1. Introduction

Due to limited wet chemical etch results for the group-III nitrides, a significant amount of effort has been devoted to the development of dry etch processing [5] [6]. Dry etch development was initially focused on mesa structures where high etch rates, anisotropic profiles, smooth sidewalls and equirate etching of dissimilar materials were required. For example, commercially available LEDs and laser facets for GaN-based laser diodes were patterned using reactive ion etch (RIE). However, as interest in high power, high temperature electronics increased, etch requirements expanded to include smooth surface morphology, low plasma-induced damage and selective etching of one layer of another occurred. Dry etch development is further complicated by the inert chemical nature and strong bond energies of the group-III nitrides as compared to other compound semiconductors. GaN has a bond energy of 8.92 eV/atom, InN 7.72 eV/atom and AlN 11.52 eV/atom.

2. Plasma Reactors

Chemically dominated etch mechanisms rely on the formation of reactive species in the plasma which absorb to the surface, form volatile etch products and then desorb from the surface. Since ion energies are relatively low, etch rates in the vertical and lateral direction are often similar thus resulting in isotropic etch profiles and loss of critical dimensions. However, due to the low ion energies used, plasma-induced damage is minimized. Alternatively, ion-assisted plasma etching relies on both chemical reactions and physical sputtering to yield anisotropic profiles at reasonably high etch rates. Provided the chemical and physical component of the etch mechanism are balanced, high resolution features with minimal damage can be realized and optimum device performance can be obtained.

2.1. Reactive ion etching.

2.2. High-density plasmas.

ICP offers another high-density plasma etch platform to pattern group-III nitrides. ICP plasmas are formed in a dielectric vessel encircled by an inductive coil into which rf power is applied [see Figure 1 (c)]. The alternating electric field between the coils induces a strong alternating magnetic field trapping electrons in the center of the chamber and generating a high-density plasma. Since ion energy and plasma density can be effectively decoupled, uniform density and energy distributions are transferred to the sample while keeping ion and electron energy low. Thus, ICP etching can produce low damage while maintaining fast etch rates. Anisotropy is achieved by superimposing of rf bias on the sample. ICP etching is generally believed to have several advantages over ECR including easier scale-up for production, improved plasma uniformity over a wider area and lower cost-of-operation. The first ICP etch results for GaN were reported in a Cl₂/H₂/Ar ICP-generated plasma with etch rates as high as ~6875 Å/min [7] . Etch rates increased with increasing dc bias and etch profiles were highly anisotropic with smooth etch morphologies over a wide range of plasma conditions. GaN etching has also been reported in a variety of halogen- and methane-based ICP plasmas [8]. Use of a Cl₂/Ar/O₂ chemistry produced good selectivity for GaN and InGaN over AlGaN (up to ~50), due to formation of an oxide on the AlGaN [8].

MRIE is another high-density etch platform which is comparable to RIE. In MRIE, a magnetic field is used to confine electrons close to the sample and minimize electron loss to the wall. Under these conditions, ionization efficiencies are increased and high plasma densities and fast etch rates are achieved at much lower dc biases (less damage) as compared to RIE. GaN etch rates of ~3500 Å/min were reported in BCl₃-based plasmas at dc biases <-100 V. The etch was fairly smooth and anisotropic.

2.3. Chemically-Assisted Ion Beam Etching.

2.4. Reactive Ion Beam Etching.

There was relatively little effect of either beam current or sample temperature on the RIBE removal rates of the nitride. The etch profiles are anisotropic with light trenching at the base of the features. This is generally ascribed to ion deflection from the sidewalls causing an increased ion flux at the base of the etched features.

2.5. Low Energy Electron Enhanced Etching.

3. Plasma Chemistries

3.1. Cl₂-Based.

GaN etch rates are plotted as a function of dc bias (which correlates to ion energy) for an ICP-generated BCl₃/Cl₂ plasma in Figure 6. The GaN etch rates increased monotonically as the dc bias or ion energy increased. Etch rates increased due to improved sputter desorption of etch products from the surface as well as more efficient breaking of the Ga-N bonds. Etch rates have also been observed to decrease under high ion bombardment energies due to sputter desorption of reactive species from the surface before the reactions occur. This is often referred to as an adsorption limited etch regime. In Figure 7, SEM micrographs are shown for (a) Ð50, (b) Ð150 and (c) Ð300 V dc bias. The etch profile became more anisotropic as the dc bias increased from Ð50 to Ð150 V dc bias due to the perpendicular path of the ions relative to the substrate surface which maintained straight wall profiles. However, as the dc bias was increased to Ð300 V, a tiered etch profile with vertical striations in the sidewall was observed due to erosion of the resist mask edge. The GaN may become rougher at these conditions due to mask redeposition and preferential loss of N₂.

