Efficiency limiting process
Internal quantum efficiency
The MOVPE growth of AlGaN alloys was first performed by Matloubian and Gershenzon in 1985 . The parasitic-gas-phase reaction between an Al source such as trimethylaluminum (TMAl) and ammonia (NH3) , the decomposition of TMAl in the gas phase to AL4C3 , or the very low rate of diffusion of an Al precursor in an NH3-rich gas flow, causes a reduction in the growth rate and polymerization in the gas phase. Several groups have reported the effectiveness of alternative-source gas supplies or flow-rate modulation using methods such as pulsed-atomic-layer epitaxy , the alternative-source feeding technique , migration-enhanced MOCVD , and the NH3 pulse-flow multilayer growth method . An alternative-source supply was initiated by atomic-layer epitaxy , then modified by gas-flow-rate modulation during growth by MOVPE . The effectiveness of these gas-flow-modulation techniques for the growth of AlN and the fabrication of UV and DUV LEDs is summarized in reviews [18, 19].
Another approach to the growth of high-quality AlN is to grow AlN under a relatively low NH3 partial-pressure at a high temperature. In the case of GaN growth, the ratio of the molar fraction of group V source gases such as NH3 to that of group III source gases such as trimethylgallium (TMGa), which is called the V/III ratio, is approximately 1000 or even higher. For InGaN growth, the V/III ratio should be much higher because of the high dissociation pressure of nitrogen in InN. In contrast, the V/III ratio during AlN growth can be lower [20, 21].
Recent experiments have shown that relatively high-quality AlN can be grown at a V/III ratio of as low as 10, which is two orders of magnitude lower than that for the growth of GaN. The growth temperatures of AlN and AlGaN are also an important issue. The sublimation temperature of AlN is much higher than that of GaN [22, 23]. Therefore, the epitaxial growth temperature of AlN is also expected to be higher than that of GaN.
Fig. 1.4. TEM images of AlN grown at different Tg of (a)-(c) 1200°C, (d)-(f) 1400°C, and (g)-(i) 1600°C. The cross-sectional dark-field TEM images in the top row were taken under two-beam conditions with the incident beam g//  ((a), (d), and (g)). The TEM images in the middle row were taken with g// [11-20] ((b), (e), and (h)). The TEM images on the bottom row show bright-field plan-view TEM images with g// [11-20] ((c), (f), and (i)).
Figure 1.4 shows cross-sectional dark-field TEM images observed under two different incident beam conditions and bright-field plan-view images of MOVPE-grown AlN on a 6H-SiC substrate grown at different temperatures . With increasing growth temperature, the screw-type dislocations almost disappeared, as shown in (g), and the density of edge-type dislocations was also much reduced, as in (i).
The AlN molar fraction of AlGaN is considered to be linearly dependent on the ratio of the flow rate of TMAl to the total flow rate of group III alkyls; that, is the sum of the flow rates of TMAl and TMGa. This linear relationship holds only at relatively low-growth temperatures. At high-growth temperatures, GaN desorption leads to the growth of high-AlN-molar-fraction AlGaN. The thermodynamic analysis conducted by Koukitsu et al. is very helpful in estimating the
Fig. 1.5. Comparison of the experimental and calculated Al content of AlGaN.
vapor-solid relationship of AlGaN [25, 26]. Figure 1.5 shows the results of experimental and thermodynamic analysis of the relationship between the Al content in AlGaN and the growth temperature for two different V/III ratios under a fixed TMGa/(TMGa+TMAl) supply ratio . The experimental results agree relatively well with the thermodynamic analysis.
To grow low-dislocation-density GaN, epitaxial lateral overgrowth (ELO) using a dielectric mask has been very successful. However, a dielectric mask cannot be applied to the selective growth of AlN-containing materials owing to the deposition of polycrystals on the mask. Therefore, AlGaN growth on a maskless, groove-patterned template has been investigated.
