GaN-BASED UNIPOLAR ELECTRONIC DEVICES

Performance of AlGaN/GaN High Electron Mobility Transistors at

Nanoscale Gate Lengths

J.W.Johnson⁽¹⁾, F.Ren⁽¹⁾, S.J.Pearton⁽²⁾, A.G.Baca⁽³⁾, J.Han⁽⁴⁾, A.M.Dabiran⁽⁵⁾ and P.P.Chow⁽⁵⁾

¹ Department of Chemical Engineering, University of Florida, Gainesville, FL 32611

²Department of Materials Science and Engineering,

University of Florida, Gainesville, FL 32611

³ Sandia National Laboratories, Albuquerque, NM 87185

⁴Department of Electrical Engineering,Yale University, New Haven, CT 06520

⁵ SVT Associates, Eden Prairie, MN 55344

Abstract

The dc and rf performance of AlGaN/GaN high electron mobility transistors with nanoscale gate lengths is presented.The layer structures were grown by either Metal Organic Chemical Vapor Deposition or rf plasma-assisted Molecular Beam Epitaxy.Excellent scaling properties were observed as a function of both gate length and width and confirm that these devices are well-suited to both high speed switching and power microwave applications.

Introduction

The Group III-nitrides (AlN, GaN, InN, and their alloys) were initially researched for their promise to fill the void for a blue solid state light emitter. GaN, in particular, received considerable attention because of its direct energy gap of 3.39 eV (365 nm).

The energy gaps of several common semiconductors are given in Figure 1 as a function of lattice constant. The approximate boundaries of the visible spectrum are shown as is the nature of the energy transition (direct or indirect) for each material.

Electronic devices from III-nitrides have been a more recent phenomenon .

The thermal conductivity of GaN is three times that of GaAs, as listed in Table 1[1-3]. For high power or high temperature applications, good thermal conductivity is imperative for heat removal or sustained operation at elevated temperatures. The development of III-N and other wide bandgap technologies for high temperature applications will likely take place at the expense of competing technologies, such as silicon-on-insulator (SOI), at moderate temperatures. At higher temperatures (>300°C), novel devices and components will become possible. The automotive industry will likely be one of the largest markets for such high temperature electronics. Automotive control components fabricated with wide bandgap materials could be mounted directly to an engine block, reducing signal delay. Well logging equipment, military combat systems, and aerospace components are just a few of the additional sectors in the global market for these devices.

One of the most noteworthy advantages for III-N materials over other wide bandgap semiconductors is the availability of AlGaN/GaN and InGaN/GaN heterostructures. A 2-dimensional electron gas (2DEG) has been shown to exist at the AlGaN/GaN interface, and heterostructure field effect transistors (HFETs) from these materials can exhibit 2DEG mobilities approaching 2000 cm²/ Vּs at 300K. Combined with the polarization-enhanced sheet charge density of ~10¹³ cm^-2 and large peak and saturation velocities, AlGaN/GaN HFETs are capable of handling large DC and RF currents. HFETs have been by far the most heavily investigated III-N electronic device, largely due to their promise in the radio frequency semiconductor market. Power handling capabilities of ~12 W/mm appear feasible, and extraordinary large signal performance has already been demonstrated, with a current state-of-the-art of >10W/mm at X-band and 6.6 W/mm at 20 GHz for AlGaN/GaN HEMTs on SiC substrates. These values are already far superior to the best power density achieved with GaAs devices, which is on the order of 1.5 W/mm at X-band [4]. The microwave market for III-N devices is expected to lie at moderate frequencies (X-band to Ku-band) and RF power levels greater than can be achieved with competing technologies[5,6]. At higher frequencies (and lower powers) materials such as InP are expected to dominate because of an extremely low electron effective mass.

To summarize many of the attractive features of the III-nitrides, selected material properties of AlN, GaN, and InN relevant to electronic device applications are listed in Table 1 along with those of Si, GaAs, SiC, and diamond for reference. SiC and diamond are the primary competitors of the III-nitrides for many of the electronic device applications previously mentioned in this chapter. Currently, SiC technology is much more mature than that of GaN, with some RF devices already having reached the market [7]. Power devices from SiC will almost certainly make the first progress toward the commercial importance of wide bandgap electronics. However, GaN offers a larger breakdown field than SiC, as well as the greatly increased mobility afforded by heterostructures. It is believed that III-N devices will ultimately provide performance superior to that achievable with SiC. Diamond offers an extremely large bandgap and near-ideal material properties, but lags well behind due to difficulties associated with crystal growth and n-type doping.

