for the influence of interface material interactions on the metal–semiconductor charge transfer. The charge transfer between semiconductor and metal changes the band bending within the semiconductor and the resultant Schottky barrier height between the semiconductor band edges and the metal Fermi level EF. As a result, EF relative to the semiconductor band edges changes as layer upon layer of metal deposits on the (clean) semiconductor surface. Figure 2 shows this process for different metals on InP. The final EF position after 10–20 nm metal thickness signifies the final band bending and EC—EF Schottky barrier height of the now metallic overlayer. Figure 2 shows these differences in EF energies and their evolution with metal layer thickness [16]. Each metal exhibits a rapid EF movement to different bandgap energies with only a few Å deposited metal coverage followed by additional movements to final EC—EF positions, energies which span nearly half the InP bandgap rather than a narrow range of “pinning” energies. These additional movements vary in both magnitude and direction depending on the specific metal—lattice chemical bonding.

Figure 2
figure 2

EF movements versus metal overlayer thickness for various metals on UHV-cleaved InP surfaces. Differences in metal–InP interactions are due to different interface material interactions as well as differences in metal electronegativities [16].

Together with defects in the semiconductor bulk [17], these processes provided a picture of semiconductor—metal interfaces which is extrinsic, in contrast to intrinsic models of ideal, unreactive semiconductor–metal interfaces [18,19,20,21]. This extrinsic picture could then account for why rectifying metal–semiconductor contacts did not vary systematically with the difference in work function between metal and semiconductor, i.e., Schottky–Mott ideal behavior. Rather, specific materials interactions at the atomic-scale metal–semiconductor interface could account for Fermi level movements which would otherwise be “pinned” in a narrow range due only to surface defects.

Defects at the nanoscale and interface electronic properties

These metal-specific results showed that electrically active defects could involve more than the outermost semiconductor monolayer. A relatively undeveloped technique provided a way to measure the defects both at and below the metal–semiconductor interface. This depth-resolved cathodoluminescence spectroscopy (DRCLS), extended to UHV environments from early scanning electron microscope studies [22], could probe electronic transitions involving defects within semiconductors, their spatial distribution on a nanometer scale, and their relation to the materials processes described in previous decades. Due to its ability to probe from near-monolayers to microns below the semiconductor surface and even through metal contacts [23], DRCLS provides a range of information about defects at electronic interfaces. These studies included defects induced by strain, dislocations [24, 25], ionizing radiation [26], oxidation [27], heterojunction growth and diffusion [28], surface treatments such as plasma deposition, cleaning, etching and annealing [29], quantum well interface states [30], polytype transformations [31, 32], grain boundary segregation [33], and growth – sensitive processes of next-generation dielectrics and ferroelectrics [34,35,36,37].

The DRCLS process shown in Fig. 3 involves an incident electron beam which generates a cascade of secondary electrons whose energies steadily decrease with X-ray generation and plasmon energy loss [38]. Ultimately, these electrons have only enough energy for impact ionization, i.e., incident electron collisions with atomic orbital electrons to eject those electrons, resulting in free electrons and holes within the material’s conduction and valence bands, respectively. These free electrons and holes can then recombine via several pathways to emit light, specifically, conduction band-to-valence band transitions and transitions between gap states and those bands. The depth of impact ionization and cathodoluminescence excitation depends on the lattice scattering of the cascading secondary electrons, which in turn depends on both material parameters, e.g., density ρ, atomic weight A, atomic number Z, and Avogadro’s number NA and the incident electron beam’s energy EB. Monte Carlo simulations statistically determine the depths at which a cascade of electrons generates electron–hole pairs for a selected incident beam energy [39]. For EB ranging from a few hundred volts to several kilovolts, typical semiconductor excitation depths range from only a few nm to a few hundred nanometers, respectively, depending on the material parameters and film thicknesses through which the incident electron beam traverses.

Figure 4 Left shows DRCL spectra of excitation through 30 nm thick Au vs Ta on ZnO(000–1) surfaces, resulting in excitation only 10 nm below the metal interfaces. In addition to the 3.4 eV bandgap and associated phonon replica peaks, these spectra show sub-bandgap features due to native point defects. At room temperature, only a broad peak at ~ 2.5 eV appears for both metals, associated with oxygen vacancies VO.[41] For Au, Fig. 4(a) left shows only gradual increases in this peak with annealing until 650 °C, at which temperature a second peak due to zinc vacancies VZn appears at ~ 2 eV [42]. The appearance of VZn corresponds to exceeding the phase diagram’s eutectic temperature boundary for Au–Zn eutectic formation [43] and the resultant outdiffusion of Zn into the Au, leaving behind VZn just below the interface. In contrast, the Fig. 4(b) Left Ta-ZnO(000–1) interface shows only the initial VO feature, which grows relative to the near band edge (NBE) features with increasing temperature.

