New GaN faces offer brighter emitters

The tremendous improvement in III-N material quality has driven the commercial fabrication of lasers and LEDs emitting from the ultra-violet to the green, and aided the development of high-power HEMTs suitable for numerous telecom applications. However, despite these advances, GaN is hindered by its growth in the hexagonal wurtzite structure – which consists of two intermixed hexagonal closed packed lattices – and is highly susceptible to polarization-induced charges that can adversely impact device performance. For example, these charges can limit the output of LEDs and produce a shift in the emission wavelength at different drive currents.

The standard material orientation for GaN-based LEDs, lasers and HEMTs involves growth along the [0001] direction, which is also referred to as the "c"-direction (see figure 1). What is readily apparent from this diagram, and is a direct consequence of its wurtzite structure, is that there is a fundamental difference between the crystal A and B faces. The A face, or gallium face, is only terminated by gallium atoms, which have three unsatisfied bonds, while the bottom nitrogen face is terminated in an identical manner by nitrogen atoms.

Just beneath the gallium-face surface is a nitrogen layer. The gallium-nitrogen bond is highly ionic and the charge asymmetry, coupled with the lack of inversion symmetry in the c-direction, gives rise to a large spontaneous polarization along the c-axis. (This spontaneous polarization is greatly reduced in lattices with a high degree of symmetry, such as the GaAs or InP zincblende structures.) In addition to this spontaneous polarization, piezoelectric polarization is generated at interfaces between the III-Ns due to the strain that results from the different lattice constants (see "The properties of III-Ns").

The combined (net) polarization generates a polarization-induced electrostatic charge at alloy interfaces. While GaN and InN have very similar spontaneous polarization constants, in AlN it is almost three times larger. This difference produces a large spontaneous polarization-induced charge generated at interfaces between aluminum-rich and gallium- or indium-rich alloys, such as those that can occur in GaN-based HEMTs. For LEDs and lasers containing GaInN multiple quantum wells, the induced charge is dominated by piezoelectric effects arising from the large difference in lattice constant between InN and GaN. These induced interface charges bend the profile of the conduction and valence band, and consequently impact device operation.

A simplified quantum-well band structure that contains GaN barriers and GaInN wells, and is found in many LEDs and lasers, is shown in figure 2a. Electron-hole radiative recombination within the wells generates a photon with an energy equal to the GaInN s bandgap, if we ignore energy-level shifts due to quantum effects in narrow wells. However, in the presence of polarization-induced charges at well–barrier interfaces, the bands will be bent by the electric field generated by these charges (see figure 2b).

This polarization-induced band bending leads to a phenomenon known as the quantum-confined Stark effect (QCSE). Electron and hole wavefunctions are displaced to opposite sides of their respective wells (see figure 2b), rather than residing in the center. This decreases the oscillator strength – the probability of an electron-hole pair recombining to generate a photon – because the electrons and holes are now physically displaced from each other. And the greater this polarization-induced electric field, the greater the displacement between the two different charge carriers.

The reduced oscillator strength decreases the light output from the active region. In addition, the GaInN bandgap across which the recombination process will take place effectively shrinks, because the smallest energy transition now occurs between an electron in the bottom-left-hand corner of the conduction band and a hole from the right-hand corner of the valence band. This red shift means that the energy of the emitted photon is lower than that of the GaInN bandgap.

The impact of the red shift on blue and green LEDs, as a function of drive current, is shown in figure 3. As the current increases, the emission shifts towards shorter wavelengths due to an increase in the screening of the polarization-induced field, until a point is reached at which the band structure approaches that shown in figure 2a. For LED applications, as emission is pushed to longer wavelengths by increasing the indium composition within the well, the size of the piezoelectric-induced charge also increases. As a consequence, longer wavelength green-emitting devices are more susceptible to the QCSE. In addition, for higher indium content quantum wells, the band-bending becomes steeper and the separation of the electron and hole wavefunctions further increases, which decreases the oscillator strength. These polarization effects mean that it is more challenging to make high-brightness green LEDs than blue ones, and explains why the brightest green LEDs typically generate only half the light output of their blue cousins.

A potential remedy to these polarization-induced effects is to grow GaN films oriented in other directions that have either a reduced or no polarization fields in the growth direction. Figure 1 shows that such non-polar directions do exist in GaN, such as those directions perpendicular to the gallium or nitrogen face that are still in the plane of the figure. These directions produce a surface plane with equal numbers of gallium and nitrogen atoms, and eliminate the polarization effects.

Two non-polar directions are present in GaN, both perpendicular to the c-axis: the {11–20} plane, called the "a" plane, and the {10–10} plane, called the "m" plane. Devices fabricated on these planes should not suffer from the polarization effects currently observed in c-axis-oriented GaN. In addition to these two non-polar planes, there are several "semipolar" planes at different angles between the polar c-axis and the a- and m-planes that should have reduced polarization effects.

