Fantastic Foundation Yields Great Devices
GaN substrates formed from ammonothermal growth underpin the fabrication of devices delivering outstanding levels of performance
BY PIOTR WILINSKI FROM AMMONO
If you want to make a great GaN device, you must start off with a native substrate. By doing this and thus employing homoepitaxial growth, issues associated with differences in thermal and lattice mismatch are eliminated, so the epiwafers are flat and material quality high.
A native substrate is actually the only option for growing ultraviolet, blue and green lasers – all alternative foundations lead to material quality that is insufficient for the manufacture of commercial products. Due to this monopoly, it is not that surprising that for the makers of GaN substrates, shipments to laser diode manufacturers account for a high proportion of sales. Lasers formed on these substrates feature in the read and write heads Blu-ray players and recorders, and shipments of edge-emitting chips should increase as they are deployed in optical laser displays, laser TVs, movie projectors and pico-projectors, printers, and lithography systems.
In comparison to the laser, the LED is a less demanding device, capable of delivering good performance with material of lesser quality, formed via GaN growth on a foreign substrate. However, for really high levels of performance at very high currents densities, a GaN substrate is invaluable. Soraa, Panasonic and Seoul Semiconductor use this foundation to make LEDs, and are causing something of a stir with devices delivering very high output power densities, while suffering from relatively little droop – the decline in device efficiency with increasing current. GaN substrates are also valued in the electronic domain. Armed with this foundation, engineers can turn to vertical device architectures and produce smaller chips with higher breakdown voltages, higher power densities and higher signal frequencies.
Variations in quality
It would be easy – but wrong – to think that all GaN substrates would yield the same results. While differences between the performances of devices made on different types of substrate may be bigger than the variations between those made on GaN, the quality of the native substrate does still have a big impact on the attributes of the resulting chip.
The majority of GaN substrates sold today are formed by depositing a thick layer of this material by HVPE to form a GaN crystal, and then slicing this up. Manufacture by this approach, which is employed by the likes of Sumitomo, Hitachi and Mitsubishi, produces a substrate that is not perfectly flat, because at some point GaN growth had to be initiated on a non-native substrate. What’s more, the defect density in these substrates is far higher than that found in InP and GaAs substrates – it is typically in the region of 106 cm-2.
GaN substrates with a far higher material quality are possible with the ammonothermal growth technique that we pioneer at Ammono of Warsaw, Poland. Our approach involves the growth of Ammono-GaN crystals in an autoclave containing GaN seeds, ammonia, and alkali metal amide mineralizers. Heating this concoction to around 550 °C, while maintaining it at pressures of typically 5000 atmospheres, leads to the growth of GaN crystals with a dislocation density that can be as low as 103 cm-2. They can be sliced into crystals that are exceptionally flat.
We are currently scaling up our production of GaN crystals to serve our customers, which number more than a hundred. By building bigger autoclaves, we can grow bigger crystals and increase the size of our substrates from 2-inch – our standard offering today – to 4-inch. The industry will eventually move to even larger substrates, and there is no fundamental reason why we cannot move accordingly, by producing 6-inch and then 8-inch material.
Scrutinizing our material with various common semiconductor characterisation techniques, such as X-ray diffraction, microscopy and optical methods has already proven that this form of GaN is superior to its HVPE-grown cousin (see Bulk GaN: Ammonothermal trumps HVPE from Compound Semiconductor March 2010, p.12). Also, intrinsically, for bigger wafers sizes the production cost for an Ammono-GaN wafer is lower than for HVPE GaN. To what extent this leads to superior device performance is still to be fully established, but we are making efforts to determine this by teaming up with various partners, together producing and evaluating devices built on our substrates.
In the majority cases, results obtained by these partners are confidential and covered by non-disclosure agreements. But there are some instances where we can reveal findings, because participants in the project have received external funding and there is an obligation to report results in the public domain.
