Uncovering LED's best-known secret: Feel the power inside

Efficiency droop is the main reason that has prevented GaN-based LEDs from making great strides in high current density, an important emerging application area. But researchers at RPI say this shortcoming can be overcome by using a polarity-matched epitaxial structure.

LED manufacturers are currently focusing on new markets such as automotive headlights, large-screen displays and general lighting. To some extent, lowering prices can help LEDs enter these areas, but this alone is not enough. LED chips also need to achieve high efficacy at high drive currents. This also means that the widely concerned issue of power droop needs to be addressed.

Comparing the output of electrically and optically excited LEDs can help RPI researchers study the causes of power degradation in GaN-based devices.

Power droop specifically refers to the phenomenon that the efficiency of blue, green and white LEDs decreases as the current increases. GaN-based LEDs typically reach peak efficiency only at a current density of 10A cm-2, and the efficiency drops to half of the peak at a current density of 100A cm-2. This has attracted everyone's attention, and today's high-brightness chips need to work efficiently at current densities much greater than 10A cm-2.

This is obviously a very important problem, and it has attracted the interest of a large number of researchers in industry and academia around the world, including our team at Rensselaer Polytechnic Institute in Troy, New York, who have spent the past few years exploring the root causes of efficiency droop. We believe we can find the root cause and overcome it with a fundamentally different LED design.

We worked with Mary Crawford’s group at Sandia National Laboratories in New Mexico to investigate the cause of the power droop problem. We focused on the effect of dislocation density on LED efficiency and found that dislocations reduce efficiency at low current densities, but do not affect the efficiency droop at high current densities.

At low current densities, carriers are typically lost in a trap-assisted process, known as SRH recombination, which becomes more severe with increasing dislocation density. Increasing current density initially improves efficiency by enhancing spontaneous emission, but as current increases further, another competing carrier loss mechanism causes efficiency to drop.

Leakage LED

We also collaborated with Samsung Electro-Mechanics of South Korea. This effort finally led us to the cause of the efficiency drop—electron leakage from the active layer due to the polarity mismatch between the quantum well layer, quantum barrier layer, and electron blocking layer.

Our explanation explains the main reason why efficiency decreases at high current density - when the driving voltage is increased, more injected electrons escape from the active layer and reach the p-type region of the LED, thereby non-radiatively recombining with holes at the p-electrode (Figure 2).

This theory has been experimentally verified by comparing the light output of LED devices when they are electrically biased and optically excited (see ‘Exploring the structure of LEDs’). We have previously shown that recombination mechanisms also occur outside the quantum wells, using numerical simulation tools to link carrier leakage to polarity mismatch.

Interface issues

Our efforts have focused on growing LEDs on the c-plane, the traditional crystal face of GaN. These devices typically have a strong built-in electric field, which produces a fairly strong surface charge at the interface (see "Feeling the force inside").

Figure 1a. Measurements of a conventional GaInN LED chip with an area of ​​1 mm × 1 mm reveal that the output efficiency decreases when the drive current is greater than 10 mA. If the mechanism causing the efficiency drop can be completely avoided, the efficiency of the LED chip will continue to increase with the increase of current.

Figure 1b. Nonradiative mechanisms dominate at low drive currents, giving way to radiative recombination as current increases, after which efficiency decreases.

Interfacial surface charges hinder LED performance in two ways: They increase the barrier for electron injection into the multi-quantum well region and reduce the barrier for electron leakage from the quantum wells and current blocking layers.

Our simulations support this hypothesis, as surface charge degrades LED performance through electron leakage, suggesting that reducing surface charge can mitigate carrier loss. Calculations also show that the lack of heavy doping of the p-type structure exacerbates efficiency degradation, especially in the electron blocking layer.

As is well known, the heavy doping of the emitter compared to the base of a bipolar junction transistor can prevent minority carriers from injecting into the emitter. In LEDs, the low p-type doping concentration of GaN and AlGaN layers prevents holes (which are majority carriers in the p-type region) from injecting into the active region - which in turn exacerbates the leakage of electrons.

Our explanation for the drop in LED efficiency due to electron leakage is not widely accepted; in fact, several different mechanisms have been proposed by researchers, with the Auger recombination theory proposed by Philips Lumileds researchers (Shen et al. 2007) being the mainstream.

Several photoluminescence experiments were conducted on c-plane GaInN/GaN double heterojunctions, and it was observed that the efficiency decreased at high photoexcitation density. The use of rate model analysis allowed them to identify Auger recombination as the cause of the reduced efficiency of multi-quantum well LEDs.

