Comparison of wide band gap and IIIV semiconductor devices

Wide band gap semiconductor materials allow high voltage swings to be sustained across the device, giving the potential for high RF power densities. This paper by C.H.Oxley and O.Buiu reviews the characteristics of SiC,GaN and InN in direct comparison to those of GaAs and InP in relation to their use in RF devices.

W ide band gap (WBG) semiconductors have been considered ideal for high power microwave devices since they were studied over 30 years ago [1]. These materials have a high critical breakdown field, high saturation drift velocity and good thermal conductivity. The material until recently has not been of sufficient quality to fabricate viable microwave devices but this is now rapidly changing.

The materials included in this review are silicon carbide (SiC), gallium gitride (GaN), and indium nitride (InN) and will be directly compared with devices on gallium arsenide (GaAs) and indium phosphide (InP).

The wide band gap materials will enable much higher voltage swings to be sustained across the device leading to high RF output power densities. Gallium Nitride gives the highest power density recorded to date, 9.2W/mm of gate width at 8GHz [1], which is 6 to 9 times greater than recorded for gallium arsenide.

The requirement for highpower solid state amplifiers has exploded with the rapid development in radio communications and the need to replace outdated or unsupported high power vacuum tubes.

Devices are also required for a wide range of other applications including efficient and high power Tx/Rx modules for phased array radar, transponders and sensors, an example being automatic cruise control (ACC). The needs placed on the microwave device are multifold and will include high frequency, high power, widebandwidth, high gain, high efficiency, high linearity and manufacturability at a low cost. To meet these demands a range of power devices are becoming available and include Silicon Lateral Double Diffused MOSFET (LDDMOSFET), GaAs/InP High Electron Mobility Transistors (HEMTs) and a new range of devices on wideband gap materials. Perhaps the most exciting and fastest growth area is with devices being developed on gallium nitride.

This paper will review the present state-of- the-art wide band gap microwave devices, in particular GaN, relating where possible the material parameters to the device performance and showing the rapid expansion of work in this area.

Material growth and device fabrication
The work on GaN is mainly concentrated around the HEMT structure. As GaN substrates are not readily available the epitaxial layers are grown on silicon [2], sapphire [3] or silicon carbide [4, 5].

SiC layers (in the 6H form) are a promising substrate for GaN heteroepitaxial growth due to a low lattice mismatch with GaN (3.5%) and very interesting physical properties, including an excellent thermal conductivity. However, due to its cost and the difficulties of nucleating GaN on SiC surface, this substrate is not commonly used. Sapphire has been the most used substrate up to now; in spite of a large mismatch with GaN (~ 16%) and poorer thermal conductivity. The large mismatch is leading to a high dislocation density of 10 7 to 10 8 [6], which may be a contributing reason to the lower cut off frequency (Ft ) than expected. Basically, good quality GaN layers can be grown using a two step process including a low temperature buffer layer [7].

Both MOCVD (metal-organic chemical vapour deposition) and MBE (molecular beam epitaxy) growth techniques are used and recent work [8] suggests MBE will give improved control of the layer uniformity.

Other material growth techniques are being explored and include: hydride vapour phase epitaxy (HVPE)) metal organic vapour phase epitaxy (MOVPE); sometimes is called organometallic vapour phase epitaxy (OMVPE) A typical GaN HEMT structure consists of substrate (either Si, Sapphire or 6H SiC), buffer layer (either AlN, undoped GaN), a thin AlGaN nucleation layer, followed by an AlGaN (10 18 atoms cm 3 ). Some structures have an additional high doped layer of AlGaN (10 19 atoms cm 3 ) to improve the ohmic contacts [9]. All MBE grown wafers have a 10 to 25ý undoped GaN cap layer to minimise gate leakage [10]. Both doped and undoped AlGaN HEMT structures have been measured; there is strong evidence to suggest that a significant part of the 2D electron gas density at the AlGaN/GaN interface (10 13 cm 2 compared with 10 12 cm 2 for GaAs based HEMT structures) is due to the strong piezoelectric effect seen with this material [11].

To date most of the device processing has used dry etch technologies, as chemicals will etch the AlGaN preferentially where the dislocations meet the surface and is a reason for the majority of devices not having a gate channel recess. The ohmic contacts are typically Ti/Al with a contact resistance of between 0.2 to 15/mm [12]. Different metals have been used for a Schottky contact to the gate, including Ni, Pt, and Au. The evidence of trapping effects has been observed in the majority of GaN FETs and is seen by limited power density, transconductance frequency dispersion etc. These effects are being investigated and are akin to the early work on GaAs structures.

