View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Elsevier - Publisher Connector Dr Alan Mills The Fall MRS Meeting was held in Boston Pacific Rim attendance. However, the quality PO Box 4098, Mountain View, last November, although to smaller audi- of the papers and the improvements in Ill- CA 94040. USA ences than usual. Even the normally popular Tel/Fax: +1-6~0-96B-z~8~/8k16 nitride and silicon carbide device process E-mail: akilk%nr~a&.cbm wide gap symposium attracted about 20% technology was not diminished, and many fewer attendees, with a noticeable decline in papers were worthy of review. Advances in wide bandgaptechnology Analytical advances - micro- range.Thus, much more detailed temperature temperature variations data for micron and sub-micron gate devices can be derived than was previously possible from the Most Group III-nitride compound devices are approximately 10 micron resolution limit avail- still epitaxially grown on hetero-substrates such able from infrared measurements. Micro-Raman as sapphire, but the development of peak per- spectroscopy has also been recently used to formance, high frequency and high power determine the localized chemical reactions in devices depends on better lattice matched, heat organic light emitting diode active layers. conductive substrates and to a large extent on In this research,AlGaN/GaN HFET heterostruc- the ability to accurately monitor and thereby ture devices (28nm of AlGaN with 23% of Al on control temperature extremes in an operating 1.2 microns of GaN) were grown by MOCVD chip. Martin Kuball from the University of processes in a showerhead-type reactor on both Bristol, in cooperation with Trevor Martin and sapphire and silicon carbide substrates.The sili- Mike Uren (for growth and device design) from con carbide HFET was quite respectable with a QinetiQ Ltd., Malvern, described the use of mobility of 1440cm*/Vs, an ft of 9.8GHz and an Figure 1. Current/voltage micro-Raman spectroscopy in an elegant method fmax of 39GHz. Devices on both substrate types plots for AIGaNIGaN HFET to determine device self-heating effects and the heterostructure devices were analysed by micro-Raman spectroscopy temperatures generated with respect to local with and without the laser in using a 3mW 488nm laser spot, with an approxi- operation (Kuball et al) device geometries in the sub-micron dimension mate diameter of one micron. Operating under DC drain bias voltages of up to 2OV,the HFETs were scanned on an XYstage in 0.2 to 2 micron (b) steps and their gallium nitride E2-phonon fre- With laser on With laser on quency shifts were recorded with an accuracy of / / better than O.lcm-l wave numbersThese fre- ’ . v,, = ov quency increments correspond to temperature changes of less than 10°C and Fig. 1 provides F 80- current/voltage plots for the devices with and g_ 30- g without the laser in operation. Since the sub- _s -IV 2 60- strates are transparent to the low power, sub- 20 - -2v bandgap laser radiation, it can be seen from this -3v figure that very little change in the readings is 10 - observed when the laser is activated and that any -4v temperature effect related to reading the Raman 7 O- E -5V spectra will be minimal.As an additional control, I I , I I 0 5 10 15 20 0 5 10 15 20 the devices were also tested at zero drain bias vD, lvl v,6 Lvl with the laser on, but no laser heating was observed. Typical temperature map results are shown in Figure2. Typicaltemperature map results far an HE T on Figure 2. for an HFET on sapphire with an active sapphire with an active region region layout indicated by the inset, where the /ayout indicated by the inset. dimensions are for a 10 micron source-drain gap and a 4 micron gate, centered between the source and drainAs may be anticipated, the sili- con carbide supported device exhibits lower device temperatures than does the sapphire device, a result presumed to be due to the much lower heat conductivity of sapphire (about l/lOth the 3.3W per cm per degree Kelvin of sili- con carbide). AlGaN/GaN high electron mobility transistors At the maximum applied bias levels of 2OV,the (HEMTs) grown side by side on sapphire and on maximum temperatures for the sapphire support- an aluminium nitride template deposited on sili- ed devices were in the 180°C range but only con carbide (6-H on-axis). In efforts to reduce 120°C for the silicon carbide supported device. the lattice mismatch to silicon carbide, an insulat- (See Figure 3.) where the calculated power dissi- ing aluminium nitride template (- 100nm) was pation is plotted against the device temperature deposited on the silicon face by hydride vapour and the actual experimental temperatures are phase epitaxy. Both substrate-types then received slightly higher than the theory). Other results a thin aluminium nitride buffer layer followed by indicated that substrate heat dissipation, rather the MOCVD growth of the active layers at