Ca as an unintentional impurity has been investigated in III-nitride layers grown by molecular beam epitaxy (MBE). It is found that Ca originates from the substrate surface, even if careful cleaning and rinsing procedures are applied. The initial Ca surface coverage is ∼1012 cm−2, which is consistent with previous reports on GaAs and silicon wafers. At the onset of growth, the Ca species segregates at the growth front while incorporating at low levels. The incorporation rate is strongly temperature dependent. It is about 0.03% at 820 °C and increases by two orders of magnitude when the temperature is reduced to 600 °C, which is the typical growth temperature for InGaN alloy. Consequently, [Ca] is as high as 1018 cm−3 in InGaN/GaN quantum well structures. Such a huge concentration might be detrimental for the efficiency of light emitting diodes (LEDs) if one considers that Ca is potentially a source of Shockley-Read-Hall (SRH) defects. We thus developed a specific growth strategy to reduce [Ca] in the MBE grown LEDs, which consisted of burying Ca in a low temperature InGaN/GaN superlattice (SL) before the growth of the active region. Finally, two LED samples with and without an SL were fabricated. An increase in the output power by one order of magnitude was achieved when Ca was reduced in the LED active region, providing evidence for the role of Ca in the SRH recombination.
White light emitting diodes (LEDs) based on III-nitride materials are currently revolutionizing the lighting market as they allow huge energy savings.1 The luminous efficacy of commercial LED lamps is higher than 150 lm/W, which significantly outperforms conventional incandescent bulbs and even compact fluorescent lamps. The success of III-nitride semiconductor compounds is above all the result of a long quest aimed at improving the material quality, particularly in light of the lack of GaN native substrates. Various growth techniques were investigated, but eventually metal organic vapor phase epitaxy (MOVPE) became the most successful one. MOVPE allows the growth of high-quality GaN layers on sapphire substrates, a prerequisite for high brightness blue LEDs.2 In the late 90s, the availability of GaN substrates changed the situation and molecular beam epitaxy (MBE) proved able to produce high-quality materials, as exemplified by the demonstration of electrically injected low threshold laser diodes (LDs).3,4
MBE possesses inherent advantages such as a low growth temperature that is key to synthesizing high In content InGaN alloys or InN,5 the absence of memory effects, and perfect control of layer thickness. Furthermore, MBE enables high p-type doping levels, without the need for post-growth annealing.6 This enables the achievement of tunnel junctions (TJs) with excellent characteristics.7 The MBE-grown TJs have recently been implemented in LEDs,8 vertical cavity surface emitting lasers,9 and LDs.10,11 Such realizations perfectly illustrate the potential of MBE for enabling complex optoelectronic devices, provided the active region is grown by MOVPE.8–11 Indeed, MBE has failed so far in producing high efficiency LEDs based on the InGaN quantum wells (QWs). The maximum external quantum efficiency (EQE) of the MBE grown InGaN based LEDs has been limited to few percent, despite the excellent I–V characteristics.12
Whereas low growth temperature is often presented as the main advantage of MBE over MOVPE, this parameter could actually be responsible for higher concentration of impurities and/or native point defects. Secondary ion mass spectrometry (SIMS) measurements performed on GaN and InGaN layers grown by NH3-MBE indicate [O] and [C] in the low 1016–1017 cm−3 range, comparable to the MOVPE standards.13,14 If those usual contaminating species can be disregarded, other impurities or native point defects must be responsible for non-radiative Shockley-Read-Hall (SRH) recombination, for example, a complex between gallium vacancies and oxygen.15
Another potential source of SRH recombination is metal impurities. These species could arise not only from the group-III element sources, crucibles, effusion cells, and process gases but also from the substrate surface. This is why wafer cleaning and surface preparation have been carefully carried out in semiconductor technology. However, even applying stringent surface cleaning, some impurities are very difficult to remove. This is the case of Ca, which has been routinely observed on silicon wafers after surface preparation.16,17 In the case of III-arsenides, a Ca spike as high as 1015 cm−3 was observed at the interface between the GaAs substrate and the epilayer, even after surface cleaning.18 In this paper, we show that Ca is present in the MBE grown GaN epilayers and investigate its incorporation as a function of growth conditions and its role in photo- and electroluminescence (EL) properties of InGaN quantum wells.
