Inverted epi - Bolstering carrier capture in nitride quantum wells

LEDs only deliver strong performance when they have high internal quantum efficiency – a high ratio of photons generated to carriers injected. However, ensuring that this is the case in nitride LEDs operating over a wide current range is very challenging because in this class of device the internal quantum efficiency tends to plummet as the drive current is cranked up.  Uncertainty surrounds the cause of this efficiency decline that goes by the name of droop. Its origin is the subject of fierce debate, because understanding what causes it will help to spur the fabrication of droop-busting LED architectures that can underpin a solid-state lighting revolution. At present, some groups are attributing the decline in internal quantum efficiency to various Auger mechanisms, while others are blaming defect recombination or a leakage current. However, no one is disputing that the inclusion of an AlGaN electronblocking layer (EBL) substantially improves the performance of GaN LEDs.  One way to understand the benefits of the EBL is to first see how it works in a typical LED structure (see Figure 1). In this particular device, just like any other LED, efficiency is maximised by injecting as many electron and holes as possible into the quantum wells (QWs) and enabling incredibly efficient recombination in that trench. However, because electrons have a smaller effective mass and a higher mobility than holes, there is high likelihood that they cross the QW region and reach the p-doped region, rather than recombining to emit light in the QW. This current leakage does not generate any useful photons, and most of the recombination outside the QW is non-radiative.  To prevent electrons escaping over the QW region, engineers can insert a p-type AlxGa1-xN EBL to impede the progress of these carriers: The larger band gap of AlxGa1-xN acts as a road-block to electrons travelling outside the QW region. However, in nitride materials holes are about ten times heavier than electrons, and their mobility is roughly 20-30 times lower. Consequently, holes rarely cross the MQW region, leading to nonuniform distributions of hole density in the QWs and ultimately a low internal quantum efficiency. Improvements in this key figure of merit are possible by optimising the thickness, doping density and aluminium composition of the EBL. Exposing the perfect combination by experimental efforts would be very costly and timeconsuming, and a modelling-based effort has far more appeal. Such an approach is possible using numerical tools that we have developed at Crosslight Software, which is located in Burnaby, British Columbia, Canada. Our software, which has been used by many academic and industrial researchers to design the active region and passive layers of LEDs, determines the carrier transport and optical generation in these devices using multi-dimensional finite-element analysis. All salient features for nitride semiconductors are catered for, including polarization charge at heterojunctions and the influence of different crystal orientations on QW properties. The core of the LED – the multi-quantum well, EBL and contact layer – is usually designed with one-dimensional simulations; two- and three-dimensional simulations tend to be employed for uncovering and understanding issues related to packaging, such as thermal effects, current spreading and optical extraction. We have simulated LED performance for a range of devices with different AlGaN EBLs. These efforts reveal that increases in the aluminium composition and the thickness of the EBL can block electrons more effectively. But there’s a catch – these measures impede hole transport, and the upshot is a degradation in LED efficiency. One theoretical solution to this problem is to ramp up the p-type doping density of EBL, a step that effectively increases the barrier for electrons and lowers that for holes. This is impractical, however, because it is difficult to obtain very high p-doping concentrations in GaN-based materials. The energy level for the commonly used acceptor dopant, magnesium, is very deep – at room temperature only about 1 percent is ionized and contributing to the hole density. Adding substantial amounts of magnesium into the structure is very difficult, and even if this were possible, it would degrade the device.  Debdeep Jena and co-workers from the University of Notre Dame, Indiana, have uncovered a possible way to overcome this hole doping issue. They have shown that an AlGaN layer with a properly graded aluminium composition can induce hole doping due to the intrinsic polarization of these materials. This technique promises to circumvent many of the difficulties associated with magnesium doping because polarization-induced hole doping is not thermally activated.

Traditionally, LEDs are fabricated by growing an n-doped region on top of the substrate and adding QWs, an EBL and a p-doped region (see Figure 1). This configuration has a major downside: The electric field that stems from interface polarization charges between the quantum barrier and the well sucks the carriers out of this region, hampering efficient carrier capture in the well and leading to a high leakage current.

We have shown that a novel, p-side down LED architecture can address this issue. Such a device, which features a p-type doped region adjacent to the substrate and an ntype region on top, operates with reversed current injection compared to the conventional structure (See Figure 2 fo details of this LED’s bandstructure). This design has two benefits: Carrier capture is now more favoured than their escape, thanks to a switch in the direction of current flow; and hole injection is improved without resorting to excessively high levels of magnesium doping, because by putting the EBL layer at the bottom of the structure, it can be graded in a way that enables polarization-induced doping to increase hole density. Our simulations show that this inverted architecture enables a hike in the LED’s output power, as well as a cut in its forward voltage.  Unfortunately, realizing such gains in real devices is not trivial. Growing high-quality p-side down devices is tricky due to diffusion of magnesium impurities, which are the most common p-type dopants associated with MOCVD. It should also be noted that many of the techniques for producing p-side down devices, such as lift-off, will not deliver the results promised by our simulations, because they flip the entire crystal upside down – this fails to reverse the direction of the current injection with respect to the growth orientation of the crystal. Nevertheless, given the rapid improvements of epitaxial technology in nitride compounds, we are optimistic that high-quality p-side down devices will make their debut in the next few years.