1. Introduction

As there is a great demand of high brightness InGaN/GaN based light emitting diodes (LEDs) for solid state lighting and backlight modules of flat panel displays, understanding the fundamental physics of GaN epi-structures and its influence on the device performance have therefore become an interesting research topic in recent years. For a GaN based material structure grown on the c-plane sapphire substrate, the peak wavelength and efficiency of emission depend on the built-in internal electric field arisen from the divergence effect of spontaneous and piezoelectric polarizations in the quantum wells (QWs). The strong built-in electric field can cause low internal quantum efficiency due to a reduced overlapping between the electron and hole wave functions. Such internal field can further introduce a quantum-confined Stark effect (QCSE) in which the interband transition energy between the QW ground states decrease quadratically with the field strength. For an InGaN-based LED under the current injection condition, there are two competing mechanisms that determine the QW emission wavelength. That is, the red shift of the spectrum due to carrier induced bandgap renormalization tends to be counteracted by the blue shift due to charge screening. The carrier-induced emission peak wavelength shift in the c-plane grown InGaN LED therefore becomes a concern for device applications. To overcome such an obstacle, several methods have been proposed to release the strain induced piezoelectric effect in the nitride based QWs. For example, the effect of QCSE can be mitigated by growing a p-InGaN layer on top of the active region [1–2], or by inserting a pre-strain layer prior to the growth of QW structures [3–5]. Furthermore, it can also be reduced by growing GaN based epi-layers on patterned sapphire [6–8], or m-plane sapphire substrates [9–10].

On the other hand, reduced spectral shifting in the PL (photoluminescence) wavelength of the GaN-based QW has been observed by using low-dimensional structures such as nanorods, nanocolumns, nanotips, nanoposts et. al [11–12]. There are also additional advantages inherent to the nanostructure approach. For example, the low-dimensional light emitting structures have the potential of high emission efficiencies due to the quantum confinement effect [13]. There are generally two approaches, viz. bottom-up and top-down, that have been applied to fabricate the GaN based nanostructures. For the bottom-up method, the common approach is to use metal nanoparticles such as Fe, Au and Ni as catalysts in the vapor-liquid-solid (VLS) growth process to control the critical nucleation and the subsequent elongation steps for nanowire formation [14–16]. There are also methods to grow GaN nanowires directly without any extra existence. For example, self-assembled dislocation-free vertical GaN pillars are demonstrated on Si surface [17]. As for the top-down approach, nanorods can be realized by using nano-scale etch masks followed dry etching to achieve a desirable pattern transfer [18–19].

Among many reports related to the synthesis and characterization of nanostructures, only few of them investigated the device properties of light emitting nanorod arrays since p-type ohmic contacts are difficult to be deposited on nanorod tips. Chiu et al reported the fabrication of InGaN/GaN nanorod LEDs using inductively coupled plasma reactive-ion etching (ICP-RIE) via self-assembled Ni nanomasks followed by photo-enhanced chemical (PEC) wet oxidation process [20]. However, it has been noted that the light emission characteristics in the strained InGaN QWs activated by the electrical pumping are different from those of optical pumping due to different carrier transportation and recombination mechanisms between quantum wells [20–21]. On the other hand, although the mechanism of strain relaxation has been previously observed in the low-dimensional GaN/InGaN structures [22–23], the corresponding effect on the emission peak wavelength from nanostructure LEDs has not yet been reported.

In this work, we report the fabrication and electroluminescence (EL) characterization of InGaN nanorod LED and structures by current injection and Raman scattering. The p-type contact electrodes of nanorods are connected in parallel by inserting a spin-on-glass (SOG) space layer between rods to avoid electric shorting to the n-type GaN layer. In our nanorod device, the peak wavelengths of EL spectra remain nearly as a constant as one increases the injection current. This observation indicates a balanced mechanism of carrier induced bandgap renormalization been offset by the screening of the QCSE and the band filling effect. It is also related to a reduced strain effect in these nano-structured InGaN QWs. The latter phenomenon is further analyzed by Raman scattering measurements by comparing the spectra of conventional LEDs and nanorod LEDs.

2. Fabrication of Nanorod LED Arrays

The LED samples were grown by metal organic chemical vapor deposition (MOCVD) on a c-plane sapphire substrate. The epi-structure was shown in Fig. 1(a) and composed of a 25nm GaN buffer layer, a 2µm Si doped n-type GaN layer, a five period of In_xGa_1-xN/GaN multiple quantum well (MQW) structure in which each period is 17nm, and a 160nm Mg doped p-type GaN layer. The average indium (In) composition x was around 0.2 for a quantum well of 3nm thickness. The fabrication process of the nanorod LED was illustrated in Figs. 1(b)–1(d). First, a p-type mesa area was defined by the use of photolithography and ICP-RIE etching. We then deposited a 200nm thick SiO₂ film by PECVD (plasma enhanced chemical vapor deposition) across the sample but with vias left open on the p-type mesa.

