Introduction

Within the last decade, the III-N semiconductors have experienced rapid development to the point that commercial optical devices are now available. Recently the importance of the piezoelectric properties of the III-N has come to light. With this realization, the knowledge and control of the polarity of III-N thin films has become paramount for device design. The piezoelectric and spontaneous polarizations in this system are large enough to induce a two-dimensional electron gas (2DEG) at Al_1−xGa_xN/GaN heterojunctions [Reference Hangleiter, Jin Seo, Kollmer, Heppel, Off and Scholz1-Reference Yu, Sullivan, Asbeck, Wang, Qiao and Lau7]. For Molecular Beam Epitaxy (MBE) material grown on sapphire substrates using a plasma nitrogen source, the nucleation layer determines the polarity of the resulting layers [Reference Murphy, Chu, Wu, Yeo, Schaff, Ambacher, Smart, Shearly, Eastman and Eustis8]. With this control of the polarity of the material, high quality MBE High Electron Mobility Transistor (HEMT) devices have been achieved [Reference Murphy, Chu, Wu, Yeo, Schaff, Ambacher, Smart, Shearly, Eastman and Eustis8]. In this paper, the composition and structure of the AlN nucleation layer in a HEMT device structure studied in the Cornell University Scanning Transmission Electron Microscope (STEM), will be presented.

The Cornell University STEM produces an electron probe approximately 2.1Å in diameter and thus is distinctly qualified for near atomic resolution interface investigations. As this probe is scanned across a TEM type sample, the scattered electrons are collected. Electrons scattered to high angles are collected by the Annular Dark Field (ADF) detector producing an image where the contrast depends mainly on atomic number (Z) and thickness. Thus ADF imaging is commonly referred to as Z-contrast imaging. Electrons scattered to small angles are either detected by the Bright Field Detector, producing an image analogous to normal bright field TEM, or are recorded as EELS by either the Serial or Parallel detector.

EELS provides information about the local bonding and electronic structure by probing unoccupied electronic states. The fast moving electrons from the STEM probe interact with the core electrons in the specimen, which are excited from their ground state to unoccupied states. Since the energy lost by the fast moving electrons in the probe is equal to the difference in energy between the core level and excited level, the measurement of the intensity of the probe electrons as a function of energy loss provides substantial details of the electronic structure and chemistry.

Experimental Details

The III-N HEMT sample with an AlN nucleation layer was grown in a Varian GenII MBE at Cornell University. The specifics of the growth conditions, electrical characterization, and device results are given elsewhere [Reference Murphy, Chu, Wu, Yeo, Schaff, Ambacher, Smart, Shearly, Eastman and Eustis8]. The sample was prepared by standard tripod polishing techniques to form a wedge specimen [Reference Benedict, Anderson and Klepeis9]. The sample was ion milled with a BAL-TEC Res 010 for final thinning.

The Cornell University STEM has a maximum energy resolution of ∼0.7eV over a large energy range (from 0 to 2keV). With high spatial and energy resolution, the STEM is particularly suited for sub-nanometer chemical and structural studies. With such a focused small probe, high spatial resolution EELS can be obtained. Thus compositional and structural information can be acquired with core-loss EELS. In addition, thickness information can be obtained on a nanometer scale from the ratio of the first plasmon intensity to the zero loss (electrons that have lost no energy) intensity. More detailed information concerning the STEM can be found elsewhere.

EELS spectra were acquired using the Parallel EELS detector. Specifically the nitrogen and oxygen K-edges were obtained simultaneously and individually by stepping across the GaN/AlN/Sapphire interfaces. The K-edges correspond to transitions from atomic 1s states to empty local and conduction band states with p-character. Spectra of the aluminum L₂₃-edge were also obtained. The L-edge corresponds to transitions from atomic 2p to empty local and conduction band states with d and s-character. The ADF signal is acquired in conjunction with the spectra. In addition, ADF images of the area of interest collected before and after spectrum acquisition indicated no significant spatial drift.

All data and images are acquired digitally. After acquisition, the spectra were smoothed and shot noise was removed. The background was subtracted using a standard power-law curve fit to the pre-edge. The intensities of the nitrogen and oxygen K-edges were integrated over 50 eV starting at threshold. Ratios of O/N+O were obtained using the standard equation involving integrated intensities and cross section [Reference Egerton10]. Acquisition times for the K-edges were selected in an attempt to balance electron beam damage [Reference Serin, Colliex, Brydson, Matar and Boucher11] and signal to noise. Thus, for the acquisition times used, some beam damage was observed in the sapphire substrate. The damage results in a very small artificial nitrogen signal in the sapphire.

Discussion

The presence of oxygen is observed in the AlN nucleation layer upon examination of Figure 1. The oxygen signal is constant within the sapphire substrate. Moving across the AlN/sapphire interface (based on the ADF image in Figure 1(a)), the oxygen signal drops sharply. Once across the interface, the oxygen signal decays and reaches zero at approximately the AlN/GaN interface.

