On the efficiency degradation in InGaN-based LEDs: Mechanisms and remedies
BY SAIKAT DAS, VITALIY AVRUTIN, ÜMiT ÖZGÜR AND HADiS MORKOà‡ FROM VIRGINIA COMMONWEALTH UNIVERSITY AND ARVYDAS MATULIONIS FROM CENTER OF PHYSICAL SCIENCES AND TECHNOLOGY, VILNIUS
LEDs have become indispensable in a variety of applications including displays, indicator lights, signs, traffic lights, printers, telecommunications, and the biggest of all, general lighting. In the centre of this blazing journey is the GaN based LED, which is somewhat mysterious in that the output does not scale linearly with the drive current at high injections aside from thermal issues. The injection route to increase the light output runs aground because of efficiency degradation, the causality of which has received a good deal of attention. More specifically, the internal quantum efficiency, IQE, typically peaks at levels as low as a few A/cm2 and then drops with increasing injection to in some cases as low as 50 percent of the peak value. The reduction in IQE is also observed under pulsed injection with pulses having duty cycles and widths for which the heating effect can be ruled out as an origin1, thus implying that the reduction in the IQE is related to the internal properties of the LED. Although this loss of efficiency is observed in UV LEDs, blue and, more severely, green LEDs are more prone to efficiency degradation at high injection.
For over a decade, some research efforts have been put forth for understanding the origin of the degradation mechanism and exploring possible remedies. Consequently, several technological modifications to the GaN-based LED structure has been proposed, some of which have been successfully implemented for laboratory prototype and commercial manufacture. The IQEs of blue LEDs are already in the high 90 percent range and attainment of packaging with efficient heat removal and cutting cost while retaining or even improving the performance are currently the key focus areas. However, unequivocal design rules are yet to be established to overcome efficiency degradation. This article reviews both the mechanisms of and remedies for the efficiency degradation, with the goal of providing the reader with an as complete as possible and updated picture of the present status of our current understanding of the efficiency loss and the mitigation efforts.
One characteristic of efficiency degradation in nitride LEDs is that it is more pronounced as the emission wavelength increases (see Figure 1). Possible explanations for more severe efficiency loss with increasing indium mole fraction in the active region are widely ranging and include a higher density of defects in the active region, more significant fluctuations in indium content, higher kinetic energy of the electrons due to increased conduction band offset, and the tendency to shift from direct to indirect Auger recombination processes.
Efficiency degradation is clearly caused by a non-radiative carrier loss mechanism that can be effective inside and outside the quantum wells (QWs) of the LED active region. If it is inside, the culprit might be a process governed by defect-related Shockley-Read-Hall recombination and Auger recombination, and enhanced by carrier delocalization and a reduction in the effective volume of the active region for injection. If it is outside, the origin of efficiency loss is either carrier leakage due to current crowding, inefficient hole injection, asymmetry of doping polarity, polarization charges or electron overflow from the active region. An overview of all these theories is provided in Figure 2, while an equation for internal quantum efficiency that accounts for all these mechanisms is provided in the panel "internal quantum efficiency".
Figure 1: Efficiency versus current density curves of GaN-based UV, blue and green LEDs showing a decrease in efficiency with increasing injection current. Green LEDs have the largest efficiency degradation. The numbers in the brackets in the legend denote the efficiency degradation at 1000 A/cm2. (Simulation results for illustration purpose only, not actual experimental data).
Auger recombination is one of the proposed mechanisms that lead to efficiency degradation in InGaN/GaN LEDs. This process involves recombination of an electron and a hole, with the energy that is released consumed by the excitation of a third carrier − an electron or a hole − rather than the emission of a photon.
In the GaN-based LEDs different Auger recombination processes are at play. In a direct Auger process, such as the eeh variety (see Figure 3, left panel), the energy given off by an electron dropping from the conduction band to the valence band is consumed to excite another electron to a higher state within the conduction band. This chain of events satisfies the requirements for the conservation of energy and momentum. The other direct Auger processes are less likely to occur, so they can be neglected2.