	GaAs	InP	GaN	SiC	InN
Band-gap, Eg (eV)	1.43	1.34	3.39	3.26	1.89
Saturation velocity Vs (m/s) @ 300K	10 5	0.9x10 5	2.7x10 5	2x10 5	2.5x10 5
Critical breakdown field Ea (MV/m)	40	50	330	200
Relative dielectric constant, r (static)	12.8	12.5	8.9	10	15.3
Piezoelectric constant, e14 (C/m 2 )			-0.035	0.375	0.375
Thermal conductivity, (W/m K) @RT	50	69	170	380
Linear thermal expansion coefficient, x10 -6 (1/K)	5.73	4.6	3.5/5.0
Table 1: Comparison between material parameters for GaAs/InP and wide band-gap materials

Device performance
The work reviewed in this section of the paper will primarily look at GaN power transistor with respect to frequency of operation, gain, output power, thermal characteristics, and efficiency, with summarised in Table 1.

Very recent Monte Carlo modelling work at Cornell University [10] has shown that GaN and InN have significant velocity overshoot effects with peak electron velocities of 10 6 m/s over short distances when the applied field is greater than 50MV/m (GaN) and 25MV/m (InN). The upper limit of the frequency performance of an HFET (High Frequency FET) can be estimated by its cut-off frequency Ft (Ft =1/(2 t)=Vsat /(2 Lg )) where t is the transit time under the gate of length, Lg ,and where Vsat is the invariant high field velocity.

The theoretical upper-limit to the frequency performance of the HFET devices fabricated on GaN and InN [10] have been included in Figure 1 and are substantially greater than for GaAs based devices.

The available Ft from published work on GaN MESFET [13] and HEMT [14-24, 4, 5, 30] structures have also been plotted (Fig reference to material parameters.

The materials and their parameters are ure 1) as a function of gatelength (Lg).

From the Figure the experimental Ft results fall far below the predicted values using results from Monte Carlo analysis [10]. The effective velocity Veff is the average of the velocity field characteristic up to the peak velocity of the carrier. This is known as the Equal Areas Rule and from recent results [10] gives a value of 1.5x10 5 m/s. Experimental results (Figure 1) indicate that presently the GaN HEMT has a similar Ft to the GaAs MESFET. Whereas, with its higher predicted effective velocity the GaN HEMT should have a similar Ft to the state of the art GaAs based transistors for example GaAs metaphorphic and InP PHEMTs [25 - 27]. Eastman [10] has suggested that the present low Ft indicate the high field region is extended well beyond the gate metallization. It is expected that with continued material and device improvements, the Veff will be increased to reflect the magnitude of Ft given in Figure 1.

For example, work has shown that devices with high Al molar content [3] appear to have both a higher Vs and RF power density.

The gain of the transistor is dependent on the magnitude of the transconductance, gm, which is proportional to the product of input capacitance, Cgs, and the cutoff frequency, Ft. Therefore the gm is increased with short gate lengths and materials with a high effective velocity, Veff .

To a first approximation the linear output power (Pmax ) of a FET is dependent on the product of the open channel current, Is, and the gatedrain breakdown voltage (Vbgd ).

The Pmax Ft figure of merit [1] is totally dependent on the material parameters of the semiconductor. Using the material parameters given in Table 1, the Pmax Ft figure can be calculated for the different materials. GaN is in excess of 100x greater than for GaAs and InP and has the highest Pmax Ft product of all the materials in Table 2.

In the above calculation the channel is assumed to be uniform and like gallium arsenide and indium phosphide, bandgap engineering can be readily applied to GaN [6] and InN; this will result in heterojunction FET structures giving improved charge confinement enabling higher frequencies and currents to be realised. This property gives a substantial benefit over SiC in which bandgap engineering has not been applied. The carrier mobility in SiC is much lower when compared with GaN/InN and restricts the saturation current density to approximately 300mA/mm, whereas in an AlGaN/GaN HEMT published figures in excess of 1000mA/mm have been recorded [16]. The high current density enables high power added efficiencies (PAE) to be realised at higher frequencies.

For example at 10GHz GaN CW devices give PAE of 50% while SiC CW devices achieves only 20% [41].

The power density has been measured by a number of workers and Figure 2 gives experimentally published results for GaN, GaAs, InP, and SiC as CW power density (Watts/mm) against frequency for small total gatewidth devices [4, 8, 14, 16, 18, 29, 30]. The experimental results show GaN gives the highest power densities recorded for any solid state transistor, and at 8GHz is 9.2W/mm which is 6 to 9 times higher than for GaAs. Most of the work to date on GaN has been on small total gatewidth devices.