than the device layout, is the main influence on 1040”C.The structure diagram is shown in Figure the fmal temperature.The chip temperatures 4. It is interesting to note that the device mobili- were also verified by heating an HFET on a hot ty is higher for the device layers grown on sap- stage. However, as illustrative as the data is in phire, even though the defect levels are lower for Figure 2., the actual localized in-layer tempera- the silicon carbide grown device. tures will be higher than those measured by this Wayne and his team also found higher transcon- method, because the active regions are smaller ductances from the sapphire based HEMTs, 200 than the actual areas measured by this technique, versus 135mS/mm (for the larger gate devices), (possibly only 0.04 microns out of a one micron although this difference was thought to be due layer).Therefore, the local active-region tempera- to non-optlmisation of the growth process for tures were estimated to be up to 12% higher than the nitride templates.The transconductance does those recorded by this micro-Raman technique. fall off faster for sapphire substrates as the self- This research has shown that for the first time, heating effect and the junction temperatures Figure 3. Calculated power increase with increased drain-source voltages up the analysis of micro-Raman spectra has provided dissipation is plotted against accurate micron-area comparisons of the device to 40V (see the normalized data in Figure 5.) but the device temperature. temperatures of gallium nitride devices (AlGaN/GaN HFETs) grown on sapphire and sili- (a) 03 con carbide substrates.The demonstrated active --- Modelling Modelling . - GaN top surface 120 _ - GaN top surface region temperatures reinforce the importance of 180 - - - Average GaN layer l - - - Average GaN layer heat extraction and dissipation for real devices, Experiment: Experiment: 160 l Drain- gate l Drain- gate l - * - Edge of chip lOO--e-Edgeofchip where local hot spots could ruin both perform- i l k ance and lifetime behaviour. In the future this micro-Raman technique should be a valuable tool for the analysis of both developmental and com- mercial high power devices. Other substrate comparisons In other substrate comparisons, Wayne Johnson from the University of Florida, part of a large 20- I I I cooperating team (Yale University, Sandia ‘OOW 8 0 0.5 1.0 1.5 2.0 National Laboratories, Agere Systems and TDI Power dissipation [W] Power dissipation [W] Inc.) reported on the DC performance of Ill-Vs REVIEW THE ADVANCED SEMICONDUCTOR MAGAZINE VOL15 - NO 2 - MARCH 2002 1 HEMTs to operate at high temperatures. However, the AlN/SiC substrates had good pinch- 3 nm ~UIDDaN- off characteristics for all the temperatures tested 10 nm UID AIGaN, -20% Al up to 300°C and showed little in the way of self- AI,O, substrate devices: heating effects. On the other hand, the indication N, = 8.54E12 cm-* may be that the use of a thin aluminium nitride k = 1045 cm2 V1.sl template may have little beneficial effect on the lattice mismatch between silicon carbide and -1 km GaN SIC substrate devices: N, = 1.38Ei 3 cm-* AlGaN and also on the comparative heat loss p = 731 cm2 V’s_’ between the two substrate types. Operating temperature determinations were also reported for aluminium gallium nitride lasers on silicon carbide by V Ktimmler from Osram Opto Semiconductors.The AlGaN ridge lasers (see Figure 4. Structure diagram both substrate tvnes retain about 50% of their Figure 7. for the design) were MOCVD grown on ,f AIGaNIGaN HEMTs grown transconductance at 300°C. Normalised drain on sapphire and A/N/SK high conductivity substrates and packaged with subs&& (Johnson et al.). currents are shown in Figure 6. for 0.25 micron different heat sink capabilities.This laser design gate lengths, with slightly higher absolute values produced the lowest CW threshold currents for prevailing for the sapphire devices. Osram, of 9.7kA.cm2, in anticipation of lower Apart from the inability of the sapphire-based active layer heating effects and longer operating devices to pinch off above 25°C they saw little lifetimes. For a 15% indium QW-composition, the difference in performance between the two sub CW lasing wavelength was 410.4nm with a tem- strate types indicating the ability of the sapphire perature wavelength-sensitivity of 0.058nm per “K. CW active layer temperatures were deter- mined by the time to failure and FEM modeling of heat spreading in a ridge wave guide.A tem- perature difference of 20°C was found between p-side down mounting (80°C) and bottom con- tact mounting with the p-side up (lOO”C), with the heat sink being maintained at 25°C in both cases.At 1mW optical lasing powers, the active layer heating caused the p-side down laser to cease operation after two minutes whereas the reverse mounting, with the heat sink being on the opposite side of the chip from the ridge, last- ed only 25 seconds.