Indirect Auger recombination, which can occur in parallel with the direct variety, encompasses phonon-assisted processes and/or many other scattering mechanisms (Figure 3, right panel). While the direct variety depends exponentially on temperature, the indirect variant exhibits a power law dependence.
In general, the rate of Auger recombination is proportional to the third power of the free carrier density n. Consequently, if one assumes that the coefficients for Shockley-Read-Hall and radiative recombination do not depend on injection, the Auger recombination can be invoked to account for the efficiency loss with increasing injection depending upon the magnitude of the Auger coefficient C.
If this approach is taken, Auger coefficients should be greater than 10-31 cm6s-1 to cause notable efficiency degradation3,4. However, such a large Auger recombination coefficient for III-Nitride systems has not been predicted so far by theoretical calculations except for the indirect and resonant processes that would take place only in LEDs emitting at 2.5 eV5, assuming their presence. Yet, direct observation of Auger electron was made and it was postulated that the observed drop in electroluminescence efficiency corresponded to the detected high-energy Auger electrons, suggesting Auger recombination to be the sole contributor to the efficiency degradation6.
One would then surmise that, if the nonradiative recombination due to Auger effect is to be the major (if not the sole) contributor to the observed efficiency degradation, indirect Auger processes have to be dominant as the direct variety is a very unlikely possibility. The likelihood of this is higher for relatively low quality material with alloy and compositional fluctuations. In general, the Auger coefficient extracted from the overly simplified ABC model is strongly dependent on quantum well properties such as the relative density of electrons to holes, the net polarization field, and the hot carrier escape ratio. A more rigorous approach than simple curve fitting plots of efficiency as a function of current density is required to obtain a consistent value for the Auger coefficient7.
Figure 2: Mechanisms leading to efficiency degradation in InGaN LEDs.
The proponents of Auger recombination as the major, if not the sole, mechanism for the efficiency degradation, have proposed different theories to reconcile the relatively small Auger coefficients with the strong efficiency loss effects. The most popular conjecture is that the effective optically-active volume is significantly reduced with respect to the nominal one. This could arise from strong non-uniformity of carrier-density distributions, which results in significantly higher local carrier densities in the optically active region than in the case of uniform distribution for the same driving current.
A relatively small C (»10-31 cm6 s-1) can then give rise to strong efficiency loss. Several explanations have been put forward as the genesis of the strongly non-uniform lateral carrier density distribution: polarization-field-induced electron-hole wave-function separation8,9, carrier localization in the indium-rich areas of the active region10, and lateral current crowding, which occurs in LEDs supporting lateral ohmic contact schemes11.
Figure 3: Direct (left) and indirect (right) eeh type Auger recombination processes.
An internal non-radiative loss process that has been proposed is associated with carrier delocalization. The premise assumes carrier confinement at low current densities, associated with randomly distributed potential minima − these could be caused by energy-activated defects, or fluctuations in the indium content or width of the quantum well. These localized regions have low defect recombination and can be represented by a SRH recombination term characterized by a relatively small A coefficient, typically on the order of 107 s-1. When the LED is operated at low current densities, carriers are confined in dislocation-free regions and have long non-radiative lifetimes, and the radiative recombination efficiency is relatively high; but as the current is increased, potential minima fill, with carriers spilling over, delocalizing, and diffusing to regions with a far higher defect density, leading to increased non-radiative recombination. A density-activated defect recombination (DADR) mechanism was proposed, wherein the loss rate is negligible below a certain threshold carrier density but rises with increased injection according to the quadratic dependence of the electron-electron scattering rates on the carrier concentration12. Detailed microstructure analyses seem to indicate fairly uniform quantum wells, however, in high quality LED structures. Therefore, carrier delocalization can be considered negligible in such structures.