The extremely high current density achievable with AlGaN/GaN heterostructures (up to 1.6 A/mm has been reported) is a result of the large polarization-induced field and large conduction band offset in the AlGaN/GaN system. This polarization field has both a piezoelectric, strain-induced, as well as a spontaneous polarization component. For the AlGaN HEMTs most widely studied to date, the spontaneous polarization field dominates. Even in the absence of modulation doping, this built-in field induces a two-dimensional electron gas (2DEG) that is linearly proportional to the Al-mole fraction, x (0 ≤ x ≤ 1) as n_s = xּ((5 ´ 10¹³ cm^-2). For the ~30% Alomparison most widely studied, a channel sheet electron density of ~1.5 ´ 10¹³cm^-2 can be realized that is a factor of 5 - 10 times higher than typical GaAs or InP pHEMTs. The associated mobility at this high current density is in the range of 1000 - 1500 cm²/ Vּs, which is well below that achieved in the conventional III-V material. However, the combined μ_cn_s product remains competitive. It is the μ_cn_s product that enables this material system to be applicable to low noise amplifiers as well as power amplifiers.

The high voltage capability of AlGaN/GaN HEMTs is the result of a critical electric field of ≥3 MV/cm. This value is 10 times that of Si and 5 times that of GaAs. For typical AlGaN/GaN HEMT structures, this translates to gate-to-drain breakdown voltages of approximately 100 V/μm. While record breakdown voltage over 500 V have been demonstrated for large gate-to-drain spacing, the more important result is for a device layout consistent with good microwave performance (i.e. gate-to-drain spacing of 1-2 μm). In this latter case, a breakdown voltage of ~80 - 100 V can be maintained in a microwave transistor that also has the high channel current mentioned previously. Figure 2-15 clearly illustrates these advantages compared to conventional high-speed compound semiconductor technologies. Due to the high breakdown field, III-N HEMTs are able to maintain large breakdown voltages in smaller geometry (i.e. short channel) devices that retain excellent high-speed performance.

To make full use of the high voltage capability, a transistor must not be thermally limited. GaN also has an advantage over GaAs in this regard, with thermal conductivity up to 2.0 W/cmּK for GaN versus 0.46 W/cmּK for GaAs. Furthermore, GaN HEMTs can be grown on semi-insulating SiC with a thermal conductivity of 3.3 W/cmּK. The high breakdown field and good thermal conductivity allow these devices to be used in high efficiency (theoretical efficiency of 78%) class B push/pull amplifiers at full power rating. GaAs microwave devices, on the other hand, can only be implemented in a class B push/pull amplifier by backing off the voltage bias (and hence the power level) to accommodate the higher voltage swing in this configuration as compared to class A, single-ended operation.

Over the past few years, dramatic progress has been made in understanding the AlGaN HEMT device physics and in demonstrating record microwave power performance. The potential of any microwave device technology is first realized in small periphery devices with microwave power density being the relevant figure of merit. These results are summarized in Figure 2,where state-of-the-art output power densities at X-band are shown as a function of time. Note the steady progress in power performance since the mid-1990s. The best results to date are 9.8 W/mm with 47% power added efficiency (PAE) at 8 GHz [8]. This power density is close to seven times the best result obtained with any GaAs technology and over twice that achieved with SiC MESFETs at this frequency [9]. Compilations of reported power density and PAE from AlGaN/GaN HEMTs and GaAs pHEMTs are given in Figures 2 and 3, respectively. Power added efficiency values are comparable between the two technologies, but the power densities achievable with AlGaN/GaN HEMTs are over an order of magnitude higher. With continued improvements in material quality and device design, a power density of 12 W/mm or higher appears plausible for AlGaN/GaN HEMTs. Other impressive results include 6.6 W/mm with 35% PAE at 20 GHz [10]. The 20 GHz performance clearly illustrates the potential for high power III-N electronics at K-band and beyond.

It has also been predicted and shown that AlGaN/GaN HEMTs can achieve low microwave added noise figures (NF = 0.6 dB at 10 GHz) while maintaining a large breakdown voltage (>60 V) and hence a large dynamic range [11,12]. These results imply that these devices can be used to perform the active transmit and receive functions in more robust, higher dynamic range modules.

One of the most noteworthy advantages of the III-nitrides over other wide bandgap semiconductors, such as SiC, is the availability of AlGaN/GaN heterostructures. The Type I band alignment between AlGaN and GaN has been shown to form a potential well and a 2-dimensional electron gas (2DEG) at the heterointerface . When these materials are brought into contact, thermal equilibrium requires alignment of their respective Fermi levels (E_F). This induces conduction (E_c) and valence (E_v) band bending in both the AlGaN and GaN layers and can cause the GaN conduction band at the interface to drop below E_F, as illustrated in Figure 4.

