European efforts propel nitride devices to a new level
A tried and tested route for improving device performance involves the introduction of new, superior material combinations. This can pay dividends with nitride electronics, which are traditionally based on the pairing of GaN and AlGaN. The introduction of InAlN in place of AlGaN reduces stress, leading to enhanced output powers and superior high temperature capability. What’s more, thanks to far stronger spontaneous polarization, the InAlN/GaN heterojunction produces twice the charge density of its Al0.25Ga0.75N/GaN cousin, enabling devices with higher frequency capabilities. And on top of this, it is possible to wring out further improvements by switching to diamond substrates, a step that aids thermal management of the device. Advances such as these have just been realized in a three-year project entitled Materials for Robust GaN (MORGaN). This effort, which kicked-off in November 2008 and was backed by €9.2 million (about $13 million) of funding from the European Commission, involves 23 industrial and academic partners from 11 nations (see Figure 1 for details). Figure 1. The European Commission funded project MORGaN –Materials for Robust GaN – involved 23 industrial and academic partners from 11 nations The great promise of the GaN/InAlN heterostructure came to light in a forerunner of MORGaN, a project known as UltraGaN, which started in 2005. Thanks to the success of both these projects, the InAlN/GaN heterostructure is now under close investigation by major research laboratories worldwide for its specific advantages in optoelectronic and microelectronic applications. Another goal of the MORGAN project is to exploit the thermal and physicochemical robustness of diamond. The former target has involved the development of new silicon/polydiamond composite substrates, along with the deposition of a nanocrystalline diamond coating on top of the wafer to enhance heat removal. Diamond coatings can also protect GaN, which is a major asset when this wide bandgap material is used to make sensors operating in aggressive electrochemical solutions. One of the hallmarks of the MORGaN project has been its focus on materials. More than half of all resources have been devoted to substrate development, strain management, heterogeneous semiconductor growth, refractory metals, three-dimensional metal manufacturing and ceramic packaging. This approach has born much fruit: Europe’s first 2-inch siliconpolydiamond composite substrates, the world’s first GaN HEMT grown on single crystal diamond operating in the microwave region, and InAlN/GaN devices operating in continuous wave at 3.5 GHz with an output of 6.6 W/mm and a power added efficiency of 70 percent (see the box “MORGaN’s milestones for more details”). Many applications are set to benefit from the successes of the MORGaN project. High-power, high-efficiency amplifiers based on InAlN HEMTs promise to cut the carbon footprint of mobile communication base-stations and various forms of power electronics used in consumer applications. In addition, first version pressure sensors capable of operating at temperatures up to 700 °C under pressures of several tens of bar have been fabricated, which should aid oil exploration and space missions, and also permit measurements in automobile and jet engines. And last but by no means least, MORGaN has spawned incredibly robust chemical sensors offering pH measurements over a large dynamic range. MORGaN’s milestones
Building on diamond Leading the development of ultra-high-conductivity platforms is the firm Element Six, which has optimised monocrystalline diamond substrates for the direct growth of GaN on diamond. Although the size of the substrates is relatively small – just 4 mm by 4 mm – electrical performance is promising. Larger sizes are possible by depositing diamond layers on silicon, and this approach has enabled Element Six to produce 2-inch free-standing wafers featuring a 2 μm-thick (111) silicon surface on a 70 μm polycrystalline diamond layer. Scaling to larger substrate sizes, such as 100 mm, will require additional work. One great attribute of the 2-inch polycrystalline composites is a thermal conductivity that reaches 1000 W m-1 K-1. Figure 2. A scanning electron microscopy image of a crack induced by tensile strain that terminates inside the silicon (111) layer Researchers at the University of Bath, UK, have deposited nitride layers on these complex silicon (111)/ polycrystalline diamond composite substrates. Highlights include the formation of high-quality, crackfree AlN and GaN layers, the latter of which is 350 nm thick (see Figure 2). MBE growth, using either ammonia or RF sources, has been used by engineers at EPFL, CH and FORTH to deposit nitride epi-structures directly onto single crystal diamond. A low-temperature AlN buffer layer was deposited first, followed by strain-engineered interlayers that allowed the subsequent GaN layer to be formed under compressive strain, prior to the growth of an 800 nm-thick GaN layer. On top of this went a HEMT structure, composed of a 24 nm-thick Al0.28Ga0.72N layer followed by a 2 nm-thick GaN cap (see Figure 3). Figure 3. (a)(left) The design of the HEMT structure grown on single-crystal diamond (b) (centre) An optical image shows the processed HEMT on single crystal diamond (c) (right)Measurements of current gain (h21) and power gain (MSG/MAG) as a function of frequency for a 2 x 0.2 x 75 x μm3 HEMT with 1μm thick nanocrystalline diamond heat spreader. The bias conditions were fixed at Vds=10V and Vgs=-2.5 V Hall Effect measurements revealed room-temperature electron mobility of 731 cm2 V-1 s-1 (1740 cm2 V-1 s-1) and