 

Fig. 1. (a). The as-grown LED structure. (b). Illustration of a SiO₂ layer deposited by PECVD. The SiO₂ thin film is served as the etch mask and the via hole is for silica nanoparticle coating at a latter step. (c). Formation of nanorods by spin coating a monolayer of silica nanoparticles followed by ICP dry etching. (d). Device profiles of nanorod LEDs. SOG (in pink) is coated between rods as a space and sidewall passivation layer.

Download Full Size | PDF

The latter approach allows a spin-cast mono-layer of silica nano-particles to be prepared as the etch mask for forming a nanorod structure on the p-GaN opening by dry etching. We note the silica particles were self-organized and firmly bounded to the GaN surface by the van der Waals force. [24]. Here the SiO₂ layer is meant to prevent surface and sidewall damage on the GaN materials at the subsequent ICP etching step. To investigate the effect of strain relaxation on the GaN based nanorod LED structures, a monolayer of self-organized silica nanoparticles of diameter 100nm and 50nm, respectively, was employed as the etch mask. Our ICP dry etching condition consisted of a gas mixture of SiCl₄/Cl₂/Ar (1/20/15 sccm) and source power 200W and bias power 100W at 13.56MHz. The etching rate was about 400nm/min. After completing the etching process, the sample was dipped in buffer oxide etchant (BOE) to remove the silica nanoparticles and SiO₂ layer on sample surface. The SEM (scanning electron microscopy) images in Figs. 2(a) and 2(b) indicate that nanorod structures with diameter 100nm and 50nm, respectively, and a height of 300nm can be achieved. In order to prevent the occurrence of electric shorting between metallic contact to the p-type and n-type layers, we refilled the nanorod spacing by a 300nm-thick SOG layer using a reflow process at 130°C for 10 mins. The insulating property of SOG can also serve to compensate the sidewall defects as induced by the ICP etching. After the passivation, we then deposited a thin Ni/Au (5nm/5nm) metal layer to connect each of the nanorod LED units in parallel. Finally, thick Ni/Au (10nm/120nm) and Ti/Au (10nm/120nm) were deposited as the p-type and n-type probe contact pad, respectively. The finished GaN nanorod LED structure was illustrated in Fig. 1(d), with a cross section view of a100nm-diameter nanorod device shown in Fig. 2(c). From the image, the p-type current spreading layer was shown on top of the nanorod structure along with the SOG space layer.

3. Results and discussions

3.1 Characterizations of nanorod LEDs

The electrical properties of nanorod LED devices were characterized by using Agilent 4155C semiconductor parameter analyzer. The current-voltage curves of LED arrays with rod diameter 50nm and 100nm are shown in Fig. 3. We note both I–V curves do exhibit the rectifying behaviors in the forward bias regime. However, there exists noticeable leakage current at reverse bias. The latter can be ascribed to the surface defects caused by the ICP dry etching. At a bias voltage of -5V, the reverse leakage current of a 50nm and a 100nm nanorod LED is 5.2 mA and 1.2mA, respectively. The higher leakage current of the smaller-size nanorod array is due to a larger sidewall area of such a device.

 

Fig. 2. (a). An SEM image of 100nm-diameter (a) and 50nm-diameter (b) nanorods. (c) The cross-section SEM image of nanorod LED array.

Download Full Size | PDF

 

Fig. 3. I–V curves of the 100nm and 50nm nanorod LEDs.

