LED breakthrough highlights AlN promise

Two developments within the last month promise to unlock the potential of a relative newcomer to the compound semiconductor business - aluminum nitride (AlN). At just over 6 eV, the wurtzite AlN structure has the widest bandgap among semiconductors, and obvious potential for light emitters and detectors operating in the deep-ultraviolet portion of the spectrum.

But AlN is a very tough nut to crack. An electrical insulator with a melting point of 2200 °C and a boiling point only 317 °C higher, forming usefully-sized single crystals of the material for electronic applications has proved impossible until recently.

In the May 18 issue of the leading academic journal Nature, researchers at Japan s NTT reported the shortest-wavelength LED ever seen. Yoshitaka Taniyasu and co-workers doped epilayers of AlN to form the p-type and n-type parts of the device, which emitted 210 nm photons.

Although extremely inefficient and with a hefty operating voltage of 25 V, the LED made at NTT s Basic Research Laboratories represents a crucial first step towards the development of very low wavelength emitters that could be used to detect or destroy harmful biological species. So, just how did the NTT team do it?

In the Nature paper, the group describes how it used a refined doping strategy using silicon and magnesium to make the PIN LED. Their precise approach is the critical element of the work - until now, researchers had been unable to control the doping of the n-type and p-type layers of AlN precisely enough to demonstrate an LED.

"By reducing the dislocation density and finely controlling the silicon doping level, we were able to boost the room-temperature electron mobility," wrote Taniyasu.

III-N expert Asif Khan at the University of South Carolina (USC) in the US explains that this careful doping control means that sufficient conductivity can be imparted to the layers of AlN to form both p-type and n-type layers, which sandwich the undoped AlN layer in the device. As a result, enough electrons and holes can recombine to produce a useful number of deep-ultraviolet (UV) photons.

LEDs based on AlGaN that emit down to 244 nm are already available. In fact, Khan s own research group at USC has made such devices using a technique called migration-enhanced MOCVD, developed in collaboration with Sensor Electronic Technology, and is now working to refine their extraction efficiencies.

But doping AlN has proved much more troublesome. "Unfortunately, as the aluminum fraction increases, so too does the doping difficulty," wrote Khan in Nature. "It is hardest of all for AlN - which is in fact an insulator."

At present, the 210 nm LED reported by the NTT team is nowhere near good enough for any real-world applications such as water or air purification systems. The primary challenge facing the researchers now is to increase device efficiency and improve on the tiny output power - just 0.02 μW.

Reducing the number of dislocations in the structure will be the most effective way to do this. Since the NTT LED was fabricated on a SiC substrate, the lattice mismatch between the two materials yielded relatively high dislocation densities of 109cm-2.

That level ought to be greatly reduced by using native AlN substrates, which coincidentally have recently become available on the open market thanks to the US companies Crystal IS and The Fox Group. They hope that this will help to spur the commercialization of AlGaN-based devices for both optoelectronic and microelectronic applications. According to Tim Bettles, the new vice-president of sales and marketing at Crystal IS, the huge market for water filtration products could become a lucrative one in the long term, while swimming pools may also one day use deep-UV semiconductor emitters to ensure sterilization of nasty bugs.

The release of 2 inch single-crystal AlN substrates with a 50% usable area from Crystal IS should prove to be particularly beneficial for AlGaN-based optoelectronics, if not necessarily for LEDs with active AlN epilayers.

In the very long term, other applications such as optical data storage may present an opportunity for deep-UV lasers based on similar structures, although the pace at which hard disk, flash memory and Internet storage technologies are advancing suggests that a sufficiently large market to justify their costly development may never materialize. Either way, making reliable lasers based on AlN would require epitaxial growth of semiconductor layers with even lower defect densities.

Crystal IS CEO Ding Day told Compound Semiconductor that his latest substrates have a defect density of less than 104 defects per cm2, although Crystal IS also produces custom material for its clients with different specifications, and levels as low as 103 defects per cm2 have been shown.

"Alternative techniques to produce quasi-bulk AlN substrates do exist, but these all involve growing on non-native materials and result in high defect densities that are more than 100,000 times that of the native substrate," explained Day. "We are following in the footsteps of silicon carbide, [but] aluminum nitride is a very challenging material."

Growing a single crystal of AlN to form a boule from which AlN wafers can be sliced is the biggest challenge. Crystal IS has been awarded US patents that detail the use of a tungsten crucible to do this. The method quoted in US patent 6,770,135 describes how the growth chamber is first evacuated and then pressurized to about 1 bar with a gas mixture consisting of 95% nitrogen and 5% hydrogen. A polycrystalline AlN seed is then placed in the growth chamber and heated to around 1800 °C, resulting in a crystal that can be grown at 0.6-0.9 mm per hour.

Although using a native substrate should help to increase the internal conversion efficiency of the 210 nm LED by up to two orders of magnitude, Khan reckons that a great deal more work is required before any commercial application becomes plausible, and warns that it will not be a straightforward development path.

"For high-aluminum compounds, one can surely benefit from AlN substrates," said Khan. "For [AlGaN-based] 245 nm devices, efficiency can potentially be doubled by reducing defects by a factor of 10."

"However, for AlN LEDs the problem will be to find a compatible material with a bandgap larger than AlN," he added. "This will be needed to form quantum wells for efficient carrier capture."

Apart from an increase in efficiency by a factor of at least a million, the 210 nm LEDs need a reduction in operating voltage to well below 25 V for commercial use. To achieve the latter, the room-temperature conductivity of the AlN layers must be increased - probably by developing more efficient doping methods.

Taniyasu suggests that this could be achieved partly by using elements such as carbon, beryllium, zinc or cadmium instead of magnesium to dope the p-type layer, and partly by depositing carrier confinement structures such as quantum wells.

Although these are very early days for AlN, the recent advances made by both the NTT and Crystal IS teams suggests that its time will come.