The simulation, fabrication and optical characterization of InGaN/GaN MQW-LEDs grown by MOVPE over embedded photonic quasi-crystals (PQCs) are reported. Fully coalesced GaN layers as thin as 140 nm were grown over 1200 nm high air-gap PQCs using an intermittent pulsed/normal growth method. Simulations and angle-resolved photoluminescence measurements reveal there are strong interactions between the embedded PQC and a wide range of trapped modes as well as the dominant low order mode. The interaction with a dominant low order mode is most pronounced when the LED cavity above the embedded PQC has fewer guided modes.
1200 nm high embedded air-gap PQC overgrown with a 280 nm thick coalesced GaN epitaxial layer by MOVPE.
Photonic crystals (PhCs) have been widely investigated as a means of overcoming light trapping inside light emitting diodes (LEDs) 1–3. Their diffractive efficiency in extracting light critically depends on their overlap with the waveguide modes in which light is conventionally trapped inside the chip. Optimizing the epitaxial structure, the location and etch depth of the PhC presents a challenge complicated by any additional beam shaping requirements, for example enhancing the surface-normal emission of an LED. Whilst increasing the depth of a PhC etched into the surface of the chip offers a route to increasing the overlap, factors like etch damage and increased access resistance to the LED active region can undermine the effectiveness of this option.
Embedding the PhC into the epitaxial structure using epitaxial lateral overgrowth has been demonstrated as a potentially effective method for overcoming some of the limitations of surface PhCs 4–6. To date, work in this area has focused mainly on the interaction of embedded PhCs with the lowest order modes of the LED chip, to overcome their poor extraction efficiency, with little work done on the interactions with higher order modes.
This paper reports an investigation of the effect of epitaxial structure on the properties of air-gap photonic quasi-crystals (PQCs) embedded below the MQW region of a p-side-up LED. The interactions between the PQC and both low and high order modes are considered. PQCs have been used because of the reduced dependence of the emitted light intensity on azimuthal angle 7, 8.
The light extracting performance of a PhC can be characterized by its contribution to the loss experienced by those parts of the guided modes of the composite substrate-chip waveguide confined to the LED itself, i.e. where the optical power in the i-th mode is given by
In Eq. (1) Pci accounts for the coupling of light from the MQW, Γi is its optical confinement factor within the LED chip, is the wave vector component in the plane of the chip and its total loss coefficient is given by
All the terms on the right-hand side of Eq. (2) depend on the mode order, where γPhC is the extraction efficiency of the i-th mode by the PhC, α is the effective absorption coefficient experienced by this mode, and γs accounts for other scattering loss. The impact of scattering/absorption loss on the light extracting properties of PhCs has been described elsewhere by the authors 9, and the influence of absorption in the active layer has since been considered in detail by Matioli and Weisbuch 10.
From a bulk optics perspective embedding a PhC or PQC in the epitaxial structure has the same effect as adding a layer of lower refractive index. This changes the shape hence optical power distribution of the guided modes. A model has been proposed in which the trapped modes can be characterized as either cap layer modes largely confined to the active layer above an embedded PhC, buffer layer modes largely confined below the PhC and higher order modes which extend though the whole structure 6, 10.
However, the geometry of the whole structure does not always lend itself to such a simple categorization of the guided modes. To capture the interaction between all guided modes and the PQC, finite difference time domain (FDTD) calculations 11 have been performed to elucidate the interaction between embedded air-gap PQCs and all guided modes of the total LED structure.
Figure 1 shows the calculated light output as a function of the original direction of the light emission from inside a p-side up LED with a buried 450 nm pitch 12-fold symmetric buried PQC. The coalesced layer thickness was assumed to be 300 nm, which was capped by a 50 nm wide MQW, a 20 nm wide electron blocking layer and a 110 nm wide p-GaN layer. The height of the PQC is varied from 200 to 650 nm and the figure includes a comparison with an un-patterned LED having the same upper layer structure.
For emission directions close to vertical (90° ± ∼3°), the light output is little different from that of the un-patterned LED. However, as the emission angle drops outside this near-vertical range, the light output increases for emission directions in the range 65–87° (93–115°) for all PQC depths over that of an un-patterned chip and the contrast of the Fabry–Perot fringes (the oscillations in output for the un-patterned LED) is diminished. Reduced fringe contrast is evidence of a strong interaction with a PhC as the cavity loss is increased 8. The enhancement of γPhC along these emission directions corresponds to interactions between the PQC and the high order waveguide modes.
There is also a significant increase in γPhC for light emission from the MQW at angles in the range ∼40–65° (115–160°). This derives from interactions between the PQC and low order guided modes of the LED structure. The peak in light output at 40° (140°) corresponds to light extraction by the dominant cap layer mode identified in Ref. 6. The height of this peak decreases with increasing height of the PQC, but only slowly, confirming that efficient light extraction from the lowest order mode requires only a shallow PhC 6, 10. However, the enhancement in the light output relative to the control (lighter dashed line in Fig. 1) for other low order modes corresponding to emission directions in the range ∼45–65° (115–135°) and for high order modes is strongly sensitive to the PQC height. This must be taken into account in considering the total light extracting behaviour of the PQC.
Overall, the total light extraction is improved by 42–49%, dependent upon the PQC depth. This is a much larger than the γPhC∼ 10% values calculated for similar LED structures with a PQC etched into the top p-type layer 11. This implies that the buried PhC layer is able to interact more strongly with trapped modes in the heterostructure.