Figure 3
figure 3

DRCLS excitation process consisting of an incident electron that generates a cascade of secondary electrons, producing impact ionization, and the recombination of free holes with electrons which generates optical emission characteristic of lattice defects and the bandgap.

Figure 4
figure 4

Left: 5 keV, 10 K DRCLS of 30 nm (a) Au and (b) Ta deposited on low defect (000–1)ZnO. For Au, annealing induces only slight changes for T < 650 °C but a large ~ 2 eV peak at 650 °C. Initial Ta deposition triples 2.5 eV emission, but annealing induces no ~ 2 eV emission or further 2.5 growth. Right: (a) Low defect Au contacts remain rectifying with annealing up to 550 °C whereas (b) high defect Au contacts are first ohmic then blocking with annealing. High defect I-V characteristics show (c) Au Schottky barrier steadily degrading starting at 350 °C and (d) Ta contacts that become blocking at 450 °C with leakage increasing at 550 °C. Both metals show lower thermal stability vs low defect ZnO [40].

These defect changes correlate with Schottky and ohmic I–V properties of their respective interfaces. For the left-side low defect DRCL spectra, I–V characteristics show that: (a) Au–ZnO Schottky contacts remain rectifying up to a 550 °C anneal and (b) the more reactive Ta contacts are initially ohmic, then blocking when annealed and stable up to 550 °C. This blocking corresponds to formation of an insulating Ta2O5 interface layer, whose bandgap emission DRCLS indeed detects [40]. For high defect ZnO with orders of magnitude higher defect intensity, the Au-ZnO(000–1) interface is now ohmic at all temperatures, due to defect-assisted tunneling through the Schottky barrier, whereas the Ta–ZnO(000–1) interface is initially ohmic, then blocking at 450 °C as the Ta2O5 forms, then less blocking as the higher density defect-assisted tunneling begins [40]. In general, these experimental results show that macroscopic electrical measurements correlate directly with nanoscale measurements of defects at intimate semiconductor interfaces. Furthermore, they establish that macroscopic Schottky barrier and ohmic behavior has a nanoscale physical basis.

Scanning electron microscopy (SEM) enables 3-dimensional DRCLS measurements of native point defects inside state-of-the-art device structures. Figure 5(a) shows an AlGaN/GaN high electron mobility transistor (HEMT) structure illustrating the ability of DRCLS to probe laterally between source and drain, laterally along the gate-drain interface, and in depth across and below the 2DEG AlGaN interface layer [44, 45]. Besides defect peak intensities, the semiconductor bandgap energy is sensitive to temperature T, decreasing with increasing T, and crystal lattice strain σ, increasing with increasing σ, so that these critical materials parameters can be measured in 3 dimensions while the device is operating. Three-dimensional DRCLS hyperspectral imaging (HSI) maps of T and σ at conventional power levels (not shown) [46, 47] show increases at the drain-side gate edge, consistent with finite difference simulations and Raman spectroscopy [48]. Figure 5(b) shows DRCLS defect measurements vs depth through the drain-side gate, revealing 2.8–3.0 eV blue band (BB) and 3.75 eV AlGaN deep-level defect densities localized in the AlxGa1-xN strain layer [44]. The increase in defect density in this layer is consistent with the inverse piezoelectric effect, now believed responsible for AlxGa1-xN/GaN HEMT degradation due to current leakage through strain-induced defects [49, 50].

Figure 5
figure 5

(a) DRCLS of AlGaN/GaN high electron mobility transistor (HEMT). Arrows signify probe location and depth of excitation. (b) AlGaN/GaN HEMT structure with dashed rectangle indicating region near the drain-side gate edge where yellow band (YB), blue band (BB), and 3.75-eV AlGaN deep-level defect densities are measured versus depth [44].