AlGaN growth on a groove-patterned GaN template in combination with a low-temperature-deposited AlN interlayer (LT interlayer) has been found to be very useful for the fabrication of UV LEDs and LDs with emission wavelengths longer than 350 nm , whereas AlGaN growth on a patterned AlN template is useful for the fabrication of UV/DUV LEDs with emission wavelengths shorter than 350 nm [28-30]. Figure 1.6 shows a cross-sectional TEM image of AlN grown at 1,450°C on a Si-face 6H SiC substrate . There is a lattice mismatch between the AlGaN and the underlying AlN layer. Therefore, it is not easy to grow low-dislocation-density AlGaN on AlN even though dislocation density in underlying AlN is low.
Figure 1.7 summarizes the compositional dependence of the threading dislocation density (TDD) in AlGaN grown on an AlN template with a dislocation density of 107 cm~2 . The lattice mismatch increases with increasing Ga content. Therefore, in the initial stage, the dislocation density increases with the Ga content. However, after a certain thickness is reached, island and lateral growth dominate, especially in high-Ga-content AlGaN, so the dislocation density tends to decrease. A grooved template is effective for further reducing the dislocation density.
Fig. 1.6. Cross-sectional TEM image of AlN grown at 1,450°C on patterned 6H SiC substrate.
The IQEs of UV/DUV MQWs with different TDDs were characterized by measuring the excitation density dependence of photoluminescence (PL) using an ArF excimer laser with a wavelength of 193 nm as the excitation source, which is known as Shockley-Read-Hall (SRH) analysis [33-36]. In the case of UV/DUV devices, the band structure indicates that the ratio of Auger recombination
Fig. 1.7. Compositional dependence of AlGaN on planar AlN and grooved AlN. Upper line: TDD estimated from lattice mismatch. Solid circles: TDD in 0.25-pm-thick AlGaN on planar AlN. Solid squares: TDD in AlGaN of thickness greater than 2 pm on planar AlN. Open circles: TDD in 2-pm-thick AlGaN on grooved AlN. Solid squares: TDD in AlGaN thicker than 10 pm on grooved AlN.
should be negligibly small. When the generation rate (G) is equal to the total recombination rate (R), G = R and the PL integrated intensity are given by the following formulae.
Here, n is the excited carrier density, A is the non-radiative recombination constant, B is the radiative recombination constant, and n is a constant determined on the basis of the PL collection efficiency. Differences in the B value cause an error in the absolute IQE value. It is currently unclear which value of B should be used. However, the relative trend of the TDD dependence of IQE can be discussed regardless of the chosen value of B .
By combining the above two formulas, G can be rewritten as
Then, the non-radiative coefficient A is deduced by curve fitting, and the IQE can be expressed as
Figures 1.8 and 1.9 show examples of the relationship between the generation rate G and PL intensity (Ipp) (Fig. 1.8) and IQE as a function of n (Fig. 1.9) for QWs emitting 230 nm UV-C with different TDDs . The same procedure was
Fig. 1.8. Experimental results and fitting curves of the G and IPL for 230 nm QWs with different TDDs. The non-radiative coefficient A is deduced from this fitting curve using eq. (1.3).
Fig. 1.9. Calculated IQE using eq. (1.4) as a function of n for samples shown in Fig. 1.8.
performed for every sample to deduce the non-radiative recombination coefficient A. Figure 1.10 summarizes the TDD dependences of A for 230, 250, 300, and 350 nm QWs. Those of 420 to 495 nm c-plane QWs grown on free-standing GaN with very low dislocation density, and semipolar QWs on a patterned Si substrate are plotted for comparison. From the results, it is clear that A is linearly dependent on TDD, A = 0.05 x TDD (cm-2) + 8.1 x 106 (s-1), and independent
Fig. 1.10. TDD dependences of non-radiative coefficient A for QWs emitting different wavelengths from 230 to 495 nm.
of the emission wavelength. Again, it should be emphasized that this A is deduced assuming B to be 0.2 x 10~10 cm3/s. In this assumption, a non-radiative coefficient A of approximately 107 s-1 appears to be the limit of accuracy of the measurement system.