The III-nitride epilayers were grown in a Veeco 930 MBE reactor with NH3 as nitrogen source. We used ultra-high purity NH3, which is further passed through a gas purifier. Effusion cells were outgassed at 1200 °C before loading with high purity (6N) Ga and In solid sources. The substrates were c-plane oriented GaN/sapphire templates grown by MOVPE. The cleaning procedure prior to growth consisted of solvent cleaning, followed by degassing in ultrahigh vacuum at 400 °C for several hours. The substrate temperature was monitored with a calibrated pyrometer. The backside of the substrates was coated with 500 nm of Ti to ensure thermal coupling with the heater. The growth temperatures of GaN and InGaN were 820 °C and 600 °C, respectively. Growth rates and In compositions were calibrated by high-resolution X-ray diffraction. The SIMS measurements were performed with a Cameca IMS 7f system. Acquisition parameters were optimized to ensure high mass resolution, which is required to discriminate 40Ca from the molecular ion silicon carbide 28Si12C+, as well as 24Mg16O. The mass resolving power of the SIMS was set to 3200, and Ca concentration was calibrated using an ion implantation standard. Two LED samples were prepared on GaN templates. The structure consisted of 100 nm Si-doped GaN, 3× In0.1Ga0.9N (2.5 nm)/GaN (5 nm) active region, 100 nm p-type GaN, and 5 nm p++-GaN contact layer. Note that the growth temperature of p-type GaN layers was reduced to 740 °C, which is the optimum temperature when using NH3-MBE.19,20 In one of the two LED samples, low temperature (LT) layers were introduced to reduce Ca contamination at the surface by forcing its incorporation (details about the interlayer are given hereafter). The I–V, electroluminescence (EL) and light output power of the LEDs were tested using indium contacts.
Figure 1 displays the SIMS profile recorded on a commercial LED epiwafer grown by MOVPE from which processed devices exhibit peak external quantum efficiency higher than 60%. In this sample, no traces of transition metal species Fe, Cu, Ni, Ti, or Cr were found. The [Ca] was on the order of the detection limit, i.e., ∼1013 cm−3. Subsequently, SIMS was performed on the MBE films deposited on GaN/sapphire template. A reference sample with an LED-like structure (no p-doped layer) was first prepared. It consisted of 500 nm n-type (1017 cm−3) GaN:Si, 3× InGaN (5 nm)/GaN (5 nm) MQWs, and 85 UID GaN cap layer. Like the MOVPE material, no traces of transition metals Fe, Cu, Ni, Ti or Cr were detected with high mass resolution SIMS. However, the corresponding [Ca] profile exhibited several striking features (Fig. 2(a)). [Ca] was below the detection limit in the GaN template grown by MOVPE, as observed on the commercial LED epiwafer. At the regrowth interface, [Ca] peaked at 7 × 1016 cm−3 and then remained constant in the n-type GaN layer at a relatively high level (8 × 1015 cm−3). When the growth temperature was decreased down to 600 °C to allow for the growth of the InGaN/GaN MQWs, [Ca] spikes of ∼1018 cm−3 were observed. An individual [Ca] peak was observed in each of the 3 QWs, with a decrease corresponding to the growth of the GaN barriers at higher temperature. These observations suggest that [Ca] depends on the growth temperature but the origin of the contamination was unclear. However, a closer look at the SIMS profile proved informative. [Ca] in the n-type GaN was 8 × 1015 cm−3 which reduced to 2 × 1015 cm−3 in the UID GaN cap layer, whereas the growth temperatures were the same (820 °C). This evolution is inconsistent with contamination originating from the growth chamber environment. If it was the case, [Ca] should only depend on the substrate temperature (keeping the same growth rate and V/III ratio), i.e., fixed by the residence time of Ca adatoms, and therefore [Ca] should be constant. This is a first indication that the source of Ca is at the growing surface. In addition, we observed the [Ca] impurity in material grown by NH3 MBE both at UCSB and at EPFL, as well as in material grown by plasma assisted MBE at UCSB.