Since (for n-type material) the Fermi level can be viewed as an electrochemical potential for electrons, majority electrons will accumulate in the narrow gap material just below the heterointerface to fill the quasi-triangular potential well between E_c and E_F. These electrons are confined by a distance shorter than their deBroglie wavelength, causing quantization of the allowed energy levels in the potential well. Depending on the structure, there may be more than one allowed energy level below the Fermi level, although only the lowest allowed level will be substantially populated at room temperature. With the heterointerface on one side and a potential barrier on the other, electrons in the 2DEG are only free to move in along the plane of the interface. A thin ‘sheet’ of negative charge (electrons) results.

Modulation doped field effect transistors (MODFETs) are a class of heterostructure FET that use selective barrier doping to spatially separate ionized donors from the electrons in the 2DEG, leading to an increase in channel mobility. For this reason, these devices are also known as high electron mobility transistors, or HEMTs. The AlGaN barrier layer is typically grown with the same Al content throughout, but with varying doping levels. A heavily doped donor layer supplies electrons to the 2DEG. The ionized donors remain in the AlGaN, while the electrons are transferred as mobile carriers into the GaN. An unintentionally doped (UID) AlGaN layer between the donor layer and the gate metal allows formation of a Schottky contact to the wide bandgap barrier. A thin AlGaN spacer layer between the donor layer and channel (~30 Å) is used to screen the Coulomb potential of the ionized impurities. Without this spacer, Coulomb scattering from positively charged impurities in the donor layer would lead to reduced electron mobility in the 2DEG.

AlN, GaN, and their alloys are polar crystals. Polarization occurs due to the lack of inversion symmetry for typical growth along the polar (0001) axis of the wurtzite crystal structure. The strong spontaneous and piezoelectric polarization effects in the III-nitrides can strongly influence the electron density and potential profile of heterostructure devices. [14,15]. The polarization field is such that sheet electron densities on the order of 10¹³ cm^-2 can be realized even in undoped HEMTs [16].

Experimental

All HEMTs in this study were fabricated by a 4-step process sequence consisting of dry etch mesa isolation, Ohmic metallization, Ohmic anneal, and gate metallization. The mesa etch was performed in a load-locked Plasma-Therm SLR 770 with a 2 MHz, 3 turn coil ICP source. A conductive adhesive was used to mount the samples to an anodized Al carrier plate which was cooled from the backside by He gas. A 13.56 MHz rf bias was superimposed on the substrate. The clear field etch mask was photoresist AZP 4330, which was spun at 4000 rpm for 30 sec., baked at 90°C for 90 sec., exposed for 6.5 sec. at 20 mW/cm², and developed for ~125 sec. in MF319. Surface profilometry indicated a photoresist (PR) thickness of ~3.75 μm, which was highly reproducible across the surface of the sample. The ICP process gases and flowrates were 8.0 sccm BCl_3,32.0 sccm Cl_2,5.0 sccm Ar. The addition of BCl₃ to the Cl/Ar chemistry was important for maintaining smooth surface morphology of the etched surface [17]. The chamber temperature and pressure were maintained at 2 mTorr and 25°C, respectively. An ICP source power of 500 W and a substrate RF power of 40 W was used to minimize surface damage. By using high ICP and low RF powers, a high density plasma is created which maintains fast etch rates with reduced ion energy. The DC self bias was below -90 V under these conditions. Typical etch rates were ~1300 Å/min. Etch depths (mesa step heights) were 1300 - 1700 Å. The stability of the 4330 photoresist was excellent.