One such effort has focused on the fabrication and testing of laser diodes. This has showcased the virtues of working with GaN substrates formed from ammonothermal crystal growth.
When nitride lasers can deliver an output of a Watt or more, they can be used in laser displays delivering exceptional brightness and colour resolution. What’s more, these sources can be used to construct high-power laser arrays that form the heart of efficient white-light sources involving the pumping of a phosphor. This technology is being pursued by BMW for use in car headlights.
Working in partnership with TopGaN Lasers and the Institute of High Pressure Physics in Warsaw, Poland, powerful arrays of InGaN laser diodes have been fabricated. By forming several stripe emitters on a single chip, the power density per emitter is reduced. This avoids catastrophic optical damage that leads to facet melting – which can occur above a threshold current density of around 50 MW/cm2 – and ultimately creates devices with long lifetimes.
Violet-emitting chips formed in this manner with three and 16 emitters have delivered a maximum output power of 2.5 W and 4 W, respectively (see Figure 1). These laser, which deliver a performance that is ‘top of the class’, typically operate at 408 nm - 412 nm. Threshold current for three-stripe emitter is 1.2 A, corresponding to a current density of 5 kA/cm2, and threshold voltage is 6.5 V. Meanwhile, slope efficiency is 0.76 A/W, and for a 10 mm stripe-width device the line-width is very narrow – it is just 0.25 nm. Another impressive attribute of these chips is that their light-current characteristics for individual stripes are fairly similar, with emission comparable within an experimental error of ±0.4 nm. This means that the emission wavelength is very uniform within one chip.
Figure 1: UV Laser Matrix emission spectrum (a) and the dependence between the optical power and current (b) (courtesy of TopGaN).
As explained previously, this multi-emitter chip design avoids catastrophic optical damage at the facet, leading to a long lifetime. For our 16-emitter device, which delivers an output of 4 W at a 6A drive current, the operating current density is just 18 kA/cm2. This enables a reasonable lifetime that exceeds 6000 hours.
We attribute the world-class performance of these lasers to the exceptional quality of the Ammono-GaN substrates. If the defect density in a laser is high, or non-uniform, it is well established that this impairs performance, shortens lifetime, drags down efficiency and hampers reliability. These issues do not exist in our partner’s lasers − one of the great strengths of our substrates is their very low, homogeneous dislocation density that is replicated in the epitaxial structure.
A further virtue of our substrates is their high level of conductivity, which stems from an electron concentration of the order of 1019 cm-3. This high concentration produces a lowering of the refractive index in the GaN substrate, enabling a so-called plasmonic effect. This is highly beneficial, because a higher output power results from a lower level of laser mode penetration into the substrate, and enhanced wave-guiding and light confinement within the laser chip. Consequently, it is possible to trim the AlGaN cladding layers while maintaining high-performance. In turn, this leads to a simpler laser diode structure that is: less bowed, thanks to smaller stress generated by lattice mismatch between the AlGaN cladding and the InGaN/GaN quantum wells; and has a lower defect density than conventional devices, which have thicker AlGaN claddings.
Figure 2: Epistructure of the GaN Schottky barrier diode and a corresponding scanning tunneling electron microscopy image. The growth interface between bulk substrate and epitaxial layer is invisible. (Courtesy of University Notre Dame)
Trimming the AlGaN cladding thickness has proven to be highly beneficial for the fabrication of Ammono-GaN 460 nm, blue laser diodes grown by plasma-assisted MBE. This growth technology is ideal for growing quantum structures, because it can realise very sharp interfaces and provide very good control of epitaxy parameters at relatively low temperatures in a non-hydrogen environment. By reducing the thickness of the AlGaN cladding, lifetime of the laser can top 2000 hours. This Ammono-GaN based device, which delivered a maximum output power of 80 mW, delivers outstanding performance for an MBE-grown blue laser diode.
TopGaN has produced lasers with Ammono’s GaN substrates that deliver state-of-the-art output powers.