A temperature-dependent incorporation of Ca impurity was observed in GaAs as early as 1982.21 [Ca] was much lower than what we measured in the present III-nitride layers, typically 1013 cm−3 for GaAs growth temperatures above 500 °C. The authors of Ref. 21 tentatively proposed the effusion cells as the origin of the layer contamination. More recently, Ptak et al. carried out the SIMS measurements on various epitaxial layers (GaAs and InGaAsN) grown by both MBE and MOVPE and in different laboratories.18 The corresponding [Ca] profiles in those layers exhibit features very similar to those we observed: (i) a Ca spike is present at the regrowth interface, (ii) Ca is present in the epilayers, and (iii) Ca incorporation is enhanced when the epilayers contain nitrogen. This last observation suggests that Ca has a tendency to incorporate more in GaN than in GaAs, which is explained by a stronger Ca-N interaction.18 The main conclusion of Ref. 18 is that Ca is present at the surface of the GaAs epiwafers and that cleaning procedure improved but did not completely remove the contamination. Actually, the contamination of semiconductor wafer surface by Ca is quite common and well documented in silicon technology.16,17,22–24 Thus, we can postulate that Ca impurities in the MBE grown layers originates from surface contamination of the GaN templates, despite the careful cleaning procedures. Actually, it is very likely that deionized-water is the source of Ca contamination, as previously reported in silicon technology.24
To confirm the origin of the Ca contamination and with the aim of reducing [Ca] in the active region of InGaN/GaN QW based devices, we grew a similar structure to the one in Figure 2(a) but with an LT step in the GaN buffer. The idea is straightforward: as a complete removal of Ca species from the substrate surface is challenging, we should instead facilitate its incorporation in an LT layer to be buried away from the active region of a light emitting heterostructure. Figure 2(b) displays the SIMS profile of an InGaN/GaN MQW structure deposited on a GaN buffer layer, which features 50 nm of GaN grown at 600 °C. A Ca spike was clearly seen at the regrowth interface and ascribed to the wafer surface contamination. Only a small fraction of Ca was buried at the interface and the rest segregated at the growth front while slightly incorporating. This surface segregation process is responsible for the background concentration of about 5 × 1015 cm−3. Note that the nearly constant [Ca] is an indication of very strong segregation phenomenon. When the LT-GaN step was performed, Ca incorporated rapidly, as seen in the SIMS profile. As a consequence, the Ca at the growth front was markedly reduced, corroborated by a much lower [Ca] in the subsequent n-type GaN layer. The reduction of Ca contamination is even more pronounced in the InGaN/GaN MQW structure. There is a decrease of [Ca] by nearly 3 orders of magnitude, as compared to the reference sample (Fig. 2(a)). Despite this drastic reduction, [Ca] is still higher in the MBE grown materials than in the MOVPE layers. If one considers that Ca may be a strong SRH recombination center, then such a concentration is high enough to plague the LED efficiency.25 In addition, even with the use of an In surfactant, the 50 nm thick LT-GaN creates a significant roughening of the growth front, which might be detrimental for the growth of the InGaN/GaN MQWs. A similar set of InGaN/GaN MQW samples to those shown in Figures 2(a) and 2(b) was grown but with only 10 nm of LT GaN, as opposed to 50 nm. The 10 nm LT GaN layer was much less effective than the 50 nm LT GaN layer at reducing the [Ca] but resulted in a reduction of [Ca] in the QWs to ∼1017 cm−3, as compared to the reference sample that had ∼1018 cm−3 [Ca]. Room temperature photoluminescence (PL) was performed using 325 nm HeCd laser excitation and showed an increase in PL intensity at 405 nm of 25% for the sample with the 10 nm LT GaN interlayer, as compared with the reference.
With the intent of further reducing [Ca] while maintaining a smooth surface morphology, the LT-GaN layer was replaced by a superlattice (SL) with a period that consisted of 5 nm GaN grown at 600 °C and 20 nm GaN deposited at high temperature (HT), i.e., 820 °C. In addition, a low In flux (∼1 × 10−8 Torr) was continuously supplied during the LT growth to induce a surfactant effect, i.e., to promote the surface diffusion length.26 The sample structure is as follows: GaN template, 100 nm HT-GaN, 10× LT-GaN (5 nm)/HT-GaN (20 nm) SL, and 120 nm HT-GaN, and the smooth surface morphology on the order of 1 nm rms roughness was observed by the atomic force microscopy. The corresponding SIMS profile is displayed in Figure 3(a). As already seen, [Ca] dramatically increased at the regrowth interface and slightly decreased within the GaN layer grown at HT. There is a [Ca] peak for each LT-GaN layer in the SL due to a much higher incorporation rate. After the SL growth, [Ca] background in HT-GaN was below 1014 cm−3, close to the residual concentration measured in the MOVPE materials. However, [Ca] in LT-GaN is still high, which indicates that more SL periods are required to eliminate Ca atoms at the growing surface. From the SIMS profile, one could extrapolate 15 periods to allow for complete incorporation.