Ohmic lithography was performed with AZ 5214 PR in both positive tone (dark field mask) and image reversal (clear field mask). The linewidth resolution was approximately the same for both processes, while the thickness of the positive tone PR was slightly larger (1.46 μm vs. 1.35 μm for 5000 rpm spin speed). A hexamethyl disilazane (HMDS) treatment and dehydration bake was used prior to PR application to improve PR adhesion. The PR procedure for positive tone was: 5000 rpm spin for 30 sec., bake 90°C for 90 sec. on hotplate, pattern expose ~6.5 sec. at 20 mW/cm², and develop ~110 sec. in 1:1.4 MF312:H₂O_. Immediately before loading the patterned wafers for metallization, a 60 sec.1:10 HCl:H₂O dip was used for removal of the native oxide. Electron beam evaporation was initiated at chamber pressures of ~3 ´ 10^-7 torr. An Ohmic metallization scheme of Ti/Al/Pt/Au (250/1000/450/1500Å) was deposited at rates from 2 - 10 Å/sec, depending on the metal. This quad-layer scheme was annealed in a rapid thermal annealing (RTA) furnace at 850°C for 30 sec. under a N₂ ambient. Four-probe transmission line method (TLM) analysis resulted in specific contact resistances from high 10^-7 to low 10^-6 Ωּcm² with transfer resistances of 0.3 - 0.4 Ωּmm. From such a plot, sheet resistance, transfer resistance, and specific contact resistivity can be simultaneously determined. Figure 5 illustrates the very rough morphology of the Ohmic metallization after the high temperature anneal. The top picture is unannealed Ti/Al/Pt/Au for comparison. Note, however, that the edge acuity of the contacts (which affects the channel length) appears to be quite good. Some reports claim a tradeoff between contact resistance and surface roughness [18], which can be tailored by varying layer thicknesses or annealing conditions. For this study, an Ohmic metallization experiment was performed in which Ti/Al/x/Au (250/1000/450/1500Å) was deposited on AlGaN/GaN HEMT structure, where x = Ni, Pt, or Ti. Each of the metallizations was annealed at 800, 825, 850, 875, and 900°C for 30 sec. in N₂. The general trend is for decreasing contact resistance at higher annealing temperature. However, as temperatures were increased above ~850°C, the contact morphology significantly degraded. An 850°C-annealed Ti/Al/Pt/Au Ohmic metallization was chosen for all devices presented in this work, unless otherwise noted.

The DC characteristics were measured in common source mode with an HP 4145A Parameter Analyzer. Maximum drain current densities of 0.5 A/mm were obtained for 0.8 ´ 100 μm² gate dimension devices. For the same devices, the extrinsic transconductance was ~135 mS/mm at V_G = -1.5 V and V_DS = 3 - 5 V. Slight self-heating effects were evident for the higher power levels associated with wider gate devices (i.e. 200 μm devices) biased to V_DS=10 V. The Schottky gate contacts exhibited leakage current on the order of 0.3 - 3mA.

Scattering parameters were measured with an HP 8510 Vector Network Analyzer calibrated via the short-open-load-thru (SOLT) method to ~40 GHz.

(a)Nanoscale Gate Devices

Nanoscale gate length devices were fabricated using the processing sequence on samples with 230 Å or 430 Å AlGaN layers (Figure 6). Scanning electron microscopy (SEM) was used to examine the quality of the lithographic processing steps and estimate the gate lengths of the submicron devices. Figure 7 shows an SEM micrograph of a 0.1 μm gate crossing the mesa. Note the use of an anchoring technique at the end of the gate to improve adhesion. The anchors were necessary for the 0.1 μm devices, but not for the longer gate lengths. Note the excellent edge definition on both the gate and the source/drain contacts. The surface morphology of the MOCVD-grown epilayers is also excellent, with no evidence of defects or tilted growth boundaries.

The maximum drain current density for the 430A devices approached 1 A/mm before self-heating effects caused a decrease in the output current. Characteristic drain I-V curves are given in Figures 8(top and bottom) for 0.1 and 0.25 μm gate length devices, respectively. The 0.1 μm devices do not maintain complete pinch-off and exhibit significant output conductance due to short-channel effects. These are caused by the 2-dimensional nature of the electric field for short gate lengths and large drain bias, and are common in GaAs- and InP-based HEMTs. For the shortest gate length devices, the extrinsic transconductance exceeded 200 mS/mm. The transfer characteristics are given in Figure 9. The peak of the extrinsic transconductance curve shifted toward more positive voltage with increasing gate length and the magnitude of g_m scaled with inverse gate length (Figure 10). For these devices, high temperature operation with excellent pinch-off characteristics has been verified to 400°C, the temperature limit of the probe station heater. Drain current density decreases monotonically with increasing temperature over the range 25° - 400°C.

Maximum gate-source breakdown voltages were ~25 V for 430A and > 60 V for 230A devices with 0.25 ´ 150 μm² gates measured in 3-terminal mode with source and drain shorted to ground . The gate leakage current for 1214D was ~200 μA at a reverse bias of -60 V. The leakage from 430A devices was much higher (~1 mA prior to hard breakdown). The reverse current density of the 0.25 and 0.5 μm gate length devices was similar, but leakage from the 0.1 μm device was significantly higher, as shown in Figure 11. This may be due to the higher dose (~800 μC/cm²) during e-beam writing of the shorter gate lengths. This effect should be investigated in additional detail to optimize the submicron lithography process.

