UK Cracks GaN-on-silicon LEDs
The US Department of Energy believes that widespread adoption of LED lighting will require cheaper, more-efficient commercial devices. Both of these criteria are captured in its LED roadmaps, which include 150 lm/W lamps by 2015 that cost less than $5 per kilolumen.
Today LED lighting costs are more than twice that, with the cost of the die alone contributing nearly 50%. But chip costs could fall by an order of magnitude by switching to larger, lower-cost substrates that improve processing efficiency.
Our UK consortium – which is headed by RFMD (UK) and involves contributions from Cambridge University, Aixtron (UK), QinetiQ and Forge Europa – is working towards that goal with a government-funded project that started in April 2007. Backed by Â£3 million ($4.5 million), our effort is focused on the development of high-quality LEDs on 150 mm silicon. It has already produced devices with internal quantum efficiencies (IQEs) of nearly 40%.
Development of the growth processes for producing good-quality LED material is being led by Cambridge University and supported by growth and processing at QinetiQ, which uses its broad semiconductor expertise to validate material developments prior to process scale-up for LED manufacture at RFMD. When device manufacture commences, marketing will be carried out by LED applications specialist, Forge Europa.
Silicon substrates have been selected for nitride growth because they combine stability at typical growth temperatures with low cost and are available in diameters of up to 300 mm with a surface suitable for epitaxial growth. Nitride-on-silicon epiwafers also enable processing with standard manufacturing tools employed in the silicon industry, leading to cost-effective production of chips that can be bonded and converted into packaged LEDs.
Although silicon has key advantages over the existing substrate materials used for LED manufacture – sapphire and SiC – it has an Achilles heel: high lattice and thermal expansion mismatches with GaN. Growth directly onto silicon produces tension in the film from the moment that GaN is deposited at a typical growth temperature of 1000 °C and this is magnified as the wafer cools to room temperature, due to differences in thermal expansion. Unless managed correctly, this strain can even crack the GaN film. More often it causes the wafers to bow, making them awkward to process in automated tools that are designed for silicon processing. In comparison, when nitrides are grown on SiC the thermal expansion and lattice parameter oppose each other, while on sapphire both terms cause compression of the GaN film, but this does not cause cracking.
The large lattice mismatch between GaN and silicon leads to a relatively high dislocation density for the epiwafers. Nitride LEDs are resilient to surprisingly high levels of defects, but the IQE of blue LEDs declines as dislocation densities exceed 109 cm–2.
Silicon s other weakness is its reaction with gases in the growth chamber during the initial stages of growth. Defects form on the surface, leading to a morphology unsuitable for subsequent GaN growth.
The growth process developed at Cambridge University for producing blue-emitting LEDs addresses all of these issues. Epitaxial structures are produced in an MOCVD Aixtron/Thomas Swan close-coupled showerhead reactor that accommodates a single 150 mm wafer (or multiple 2 inch wafers) and is equipped with in situ instruments for measuring wafer bow and temperature. Growth on 150 mm silicon (111) includes the deposition of a complex buffer structure to control strain and wafer curvature. This is followed by the growth of a multiple quantum well (MQW) LED structure with InGaN wells and GaN barriers that emits at 460 nm, and a magnesium-doped p-type GaN (figure 1a).
Substrates are prepared for growth by annealing in hydrogen to remove the native oxide layer and create a terraced, reflowed silicon surface. Growth begins with deposition of an AlN nucleation layer to ensure that the silicon surface does not degrade, followed by a complex buffer structure. Careful control of the composition and thickness of this buffer balances its strain with that induced by thermal expansion mismatch on cooling from the growth temperature.
To improve LED performance, GaN and AlGaN layers are added on top of the buffer to reduce dislocation density. Inserting a SiNx layer, a technique used for nitride film growth on sapphire, further cuts threading dislocation density.
In situ tools
Continual monitoring of wafer temperature and bow holds the key to successful, reproducible growth of flat, uncracked material. The substrate temperature in the reactor at Cambridge University is profiled with an Aixtron Argus tool and a LayTec Epicurve provides real-time wafer-bow measurements.
