Improving p-type nitrides

Figure 1.  A transmission electron microscope image reveals the absence of inversion domains and magnesium precipitates in p-type GaN films grown by MME. Magnesium concentration is 2.6 x 10²⁰ cm^-3.

BY BRENDAN GUNNING AND ALAN DOOLITTLE FROM GEORGIA INSTITUTE OF TECHNOLOGY

Sales of III-nitride chips are soaring, and will continue to do so as new power electronics and RF products flood the market. However, this commercial success should not obscure a major weakness in the III-nitride device portfolio − realizing high-levels of p-type doping. Even in state-of-the-art devices, hole concentrations are limited to the low-10¹⁸ cm^-3 range, a doping level that is trivial to reach in traditional III-Vs.

Poor p-type doping cannot be taken lightly, because it holds back the performance of every leading III-nitride device. In LEDs, it is to blame for weak hole injection, as well as high series and contact resistances; and in HBTs, which have the potential for massive power handling for RF applications, it is the cause of high base and base access resistances. Meanwhile, when it comes to solar cells, the lack of a p-n tunnel junction hampers the development of multi-junction cells providing strong absorption over a very wide spectral range. So it is clear that if III-nitride devices are to continue to evolve, improvements in p-type doping are essential.

The workhorse for the manufacture of III-nitride devices is the MOCVD reactor. Engineers operate this at temperatures of 900°C or more to ensure sufficient cracking of the source of nitrogen, the ammonia precursor. Such high temperatures are a double-edged sword, ensuring high crystal quality but also making p-type doping challenging.

These p-type doping issues are associated with the high density of point defects, such as nitrogen vacancies that compensate the magnesium acceptor. If high levels of magnesium doping are introduced, the nitrogen vacancy is even more energetically favourable "“ and this is disastrous, because it forms an electrically inactive complex with the magnesium dopant. What's more, at high magnesium doping levels, crystal quality tends to degrade, due to the combination of inversion domains and magnesium precipitation, which can even result in cubic inclusions and stacking faults.

Adding to the list of drawbacks, the hydrogen-rich atmosphere employed for MOCVD growth produces Mg-H complexes that lead to highly resistive p-type films in the as-grown state. A post-growth annealing step is needed to address this, with Mg-H complexes broken to yield conductive, p-type films.

Circumventing some of these problems is MBE. The two common forms of this growth technology, plasma-assisted MBE and ammonia MBE, use substantially lower substrate temperatures that typically range from 600 °C to 900 °C. Thanks to this, vacancies are less energetically favourable, magnesium incorporates more freely, and films can be conductive and measurably p-type without post-growth annealing steps, simplifying fabrication processes and reducing the thermal budget.

However, turning to traditional MBE is by no means a perfect solution. Inversion domains are still an issue for magnesium concentrations of 5 x 10¹⁹ cm^-3 or more at growth temperatures of 700 Â°C to 750 Â°C, and this places an upper bound on hole concentrations of the low-10¹⁸ cm^-3 range.

So, to sum up, mediocre p-type doping is a thorn in the side of MOCVD and traditional MBE techniques. So what is the way forward? Well, certainly not higher temperatures, because this regime has been exhaustively explored already. Turning to lower temperatures might pay dividends, however, and it is there that our team from Georgia Institute of Technology believes that the most significant advances can be made.

Going down in temperature

We have been investigating growth at lower temperatures for several years. One of our first breakthroughs has been to demonstrate magnesium concentrations of greater than 10²⁰ cm^-3 in films of GaN that are free from inversion domains and magnesium precipitation. This involved the use of low-temperature, plasma MBE under nitrogen-rich conditions, which resulted in rough surfaces and poor crystal quality.

To address these weaknesses, we then moved to a variant of MBE that is known as metal-modulated epitaxy (MME). With this approach, growth of nitrides at around 600 °C involves periodic opening and closing of shutters for the group III and dopant sources, while the nitrogen plasma continuously impinges on the surface. When the metal shutters are open, the incident metal flux exceeds that for nitrogen, with excess metal accumulating on the surface; and when the metal shutters close, nitrogen, which continues to flow, consumes this excess metal.

