Cranking Up The Efficacy Of Green LEDs
Green LEDs don't deliver the same level of performance as their red and blue cousins. However, by decreasing the current density with a larger chip and optimising growth conditions to reduce dark spots it is possible to close that gap with LEDs that hit 190 lumens per watt at a 100 mA drive current, says Osram�s Andreas L�ffler and Michael Binder.
After price, imperfect colour quality is the biggest criticism levelled at the LED light bulb. This downside stems from the way that the white light is generated: A GaN-based, blue-emitting chip pumps a yellow phosphor, and the mixing of these two colours produces white light. With this approach, the output does not feature a significant contribution from the red region of the visible spectrum.
A superior approach for making lighting products – that is also an option for solid-state projection displays – is to generate white light by mixing the emission from red, green and blue LEDs. Advantages of this approach are not limited to a higher colour-rendering index, and also include the opportunity for higher efficacy and flexible colour steering.
To produce a high-efficacy system with this form of colour mixing, efficient sources must be employed. Blue and red LED performance is already impressive, with recent improvements spurring peak power conversion efficiencies beyond 81 percent and 70 percent, respectively, but the green cousin is lagging far behind. This particular species suffers from a problem commonly known as the green gap.
The big challenge associated with trying to make an efficient green LED is that there is not an ideal, mature material system to work with. The III-N family that is used to create powerful blue LEDs is far less efficient at longer wavelengths, and a similar problem plagues the III-phosphides that are very efficient in the red range; extend emission of this class of LED to shorter wavelengths and efficiency plummets. So, in short, both material systems present low efficiencies in the green-yellow spectral range (see Figure 1).
Figure 1: Efficacy of III–nitride (green data points) and III–phosphide (red data points) LEDs with different wavelengths (data taken from recent publications). The blue lines represent the CIE 1924 luminosity function multiplied by the corresponding value of the wall plug efficiency (WPE). Marked in yellow is the green–yellow range, which is not adequately covered by either the III–nitrides or the III–phosphides. This is the essence of the green gap problem
With the III-phosphides, the falling efficiency as emission is propelled to the green is a fundamental limitation of the material system. Altering the composition of AlInGaP so that it emits in the green – rather than red, orange, or yellow – leads to insufficient carrier confinement, due to the relatively low band gap of this material system. This rules out efficient radiative recombination.
In comparison, for III-nitrides, the barriers to high efficiency may be very tough, but they are not insurmountable. With this material system, two factors are behind the decline in efficiency as emission stretches to the green: a fall in external quantum efficiency and a decrease in electrical efficiency.
The first weakness has its origins in the need to apply an extraordinarily high forward voltage to green LEDs. These devices feature extremely high internal piezoelectric fields. So, for a given current, the voltage that has to be applied to this type of LEDs is even higher than that for a blue variant, despite the lower bandgap. This higher drive voltage drags down power conversion efficiency. The second weakness is given by the fact that green LEDs are plagued by droop, the steady decline in internal quantum efficiency at increasing current densities. Droop occurs in blue LEDs, but its impact is far greater in green ones, leading to very low efficiencies at common operating currents (see Figure 1 and 2).
Figure 2: Comparison of the current dependent external quantum efficiency (EQE) of a blue and green 1mm2 InGaN/GaN LED emitting at 442 nm and 530 nm, respectively
The cause of droop is a hotly debated topic within the nitride community. Since the loss rate that causes droop exhibits a cubic dependence on charge carrier density under both electroluminescence and photoluminescence excitation, Auger recombination (direct or phonon-assisted) in the active layer is one of the main suspects.
However, this is by no means the only conjecture for the cause of droop – there have been attempts to explain the origin of this mysterious malady with theories involving either dislocations, carrier spill-over (thermionic, ballistic and Auger-induced spill-over) or electron leakage. The latter is enhanced by high internal piezoelectric fields.
Moving in the right direction
At Osram Opto Semiconductors of Regensburg, Germany, we have been steadily improving the efficacy of our� green LEDs. In 2008, our colleagues, led by Matthias Peter, reported a 1 mm2, ThinGaN 527 nm chip producing 100 lm at 350 mA. This corresponds to 73 lm/W. Two years’ later, we increased efficacy to 100 lm/W at 350 mA, using an optimised 1 mm2 chip housed in a Golden Dragon Plus package. At this drive current, luminous flux is 117 lm, while cranking up the current to 1 A increases output to 224 lm.
More recently, we have raised the bar for green LED performance again. Higher efficacies have been possible with MOCVD-grown LEDs formed on c-plane sapphire that feature an active region with five to seven InGaN quantum wells embedded in GaN barriers. A 5 �m-thick, silicon-doped GaN buffer layer underpins this active region, which is covered with a 30 nm-thick, p-type magnesium-doped AlGaN electron-blocking layer and a 140 nm-thick, magnesium-doped GaN contact layer.
We have compared the photoluminescence produced by the active region of this structure with that of a device coming off our production line (see Figure 3). With the high-volume device, micro-photoluminescence reveals strong inhomogeneity in intensity, with a pattern of dark spots appearing against a meandering, bright background. The density of the dark spots corresponds to the hexagonal crystal defect (V-pits) density, leading us to suspect that there is a strong correlation between these spots and the V-pits. This view is supported by several studies by other groups, which confirm a point-to-point correlation.
Figure 3: Microphotoluminscence images of a device from a production (a) and a research and development sample (b). For a better contrast, the lower part of the images are shown in levels of grey only
Lowering the growth rate in the active region significantly improves quantum well material quality, according to micro-photoluminescence imaging. The density of the dark spots is similar to that of the sample from the production line, but the affected area is far smaller. This increases the proportion of bright areas, leading to a more homogeneous luminescence pattern.
