The cross-section structure of AlGaN-based DUV LEDs manufactured by metal-organic chemical vapor deposition (MOCVD) is illustrated in Figure 1a. An aluminum nitride (AlN) layer working as a buffer layer to reduce dislocation density in following AlGaN layers was grown on a sapphire substrate, then two layers of AlGaN with different Al concentrations, multiple quantum well (MQW), electron-blocking layer (EBL), and finally a GaN layer.

This thin film structure was then ICP etched by lithography to form a mesa structure on the top, as shown in Figure 1b. Oxide layers formed on both the top GaN layer and side walls of GaN layer and AlGaN layer after thermal annealing in high temperature with oxygen rich atmosphere. Finally, A layer of high purity SiO₂ of 100 nm thick was deposited on the mesa of unannealed and annealed samples by magnetron sputtering ^[7], as illustrated in figure 1c.

Cross-section TEM (X-TEM) specimens of AlGaN-based diodes of unannealed and annealed were made by focus ion beam (FIB) on selected positions, they were then analyzed by TEM/STEM, image and diffraction, to characterize those oxides formed on the surface of GaN and side-walls of GaN and AlGaN during high temperature annealing.

Two TEM BF images of low-and-medium magnification in figure 2 show the cross-section structure of unannealed and 900℃/20 min annealed AlGaN DUV LEDS respectively. Three new phases, marked 1, 2, and 3, on the surface and the side-wall of the GaN layer and the side-wall of the AlGaN layer were found after comparing these two BF images carefully.

A pair of STEM BF image and ADF image of the 900℃/20 min annealed AlGaN DUV LEDS are shown in figure 3. From the TEM images and STEM images in figure 2 and figure 3, we find that the best TEM imaging technique for this material system is the STEM ADF image.

This can be further confirmed by two sets of STEM images of high magnification in figure 4. STEM ADF images give better contrast to distinguish those three phases. The characteristic image contrast of each phase tells that phase 1 and phase 3 may be polycrystal materials with grains of nano size, and phase 2 may be a single crystal.

Figure 5 shows a low magnification STEM BF image and a set of selected area diffraction patterns (SADPs). These SADPs are corresponding to GaN layer, AlGaN layer, and three products (figure 5a). Both GaN and AlGaN are epitaxial layers, corresponding SADPs indicate the view direction is along [1 1 -2 0] zone axis in this TEM analysis. All these three SADPs of oxides are from one zone axis only and not enough to be solved completely at present. However, they are all spot patterns of single array. This type of SADPs indicates that three analyzed areas are all single crystals. These results are somehow contradictory with those results from TEM/STEM image data shown in figure 3 and figure 4.

To solve this contradiction, we checked each new phase using its TEM images, STEM ADF image, and SADP, back and forth, and made sure that all these three products were single crystals finally. Black and white image contrast in product 1 and product 3 are atomic number contrast rather than diffraction contrast which is dominant in samples of poly crystals.

We can identify phase 1 and phase 3 to be porous structure definitely by higher magnification STEM ADF images, as shown in figure 6. Dark areas (bright areas in BF image) in these two products are pores. Those pores in phase 1 are obviously larger than those in phase 3. Phase 2 is a dense single crystal with some big pores inside, one of them is pointed by a white arrow in figure 6. The mechanism of forming homogeneously distributed nano pores in phase 1 and phase 3 is considered to be that the diffusion rate of oxygen and gallium atoms is higher than the stacking rate of atoms to crystals in high temperature annealing.

Generally, we need SADPs of several zone axes to solve the crystal structure. It is hard to identify the product phase by using a single SADP to characterize the new phase definitely. However, the simulated β-Ga₂O₃ [0 1 0] SADP is very similar to SADPs shown in figure 5d, 5e, and 5f. We, thus, infer that these products may be β-Ga₂O₃ and β-(Al_xGa_1-x)₂O₃ on GaN layer and AlGaN layer correspondingly. β-Ga₂O₃ is a monoclinic crystal, and this makes the analysis work of its SADP complicated and hard.

Elemental maps of GaN and the product on it, shown in figure 7, are derived by STEM/EDS spectrum image technology. These elemental maps indicate that there are only oxygen and gallium in the product phase. This indicates that the product analyzed is a gallium oxide. The variation in concentration is extracted along the trace indicated by the light blue arrow in figure 8a.

Concentration of elements in line profiles shown in figure 8b is quantitatively calculated by the built-in EDS program using K factors stored in database ^[9-11]. This kind of quantitative analysis is named to be standardless quantitative analysis which has been widely accepted by journals of science and engineering.

The concentration ratio of O/Ga of oxide 1B is derived from a flat region in the 1B region, indicated in figure 8b, is 1.23. This result indicates the composition of this oxide phase is Ga5O6. This composition is calculated by the built-in EDS program which used X-ray signals emitting from the specimen and being detected by the EDS detector and K factors stored in the database.

However, there is no gallium oxide with composition Ga5O6 reported in literature.
Absorption effect is usually neglected in TEM/EDS analyses which use thin foils to be specimens. However, we found that the absorption effect is still significant and results in large error without awareness when light elements, such as C, N, and O, are included. Checking figure 8b carefully, we find that the ratio of O to Si is much smaller than 2 (left side), and the ratio of N to Ga is obviously smaller than 1 (right side). EDS line profiles in figure 8b are then calibrated by taking stoichiometric compounds of SiO₂ and GaN. The calibrated EDS line profiles are then modified as shown in figure 8c. The ratio of O to Ga of the oxide 1B then changes to be 1.53, it means that the gallium oxide is Ga₂O₃, as reported by Oshima et.al. ^[12].

In TEM(STEM)/EDS analysis, the EDS line profiles calculated by the built-in program can be further calibrated by known phases around the analyzed target. We call this technique to be self-calibration EDS quantitative analysis.

This technique is one of TEM analysis techniques developed by iST. It gives better accuracy in EDS quantitative analysis in materials containing light elements, such as C, N, and O. The absorption effect becomes significant when energies of X-rays of these light elements are very low. Dealing with materials containing light elements discussed above, results of TEM/EDS analysis of iST give better accuracy by using the self-calibration EDS quantitative analysis technique.