To this end, they introduced the concept of effective recombination thickness, and the physical thickness selected when designing the double heterojunction is smaller because of the smaller overlap between electrons and holes in the quantum well.

Substituting this recombination thickness into the rate equation shows that the spontaneous emission rate of the quantum well is higher in the presence of an electric field than in the absence of an electric field. However, this is contrary to the fact that the electric field in GaN quantum wells weakens spontaneous emission. Therefore, we feel that Lumileds has overestimated the importance of Auger recombination in multi-quantum well LEDs at high current densities.

Of course, the proof of the pudding is in the pudding. Finally, we put our simulation results into practice and grew LEDs with AlGaInN barrier layers. Replacing the traditional GaN barrier layer and AlGaN electron blocking layer with AlGaInN layer allows us to freely adjust the band gap width and polarity, and ultimately reduce the polarity mismatch and surface charge between the active region interface.

Figure 2. RPI researchers believe that electron escape is the main cause of LED efficiency degradation. These escaped carriers non-radiatively recombine with holes in the p-type GaN region or p-electrode.

Figure 3. LEDs with matched quantum well and barrier polarities can exhibit excellent performance at high drive currents by reducing electron leakage.

For the quantum barrier, we will grow a quaternary compound with the same bandgap width as GaN and match the polarity of a typical quantum well. This is a difficult task because it is very difficult to grow AlGaInN layers with high indium content and high aluminum content. But by reducing the imbalance in polarity, the performance of the device can be significantly improved. The same applies to the electron blocking layer.

Our simulations show that adjusting the polarity match slightly can achieve almost all of the benefits, while using barriers and wells can only reduce the polarity mismatch by half. The key point is that slightly reducing the bandgap width of the barrier can create additional carrier confinement in the active region.

These adjustments have already produced some encouraging results. Light output has increased by 20% at high currents, due to slower efficiency degradation (Figure 3). The forward voltage has also been reduced, due to a lower barrier height for carrier injection into the quantum well region. Another benefit of reducing the surface charge in the quantum well region is a 25% increase in overall efficiency. The adjustments also have the added benefit of reducing wavelength drift with current, due to lower electric field strength in the quantum well.

Clearly, there is growing interest in the efficiency droop problem, which is itself a positive for LEDs. We are learning more about it, and existing designs appear to address it. Commercial devices with these characteristics should be poised to expand into new markets, with the ultimate goal of replacing general lighting bulbs.

Exploring the structure of LEDs

By comparing the output of light when excited and when electrically biased, researchers at Rensselaer Polytechnic Institute studied the efficiency droop in GaInN blue LEDs. The comparison assumes that the generation rates of electrons and holes are the same, and the results show that the efficiency droop is caused by carrier transport.

Under steady-state photoexcitation conditions, there should be no exceptional escape of electrons or holes. If it does, a carrier is lost - this means that an electric field distribution is formed in the quantum well, resulting in additional carrier leakage. Spontaneous escape of electrons or holes is also unlikely because holes are strictly confined in the potential well.

In any case, some carriers do escape from the potential well. Bombarding this structure with a 405nm laser produces a non-zero open circuit voltage. This is because the forward bias voltage of the device requires a reverse current to compensate. The net current is still zero, which is what any device that is not connected to an external circuit must comply with.

Forward bias changes the carrier transport of the LED. The injection and escape rates of holes and electrons into and out of the quantum wells change, often causing electrons to leak out of the potential wells. This is evidenced by the different light output efficiencies of the LED when it is optically excited and electrically biased. Measurements performed at RPI confirm this. A drop in efficiency was observed in the electrically pumped device, but not in the optically excited case. This implies that some form of carrier recombination mechanism occurs in the region outside the quantum wells, causing the drop in efficiency.

Feel the force within

GaN has a wurtzite crystal structure and has a polarization electric field in the c-crystal direction. When LEDs are grown in this crystal direction, large surface charges are generated between interfaces due to differences in spontaneous polarization and strain-induced piezoelectric polarization between adjacent layers. These charges reduce the efficiency of LEDs by reducing electron confinement in the active region and increasing electron leakage in the active region.

In traditional blue LEDs, GaN barriers surround the GaInN quantum well on both sides, and the surface charge on the n-type side of the potential well is negative. These charges repel electrons and hinder electron injection. The surface charge on the p-type side of the potential well is positive, which absorbs electrons and increases the possibility of electron escape (Figure a).

AlGaN electron blocking layers have the opposite effect. They generate positive charges on the n-type side, making it easier for electrons to cross the barrier (Figure b). Increasing the bandgap width of the electron blocking layer by increasing the Al composition is not a good solution because the surface charge will also increase at the same time.