a sheet carrier density of 1.3 x 1013 cm-2. Cooled to 77K, electron mobility and sheet carrier density in the structure increased to 1740 cm2 V-1 s-1 and 1.4x1013 cm-2. Thermal characterization shows a thermal resistance, Rth, of 3.5 K mm W-1. This is twice as high as that for a HEMT grown on more conventional 4H SiC substrate. Engineers at Technical University of Ulm have fabricated transistors with a 0.2 μm gate length from these wafers (see Figure 3(b)). Small signal measurements show a current gain cut-off frequency, ft, of 21 GHz and a power cut-off frequency, fmax, of 42 GHz (see Figure 3(c)). Although these values are slightly inferior to those obtained for AlGaN/GaN devices built on SiC, this is a very impressive result for a first trial. Additional work could be carried out to optimise the heterostructure and device processing. However, once this feasibility study was complete, the MORGaN consortium did not want to pursue this direction due to the limited current size of single crystal diamond as planned in the project. Handling pressure Pressure sensors were also developed during the MORGaN project. Some of these involve GaN cantilevers, which are self-supporting beams based on high-quality extended lateral overgrowth (ELOG) GaN, selectively etched from a silicon substrate. Other pressure sensors are ‘drumskin’ devices, combining a membrane made from a sapphire substrate with a HEMT sensor. All forms of sensor are housed in sophisticated packages made by FCubic, Swerea IVF, and Gooch & Housego. Advanced housing enables these sensors to operate at up to 1000 °C and 100 bar. The ELOG-based cantilevers, which are made at the University of Bath, produce deflections of several microns: Beams 5 μm wide and 500 μm long deflect 6 μm, and double-clamped variants that are 15 μm long deflect by 9 μm (The cantilevers are deflected by a piezo-controlled nano-positioner, and their deflection is determined from an interference fringe shift). Fabrication of these structures involves lateral overgrowth, which is based on a lithographic process. A thin sacrificial SiN layer is structured, before it is removed by buffered oxide etch after overgrowth to release the beams. It is possible to use a ‘double dogleg’ growth window with this lithographic process that enables ELOG growth along optimum crystallographic directions. Simulations to optimise the design of the drumskin sensor have been performed by the University of Bath and the University Joseph Fourier. These universities have developed mechanical models for the sensor, which has six sensing elements (see Figure 4). Modelling enables fine-tuning of the thickness of the sapphire substrate for a particular pressure range. Figure 4. Drumskin pressure sensors use several HEMTs as sensing elements. The sensors are capable of measuring pressures of 80 bar and above Another contributor to this effort is the IEE Slovak Academy of Sciences, which has developed a high temperature compatible fabrication process for this sensor. Conductive metal oxides are formed on the gate interface through thermal oxidation of evaporated and patterned thin nickel and iridium interfacial layers. Conductance is increased with this novel gate metallization process, and the device’s impressive transport characteristics are maintained after device annealing at 800 °C. Testing of these sensors in an environment capable of producing pressures up to 60 bar revealed that the drain-source voltage decreases in a linear fashion with increased voltage, diminishing by 0.02 percent per bar. This agrees with finite element models developed in the MORGaN project, which reveal that any deviations from linearity result from piezoelectric relaxation and imperfect clamping. When the sensor is placed in a bespoke package, it can operate at up to 80 bars and at 400 °C. AlGaN/GaN cantilevers have also been fabricated on silicon substrates, thanks to efforts at MicroGaN. In this process cantilevers are defined, before dry etching selectively removes the silicon substrate from underneath. The chip design includes a temperature sensor and two cantilevers in a Wheatstone bridge configuration. The sensor operates by measuring the deflection of one cantilever – the other one serves as a reference in order to compensate for temperature effects (see Figure 5). Measurements have been made on this sensor at temperatures up to 300 °C and a range of deflections. Figure 5. Optical microscopy image of a GaN-based cantilever Acid or base? Another aspect of the MORGaN project is the development by the Technical University of Ulm of ionsensitive FETs that feature diamond electrodes on AlInN/GaN HEMTs. These chemical sensors can be built by either monolithic integration on one chip or by hybrid integration. With both designs the nano-crystalline diamond electrode is exposed to the electrolyte, and when hybrid integration is used this electrode is also connected to an external AlInN/GaN HEMT. This heterostructure must have resistance to the nanocrystalline diamond thermal deposition budget, which is a temperature of 800 °C for many hours. Figure 6. (a) A packaged InAlN/diamond ion-sensitive FET value (b) Steady-state values of the drain current of InAlN/nano-crystalline diamond electrochemical sensor as a function of the pH Constructing these chemical sensors involves the mounting of FET chips in the centre of a PCB board with Cu/Au-based conductive pathways. Gwent Electronic Materials gave support to that task. These transistors are adhered to 48 pads on a PCB board by ultrasonic wedge bonding, a process that enables fast connection of the nano-crystalline diamond electrode to transistors with different geometries. Covering this