This improvement, which results from an increase in material quality that enhances internal quantum efficiency and transport characteristics, leads to better-performing LEDs. Recent samples that have been mounted in a Dragon package with a spherical lens produce 114 lm at 350 mA, corresponding to an efficacy of 100 lm/W (see Figure 4). In comparison, devices from the production line emit just 108 lm under the same drive current. Even better results are possible by by removing quantum wells that do not contribute significantly to light generation. In our case, that means trimming the number of wells from seven to five, and therefore improving carrier transport. With this refinement, output from the 532 nm-emitting, 1 mm2 ThinGaN chip hits 134 lm at 350 mA, corresponding to an efficacy of 108 lm/W.
Figure 4: Electro-optical characteristics of a 1mm2 ThinGaN chip in a Dragon package: device from production (blue), device with improved transport (black) and optimized epitaxial structure (orange)
The key to further improvement in these green-emitting LEDs is to combat droop by cutting carrier density, through either an increase in chip size or the number of emitting quantum wells. The efficacy curves of Figure 4 allow us to estimate that cutting the current density by a factor of two or four increases efficacy by25 percent or 60 percent.
We have adopted this approach, increasing chip size to 2 mm2. This increased output power for a green LED to a record 150 lm (280 mW) at 350 mA (see Figure 5 for plots of this 533 nm LED). This corresponds to an efficacy of 135 lm/W – compared with 108 lm/W for the 1 mm2 chip.
Increasing currents to higher values leads to far greater output at slightly shorter wavelengths: Driven at 700 mA, the chip emits 248 lm and 480 mW at a peak wavelength of 531 nm; and cranking up the current to 1A propels the output to 313 lm and 620 mW, with peak wavelength shifted to 529 nm. The latter figure, which equates to more than 310 lm (600 mW) at a current density of 50 A cm-2, is an enabling technology for high-performance projection systems based on red, green and blue LEDs. Efficacies at very low drive currents are particularly impressive. They exceed 190 lm/W at 100 mA, and are in excess of 300 lm/W below 2 mA.
Figure 5: Electro-optical characteristics of a 2 mm2 ThinGaN chip in an OSLON package with an improved carrier transport and an optimized epitaxial structure
An alternative approach for making a green emitter is to take a blue LED and add a green phosphor. We have investigated this, using a ceramic platelet of the green phosphor lutetium aluminium garnet (LuAG). This pumping approach creates a significantly different green emission profile: the emitter features a 531 nm peak wavelength, a Gaussian peak at 525 nm and a full-width half maximum (FWHM) of 33 nm; while the chip-phosphor combination produces a peak wavelength of 529 nm, has a central wavelength of 557 nm and produces a FWHM of 99 nm (see Figure 6).
A broader emission profile has its pros and cons. It’s favourable for general lighting because it offers a high CRI, but a narrower emission is preferred in applications such as projection. There, the smaller spectral bandwidth of direct-green LEDs quashes cross�talk, leading to higher system efficiency. What’s more, if direct-green LEDs are used for projection, they can cover a wider colour range than a converted green solution (see Figure 7).
Fig. 6: Spectra of two different approaches for green LEDs. The emission generated by the phosphor is broader than that resulting from a direct green, InGaN-based LED.
However, a blue LED and a green phosphor is still an attractive option, because it avoids issues associated with the green gap. Although there are inevitable losses associated with the Stokes shift, pumping a phosphor with a blue chip leads to higher efficiencies, because droop is not as strong at shorter wavelengths (see Figure 8). In addition, internal piezoelectric fields are weaker�in blue LEDs, leading to lower electrical losses. We have compared the luminous flux and the efficacy of the two different approaches, using ThinGaN chips 1mm2 in size. At lower current densities, the green LED is more efficient than its blue cousin, there are no conversion losses, and efficacy is 291 lm/W at 1 mA. However, efficacy falls rapidly as current increases, and is just 108 lm/W and 66 lm/W at 350 mA and 1 A, respectively. Blue LEDs, in comparison, are more efficient at higher current densities, with efficacy peaking at a current of 20 mA. Driven at 350 mA, the blue LED and green phosphor combination emits194 lm at 191 lm/W, and at 1A delivers 462 lm at 145 lm/W.
Fig. 7: The CIE 1931 colour space chromaticity diagram shows two red-green-blue approaches with given red (610 nm) and blue (450 nm) LEDs used in combination with a direct green InGaN LED or a phosphor-converted green LED. The narrower emission spectra of a direct green InGaN LED, compared to green generated by phosphor conversion, makes this device better suited to projection.
Several routes are available for increasing the efficiency of the direct green LED, so that it closes the performance gap with the blue-chip-and-phosphor combination: Carrier density could be cut by increasing the volume of the active region via the addition of more wells; internal quantum efficiency could be increased through improvements to material quality; and the active area could be increased by optimising the design of the chip and its dimensions. In our view, the pathway with most potential is to improve the epitaxial growth process, because this could lead to a lower forward voltage and superior carrier transport.
Fig. 8: Current-dependent luminous flux and efficacy of two different approaches to generate green light. Whereas the green InGaN/GaN LED shows significant droop at high operating currents, a blue LED in combination with a phosphor converter yields higher efficacy and luminous flux at a typical driving current.
We gratefully thank the German Federal Ministry of Education and Research (BMBF) for financial support (grant number 13N9974, ‘‘High Quality LED’’) for the development of direct green LEDs. We also appreciate the support of numerous colleagues at Osram and Osram Sylvania.