The silicon that we use has a slight convex bow, which switches to a concave shape after heating and in situ annealing because the bottom of the substrate is then at a higher temperature than its top surface (figure 1b). Concave bowing increases with the addition of the AlN nucleation layer, but a convex profile returns with the growth of the buffer and the silicon-doped GaN layer, which introduce increasing levels of compressive stress. The growth of quantum well and barrier layers produces detectable, small changes in curvature, before the wafer becomes even more convex as the reactor temperature is increased for magnesium-doped GaN deposition. Optimizing the magnitude of this bow during buffer growth can produce a perfectly flat wafer after cooling, by providing a match for the tensile stress of the film, resulting from thermal expansion coefficient differences between GaN and silicon.
Development of the growth processes has led to production of 150 mm diameter epiwafers with a surface height variation across their full diameter below 50 µm. These wafers are suitable for processing with RFMD s high-volume production equipment.
Bow must be introduced at the growth temperature to ensure a flat wafer on cooling, but it can also lead to significant temperature variations across the surface caused by differences in the distance between the substrate and the susceptor. Temperature variations are detrimental to InGaN LED growth because they alter the indium content in the quantum well and its emission wavelength. Fortunately we can detect these temperature variations with our Argus profiler and minimize them by adjusting the power delivered to three heater zones.
Cross-secional transmission electron microscopy (TEM) images reveal that the low dislocation density in the device layers is due to a SiNx layer that causes these defects to bend over and annihilate (figure 3). This also occurs at the AlGaN/GaN interface, thanks to a compressive stress that results from the larger in-plane lattice parameter of GaN compared with AlN.
The dislocation density in the epiwafers has been assessed with plan-view TEM images using diffraction conditions that reveal all of the dislocations. These images, and those taken with an atomic force microscope after exposing the surface to a silane flux at 860 °C to highlight pits, give values of less than 109 cm–2 for the best GaN-on-silicon material.
The IQE of the material has been evaluated by temperature-dependent photoluminescence measurements at the University of Manchester. This approach, which makes the assumption that non-radiative recombination becomes negligible close to absolute zero, gives a room-temperature IQE of almost 50%. Similar structures grown on sapphire with a dislocation density of 108 cm–2 have typical values of 70%, indicating that the higher dislocation density of nitride material on silicon may not be a major problem in the production of high-performance LEDs on this substrate.
We have gone on to fabricate LEDs at QinetiQ. Etching formed a mesa that exposed the n-type GaN layer. Deposition of a Ti/Al/Pt/Au alloy followed to produce the n-type contact and the p-type contact was created through addition of semitransparent, annealed NiAu and a thicker gold contact pad.
Our best 0.5 mm × 0.5 mm LEDs have very similar current-voltage characteristics to our GaN-on-sapphire controls, including a turn-on voltage of about 2.5 V (figure 4). The light output from the top side of both devices has been measured using the same optical arrangement. The sapphire-based controls produce about twice the output of LEDs grown on silicon. Taking into account the highly absorbing silicon substrate, the IQE of LEDs on silicon is 37% according to calculations based on measurements of the total light emitted in the forward direction.
GaN-on-silicon LEDs are still in their infancy, but our initial results are very encouraging. Removing the silicon substrate to prevent light absorption will not be an issue for manufacturers because high-brightness LED production often involves flip-chip mounting and substrate removal. With state-of-the-art packaging and appropriate phosphors, our current devices could deliver 70 lm/W, based on a comparison with LEDs on sapphire. Thanks to the affordability of silicon, this means that GaN LEDs that are grown on silicon will approach the Department of Energy s 2012 target for the cost per kilolumen. This makes our emitters genuine contenders for solid-state lighting and their performance will only get better as in situ growth techniques improve, leading to even higher quality material.
K Cheng et al. 2008 Appl. Phys. Lett. 92 192111.
C J Humphreys 2008 Solid-State Lighting, MRS Bulletin 33 459.
M J Kappers et al. 2007 J. Cryst. Growth 300 70.
A Krost et al. 2002 Phys. Stat. Sol. (a) 194 361.
View pdf of article