Bridging the green gap with a new foundation

Relaxed InGaN substrates open the door to high quality lighting and full colour displays based on III-nitride LEDs

BY AMà‰LIE DUSSAIGNE FROM CEA/LETI AND DAVID SOTTA FROM SOITEC

The majority of commercial white LEDs are based on a blue-emitting chip that pumps a phosphor. This combination has generated tremendous sales, but it does have two major weaknesses: an energy loss of typically 25 percent, associated with the phosphor down conversion; and a colour rendering index that is not that high.

Both these drawbacks can be addressed by direct colour mixing, using LEDs with a variety of emission wavelengths. The ideal way to accomplish this is to use one material system that spans the blue to red, because this simplifies colour control. Do this, and the technology could also find deployment in high-brightness displays.

Fulfilling this dream is not easy, however. The main issue is the material. Although the InGaN alloy can theoretically cover the whole visible range, its quantum efficiency plummets above 500 nm as emission shifts from green to red. Switching to the AlGaInP material family allows the production of high-performance amber and red LEDs, but in this case efficiency takes a nose dive as emission stretches to wavelengths shorter than 570 nm.

The 500 nm to 570 nm spectral range, where it is difficult to realise efficient emission, has its own name: the green gap. For GaN-based LEDs, lack of success stems from a reduction in material quality with increasing indium content in the InGaN alloy. This issue can be traced back to low miscibility of indium in GaN, and the high lattice mismatch between the GaN buffer layer and the InGaN quantum wells.

Our team at CEA/Leti and Soitec has tackled this issue head on with a novel substrate technology that allows InGaN quantum wells to combine a high indium content with great material quality. This is accomplished with substrates that are better suited to the growth of InGaN than the most common one used for nitride LEDs, sapphire.

Growth of GaN-based LEDs on sapphire often begins with the deposition of a GaN buffer layer. That's not a great start, as the buffer layer leads to strong compressive strain. Making matters worse, when InGaN quantum wells are added, the strain gets even higher. Compounding these issues, internal electric fields in the InGaN/GaN heterostructures hamper LED efficiency when wells are widened, or indium content increased.

The electric field can be eliminated by switching to non-polar GaN planes. But it's not a great solution, as more indium is needed to realise the same emission wavelength.

If concerns over electric fields are put to one side, and the focus placed on limiting the strain, one way forward is to insert AlGaN layers in an InGaN buffer or an InGaN active zone. This has been tested, but no reports of highly efficient LEDs have followed.

Our idea is to grow the LED on a relaxed InGaN substrate, as this reduces the strain between well and buffer. Additional merits of this platform are an increase in the indium incorporation rate and a reduction in the internal electric field for the same emission wavelength.

This particular substrate is not available - but we have something similar, an InGaN pseudo-substrate called InGaNOS, based on Soitec's Smart Cut technology. Development of this foundation, formed by transferring a thin layer of InGaN to a sapphire substrate, has been motivated by efforts to reduce the efficiency droop in high-brightness blue LEDs. However, through our work, led by CEA-LETI, we are showcasing the capability of this platform in another arena "“ its potential to form the base for structures emitting at long wavelengths. Emission can span blue to amber when InGaN LEDs are grown on an InGaNOS substrate, thanks to the increase in the indium incorporation rate.

Forming the foundation

Our starting material for making InGaN pseudo-substrates is an InGaN donor template. This template is formed by growing a 3 mm-thick GaN buffer layer on a 100 mm sapphire substrate, followed by an In_xGa_1-xN layer that is typically 200 nm-thick and has an indium content, x, of between 1.5 percent and 8 percent. This InGaN donor template, which is strained due to lattice mismatch between GaN-on-sapphire and the In_xGa_1-xN layer, is hydrogen implanted. In parallel, a compliant layer is deposited onto a sapphire substrate, before molecular bonding unites the InGaN donor template and sapphire handle substrate.

The next step is to transfer the donor strained In_xGa_1-xN layer to the compliant layer of the sapphire handle substrate, using Smart Cut technology. Photo-lithography and dry etching follow to create strained patterns, which are then relaxed, thanks to the compliant layer. This relaxation process proceeds via successive thermal annealings (see Figure 1 for an overview of the entire process).

Figure 1. The fabrication process for making InGaNOS substrates helps to increase the emission wavelength of InGaN-based quantum wells.

It is not easy to produce a uniform relaxed In_xGa_1-xN layer while maintaining a flat surface. A competitive relaxation mechanism is at play that can cause buckling, and lead to wavy, cracked surface. But we can avoid this by optimising the process, to produce flat patterns from 100 x 100 mm² up to 1000 x 1000 mm² and a partially relaxed In_xGa_1-xN layer with a value of x of up to 8 percent. It is also possible to process patterns down to 10 x 10 mm² with this technology.

The relaxation process produces In_xGa_1-xN patterns with N-face polarity. That's not ideal, as a top Ga-face is needed for standard GaN growth. So the full patterned wafer is bonded to a second sapphire handle substrate by molecular bonding, before laser lift-off removes the first sapphire substrate.

To evaluate the roughness of our partially relaxed In_xGa_1-xN layers, we have inspected them with atomic force microscopy. This reveals that the roughness is equivalent to that of the donor (see Figure 2(a) for a microscopic view of a 300 x 300 mm² pattern with a top partially relaxed In_xGa_1-xN layer).

Figure 2. (a) Microscopic view of a 300 x 300 mm² pattern with a top partially relaxed In_xGa_1-xN layer. (b) An 100 mm InGaNOS substrate with an In_0.08Ga_0.92N layer and 300 x 300 mm² patterns at 3.205 Angströms. The brown area is the patterned In_0.08Ga_0.92N layer.