3 Practical implementation
N-type GaN-on-sapphire substrates pre-patterned with a 450 nm period 12-fold symmetric PQC by nano-imprinting were sourced commercially (Luxtaltek, Taiwan). A Ni etch mask was then formed by evaporation and a novel lift-off process 12, 13 prior to ICP etching the PQC using Cl/Ar plasma (Oxford Instruments ICP 180). Subsequent KOH etching at 80 °C was used to form more regular shaped air holes with vertical side walls. The insides of the holes were then passivated with a thin layer of SiO2 to prevent epitaxial growth on these surfaces during the overgrowth stage.
Fully coalesced GaN epitaxial layers were overgrown using a pulsed MOVPE technique 14. Figure 2 compares the measured and simulated optical reflectance of one of the embedded PQCs formed by this process. The reflectance curve contains two types of Fabry–Perot fringe: rapid fluctuations caused by interference between light reflected from the upper surface of the coalesced GaN layer and the interface with the sapphire substrate, and a slow variation due to reflections from the upper and lower surfaces of the coalesced layer. From the latter a coalesced GaN layer thickness of just 140 nm was inferred. This result confirms that ultra-thin coalesced epitaxial films can be achieved using pulsed growth and that the embedded PQC did not decouple completely the microcavities above and below it.
Figure 3(a) shows a SEM image of an embedded PQC covered with a ∼340 nm thick fully coalesced GaN layer. The AFM data in Fig. 3(b) shows the upper surface has a roughness of only 0.99 nm rms. The capability of the technology described here is demonstrated in the Abstract figure which shows a ∼1200 nm high PQC overgrown by a ∼280 nm thick fully coalesced GaN layer with a smooth, device ready surface. This compares favourably with the 50–100 nm PhC heights reported elsewhere 4, 6.
The thickness of the coalesced layer is known to have a significant effect on the extraction efficiency of an embedded PhC 4. To investigate this effect for embedded PQCs, overgrowths were performed such that the coalesced layer thickness varied from ∼300 to ∼1500 nm across 50 mm diameter wafers. The PQC pitch was 450 nm with 0.32 air fraction and its height was ∼1200 nm. LED structures consisting of 3× In0.03Ga0.97N/In0.16Ga0.84N quantum wells, 27 nm Al0.17Ga0.83N and 80 nm p-GaN were then grown by MOVPE over these templates. Reflectance measurements similar to those in Fig. 2 revealed the optical coupling between the LED cavity and the underlying buffer cavity was still strong. Thus, the LED chips are still multimode waveguides despite the 1200 nm height and large air fraction (0.32) of the PQC.
Figure 4(a) and (b) shows the far-field emission patterns of un-polarized 475 nm wavelength photoluminescence (PL) from the LED-MQW with ∼650 nm (a) and ∼460 nm (b) thick coalesced layers measured at room temperature. The coalesced layer thicknesses were estimated from reflectance spectra. The corresponding PL at a fixed zenith angle of 30° normalized to the estimated emission band is shown in Fig. 4(c) and (d), respectively. In both cases the far-field emission contains a rich array of periodic features arising from interactions between waveguide modes with the buried PQC. For the narrower upper cavity [Fig. 4(b) and (d)], particularly strong features appear in the emission with the same 30° periodicity as the PQC. These are ascribed to interaction between the PQC and the dominant mode of the upper cavity, in line with the findings of Ref. 10. These same features appear in the emission of the wider upper cavity device but at slightly different wavelength owing to the dependence of the effective index and hence of the dominant upper cavity mode on the coalesced layer thickness.
However, in the wider upper cavity device the interaction of this dominant mode with the buried PQC is significantly weaker and is superimposed on other emission features having the periodicity of the PQC [Fig. 4(c)]. Since the overlap of the dominant mode with the buried PQC will vary slowly with the coalesced layer thickness we ascribe these other periodic emission features with light from the MQW coupling strongly with more modes than in the LED with a thinner coalesced layer.
In both Fig. 4(c) and Fig. (d) the sharply resolved features are superimposed on diffuse but periodic background emission which extends over the whole emission band. This can arise from first order diffraction of the high order cavity modes, which have a closely spaced values and hence will give rise to diffuse yet periodic enhancement of the emission, or from second order diffractive interaction between the lower order modes and the PQC. The former is the favoured explanation as the reflectance measurements show that the decoupling of the lower buffer layer cavity is incomplete despite the 1.2 µm height and large air fraction of the PQC. Work is ongoing to clarify this.
In summary, it is shown that a wide range of trapped modes can make a significant contribution to the light extracting properties of a buried photonic crystal inserted in a LED even in thin-chip structures.
The authors wish to acknowledge support for this work from the UK Technology Strategy Board via contract DTI Project No: TP K2522J “NoveLELS” and from EPSRC via grant EP/I012591/1 “Lighting the Future”.
Dr. Philip Shields received his D.Phil in Physics from the University of Oxford, UK, in 2001. In 2004 he joined the wide band-gap semiconductor group at the University of Bath, and in 2012 he was appointed lecturer in the Electronic Engineering Department. He is involved in the growth and fabrication of III-nitride devices mainly for optoelectronic applications, with a particular focus on the use of nanostructures to improve device performance.
Duncan Allsopp is a Reader at the University of Bath, UK. He is a past Royal Academy of Engineering/Leverhulme Trust senior research fellow, and previously worked at Ferranti Electronics and British Telecom Research Laboratories. He currently leads the III-nitride research group at Bath.