The mechanisms of localization of carriers inside an LED active region are no different than those discussed under Auger recombination. However, the mechanism leading to degradation is different: carriers remain in the localized regions even at high current densities and undergo strong Auger recombination according to the reduced-active-volume theory, whereas the energized carriers overflow from the localized regions and undergo defect-assisted recombination according to the carrier delocalization theory.
Another theory for explaining efficiency loss at high injection is that instead of recombining with holes in the quantum wells, electrons fly over the active region and undergo a non-radiative recombination process in the relatively low quality p-GaN, or recombine with holes in p-type GaN resulting in emission at unintended wavelengths, or collected at the p-type contact electrode. Electron overflow is by no means a new idea, having dogged LEDs since their inception, and inserting electron-blocking layers (EBLs) can reduce it, a practice that has been employed copiously in conventional III-V-based green LEDs and also in InGaN LEDs.
Inclusion of the electron overflow term into the rate equation was shown to represent the efficiency loss at high injection13, which has been directly observed in LEDs confirming it as a source of the efficiency degradation in LEDs14. Electron leakage is indeed a family of degradation mechanisms rather than a single one, encompassing several different promoting factors: electron overflow, polarization-induced charges, poor hole injection, and asymmetry in doping.
Figure 4: Schematic of electron overflow caused by ballistic and quasi-ballistic electron transport across the InGaN active region. The electrons gain a kinetic energy after being injected into InGaN, resulting in a total energy of E + Δ EC+qV (x) . Dashed lines in the band structure represent additional band bending caused by the piezoelectric field due to the EBL (redrawn after3).
The electron overflow in semiconductor heterostructures can originate from two processes. One is thermionic emission of equilibrium electrons from the bottom of the conduction band in the active region over the barrier into the p-layer. However, its effect in the InGaN system is negligible due to large band discontinuities15. The other possibility is ballistic or quasi-ballistic transport of the injected electrons, which can recombine in the p-GaN region instead of the active region, unless blocked by an EBL.
Upon injection, electrons acquire additional kinetic energy equal to the conduction band offset between n-GaN and InGaN. These hot electrons can either lose their excess energy mainly through interaction with LO-phonons; and thus contribute to recombination or they can avoid cooling altogether and leave the InGaN region, contributing to electron overflow as depicted in Figure 4.
Alongside leakage due to highly energized electrons, another culprit may be the polarization-induced sheet charge at the interface between the AlGaN EBL and the spacer and at the quantum well/quantum barrier interfaces. The weakness associated with this conjecture is that in order for simulations to replicate experimental efficiency loss data, a multiplication factor ranging from 0.3 to 0.7 has to be introduced for the polarization charges. Without this factor, numerical device simulations can only reproduce sufficient efficiency degradation when the AlGaN/GaN band offset ratio ΔEC: ΔEV is reduced from normally accepted values of 70:30 to 60:40 or 50:5016.
The AlGaN electron-blocking layer employed as a barrier for the electron overflow into the p-side of the LED device is to some extent a double-edged sword, as it results in a barrier for hole injection as well, due to the valence band offset between AlGaN and GaN. This unwanted barrier can be reduced by employing p-type doping in the AlGaN layer; however, p-type doping efficiency decreases as the aluminium mole fraction increases17. As a result, increasing aluminium content in the EBL for better confinement of electrons will lead to an increasing energy barrier for the holes. Hole injection is further hindered as compared to electron injection due to the fact that active regions are typically unintentionally or intentionally n-type doped.
The problem of limited hole transport has been found to be independent of material polarization and sheet charges at hetero-interfaces; therefore, this mechanism is relevant even in LEDs grown along non-polar and semi-polar orientations or polarization-matched LEDs grown on c-plane sapphire or (111) silicon. Moreover, poor hole transport through the active region leads to non-uniform hole distribution in a multiple-quantum-well LED. Appropriate structural designs mitigate this problem considerably.
Another proposition for the genesis of the electron overflow is the asymmetry in the free carrier concentration in the p- and the n-side of the LED device. In addition, the relatively low mobility (and thus diffusivity) of the holes compared to electrons contributes to a non-uniform hole distribution, particularly across a multiple-quantum-well LED active region.