A careful analysis of the SIMS profile in Figure 3(a) allows a more quantitative analysis of Ca contamination. One can indeed estimate the initial Ca surface substrate coverage (θ0) and the magnitude of the surface segregation, which is well-captured by a phenomenological coefficient R.27 The model proposed in Ref. 27 assumes that during growth a fraction R of Ca atoms present at the growth front segregate at the surface of the next growing layer, while a fraction (1-R) is incorporated. The surface coverage θn and bulk concentrations [Ca]n of each monolayer (ML) can then be calculated as θn = θ0 Rn and [Ca]n = θ0 (1-R)Rn, after indexing the nth ML from the interface (n = 0). One can also deduce the amount of Ca that is incorporated over N MLs: QCa = . It follows that the surface segregation coefficient R can be calculated from the SIMS data: R with [Ca]1 and [Ca]N the concentrations of the first and Nth ML, respectively. Applying this model, we found very high R values: R (600 °C) = 0.967 and R (820 °C) = 0.9997. This explains why it is so difficult to get rid of Ca at the growing surface. We also estimated an initial surface coverage θ0 of 1.3 × 1012 atom cm−2, which is in line with the data obtained on the GaAs18 and silicon16 wafers. The calculated [Ca] profile taking into account R (600 °C), R (800 °C), and θ0 is reported in Fig. 3(b). Excellent agreement with the experimental SIMS data was obtained, which fully validates the substrate surface as the main origin of Ca contamination. One can also argue that [Ca] is very low in MOVPE materials because of a much higher growth temperature and therefore an R coefficient close to unity. Notice that this assumption might be valid for c-plane Ga polar surface but not for N-polar surface, or for non-polar and semipolar surfaces due to different surface kinetics.
The reduction of [Ca] in the MBE grown III-nitride layers should have a large impact on the performance of InGaN optoelectronic devices if Ca is indeed a source of SRH recombination centers. For further insight, we grew two LED structures with and without a 15× LT-GaN (5 nm)/HT-GaN (20 nm) SL with an emission at 405 nm. We note that the LED epilayers were grown on planar GaN/sapphire templates and that the structure lacks an electron blocking layer. The light extraction efficiency and injection efficiency are therefore not optimized in these devices. Figure 4 displays the I–V and L-I curves of both LEDs. Light extraction was expected to be similar for the two samples, which had similar, smooth surface morphology. The I–V characteristics were similar with same turn-on voltage. Note that the slight variation of the series resistance and contact resistance could be due to p-type doping level fluctuations from sample to sample. The striking result in Figure 4 is the much higher output power of the LED featuring an SL, which was 10× greater at low current injection compared to the standard LED. This higher efficiency is ascribed to a reduction of [Ca] in the InGaN/GaN MQW active region by an order of magnitude from 7 × 1016 cm−3 to 5 × 1015 cm−3, as confirmed by the SIMS experiments. In addition, the maximum external quantum efficiency (EQE) measured was twice as high for the sample with the reduced Ca, and the reference sample did not reach a maximum EQE with an increasing current up to 200 mA. In general, the effect of SRH recombination on LED efficiency can be modelled via the non-radiative recombination “A” coefficient in the so-called ABC model, with a higher A coefficient (shorter radiative lifetime), resulting in a lower peak efficiency, a shift of the peak efficiency to higher current density, and the absence of efficiency droop.28 While both the preliminary LEDs studied here show evidence of SRH based on these qualitative criteria, the SL LED shows an improvement consistent with a smaller A coefficient, and the picture of Ca as potential SRH recombination center. We would expect an even greater improvement in efficiency with further Ca reduction. Studies are currently ongoing to characterize the exact nature of Ca and related energy states in the GaN band gap, and a previous study by Krtschil et al.29 has indicated that Ca incorporation may be associated with the generation of compensating defects such as deep electron traps.
In conclusion, we have shown that the MBE grown (In,Ga)N layers are not contaminated with common transition metals; however, they systematically contain significant levels of a Ca impurity. Ca originates from the substrate surface and segregates at the growth front, with a highly temperature dependent incorporation rate, as confirmed via modelling of the [Ca] SIMS profile. The initial surface coverage was estimated to be ∼1012 cm−2, which is in line with the previous report on the GaAs and silicon wafers. We demonstrated that the Ca can be buried in a superlattice, allowing a reduction of [Ca] in InGaN/GaN MQWs. LEDs with lower [Ca] in the active region were fabricated, and output power was one order of magnitude larger than devices with higher [Ca], confirming the detrimental effect of Ca on LED efficiency. Thus, suppressing Ca contamination in InGaN/GaN QW structures could lead to a renaissance of MBE in the area of III-N optoelectronics.
This work was funded in part by the Solid State Lighting Program (SSLP), a collaboration between King Abdulaziz City for Science and Technology (KACST), King Abdullah University of Science and Technology (KAUST), and University of California, Santa Barbara. Additional support for N.G. and J.S. was provided by the DOE Solid State Lighting Program under Award No. DE-EE0007096.