Reflecting layer boosts LED brightness

The silicon substrate is an interesting contender for nitride LED growth. On the one hand it is cheap and widely available in large diameters, which gives it the potential to cut LED costs and spur device deployment. Doped substrates should lead to superior device architectures and silicon is also easy to cleave, which simplifies chip fabrication.

However, these strengths have to be weighed up against high lattice and thermal mismatches between silicon and GaN, which create a tensile stress in the epilayers that degrades their quality and reduces internal quantum efficiencies. Silicon s strong absorption of blue and green light is another drawback, which makes substrate removal essential for making high-brightness emitters. This requires additional processing steps, which have to be capable of transferring large areas of nitride material.

The ideal way forward is to develop an approach that delivers the benefits of silicon, while avoiding its pitfalls. Fortunately at Singapore s Institute of Materials Research and Engineering (IMRE) – which is a member of the Agency for Science, Technology and Research (A*STAR) – our team has identified a platform that goes a long way to satisfying these demands: silicon-on-insulator (SOI) substrates. Armed with this composite we have produced LEDs that benefit from strong reflectivity at visible wavelengths, which stems from the high refractive index contrast between silicon and SiO2. This improves the device s light output, which has received an additional boost through the introduction of photonic crystal structures.

GaN-on-SOI LEDs also promise to deliver a relatively high manufacturing yield. More GaN die can be extracted from epiwafers based on SOI than sapphire because these substrates are available in larger diameters. However, a switch to growth on material with a diameter of 200 mm or more may require the development of new MOCVD tools.

One important question that hangs over SOI is its price – it is five times that of silicon (111). But 150 mm SOI substrates are still cheaper than equivalently sized sapphire and prices should fall as silicon device manufacturers purchase more silicon implanted oxide (SIMOX) SOI substrates with diameters of up to 300 mm.

We have built devices that emit in a spectral region known as the green gap, which is a range of wavelengths where it is very challenging to produce LEDs with high quantum efficiencies from either the InGaN family of alloys or the AlInGaP material system. Devices emitting efficiently at deep-green wavelengths (530–580 nm) are highly desirable because they can improve the color reproduction of displays based on red, green and blue LEDs.

The growth of nitride material on foreign substrates demands minimization of wafer bow, which is often addressed by introducing additional strain-balancing layers. Over the last nine years researchers at the Otto-von-Guericke University, Magdeburg, have been developing growth technologies for this purpose and they can now produce flat GaN-on-silicon epiwafers for LEDs and FETs. This success has encouraged the launch of a university spin-off, Azzurro Semiconductors, which now manufactures large GaN-on-silicon epiwafers.

We are currently working together to develop epitaxial structures on large SOI substrates and our goals are improvements in GaN material quality, elimination of bowing and better device performance. We have also processed InGaN/GaN LEDs from epiwafers grown at Azzurro. These MOCVD-grown devices are the first blue-green and deep-green LEDs to be produced on SIMOX SOI wafers.

The epiwafers feature a thin silicon layer on top of the oxide that creates a compliant platform for nitride epitaxy. This allows the strain produced during growth to be shared between the silicon layer and the lattice-mismatched GaN layer that is deposited during LED growth.

We have produced a range of LEDs using silicon overlayers between 45 and 200 nm thick, and a 150–375 nm thick buried oxide. A high-temperature AlN buffer layer is deposited onto this platform, followed by an aluminum-rich intermediate layer and n-type silicon-doped GaN layers with three strain-compensating low-temperature AlN interlayers. A full LED structure completes the growth, which comprises a 2.70–2.75 µm silicon-doped GaN template, a five-period InGaN/GaN multi-quantum well and a 100 nm thick, magnesium-doped GaN layer.

Material quality has been determined with high-resolution X-ray diffraction, Raman spectroscopy and transmission electron microscopy (see box "Characterizing GaN-on-SOI structures"), before processing the epiwafers into devices at the Science and Engineering Research Council nanofabrication and characterization facilities at IMRE. 300 µm × 300 µm square-shaped mesas were formed by multi-mask photolithography and inductively coupled plasma etching, and BCl3 and Cl2 gases enabled dry etching of GaN and exposure of underlying n-type silicon-doped GaN. The addition of a Ni/Au-based p-contact and Ti/Al-based n-contact completed device fabrication.