Download Full Size | PDF

Electroluminescence (EL) spectra of the conventional (planar) and nanorod LEDs are shown in Figs. 4(a)–4(c). Referring to Fig. 4(a), we note a spectral blue shift from 478nm to 461nm can take place on a conventional LED as one increases the injection current from 25mA to 100mA (corresponding voltage drop 3V to 8.5V). This blue shift phenomenon can be ascribed to the screening of QCSE and band filling of localized states. The blue shift can also be attributed to the low miscibility of GaN and InN that leads to indium compositional fluctuation and indium-rich nm scale cluster formation in InGaN [12]. The latter mechanism of composition fluctuation and cluster structures indeed are controlled by the strain condition in QWs. For our nanorod LEDs, the material strain in the planar InGaN QW structure can be relaxed in the nanorod formation process. This can therefore lead to a mitigation of the effect of QCSE. The latter is reflected in the EL spectra of the 100nm [Fig. 4(b)] and 50nm [Fig. 4(c)] nanorod LED, where only minor ranges for peak wavelength variation, e.g. 478~480nm and 474~476nm, respectively, were observed at an injection current level between 25mA and 100mA (extending bias voltages between 4.8V and 7.6V for 100nm rods, and 4.6V and 6.6V for 50nm rods) [see Fig. 4(d)]. As compared with the significant blue shift in conventional structures, the piezoelectric field is suppressed on nanorod LEDs since the strain in InGaN layers is relaxed. Or it can be attributed to the balance between effects of carrier screening and QCSE. Also, by comparing the EL spectra of 50nm and 100nm nanorod LEDs, the emission peak wavelength of 50nm nanorod LEDs is lower than that of 100nm nanorod devices. Since both samples were fabricated from the same planar LED epi-structure, we believe the shorter peak wavelength of 50nm nanorods is due to the effect of quantum confinement.

 

Fig. 4. EL spectra of a conventional LED (a), a 100nm nanorod LED (b) and a 50nm nanorod LED (c). (d) Comparison of peak wavelengths of conventional, 50nm and 100nm nanorod LEDs.

Download Full Size | PDF

Figure 5 shows the emission images of 100nm nanorod LEDs. When the device is biased at 20mA (4V), point source light emission can be seen from Fig. 5(a) as some rods are brighter than the others, which is mainly associated with the non-uniformity of contact resistance. As the injection current is increased to 30mA (4.5V) [see Fig. 5(b)] and 45mA (5.5V), light emission can be observed from more and more rods and point light sources become brighter.

3.2 Strain Relaxation Analysis of Nanorod structures

Since the QCSE phenomenon is related to the piezoelectric field and the strain between the mismatched layers, we next carry out the Raman scattering measurement to verify the strain of those devices. Figure 6 shows Raman spectra of the planar and nanorod epistrucutres. The nanorod epi-structures were obtained from etching the planar LED epi- material using either 50nm or 100nm silica nanoparticles as the mask and to the same 300nm depth as nanorod LED structures. In Fig. 6(a), two phonon modes are identified from Raman spectra. The peak near 569 cm^-1 is from the E₂ ^H mode of GaN [25–26], and the shoulder line around 560 cm^-1 is the E₂ ^H mode of InGaN [27]. The intensity of Raman scattering spectrum of the InGaN/GaN nanorods is much stronger than that of the planar MQWs structure. Moreover, the closeup view of E₂ ^H phonon mode of InGaN in Fig. 6(b) shows a lower wave number for InGaN/GaN nanorods than the planar structure.

 

Fig. 5. Light emission of a 100nm nanorod LED with an injection current 20mA (a) 30mA (b) and 45mA (c).

Download Full Size | PDF

 

Fig. 6. (a). Room-temperature Raman scattering spectra of the planar, 100nm- and 50nm-nanorod structures. (b). Closeup view of the InGaN E2 modes of these three structures under comparison.

Download Full Size | PDF

The low Raman phonon shift results from either nanorod size effect or strain relaxation in the nanostructure, or both [27–28]. Since the 50nm and 100nm diameter nanorods are not small enough for the effect of phonon confinement to be observed, the Raman shift is due to the strain relaxation. While the Raman mode shifted to a higher frequency represents the compressive strain, the lower frequency shift indicates the tensile strain. Furthermore, the phonon mode of 50nm nanorods shows a slightly lower shift than 100nm nanorods. It implies that, with smaller nanorods, the compressive strain between InGaN and GaN is further relaxed, which results in a better lattice match in nanorod structure. Therefore, the Raman scattering measurement suggests strain relaxed nanorod light emitting devices are achieved with a nearly constant peak wavelength at an injection current between 25mA and 100mA.

4. Conclusions

We present a practical process to fabricate InGaN/GaN MQW structure using silica nanoparticle nature lithography. The EL peak wavelength occurs in the range of 478~480nm and 474~476nm for 100nm and 50nm nanorod LEDs, respectively, with injection currents between 25mA and 100mA. As compared with the significant blue shift in planar structures (from 478nm to 461nm), the piezoelectric field is suppressed on nanorod LEDs since the strain in InGaN layers is relaxed. We also carried out Raman measurement to study the strain relaxation of the nanostructures. The Raman shift of nanorods is lower than that of planar MQW structure. It indicates strain relaxed nanorod light emitting devices are achieved with a nearly constant peak wavelength.