Defects in another state-of-the-art device structure involve an ultrawide bandgap semiconductor, Ga2O3, one of the latest advanced electronic materials and a prime candidate for high power applications due to its high dielectric breakdown strength. With a bandgap of 4.8 eV, Ga2O3 represents a challenge to optical excitation measurements of native point defects. However, with electron beam excitation to generate electron–hole pair creation of even such high or higher bandgap materials, together with nanometer-scale secondary electron microscopy (SEM), one can now study the defect distributions inside Ga2O3 devices and their movement under applied electric fields [51] Fig. 6(a) shows a characteristic CL spectrum of β-phase Ga2O3 with dominant features at 3.2 and 3.6 eV attributed to Ga vacancies VGa and O vacancies VO, respectively. Since VGa and VO act as acceptors and donors, respectively, in Ga2O3, they are electrically charged and subject to applied electric fields. Figures 6(b) and (c) display the HSI maps of these defects in an electrically stressed vertical trench Schottky diode in cross section involving quasi-static reverse current–voltage measurements from 0 V to breakdown (Vbr ≈—2 kV). Figure 6(b) shows an increase in 3.2 eV emission increasing away from the cathode fin structure compared with the unstressed cross section in Fig. 6(d). Conversely, Fig. 6(c) shows an increase toward the cathode fin structure compared with the unstressed cross section in Fig. 6(e). The increase in donor density, albeit a deep level unless complexed with H, can increase the effective electric field gradient around the fin structure. While other physical effects such as self-trapped excitons [52] and interstitial hydrogen [53] may also contribute to the CL features, these also involve native point defects [51]. The movement of semiconductor defects under applied electric fields adds another dimension to their importance in materials research.

Figure 6
figure 6

(a) Representative CL spectra of Ga2O3 vertical Schottky barrier diode (SBD) deconvolved with four Gaussian peaks at 2.85, 3.2, 3.6, and 3.9 eV. (b) and (c) Hyperspectral images (HSIs) of an electrically stressed Ga2O3 vertical device showing relative redistributions of 3.2 and 3.6 eV features. Highly segregated 3.6 vs 3.2 eV CL defect intensities are evident near the trench corners with pronounced gradients extending 1 μm below, while the fins exhibit only minor variation. (d) and (e) HSI of an unstressed device showing uniform peak distributions before electrical stress. (f) CL acquisition setup showing Ga2O3 vertical SBD angled cross section with parabolic collection mirror in position [51].

Combining optical, electronic, and microscopic techniques

Combining SEM-based DRCLS with scanning transmission electron microscopy (STEM), the materials science of the metal/ Ga2O3 interface can now be probed at the nanometer to atomic scale within the same near-interface volume, revealing metal diffusion, lattice distortion, new defect and phase formation, and the opportunity to understand interface behavior under extreme electric fields leading to dielectric breakdown. This can become a new direction for the study of electronic material defects and their impact on next-generation electronics.

A recent study of defects at a metal–Ga2O3 interface provides an example of the information provided by the combination of DRCLS and STEM. Defects associated with Ir, a common impurity in edge-fed growth Ga2O3, can affect carrier transport and recombination, particularly at high device processing and operating temperature and at metal interfaces [54]. Figure 7 shows DRCL spectra taken through vs adjacent to Ir diodes on Ga2O3 as the insets illustrate.

Figure 7
figure 7

Representative as-deposited CL spectra for (a) the Ir– Ga2O3 interface at EB = 1.4 keV, with Ir-related emissions (dotted lines) and (b) an adjacent bare Ga2O3 surface. Insets show the dimensions of the Schottky contacts and probe geometries on vs off the contacts. Deconvolved peak emissions due to Ga2O3 intrinsic vs Ir-induced defects are denoted by dashed vs dotted lines, respectively. The solid line envelope denotes the overall spectral fit [54].

In addition to the common 3.2 and 3.6 eV defects already shown in Figs. 6 and here in Fig. 7(b), the electron beam-deposited Ir induces new Ga2O3 defect emissions in the 2–3 eV range, consistent with charge state transitions for Ir in Ga2O3 predicted by theory [55]. These defect features extend 10–20 nm past the interface, indicative of Ir diffusion into the Ga2O3. With rapid thermal annealing (RTP) in N2, these defects extend tens of nm to almost 100 nm with a calculated diffusivity high enough to suggest that lattice inhomogeneities can provide enhanced diffusion pathways. The Ir diffusion can also account for dramatic changes in I–V, C–V, series resistance and ideality factor characteristics by increasing charge carrier recombination through these Ir-induced defects as well as dipole formation that can account for a + 1 V shift in I–V minimum [54].