Current-voltage and electroluminescence tests have been used to assess device performance (figure 1). GaN-on-SOI devices absorb far less light than our equivalents grown on silicon (111), thanks to reflection by the SOI structure. However, even with this benefit our green and deep-green LEDs only emit an average optical power output of 1.1 mW at 20 mA. But substantial improvements in radiant flux should be possible with suitable packaging, full encapsulation and surface roughening of the die.

Our 550 nm, deep-green LEDs have a striking characteristic – strong interference peaks (figure 2). This results from the SOI substrate reflectivity and multiple reflections at the AlN/SOI/SiO2/silicon (111) boundaries. We have modeled this behavior and concluded that the emission profile is governed by substrate reflectivity changes and reflections at the top GaN/air interfaces. The emission of our deep-green LEDs extends to 600 nm and beyond, which suggests that they could provide a phosphor-free, broadband light source. Maximizing light output Like many other researchers, we are developing processes for improving LED light extraction, including efforts involving "nanosphere" lithography, reactive ion etching and, most recently, nanoimprint lithography. Research in this direction includes the development of lithographic processes for making two-dimensional photonic crystal structures that enhance extraction efficiency while controlling the direction and polarization state of the emitted light. Photonic LEDs can boost output by allowing photons to escape via multiple scattering steps from rough surfaces.

Our photonic crystal structures are created by depositing a SiO2 mask over the LED and writing a pattern in this layer with electron-beam lithography. Reactive ion etching, followed by inductively coupled plasma etching into the nitride layers, defines the photonic crystals in our green and deep-green LEDs. We have investigated a variety of circular hole arrays with different diameters, including designs that arrange the holes in square and circular patterns.

The most promising design features nanocones on p-type GaN surfaces, which should increase the output of deep-green LEDs emitting at around 550 nm (figure 3). Photoluminescence studies show that the nanocones significantly increase light output, with an enhancement that depends on the internal angle of the cones. Shallow etching improves the outcoupling more than deeper etching, which penetrates into the active region. The latter approach causes plasma damage, but even this leads to a four-fold enhancement in photoluminescence intensity (figure 4).

The development of photonic crystal LEDs on SOI will continue and we have set ourselves the goal of producing electrically driven devices that triple the power output compared with planar GaN equivalents on bulk silicon substrates. Another target is to scale up the nanoimprint lithography process and we are also interested in developing phosphor-free white-light LEDs that generate their emission through careful control of the wavelength-dependent reflectivity of the substrate.

A key strength of the SOI platform is its suitability for forming high-brightness, thin LED chips through simple sacrificial etching of the substrate. The device can be transferred to a high-thermal-conductivity, high-reflectivity platform to produce bright LEDs that could feature advanced light-extraction technologies. A similar process is used to make vertical blue and white LEDs from epitaxial material that is grown on sapphire. However, this requires laser lift-off to free the device from the substrate, which is more expensive than etching.

If SOI substrates are to enjoy commercial success in the manufacture of nitride LEDs, then this platform will have to demonstrate its benefits over a cheaper alternative – silicon – which can be separated from the device by lapping, grinding and plasma flattening. The battleground will be epitaxial quality and it is impossible to predict which type of platform will triumph – both technologies are still in their infancy. However, the more suitable platform will emerge as development moves to growth on 200 mm and even 300 mm material, where wafer bowing will present tougher challenges. But even if SOI loses out to silicon for this particular application, the uptake of SOI for microelectronics, nanoelectronics and the fabrication of optoelectromechanical systems and sensors, could lead to the manufacture of large, integrated chips that benefit from an on-board LED light source. Further reading A Dadgar et al. 2006 J. Cryst. Growth 297 279.
T Detchprohm et al. 2008 Physica Status Solidi (c) 5 2207.
S D Hersee et al. 2002 IEEE J. Quantum. Electron. 38 1017.
H Sato et al. 2008 Appl. Phys. Lett. 92 221110.
S Tripathy et al. 2007 Appl. Phys. Lett. 91 231109.
S Tripathy et al. 2007 J. Appl. Phys. 104 053106.   

View pdf of article