New clues to LEDs’ efficiency droop

In the move to use LEDs not just for indicator lamps and digital displays but to illuminate buildings and neighborhoods, gallium nitride–based semiconductors have taken center stage. They are among the few materials capable of emitting brightly at green, blue, and shorter wavelengths; as such, they are key ingredients for generating white light with LEDs.

The main allure of LEDs, and of GaN-based LEDs in particular, is their efficiency. (See the article in Physics Today, December 2001, page 42 .) In theory, the energy cost of injecting an electron–hole pair into an LED is repaid nearly in full when it recombines by emitting a photon. Inevitably, though, some electrons and holes will recombine nonradiatively or not at all. In 2007 researchers at Phillips Lumileds Lighting in San Jose, California, reported experiments ¹ showing that the probability of such nonradiative events increases as the cube of the injected current density n. Because the probability of radiative recombination grows only as n², an LED’s internal quantum efficiency—the photon yield per injected electron–hole pair—falls off sharply above some threshold current density. That falloff is known as the “droop,” and in a typical GaN-based LED it sets in at a modest current density of around 10 A/cm². A GaN LED, in other words, can more efficiently light a smartphone display than a parking lot.

The Phillips Lumileds team attributed the droop to Auger recombination: Rather than emitting a photon, an electron–hole pair transfers its energy to a neighboring charge carrier. Alternatively, a group led by Fred Schubert (Rensselaer Polytechnic Institute) proposed that droop arises from a phenomenon known as drift leakage: If a sufficiently large voltage drop builds up across the LED, charge carriers can get swept in and out of the device’s light-emitting region before they have a chance to recombine. ² Both models can account for the observed n³ dependence. A third proposition—that nonradiative recombinations are instigated at defects protected by potential barriers—also predicts a droop, though with a different dependence on n.

A collaboration led by Claude Weisbuch (University of California, Santa Barbara, and École Polytechnique, Palaiseau, France), Jim Speck (UCSB), and Jacques Peretti (École Polytechnique) has turned to electron emission spectroscopy to help settle the debate. ³ Although the group’s new findings don’t close the book on efficiency droop, they appear to strengthen the case for Auger recombination as the likely culprit.

Gathering evidence

Figure 1 gives a schematic depiction of the LED that Weisbuch and coworkers used in their experiment. The device consists of two GaN slabs—an n-type layer, doped with electrons, and a p-type layer, doped with holes. Between them are an InGaN quantum well, which serves as the light-emitting region, and a thin film of insulator known as the electron blocking layer (EBL).

When an appropriate bias voltage is applied, electrons migrate toward the p-type layer, holes migrate in the opposite direction, and they both become trapped in the quantum well. There, ideally, they recombine to emit photons. The EBL improves the odds of such successful recombinations by providing a barrier to electron leakage from the quantum well.

If, however, an electron–hole pair undergoes Auger recombination, it can boost a neighboring electron’s energy by around 2.7 eV, roughly the bandgap energy, ample to surmount the EBL. That so-called Auger electron (red in the figure) rapidly begins shedding its excess energy in the form of phonons. In the process, it can become trapped in one of the conduction band’s local potential minima, termed side valleys. Previous experiments and simulations suggest that in GaN, the lowest-energy side valley lies 1.5–2.5 eV above the global minimum of the conduction band; an Auger electron that falls into that valley can retain a significant portion of its initial energy as it travels through the p-type layer.

The researchers decided to look for and measure those Auger electrons at exposed portions of the LED. By applying a cesium coating, they were able to manipulate the LED’s surface electric field and coax the device to emit some of its conduction-band electrons into an ultrahigh vacuum. There, the electrons could be collected and analyzed with a spectrometer. The technique, electron emission spectroscopy, has been around for nearly a century, but the new work marks its first use on LEDs.

A representative emission spectrum is shown at the right of figure 1. The two overlapping low-energy peaks have ambiguous origins: They could correspond to electrons that leaked or tunneled out of the quantum well, prematurely escaped the side valley, or arrived at the surface by a more exotic route. The researchers conclude that the high-energy peak, by contrast, can only be ascribed to Auger electrons carried to the surface in the side valley. Says Weisbuch, “There’s no other plausible mechanism.”

Opening arguments

Taken by itself, the high-energy spectral peak merely confirms what was never in doubt: that Auger recombination can occur in LEDs. The real controversy is over how large a part it plays in efficiency droop. To probe that question, the team had to compare their emission spectra with corresponding measurements of the LED’s light output.

Figure 2a shows the measured emission intensities alongside the expected intensity in the absence of efficiency droop; the horizontal distance between the two gives the amount of wasted, or supplementary, current. As shown in figure 2b, that supplementary current correlates almost perfectly with the Auger electron detection rates determined by integrating the high-energy peaks in the electron emission spectra.

The slope of the fit indicates that only about one Auger electron is detected for every million or so electrons of supplementary current. But the researchers argue that the disparity is of the appropriate order of magnitude for Auger-dominated droop, given the estimated fraction of Auger electrons that are either captured by the LED’s contact electrode, prematurely scattered from the side valley, or emitted in the wrong direction to reach the spectrometer. Says Chris Van de Walle, a UCSB theorist who wasn’t involved with the work, “I was expecting Auger electrons to be present, but it’s amazing that the experiment was able to detect them. This is really conclusive.”

Cross-examination

Not everyone is convinced. In a formal comment, ⁴ Boston University’s Enrico Bellotti and coauthors from the Polytechnic University of Turin in Italy challenge the notion that the observed high-energy spectral peak is produced by Auger electrons. Their ab initio simulations indicate that any Auger electrons produced in the quantum well should be scattered out of the side valley in less than one-half the time it takes them to pass through the 200-nm-thick p-type layer. Says Bellotti, “Our calculations suggest that, in the LED under study, there is no correlation between electron energies detected by the spectrometer and those in the quantum well.”

Bellotti and his coauthors think that the high-energy peak could instead correspond to electrons that gain kinetic energy in the strong near-surface electric field—depicted at the far right in figure 1 as a sharp downward bending of the band energies—and that the lower-energy peaks could be due to photoemission from Cs surface states. Jörg Hader of the University of Arizona in Tucson has another idea: The high-energy peak may be due to reabsorption of LED light by free charge carriers.

Even if Weisbuch and company have the correct interpretation, their measurements aren’t yet precise enough to rule out drift leakage as a co-contributor to droop. It’s also possible that Auger recombination is the main source of droop under some conditions but not others. The relationship between LED efficiency and injection current is complex and known to depend on temperature, pressure, and various particulars of the device construction: the bandgap and crystalline orientation, the number of quantum wells, the properties of the EBL, and so forth.

Bellotti expects that further electron emission spectroscopy experiments should help to disentangle LEDs’ complex web of interdependencies and, in turn, help designers of future LEDs to better negotiate inherent tradeoffs: “This is a very important new diagnostic tool. Instead of jumping too quickly to far-reaching conclusions, we should make sure that it becomes a launch pad for multidisciplinary collaborations that lead to a solution of the problem.”