A STEM map of a corresponding RTP Ir- Ga2O3 diode in Fig. 8 shows: (i) unusually high intensity at Ga columns, which indicates Ir substitutional atoms (blue arrow), (ii) the extra intensity in between the Ga columns, which indicates the interstitial Ir atoms (red arrow), and (iii) a group of bright atoms (cyan arrow) that indicate clustering of Ir atoms and potentially indicative of Ir-O structure. Using a quantitative atomic counting method [56], the number of Ir atoms in all three defect types was counted vs distance from the interface. Within 5 nm of the interface, Ir defect densities ranged from between 4.1 \(\times\) 1019 cm−3 and 2.1 \(\times\) 1020 cm−3 with most of the Ir defects appearing within 3.5 nm. This demonstrated that the DRCLS-derived diffusion lengths are similar to those found by direct physical observation. Indeed, these densities provide an estimate of only 1–1.8 nm spatial separation for one of the dominant Ir-induced defects in Fig. 6(a), helping to illustrate the contribution of defect-assisted tunneling to a nearly 3-order of magnitude observed increase in reverse-bias current [54].

Figure 8
figure 8

(Left). Cross-sectional, high angular dark field STEM image of 700 °C annealed Ir-Ga2O3 interface showing Ir-related defects. (Right) Number of Ir atoms as a function distance from the interface including 3 types of Ir defects [54].

These combined DRCLS—STEM studies of the Ir- Ga2O3 interface also provide evidence for new phase formation as shown by the additional CL peaks at energies above the 4.8 eV β-Ga2O3 bandgap, consistent with the presence of higher bandgap polymorphs. Ongoing DRCLS-STEM studies of other annealed metal– Ga2O3 interfaces also exhibit CL features characteristic of lattice distortion [57], new defect [58], and new phase formation associated with metal indiffusion [59]. In addition, the locally high densities of defects created near semiconductor–metal interfaces at elevated temperatures may play a role in dielectric breakdown.

New interface defect research: challenges and opportunities

The next research frontier for defects at electronic materials interface involves several significant challenges but corresponding opportunities as well. These challenges are both experimental and theoretical, while the corresponding opportunities are both fundamental and applied. Table 1 provides examples of both.

TABLE 1 Measuring and controlling defects at semiconductor–metal interfaces: challenges and opportunities.

Describing materials properties at semiconductor–metal interfaces in 3 dimensions is a challenge requiring atomic scale chemical and structural measurements of bonding and composition changes. This is an opportunity for STEM, which can now measure (i) lattice structure phase changes in both interfacial layers and nanoscale inclusions [59]; (ii) composition from EDX or ELS core level absorption thresholds; and (iii) chemical bond changes from EDX or ELS.

Measuring semiconductor band structure at nanoscale distances from metal interfaces is challenging for optical techniques, which typically involve sub-micron or larger absorption depths. The opportunity is for techniques such as DRCLS, which can measure bandgap and sub-bandgap defect emissions locally in 3 dimensions. To determine band bending, XPS and UPS can measure Fermi-level movements in the semiconductor bandgap of clean semiconductor surfaces stepwise with atomic layer deposition of metals in UHV.

Correlating macroscopic barrier measurement to nanoscale electronic features is a challenge since macroscopic barrier height measurements are affected by defect charges and electric dipoles at the nanoscale. The opportunity is for STEM with EDX, which can measure new interface lattice phases, metal indiffusion, lattice distortion, atomic interstitials, and vacancies. DRCLS can measure new defect states near interfaces nondestructively and correlate them depth-wise to STEM features. The charge associated with these features can be related to electric dipole voltages measured macroscopically to account for Schottky barrier heights.

Controlling macroscopic barriers with nanoscale materials changes is a challenge requiring controlled material processes. The opportunity is for surface science and remote plasma chemistry combined with near-nm DRCLS to introduce controlled interface dipole changes, which can be monitored in operando for any changes as device structure are operated.

A major challenge for high power devices is understanding the physical mechanisms of dielectric breakdown under extreme applied voltages since these mechanisms must be identified at the atomic scale. This is an opportunity for DRCLS combined with STEM, which together can identify atomic lattice and electrical defect changes and possible percolation pathways with extreme electric fields extending up to and beyond breakdown.

This is an exciting time for solid-state electronic materials with advanced micro- and optoelectronic properties, but native point defects at their semiconductor–metal interface can limit their operation. These defects depend strongly on interface chemical structure, and techniques are now available to learn how their presence at nanoscale dimensions impacts electronic structure. Combining these optical, electronic, and microscopic techniques can provide new directions for defect research of metal–semiconductor interfaces at the nano/atomic scale. In turn, they can meet the experimental challenges and create the opportunities for new defect research of electronic material interfaces at a deeper level.