Figure 1. (a) ADF image of the area of interest. (b) Integrated Intensity of N and O K-edge. (c) The ratio O/O+N vs. distance. The horizontal line represents where the data in (b) and (c) were taken.

As stated above, the nitrogen signal within the sapphire results from beam damage. Therefore the ratio depicted in Figure 1(c) does not level off at one in the sapphire. On moving from the sapphire to the AlN, the nitrogen signal increases but does not peak until about half way through the AlN, as seen in Figure 1(b).

Upon calculation of the O/O+N ratio using the data in Figure 1(b), thickness effects are removed. The ratio is displayed in Figure 1(c). The ratio decreases at points moving across the AlN/sapphire as the oxygen signal decreases and the nitrogen signal increases.

Figure 2 is a bright field TEM image of the AlN/sapphire interface. The interface appears fairly abrupt giving no indication of any oxygen containing AlN. Thus un-calibrated bright field TEM imaging is inadequate to detect chemical profiles. Only through strict control of experimental conditions, has chemical mapping of semiconductor interfaces been achieved using bright field TEM images [Reference Ourmazd, Taylor, Cunningham and Tu12]. Others have stated based on TEM images that their observed interface is chemically sharp [Reference Yeadon, Marshall, Hamdani, Pekin, Morkoc and Gibson13]. Without microanalysis, such claims are difficult to substantiate.

Figure 2. Bright Field TEM image of the AlN/sapphire interface.

Now we turn to examination of the ELNES in an attempt to learn about the local environment of the individual constituents. Core-loss edges at significant points across the AlN/sapphire interface are displayed in Figure 3. The N K-edge in the oxygen poor AlN and in the oxygen rich AlN are displayed in Figure 3(a). The N K-edge for AlN compares well with previously published spectrum [Reference Serin, Colliex, Brydson, Matar and Boucher11]. In Figure 3(a), notice the decrease in intensity of the first peak of the N K-edge from the oxygen rich region of the AlN relative to the other peaks as compared to the N K-edge from the relatively pure AlN. Based on the comparison of the two N K-edges with Multiple Scattering calculations, some of the oxygen in the AlN is in octahedral interstitial positions within the AlN lattice [Reference Serin, Colliex, Brydson, Matar and Boucher11]. Abaidia et. al. determined the location of oxygen in AlN based on examination of extended energy loss fine structure. It was determined that oxygen would be in octahedral interstitial sides for Al/(N+O) 0.8 and in substitution positions for Al/(N+O) 1[Reference Abaidia, Serin, Zanchi, Kihn and Sevely14]. Others determined the O in AlN was in substitution sites based on X- ray Absorption Spectroscopy [Reference Katsikini, Paloura, Cheng and Foxon15]. The possibility of oxygen substituted for nitrogen in the AlN nucleation layer of the HEMT sample examined here cannot be ruled out. Unfortunately no core edge spectroscopy simulations of oxygen substituted for nitrogen are available with which to compare. In addition, it is important to note that the oxygen containing AlN layers examined by the above mentioned groups were grown in an oxygen-containing environment.

Figure 3. EELS core-loss edges. (a) The N K-edge in the AlN and at the AlN/sapphire interface. (b) The Al L₂₃-edge in the AlN, at the AlN/sapphire interface, and in the sapphire.

Small variations in the O K-edges were observed. However, little information is gained and thus will not be presented here.

Aluminum L-edges are presented and labeled in Figure 3(b). The edges obtained in AlN and in Al₂O₃ match well with published data [Reference Serin, Colliex, Brydson, Matar and Boucher11,Reference Mo and Ching16]. The Al L-edge obtained at the interface between the AlN and the sapphire is a combination of the edges from AlN and Al₂O₃. In fact, the Al L-edges obtained by stepping across the interface using the parallel EELS detector were a smooth transition from the AlN Al L-edge to the Al₂O₃ Al L-edge. The edge displayed in Figure 3(b) was obtained at the AlN/Al₂O₃ interface. Calculations by Brydson et al. [Reference Brydson17] confirm the validity of simply performing a linear sum of Al L-edges to determine the mount of different components. Thus, the Al at the interface is approximately 72% tetrahedrally bonded Al and 28% octahedrally bonded Al.

Conclusion

Oxygen has been observed by EELS in the AlN nucleation of a high quality MBE grown III-N HEMT device. The AlN nucleation layer is key to obtaining MBE Ga-face polarity material on sapphire. The oxygen appears to have diffused in from the sapphire substrate. The concentration profile suggests a sharp drop in oxygen crossing the AlN/sapphire interface. Once across the interface, the oxygen concentration decays approaching the GaN/AlN interface. Examination and comparison of the ELNES of the N K-edge and the Al L-edge reveals that oxygen in the AlN is in octahedral interstitial sites and that Al in the oxygen rich region of the AlN has both tetrahedral and octahedral coordination. Bright field TEM images fail to identify a chemical difference in the AlN. Although some information has been gained about the Ga-face polarity determining AlN nucleation layer, it is unclear at this point what role the oxygen plays.