To produce green-emitting devices based on the nitrides, engineers have to increase the indium content in the quantum well. In conventional structures grown on the c-plane of GaN, this leads to high internal electric fields that pull apart the electrons and holes and hamper light emission.
These unwanted electric fields can be reduced, and even eliminated, by growing devices on different planes of GaN, known as semi-polar and non-polar planes. Ammono-GaN substrates with these orientations are available commercially from us.
Our partners have started to investigate the potential of devices grown on this classes of substrate, fabricating LEDs and laser diodes on the semi-polar (2021) plane. Using plasma-assisted MBE, LEDs were formed on this substrate and also on the (0001) plane. Peak emission wavelength varied considerably between the two forms of device, with emission at 387 nm for the semi-polar LED and 462 nm for the non-polar, due to considerably lower indium incorporation for growth on the semi-polar plane.
Characterisation of the semi-polar epitaxial LED structure reveals a smooth surface morphology with atomic steps, according to atomic force microscopy, and high photoluminescence intensity. Inspection of the material with a transmission electron microscope indicates that the interfaces are abrupt and there is good structural quality. Ultraviolet laser diodes emitting at 388 nm have also been formed on ammonothermal (2021) substrates. Devices with a ridge waveguide along  direction exhibited a threshold current density and voltage of 13.2 kA/cm2 and 10.8 V. These values are quite high, but should be reduced by processing the ridge along the  direction, because this should lead to higher material gain, thanks to in-plane anisotropy.
DC and RF characteristic of the Ammono-GaN HEMT(courtesy Institute of Electron Technology, Poland)
Ammono-GaN HEMT cross-section (left) and view from the top (right) (courtesy Institute of Electron Technology, Poland)
Native substrates should also aid GaN power devices. Currently, these are manufactured on either silicon or SiC, but if GaN were used, this would allow a switch from a lateral to a vertical device architecture. With lateral devices, there are many problems associated with lateral current flow near the buffer layers and overlying dielectric layers, including current-collapse, high dynamic on-resistance and an inability to support avalanche breakdown. Further downsides arise from stress in the buffer layers that cause wafers to bow, limit epilayer thickness and ultimately place a ceiling on the maximum blocking voltage, which is typically no more than 1.2 kV.
Turning to a native substrate allows GaN power devices to be constructed with a vertical architecture. Current can then flow through the epitaxial layers, leading to devices that are immune from current collapse and dynamic on-resistance.
In such devices, which exhibit true avalanche breakdown, charge trapping effects and barriers to heat removal are absent, and it is possible to grow very thick layers of GaN that produce very high blocking voltages.
Our partners have started to investigate the potential of such devices grown on native substrates, evaluating the performance of GaN Schottky barrier diodes and HEMTs. At the IWN conference held this year, we co-authored a paper detailing a 6 GHz RF power transistor formed on Ammono-GaN, following our collaboration with the Institute of Electron Technology from Warsaw. In HEMTs and GaN Schottky barrier diodes, the low defect density in the ammonothermal substrate is retained in the epilayers, which have an atomically smooth surface and flat interfaces.
Comparisons between devices built on different forms of GaN substrate are still to be carried out. However, some very impressive results were realized with devices formed on Ammono-GaN (see Figure 2). Schottky diodes produced at University of Notre Dame produce a leakage current density of the order of 10-11A cm-2, according to publications by this team.
The electric field at breakdown voltage in these devices is 3.3 MV/cm, so close to the theoretically predicted critical field of 3.5-3.8 MV/cm. The HEMTs on Ammono-GaN have a two-dimensional electron gas density of 8 x 1012 cm-2 with a mobility of 1600 cm2 V-1 s-1 at 300 K and 8000 cm2 V-1 s-1 at 77 K.
These results, plus those of the LEDs and lasers, highlight the promise of Ammono-GaN, and indicate that better substrates lead to better device results. Availability of this material will steadily increase, and in time there should be more devices based on homoepitaxial GaN on the market delivering top of the class levels of performance.