Nearly two decades have been spent researching and developing the growth of c-axis polar GaN, and this has opened the way for the commercial production of GaN-based lasers, LEDs and HEMTs. However, while much of that experience can be applied to the growth and subsequent fabrication of GaN devices using a- and m-plane materials and semipolar orientations, there is still much to do to establish the optimum growth conditions for these orientations – especially with regard to the control of threading dislocations and basal-plane stacking faults.

The materials and electrical and computer engineering departments at the University of California Santa Barbara (UCSB) are at the forefront of this research. The first challenge in the growth of non-polar GaN films is to establish the growth regime for a smooth surface, which is not a trivial task. UCSB s Paul Fini reports, "We ve found in general that a-plane GaN has a smaller growth window for planarity than m-plane, semipolar or c-plane films." The researchers have observed that the threading dislocation density is often more than 1 × 1010cm–2, and that the stacking-fault density is greater than 4 × 105cm–2, when growing a-plane GaN on sapphire or m-plane on LiAlO2. The threading dislocation density is similar to values obtained in the early days of c-axis GaN growth, but just like polar GaN, these defect levels have been reduced by the use of lateral epitaxial overgrowth techniques. Although these non-polar films do not have the same microstructural quality as c-axis polar GaN films, they can still be used to investigate the effects of non-polar orientations on device operation.

In fact, several groups have now produced devices that show the significant reduction in shift of emission wavelength that has been predicted. This 10-fold, or more, reduction in wavelength shift compared with polar quantum wells indicates the absence of polarization-induced charges in non-polar quantum wells.

The polarization-free interfaces can also have a significant impact on HEMTs. A typical AlGaN/ GaN HEMT built on polar GaN has a channel charge of more than 1 × 1013cm–2, and all this charge is polarization-generated. With the elimination of the QCSE, HEMTs can be produced with two-dimensional electron gas formed by silicon-doping an offset electron donor layer, which is the approach used in GaAs- and InP-based devices. This switch allows the doping level in the two-dimensional electron gas to be independent of polarization effects that are highly sensitive to strain. The elimination of the QCSE also aids the fabrication of both enhancement- and depletion-mode devices, greatly simplifying the GaN-based logic devices. While this has not yet been reported, it should now be possible.

One of the toughest challenges for c-plane GaN device engineers is overcoming the limitation of p-type magnesium doping levels – the best reported electrically active levels are only in the 1–2 × 1018cm–3 range. UCSB researchers have discovered that for m-plane GaN, it is possible to produce electrically active magnesium-doped levels of up to 7 × 1018cm–3. These higher p-type doping levels are expected to lead to lower contact resistance, a reduction in p-n junction turn-on voltage and series resistance, and produce LEDs and lasers with higher optical output efficiencies.

One other benefit of growth on non-polar orientations is the possibility of polarized light emission. This could be used directly for backlighting LCD displays and projectors, because a polarized light source can eliminate polarizing filters and lead to screens that are thinner, lighter and more energy efficient. UCSB and other groups have produced polarized light emissions from m-plane GaN LEDs with a polarization ratio of 0.17 (randomly oriented and totally polarized light have polarization ratios of 0.0 and 1.0, respectively). While more work remains to be done to improve this polarization ratio, initial work shows that m-plane GaN devices can target a unique niche that polar GaN simply cannot address.

While these results highlight the promise of non-polar GaN, unless output power can meet or exceed the values obtained by devices grown on polar GaN, there will be little driving force to pursue non-polar GaN. Today the light output from non-polar LEDs is still at least an order of magnitude lower than that for polar GaN. For example, Fini has reported packaged blue LEDs on m-plane material with an output of 0.6 mW at 20 mA drive current, and on-wafer a-plane blue LEDs delivering 0.25 mW at the same drive current. While these output powers are significantly lower than what is currently available with polar GaN, it is important to remember that non-polar GaN development is still in its infancy. "Until that work is completed, directly comparing non-polar LEDs to polar devices is like comparing apples and oranges," says Fini.

A Chakraborty et al. 2006 Jap. J. Appl. Phys. 45(2a) 739.
A Chakraborty et al. 2005 Appl. Phys. Lett. 86 031901.
M McLaurin et al. 2005 Appl. Phys. Lett. 86 262104.
H Masui et al. 2005 Jap. J. Appl. Phys. 44(43) L1329.
http://nsr.mij.mrs.org (for much of the early work on nitrides).
E Yu 2003 Spontaneous and Piezoelectric Polarization in Nitride Heterostructures III-V Nitride Semiconductors: Applications and Devices Eds E Yu and O Manasreh (Taylor & Francis) 161–191. (Also see http://nanolab.ucsd.edu/group/pdfpubs/GaNbkchap2.pdf.)