sensor with synthesis glass allows observation of the device with an optical microscopic while it is being used (see Figure 6 (a)). Measurements reveal that the sensor has a reproducible pH sensitivity of 55 mV/pH, a sensitivity in the range of 20 mA/mm per pH (see Figure 6(b)) and a resolution as high as about 0.05 pH. One of the great strengths of the sensor is its fast response, which is limited by the seconds required to realise complete intermixing of the electrolyte. Dissipating the heat The potential benefits of coating GaN-based HEMTs with diamond have been recognized for many years: This yields a corrosion-resistant protection layer in harsh environments and forms an ideal heat-spreading layer for dealing with forced cooling from above. Converting the promise into a reality requires deposition at 700 °C or more of nano- or poly-crystalline diamond films in an atmosphere rich in hydrogen radicals at temperatures in excess of 700 °C. This requires an extremely stable semiconductor heterostructure, device contacts and passivation, which may also act as nucleation layer. The Technical University of Ulm fabricated the first submicron HEMTs overgrown with a 1 μm-thick nanocrystalline diamond heat spreader by bias enhanced nucleation. The small signal performances of the components are satisfactory, with an ft of 16.8 GHz and an fmax 6.4 GHz (see Figure 7). This result constituted a world first for the MORGaN project. Figure 7. Current gain (h21) and power gain (MSG/MAG) as a function of frequency for a 2 x 0.5 x 75 μm3 HEMT with 1 μm thick nanocrystalline diamond heat spreader. The bias conditions were fixed at Vds=8V and Vgs=-1V Ultra-powerful HEMTs Constructing 1 kW transistor technology was another goal of the MORGaN programme. To try and hit this output power, researchers at III-V Lab fabricated HEMTs with a 36 mm periphery and a 0.7 μm gate length on semi-insulating SiC substrates (see Figure 8). These transistors featured a 2 μm-thick GaN layer grown on SiC and a 10 nm-thick InAlN layer. Optimising device topology preserved microwave power gain and addressed thermal constraints. Due to the operating frequency, these devices tend to dissipate more power than they emit. When driven by 10 μs pulses, the basic cell 2 mm device delivers 13.2 Wmm-1 with a Power Added Efficiency (PAE) of 70 percent. Switch operation to continuous wave mode, and PAE drops to about 55 percent due to thermal self-heating. To improve the thermal environment for these power bars, they were mounted on polycrystalline diamond tabs supplied by Element Six (see Figure 9 for an example). Figure 8. (a) Scanning electron microscopy image of a 36 mm periphery, 0.7 μm gate-length HEMT power bar for L to S-Band operation. (b) Pout = 41.2 dBm (13.2 W – 6.6 W/mm) with PAE of 70 percent and Gp of 13.4 dB measured on wafer at 3.5 GHz and Vds= 35 V (2 mm device & short pulse setup) Figure 9. (a) Optical photography of a 36 mm InAlN/GaN HEMT thinned down to 100 μm thick SiC, mounted on diamond tab TM180 (appear black in image); The power bar was set to dissipate 3.5 W mm-1 in CW – the assembly is mounted on a jig at a temperature fixed at 62°C (b) Thermal infrared microscopy of the power bar shown in (a). (c) Infrared imaging of a similar dice but without a heatspreader and with a SiC substrate 400 μm-thick Prior to mounting, the SiC substrates were thinned to 100 μm to improve thermal management: Thermal conductivity of SiC is 450 W m-1 K-1, which is four times lower than that of high-quality polycrystalline diamond. The benefit of a high conductivity platform is revealed with thermal infrared imaging. When a 36 mm bar was mounted on a diamond heat spreader and dissipated 3.5 Wmm-1 on a jig kept at 62 °C, its maximum channel temperature was 173 °C (see Figure 9 b). This increased to 214 °C for a similar 400 μm-thick power bar, not mounted on a diamond heat spreader. It is possible to further improve the thermal management of the device with active heat exchangers, which can operate by pushing water into a cavity with a large surface area for heat exchange. Such systems have been designed by III-V Lab and Swerea IVF and produced by FCubic using their three-dimensions machining capabilities (see Figure 10). One of the latter firm’s contributions to the MORGaN project has been the development of a new process to make high precision copper parts using layer manufacturing. This has enabled the construction of heat exchangers with a thermal resistance, estimated by experiment, of 0.16 °C W-1. This incredibly low figure helps to dissipate hundreds of watts of heat from the HEMT. Armed with this heat exchanger, researchers have constructed a 2 GHz amplifier that includes two power bars of 36 mm periphery and produces 200 W. Switch to pulsed-mode operation, and a 250 W output is possible, which is a record for devices with an InAlN/GaN heterostructure. The limit of operation is 320 W, and PAE can hit 35 percent, which equates to 55 percent at the device level, according to reversed analysis. These incredibly impressive results are just one of the highlights of the MORGaN project, which has also led to improvements in materials, sensors, microwave devices and packaging. This article has only touched on some of the breakthroughs, and anyone seeking more details should take a look at the MORGaN public website: http://www.morganproject.eu/ Figure 10. (a) Copper heat exchanger made by three-dimensional machining (b) Partial view of 200 W L-Band using InAlN/GaN HEMT devices (c) Output power and PAE of amplifier including 2 devices of 36 mm at 2 GHz. Duty cycle measurements with 10 μs pulses were used as well as CW operation. Top: Output power; bottom: PAE