It is important to realise that the liquid metal that is formed on the surface by this process is not fixed in position, but moves freely in a highly mobile pool. This produces two favourable outcomes: large diffusion lengths for both metals and nitrogen and thus, very smooth films. However, if there is no excess metal on the surface, the incoming metal atoms fail to migrate far enough, resulting in rougher films. Note that an excess of metal is particularly important for ensuring smooth films at low temperatures, because atom diffusion lengths are shorter in this growth regime.

Turning to MME has enabled us to produce p-type GaN films at 600 Â°C that have magnesium concentrations exceeding 2 x 10²⁰ cm^-3 and an absence of inversion domains and magnesium precipitation (see Figure 1). According to Hall measurements, the hole concentration of this particular film shown in Figure 1 is 4.5 x 10¹⁹ cm^-3. That is not our best result, however, as we have formed films with a hole concentration as high as 7.9 x 10¹⁹ cm^-3 and a resistivity below 0.2 Ω-cm. Although under certain growth conditions this hole concentration contains a small contribution to the charge density from a polarization induced charge at the interface between the p-type layer and the AlN buffer layer, this makes a small contribution to the p-type doping, which has a level that is more than an order of magnitude higher than that found in today's commercial III-nitride devices.  Most of the hole conduction is via impurity band conduction not 2D hole gas conduction.

When we select MME growth conditions for highly p-type GaN, we are now able to routinely produce hole concentrations of 1-3 x 10¹⁹ cm^-3, alongside a resistivity of 0.2-0.4 Ω-cm, for 100 nm-thick films held at room temperature. Mobility is just 0.5-5 cm² V^-1 s^-1, but that is to be expected, given the extremely high magnesium concentration, which results in substantial ionized impurity scattering and a mobility dominated by an impurity band, rather than being governed by the valence band.

The characteristics of these films at cryogenic temperatures are very encouraging. There is minimal carrier freeze-out, with hole concentrations of 8-9 x 10¹⁸ cm^-3 near 77K, and conductivity is retained, with a resistivity of less than 0.5 Ω-cm (see Figure 2). In comparison, traditional p-type GaN films are resistive at 77K, with hole concentration plummeting with temperature.

Figure 2.  Temperature-dependent resistivity of traditional p-type GaN (black circles represent a hole concentration 2.2 x 10¹⁸ cm^-3 at 300K) and highly magnesium-doped p-type GaN grown by MME (red triangles, represent a hole concentration of 1.9 x 10¹⁹ cm^-3 at 300K).

Formation of an impurity band accounts for the lack of carrier freeze-out in these p-type GaN films with a magnesium concentration exceeding 10²⁰ cm^-3. In a traditional p-type GaN film with a lower magnesium concentration, acceptors are isolated, rather than being close enough to interact. But when the acceptor concentration exceeds 10²⁰ cm^-3, the Bohr radii of the holes overlap, allowing magnesium acceptors to interact and form a band, rather than a single acceptor energy within the band gap (see Figure 3).

Figure 3. The Bohr orbital overlap results in acceptor band formation for highly magnesium-doped GaN (right), compared to the traditional isolated acceptor for lower magnesium concentrations (left).

This switch from a single acceptor energy to a band leads to an increase in ionization efficiency, because the lower part of the band gets closer to the valence band, lowering the effective activation energy of the magnesium acceptor. Such a situation is well known in highly doped traditional III-V semiconductors, and accounts for the change in carrier freeze-out in these materials. We have also obtained photoluminescence spectra from films with a high level of magnesium doping to provide further evidence for the formation of an acceptor band. Spectra from a 100 nm-thick film with a room temperature hole concentration of 5.8 x 10¹⁹ cm^-3 feature a 3.28 eV peak with a full width at half maximum of 240 meV. Of particular interest is the high-energy tail of the peak, which intercepts the band edge. This overlap, and the broad width of the photoluminescence, is consistent with the formation of a broad impurity band. The spectra are indicative of high-quality material, as there are no signs of yellow or other deep level defect luminescence.