In the relaxation process, a key characteristic is the lattice parameter of the In_xGa_1-xN layer. The GaN buffer fixes the strained In_xGa_1-xN layer a lattice parameter of the donor at 3.184 Angström. According to the indium content of the donor and the relaxation process, the InGaNOS a lattice parameter can be tuned up to 3.205 Angström, with a uniformity within the wafer of +/- 0.0005 Angström.

We have used our technology to produce wafers that feature partially relaxed In_xGa_1-xN square patterns on a sapphire substrate. The thickness of the In_xGa_1-xN layer in each pattern is 100 nm, and it has a standard deviation of 4 nm. This level of uniformity ensures a continuous seed layer for LED growth within the pattern. The surface is epi-ready, and it has the preferred Ga-face polarity (see Figure 2 (b) for an example). Note that at Soitec, there is a InGaNOS pilot line for 100 mm and 150 mm wafers, and the technology can be scaled to 200 mm.

Multicolour emission

To evaluate the benefits of the relaxed InGaN substrate, we have characterized full InGaN structures that are grown on the InGaNOS substrates by MOCVD. The impact of lattice mismatch on indium incorporation in InGaN quantum wells is assessed with three different InGaNOS substrates "“ they have a lattice parameters of 3.190 à…, 3.200 à…, and 3.205 à… (see Figure 3). Note that no surface preparation is needed, as InGaNOS substrates are epiready.

Figure 3.

(a) a lattice parameter of relaxed or strained GaN and of the three type of InGaNOS.

(b) GaN wurtzite structure showing a lattice parameter position.

The structure that we have grown consists of a 200 nm-thick In_yGa_1-yN buffer layer, followed by In_zGa_1-zN/In_yGa_1-yN (with z > y) active region with five quantum wells (see Figure 4).

Figure 4. The full InGaN structure on InGaNOS that is used to highlight the capability of this substrate for long-wavelength emission.

In order to assess the impact of a relaxed InGaN substrate on the indium incorporation rate, it is imperative to apply exactly the same growth conditions to all three InGaNOS substrates. To achieve this, we co-load all the samples in the same run.

We have selected an indium content in the InGaN buffer layer of 4 percent, because this equals the average indium content in the three InGaNOS substrates. Three different growth conditions are then employed for the active region: those that we use in a conventional structure, as well as those with a higher InGaN growth rate and a lower growth temperature. The latter two variants increase indium content, and ultimately shift emission to longer wavelengths. To provide a benchmark for this study, we compare the results with a reference sample "“ a conventional structure that is grown on a GaN template and features GaN barriers.

The superiority of our InGaNOS substrates at incorporating indium has been validated by plots of photoluminescence redshift as a function of the InGaNOS a lattice parameter. Using 325 nm excitation, and measuring spectra produced by samples with just an InGaN buffer layer on either an InGaNOS substrate and on the GaN template, we have found that the higher the a lattice parameter, the longer the emission wavelength. According to measurements, indium content increases from 4 percent on the GaN template to 6 percent to 7.5 percent on an InGaNOS substrate. The only possible explanation for the increase in indium content is the relaxed InGaN substrate, which has a different a lattice parameter. This is a very encouraging result, showing that the InGaNOS substrate is beneficial at the buffer stage, even for low indium content.

After this, we went on to measure the spectra produced by our In_zGa_1-zN/In_yGa_1-yN multi-quantum wells, using excitation at 405 nm, because this directly pumps the active region. Results show a red shift as high as 62 nm when the active region is grown in a conventional manner on the InGaNOS substrate, rather than a GaN template (see Figure 5 for an example in green range).

Figure 5. Photoluminescence spectra at room temperature of full InGaN structures (green conditions) grown on InGaNOS 3.190 à… (blue curve), InGaNOS 3.200 à… (green curve), and InGaNOS 3.205 à… (orange curve). The reference sample on a GaN template has been added for comparison (black curve).

Whether the quantum well is emitting in the blue, green or amber, there is a significant red-shift in emission wavelength with an increase in the InGaN a lattice parameter (see Figure 6 (a)). Note that it is easy to obtain amber emission at 594 nm with InGaNOS 3.205 à….

Figure 6. (a) Room-temperature photoluminescence of full InGaN structures grown on InGaNOS for emission wavelength versus InGaNOS a lattice parameter for the different growth conditions. (b) Photoluminescence spectra from blue to amber emission on InGaNOS 3.190 à… (blue curve), InGaNOS 3.200 à… (green curve), and InGaNOS 3.205 à… (orange curve), respectively.

With all these InGaN-based heterostructures, the internal electric fields can cause a red-shift in emission with increasing quantum well thickness, even if the indium content is maintained. According to transmission electron microscopy, our samples have well widths of around 2.8 nm. That's a value we expected, is the same as for a conventional structure, and is thin enough to prevent the internal electric field from playing a strong role on determining the emission wavelength.

What then, must be the explanation for the redshift in emission? It is the increase in indium content with a lattice parameter. As the lattice mismatch between the buffer and InGaN wells is smaller with InGaNOS than it is with GaN "“ and with InGaNOS, the value of the a lattice parameter is unusually high "“ the indium incorporation rate is enhanced. With the relaxed InGaN buffer, the indium compositional pulling effect is diminished, allowing growth on InGaNOS substrates to incorporate more indium for the same growth conditions.

To determine the quality of our material, we have undertaken internal quantum efficiency measurements. For material emitting at 536 nm, the efficiency is 31 percent, which is a very respectable value, considering that there has been no particular optimization of the structure's growth conditions. This result provides further evidence of the promise of the InGaNOS substrate and InGaN active regions, which together are capable of multicolour photoluminescence emission (see Figure 7).

Figure 7. A full InGaN structure grown on InGaNOS samples can span from blue to amber.