In addition to all the explanations for electron leakage discussed above, there are suggestions that it can result from lateral current crowding and defect-assisted tunneling. Theoretical analyses of lateral and vertical LED structures support the view that the locally enhanced carrier densities induced by current crowding enhance electron leakage18. This is more pronounced in mesa-based lateral LED structures, but not a major impediment in vertical varieties where the substrates are removed. Defect-assisted tunnelling may also contribute to electron leakage, with electrons tunnelling from the quantum well to the defect sites, if present, in the p-doped barrier19. This is more likely to happen if the device contains a single, thin quantum well.
Combatting efficiency lossThe many and varied conjectures for the cause of efficiency loss have spawned a wide range of solutions. These aim to achieve one or more of the following: (i) reducing the carrier densities in the QWs, (ii) improving the electron confinement within the active region, (iii) enhancing the hole injection into the active region. See Figure 5 for an overview of the degradation alleviation mechanisms that are discussed below.
One popular option is to reduce the carrier density in the active region by increasing its effective volume via either employment of more or thicker quantum wells, or improved lateral current spreading and increased chip area. While the primary goal is to suppress the Auger recombination, this approach reduces electron leakage as well.
Turning to thicker wells is not a trivial solution, however. That is because dislocation generation is also a factor when determining the optimum thickness of the quantum well. For example, in a near-ultraviolet single-quantum-well LED emitting at around 400 nm, if GaN is grown on sapphire, the well should be 3 nm thick. However, if it is grown on a native platform, which has far fewer dislocations, an 18 nm-thick well works best20. These results indicate that, for good quality samples with a low defect density (both in terms of quality of the GaN epitaxial layer and the quantum wells), the dominant causes for efficiency degradation are Auger recombination and electron leakage. On the other hand, for LEDs grown on non-native substrates such as sapphire or silicon or emitting at longer wavelengths, defect-related mechanisms such as carrier delocalization and defect-assisted tunneling come into play.
To reduce the carrier density with thicker quantum wells, a relatively uniform carrier distribution must be achieved across the active region. If the LEDs have wells with low indium content, making them thicker reduces the internal electric field, leading to a spread of wavefunctions and thus an enhanced electron-hole overlap, which results in mitigation of the degradation in radiative recombination. However, thicker quantum wells may introduce other issues; and if they are rich in indium, material quality may suffer. The best solution appears to be increasing the number of wells, while keeping the thickness of each of them low enough to avoid drawbacks associated with the quantum confined Stark effect due to internal electric fields.
Figure 5: Overview of the approaches to alleviate the efficiency degradation problem.
A variant of this approach is to employ short-period superlattice (SPSL) InGaN/GaN active region21. Such a structure increases the number of QWs and introduces quantum mechanical coupling between them. This results in carrier delocalization among the QWs through the formation of minibands, leading to increased uniformity of electron and hole distributions and reduced peak carrier densities.
Improved LED performance via reduction of electron leakage is promised through modifications to the design of the active region, electron-blocking layer, and electron injector. Varying degrees of success can also result from a reduction in the polarization field, which can be accomplished through the introduction of polarization-engineered multi-quantum well and electron-blocking structures, and the switch to non-polar and semi-polar planes for the LED. Although they are appealing, the technology for these unconventional planes has not yet matured to the extent of the c-plane variety. Significant emission efficiency has hinged on the use of very high quality bulk GaN substrates, which are expensive and very small, and thus unsuitable for LED production. Thanks to the superior quality of material grown on the c-plane, it is this orientation that still sets the benchmark for LED performance.
A straightforward approach to reduce the electron leakage from the LED active region is to increase the barrier height of the electron-blocking layer. Switching from AlGaN to InxAl1-xN both increases the band offset with GaN and InGaN and allows lattice-matching with InyGa1-yN layers of the active region. An InAlN EBL has been shown to be more effective than AlGaN EBL for reduction in efficiency degradation22,23, but at the expense of an increase in the barrier for holes in p-GaN, which hampers hole injection. Magnesium-doped InAlN can be used to mitigate this problem some; but, p-type doping of InAlN with sufficient thickness is difficult. One proposed solution is to use an electron blocking structure composed of a p-doped InAlN/GaN superlattice24.