Gate current-voltage-temperature characteristics were measured on a heated chuck from 25°C to 395°C. The gate leakage current decreased by nearly 3 orders of magnitude upon heating the device (in ambient air) from room temperature to 395°C . The turn-on voltage (V_on) also increased with temperature.

The cutoff frequency and maximum frequency of oscillation were extracted from measured s-parameters and are given in Table II for gate lengths of 0.1, 0.25, and 0.5 μm devices, respectively. Maximum f_T and f_max of 58 GHz and 90 GHz, respectively, were achieved for the 0.1 ´ 100 μm² devices, clearly illustrating the potential of AlGaN/GaN HEMTs for high frequency operation. The 0.25 μm devices exhibited an f_T of 29 GHz and f_max of 63 GHz, while the same parameters for the 0.5 μm device were measured to be 19 and 39 GHz, respectively. The high f_max values are encouraging, since this is one of the most important metrics of device performance for applications such as satellites where overall gain is critical.

Figures 12 and 13 give a graphical representation of the gate length dependence of f_T and f_max. A linear f_T vs. L_G^-1 relationship is observed. Calculation of the effective electron velocity gives v_eff = 6.5 ´ 10⁶ cm/s, below the theoretical value of ~2 ´ 10⁷ cm/s derived from Monte Carlo measurements.

(b) AlGaN/GaN HEMTs on RF-assisted MBE-grown Epilayers

A common feature of reported HEMT work to date is the fact that molecular beam epitaxy (MBE) material appears quite competitive with that grown by MOCVD [10]. This has not been the case with photonic devices, in which the generally better optical properties obtained by MOCVD growth have led its dominance in LED and LD manufacturing. In this section, the fabrication and DC performance of MBE-grown AlGaN/GaN HEMTs is investigated.

MBE growth was proceeded using a standard Ga effusion cell and RF atomic nitrogen plasma source. Growth details have been given previously [19-21], but in brief consist of a low temperature AlN (300Å) buffer on sapphire (0001), 2 μm of undoped GaN grown at 750°C under Ga-rich conditions, 250Å of undoped Al_0.2Ga_0.8N, and a 30Å undoped GaN cap. A growth rate of 0.5 - 1 μmּhr^-1 was used for all depositions. The layer structure is given in Figure 14, which also includes schematic representation of the source, gate, and drain contacts.

The I_D-V_DS characteristics for devices with one-tenth micron gate length are shown in Figure 15. In all cases the gate width was 100 μm, and the devices were biased with +1V gate bias (V_G), with a step of -1V. The drain characteristics are free of kinks in these unpassivated devices, with no sign of the current collapse phenomenon attributed to hot electron injection and trapping in the buffer layer [22]. In this mechanism, as a high drain voltage is reached, electrons are injected into the GaN buffer layer where they are trapped in deep states. This trapped charge acts to deplete the two-dimensional electron gas from the buffer side of the channel and thereby reduces the drain current for the subsequent traces [22]. The shortest gate length devices show a crowding of the I-V traces at high grain source voltage and do not pinch-off. Operation at >40 V_DS with >50 mA peak I_D was obtained. It is clear that GaN HEMTs on sapphire substrates will not be as well-suited for power operation as those on SiC substrates. Flip-chip bonding is an attractive option to avoid the poor thermal characteristics of the sapphire.

From these results, RF plasma-assisted MBE growth of AlGaN/GaN HEMTs appears well-suited for producing high-performance devices. The epilayers surface morphologies and uniformity of device DC performance are excellent. Reasonable current densities and excellent scaling properties were achieved over a range of gate lengths and gate widths.The first high quality films for III-N light emitting devices were all grown by CVD. However, with the development of activated nitrogen species, MBE has since become more widely used for GaN-based devices. With additional optimization, MBE may even surpass MOCVD for devices where precise control of doping profiles or minimization of interfacial roughness is critical.

Summary and Conclusions

AlGaN/GaN HEMTs offer excellent dc and rf device performance over a broad range of contact dimensions.Future refinements should focus on surface passivation techniques that address the issue of current collapse due to surface states and provide stable operation over commercially signifiicant periods.More thermally stable metallization may be needed for applications requiring high temperature operation.Finally,the use of metal-oxide gate structures can provide advantages over Schottky gates and work is needed on novel gate dielectrics.

Acknowledgments

The work at UF is partially supported by NSF CTS9901173 and DMR 0101438.Sandia is a multi-program lab operated by Sandia Corporation,a Lockheed-Martin company,for the US Department of Energy under DOE grant DEAC04-AL85000.