Work is now progressing [31] in scaling device structures to realise high power performance from a single chip. High output power will depend on optimising the unit cell packing density to obtain maximum output power at an acceptable junction temperature and stay within the overall dimensions of the chip to satisfy the phase criteria of approximately ý /16. For example, consider a single finger GaN device with a gatelength of 0.25ým giving 4W/mm of RF output power at an efficiency of 30%. The device will require to dissipate 37kW/mm 2 compared with GaAs MESFET which will need to dissipate 5kW/mm 2 (for an RF power of 0.5 Watts/mm). At room temperature (25ýC) the thermal conductivity of GaN is approximately 3x higher than for GaAs, therefore the overall junction temperature for GaN device is going to be greater than for GaAs.

Published work indicates that GaN will operate reliably [32] at higher temperatures than GaAs devices, but the output power/gain will degrade [33].

 

Material Pmax Ft	product
GaAs	18WGHz/mm
InP	24.5WGHz/mm
GaN	1940WGHz/mm
SiC	1408WGHz/mm
Table 2: Showing Pmax Ft product for different material

The temperature performance of the GaN device can be improved by reducing the thermal impedance of the substrate, flip chip attachment, and spreading the geometry of the device, for example the finger spacing. The recently published high power RF results [14] have been from devices fabricated on SiC substrates. Work published under MURI [10] indicates that the gate drain breakdown voltage, Vbgd , is restricted to approximately 40V on SiC substrate, thus restricting the maximum power capability from the device, also that the present cost of SiC substrate is high. Alternatively, the lower cost sapphire substrate can be used and the device 'flip chip bonded' to aluminium nitride substrate (a thermal conductivity =1.8W/cmýK) to reduce the thermal impedance [34]. The recently published work of B. Kopp [33] has indicated that the separation of the gate fingers for a GaN device will have to be significantly greater than for a comparable SiC power transistor designed for either S or X band to maintain the same junction temperature. These problems will be continually addressed in the rapid development and optimisation of the device.

Zolper [35] produced a plot of total CW output power against published dates which shows the rapid development of the device between May'96 and July'98. Figure 3 gives an extension of that plot to June 2000, showing the continued rapid progress with a device giving 14.1Watts at 8GHz and 29.6% power added efficiency.

Noise performance
Most of the research effort into GaN transistors has been concentrated in the area of power, due to the material supporting a high breakdown field. As already discussed, much higher RF power densities can be realised, for example 9.2W/mm at 8GHz. Little work has appeared in the literature on the noise performance of AlGaN/GaN HEMT devices. Nguyen [21] has recently published experimental work on the noise measurement of an AlGaN/GaN HEMT with a 0.15ým gatelength and obtained some very encouraging results of 0.6dB at 10GHz with an associated gain of 13.5dB. This result compares very favourably with state-of-the- art metamorphic HEMT (InAlAs/InGaAs), which had a measured noise figure of 0.28dB with an associated gain of 13.6dB at 12GHz [36]. It is interesting to note that the minimum noise bias current for the GaN HEMT [21] and GaAs FET is a similar percentage (20%) of IDSS [37]. The high Vbgd breakdown voltage 68V of the GaN devices offer a further advantage of improved protection from high RF input signals without having to include limiter diodes in the frontend circuitry [21]. This will provide two benefits, firstly lowering the frontend component count (eventually leading to a reduced cost of the front end) and secondly minimising the noise figure as the RF losses of the limiter diodes will no longer have to be included in the front end circuitry.

The Fukui noise equation [38] can be used to estimate the minimum noise (NF)min of FET structures to frequencies below which the Cgd (gate/drain feedback capacitance) provides significant cross-coupling of the gate into the drain circuit. The Fukui equation is semiempirical and contains a fitting factor, which is required before the expression can be used.

Kf for GaAs MESFET is approximately 2.5 [38] and 1.08 for AlGaN/GaN HEMT [39]. Using this figure the minimum noise figure (NF)min for a 0.25ým GaN HEMT can be estimated to 30GHz making the assumption that the total combined parasitic resistance is 7only for comparison with a similar geometry MESFET shown in Figure 4.

The minimum noise figure (NF)min can be optimised, by reducing Rg and Rs as in a GaAs MESFET. For example the parasitic source resistance Rs will depend on the contact resistance, the physical separation of the gate to source contact, the gate/channel geometry (etched well etc),and the carrier mobility of the material.