Some researchers have attributed efficiency loss to polarization effects that result in electron leakage. To address this problem, reduction of polarization charges in the MQW region by polarization-matching quantum wells and barriers have been proposed. Using AlInGaN25 or InGaN26 as barriers instead of GaN has been shown to lower efficiency degradation. These approaches as well as multilayer and graded barriers also have the potential to provide more uniform hole density across the active region27.
According to the hot electron model, another source of leakage is the high energy of electrons injected into the active region. Upon injection, their potential energy is converted to kinetic energy and while they diffuse from the n-GaN side to the active region they gain additional energy from any field present. To reduce this effect, and thus alleviate electron leakage and the associated efficiency degradation, an "˜electron cooler' has been introduced28. This layer consists of compositionally graded (from lower to higher indium content in the growth direction) InGaN known as staircase/graded electron injection (SEI/GEI) layers between the n-GaN region and the quantum wells. A great strength of this approach is that it allows the removal of the electron-blocking layer, which acts as a barrier for holes. Each energy step would ideally be equal to or slightly larger than the LO phonon energy (92 meV in GaN). The composition can also be graded continuously, in which case the step height requirement is removed.
Most of the approaches aiming to enhance the hole injection into the active region focus on the optimization of the EBL. Other usually employed approaches include linearly grading the aluminium mole fraction (increasing along the growth direction) in an AlGaN EBL. The idea behind this proposition is to compensate the band bending associated with the polarization charges at the EBL/spacer interface, leading to enhanced hole density at the interface and in the spacer; and to improved hole injection into the QWs. Experimentally, this approach has shown to reduce the efficiency degradation significantly. AlGaN/GaN or InAlN/GaN super-lattice EBLs have also been used to improve hole injection into the active region of the LED29. Ultimately, increased hole concentration beyond the current 1018 cm-3 or slightly higher would boost the LED performance.
One route to improve the hole injection into the active region is to dope the barriers with magnesium30. This delays the onset of efficiency degradation, but drags down LED efficiency, with band-edge emission diminishing as magnesium content in the active region increases. Alternatively, delta-doping the barriers with magnesium has been proposed as a solution to reduce possible magnesium diffusion into the QWs31.
There are clearly many options for reducing the impact of efficiency loss in InGaN LEDs, the origin of which has been attributed to several mechanisms. Auger recombination and electron leakage are the two leading contenders, and there is a great deal of theoretical and experimental evidence supporting both hypotheses. However, the industry operates on a very rigorous slope and continually improves the LED performance by optimizing growth and device design schemes. The LED performance is consistent with very high internal and external quantum efficiencies to the point where the mechanisms discussed in this report for degrading the efficiency can be safely assumed to be reasonably countered. After all these measures to alleviate the degradation are taken, ultimately, the hole concentration should be improved to boost the nitride LED performance.
The overall power conversion efficiency (wall-plug efficiency) in an LED is the product of voltage conversion efficiency (hvoltage), internal quantum efficiency (hIQE), and the light extraction efficiency (hextraction).
The voltage conversion efficiency is fundamentally determined by band offsets of III-nitrides and the extraction efficiency is the function of LED design. Different mechanisms leading to efficiency loss can be encompassed in the following expression for the internal quantum efficiency of an LED, albeit with a plethora of assumptions, including injection independent A and B coefficients:
where An and Cn3 are the Shockley-Read-Hall and Auger recombination rates, Ileakage is the electron leakage current, q is the electronic charge and VQW is the volume of the active region. The term k(n-n0)m accounts for the non-radiative loss mechanism associated with the carrier delocalization with n0 being the threshold carrier density triggering delocalization, k being a suitable constant, and m ≥ 2.