To demonstrate the capability of our MME technology more fully, we have grown several p-type AlGaN films on AlN templates. Using growth temperatures of 620 °C and 640 °C, we have grown 100 nm-thick films with compositions of Al_0.11Ga_0.89N and Al_0.27Ga_0.73N, respectively. Both ternaries exhibit highly p-type conduction with hole concentrations of approximately 2.6 x 10¹⁹ cm^-3. Resistivity is just 0.3 Ω-cm in the Al_0.11Ga_0.89N film, rising to 1 Ω-cm in the Al_0.27Ga_0.73N variant, due to reduced mobility.

We have also investigated the photoluminescence produced by the Al_0.11Ga_0.89N film (see Figure 5). Its spectra has similar characteristics to that produced by the GaN film, with a broader peak that again intercepts the valence band edge.

Figure 4.  Room temperature photoluminescence spectrum of highly magnesium-doped GaN grown by MME exhibits only a broad, ultraviolet emission that crosses the valence band edge marked by the dotted red line.

Figure 5.  Room temperature photoluminescence spectrum of p-type Al_0.11Ga_0.89N grown by MME. A broad ultraviolet emission (the full-width at half-maximum is nearly 440 meV) is seen and the high energy tail intercepts the band edge marked by the dotted red line.

Building prototypes

To determine whether this highly p-type material is viable for making devices, we grew a very basic, non-optimized LED and a p-i-n diode. The LED features 2 nm-thick In_0.13Ga_0.87N wells separated by 10 nm-thick unintentionally doped GaN barriers, and also a 20 nm thick p-type Al_0.1Ga_0.9N electron blocking layer and a 50 nm thick p-GaN contact layer. Both of the p-type layers in this device were grown using conditions similar to those described above, and lead to hole concentrations that are typically greater than 2 x 10¹⁹ cm^-3. Meanwhile, the p-i-n diode comprised only GaN, with an intrinsic region just 20 nm thick, followed by 50 nm of GaN that is highly doped with magnesium. 

Current-voltage measurements on both of these 660 Âµm-diameter devices revealed a proper-turn on at around 3 V, and a series resistance of 6 Ω for the p-i-n diode and 10 Ω for the LED. Note that these values are for devices subjected to no annealing whatsoever; there was no post-growth anneal, nor any contact anneal during or after device fabrication, owing to the extremely high doping of the films.

Operating at room temperature, peak emission from the LED occurs at 425 nm, while that for the p-i-n diode is at 381 nm. Emission at this latter wavelength has been seen in photoluminescence measurements on GaN (see Figure 4), indicating that it probably originates from recombination within the p-type layer. A broad visible luminescence also emanates from this p-i-n diode, and this is not seen in any GaN grown by MME, regardless of doping. We believe this emission results from hole injection into the n-type region and subsequent diffusion through the 200 nm MME-grown n-type region into the underlying MOCVD-grown GaN template.

The lack of carrier freeze out in our MME-grown GaN allows devices to emit light at cryogenic temperatures. Both types of devices can emit light while submerged in liquid nitrogen, indicating the presence of sufficient hole concentrations at 77K (see Figure 6).

Figure 6.  Current-voltage curves (left) for non-optimized p-i-n diode (black) and LED (red), and images of both devices operating while submerged in liquid nitrogen (p-i-n diode is top right, LED is bottom right).

These results on preliminary devices illustrate the promise of highly p-type material. It offers a hole concentration that is nearly 30 times that of the previous state of the art, and is a breakthrough that might seem to good to be true "“ it has been accompanied by some justifiable scepticism. Hopefully, however, the naysayers will be persuaded that these films have great p-type performance, thanks to demonstration of properly rectifying devices exhibiting electroluminescence, even at cryogenic temperatures, and the formation of ohmic contacts without contact annealing.

There is still a long way to go, despite our successes to date. Our devices are certainly not state-of-the-art − in either design or operation "“ and the function that they fulfil is to simply demonstrate the viability of this highly p-type material for device applications.

On top of this, they also suggest that there is a need to revisit MBE as a tool for making green/and ultraviolet LEDs. After all, in addition to producing great p-type film, this growth technology offers benefits of low temperature growth, such as high indium incorporation, extremely high uniformity, high source material utilization, and excellent interface and dopant diffusion control. And GaN growth rates are comparable to MOCVD, with deposition at more than 8 Âµm/hour "“ while for InGaN, MBE can top 1 Âµm/hour, enabling faster growth than that occurring in an MOCVD tool.