While the gate parasitic resistance Rg will depend on the metallisation thickness and profile (T-structure) as well as the number of gate stripes making up the total gate periphery.

Optimisation of the gate geometry channel will reduce the parasitic source resistance and increase the Vbgd breakdown voltage. Figure 5 gives a theoretical comparison between the Vbgd for AlGaN/GaN and AlGaAs/GaAs HEMT using the simple expression derived by Hikosaka [40]. The plot indicates that the Vbgd for a GaN device is approximately 7x greater than for a comparable GaAs device and is dependent on the etched channel geometry. The parasitic source resistance, Rs , [38] of a GaAs device will significantly increase with the channel depth/recess length, causing a degradation of the noise figure (NF)min . It is anticipated that similar behaviour of source resistance will be obtained for GaN.

Therefore a compromise may be required to obtain the lowest (NF)min and highest (Vbgd ) 2.

The low noise performance, high power density per unit gate width, and high Vbgd gives the device a unique position in transmit/ receive modules (Tx/Rx) for phased array radar applications. A review of wide band gap devices, for Tx/Rx modules has recently been carried out by Kopp [41].

Circuits
The material and device parameters discussed will have direct bearing on the bandwidth performance of an amplifier utilising GaN devices [35]. Bode [42] has shown that the maximum bandwidth of the amplifier is primarily determined by the input impedance of the transistor. The upper frequency limit of the bandwidth can be written as Fhigh =1/(2Zo Cgs ) where Zo is the characteristic input impedance of the circuit and Cgs is the total gate capacitance of the FET. The device gate capacitance is directly proportional to the total gate periphery.

Hence an AlGaN/GaN HEMT device with a highpower density capability will enable a shorter total gatewidth structure with a lower input capacitance to be realised but giving a similar output power levels to GaAs device, thereby significantly increasing the bandwidth capability. The gallium nitride device will have the potential of enabling travelling wave amplifier (TWA) with multioctave bandwidth to have an increased power capability when compared with currently available GaAs solid state TWA. Work by Xu [43] has experimentally demonstrated a 1 to 8GHz modified TWA topology which eliminates the backward wave, giving an output power of approximately 35dBm over 3 to 8GHz.

The work at Cornell University has led to the fabrication of a four-stage non-uniform distributed amplifier, in which the components used are compatible with a GaN MMIC process, such as silicon nitride (Si3 N4 ) MIM capacitors, mesa resistors and air bridges [44].

The high voltage swing across the device also means that the characteristic impedance of a AlGaN/GaN transistor is approximately 2 to 3 times the impedance of a similar size AlGaAs/GaAs device thereby simplifying power combining matching circuitry. Lastly the improved thermal conductivity of GaN devices fabricated on SiC and the breakdown voltage Vbdg being 5 to 7 times greater than for GaAs devices will enable high efficiency Class B amplifiers to be realised.

Conclusions
The possibility of high RF power densities from these devices is one of the major driving forces behind a continuously sustained technological and theoretical research effort in WBG materials. Higher output power for communication and radar is required. The output power for GaAs based devices is governed by the total gate periphery, which is constrained by a number of factors including low impedance, degradation of efficiency and reduced yield in fabrication. To increase the output power, powercombining techniques can be used but are limited by the number of stages that can be tolerated before the module efficiency rapidly degrades.

GaN HEMT devices would overcome the above problems leading to higher output powers. Work has also shown that these devices give low noise performance with a higher Vbgd making them less susceptible to high RF input.

These properties give the GaN device a unique position in transmit/receive (Tx/Rx) modules for phased array radar applications and may lead to significant increases in the output power providing real improvements in the radar range performance.

The present lower than expected Ft for GaN devices will restrict their use to approximately 30GHz and GaAs/InP based devices will continually be used at the higher frequencies. The state-of-the-art SiC devices have shown good power densities of 4.5W/mm and 4.3W/mm at 4GHz and 10GHz respectively and a Ft of 43GHz [45] have been recorded. In S-band, 3.1GHz, SiC devices have been measured with high CW powers (up to 80W) and PAE of 38% [29]. The low current density will restrict the PAE at higher frequencies and the highest published result is 20% at 10GHz. This is low compared with gallium nitride with PAE of 50% at 10GHz. An area not yet fully explored, is the use of the GaN high piezoelectric constant, which may enable the construction of SAW filters [46]. Combining, for the first time, a material able to give high Tx power, low noise frontend with high Vbgd and filter technology providing Rx protection and channelization.

Acknowledgements
The authors would like to express their thanks to Prof. S. N. Ekkanath Madathil for his encouragement and the University for enabling the research to be undertaken.