In Figure 8, GaN etch rates are shown as a function of ICP-source power while the dc bias was held constant at Ð250 V. GaN etch rates increased as the ICP source power increased due to higher concentrations of reactive species which increases the chemical component of the etch mechanism and/or higher ion flux which increases the bond breaking and sputter desorption efficiency of the etch. Etch rates have also been observed to stabilize or decrease under high plasma flux conditions due either to saturation of reactive species at the surface or sputter desorption of reactive species from the surface before the reactions occur. The etch profile was anisotropic and smooth up to 1000 W ICP power where the feature dimensions were lost and sidewall morphology was rough due to erosion of the mask edge under high plasma flux conditions. In addition to etch rates, etch selectivity or the ability to etch one film at higher rates than another can be very important in device fabrication. For example, optimization of etch selectivity is critical to control threshold voltage uniformity for high electron mobility transistors (HEMTs), to accurately stop on either the emitter or collector regions for metal contacts for heterojunction bipolar transistors (HBTs), and for low resistivity n-ohmic contacts on InN layers. Several studies have recently reported etch selectivity for the group-III nitrides [6] [7] [8]. For example, Figure 9 shows GaN, InN and AlN etch rates and etch selectivities as a function of cathode rf power in an ICP-generated Cl₂/Ar plasma. Etch rates for all three films increased with increasing cathode rf power or dc bias due to improved breaking of the III-N bonds and more efficient sputter desorption of the etch products. Increasing InN etch rates were especially significant since InCl₃, the primary In etch product in a Cl-based plasma, has a relatively low volatility. However, under high dc-bias conditions, desorption of the InCl₃ etch products occurred prior to coverage of the etch surface. The GaN:InN and GaN:AlN etch selectivities were <8:1 and decreased as the cathode rf power or ion energy increased. Smith and co-workers reported similar results in a Cl₂-Ar ICP plasma where GaN:AlN and GaN:AlGaN selectivities decreased as dc bias increased. At Ð20 V dc bias, etch selectivities of ~39:1 were reported for GaN:AlN and ~10:1 for GaN:AlGaN. Temperature dependent etching of the group-III nitrides have been reported in ECR and ICP etch systems [5]. Etch rates are often influenced by the substrate temperature which can effect the desorption rate of etch produce, the gas-surface reaction kinetics and the surface mobility of reactants. Substrate temperature can be controlled and maintained during the etch process by a variety of clamping and backside heating or cooling procedures. GaN and InN etch rates are shown in Figure 10 as a function of temperature in Cl₂/H₂/Ar ICP plasma. GaN etch rates were much faster than InN due to higher volatility of the GaCl₃ etch products as compared to InCl₃ and showed little dependence on temperature. However, the InN etch rates showed a considerable temperature dependence increasing at 150¡C due to higher volatilities of the InCl₃ etch products at higher substrate temperatures.

Several different plasma chemistries have been used to etch the group-III nitrides. As established above, etch rates and profiles can be strongly affected by the volatility of the etch products formed. Table I shows the boiling points of possible etch products for the group-III nitrides exposed to halogen- and hydrocarbon-based plasmas. For halogen-based plasmas, etch rates are often limited by the volatility of the group-III halogen etch product. For Ga- and Al-containing films, chlorine-based plasmas typically yield fast rates with anisotropic, smooth etch profiles. CH₄/H₂-based plasma chemistries have also yielded smooth, anisotropic profiles for Ga-containing films, however at much slower rates. Based only on a comparison of etch product volatility, slower etch rates in CH₄-based plasmas is unexpected since the (CH₃)₃Ga etch product has a much lower boiling point than GaCl₃. This observation demonstrates the complexity of the etch process where redeposition, polymer formation, and gas-phase kinetics can influence the results. As shown above, etch rates for In-containing films obtained in room temperature chlorine-based plasmas tend to be slow with rough surface morphology and overcut profiles due to the low volatility of the InCl₃ and preferential loss of the group-V etch products. However, at elevated temperatures (>130¡C), the InCl₃ volatility increases and the etch rates and surface morphology improve. Significantly better room temperature etch results are obtained in CH₄/H₂-based plasmas due to the formation of more volatile (CH₃)₃In etch products. Another example of plasma chemistry dependent etching of GaN is shown in Figure 13 for Cl₂/N₂/Ar and BCl₃/N₂/Ar ICP-generated plasmas. In the Cl₂-based plasma, GaN etch rates decreased as the % N₂ increased, presumable due to a reduction in reactive Cl. In the BCl-based plasma GaN etch rates increased up to 40% N₂ and then decreased at higher N₂ concentration. This observation has also been reported for ECR and ICP etching of GaAs, GaP and In-containing films [8]. Ren and co-workers first observed maximum etch rates for In-containing films (InGaN and InGaP) in an ECR discharge at a gas ratio of 75/25 for BCl₃/N₂ [10]. Using optical emission spectroscopy (OES), Ren reported maximum emission intensity for atomic and molecular Cl at 75% BCl₃ as well as a decrease in the BCl₃ intensity and the appearance of a BN emission line. The authors speculated that N₂ enhanced the dissociation of BCl₃ resulting in higher concentrations of reactive Cl and Cl ions and thus higher etch rates.

Additionally, the observation of BN emission suggested that less B was available to recombine with reactive Cl. This explanation may also be applied to the peak GaN etch rates observed at 40% N2 in the ICP BCl3/N2/Ar plasmas. However, OES of the BCl3/N2/Ar ICP discharge did not reveal higher concentrations of reactive Cl nor a BN peak emission. In Figure 14, OES spectra are shown for (a) 100% BCl3 (b) 75% BCl3-25% N2, (c) BCl3-75% N2 and (d) 100% N2 ICP plasmas. As N2 was added to the BCl3 plasma, the BCl3 emission (2710 Å) and Cl emission (5443 and 5560 Å) decreased while the BN emission (3856 Å) was not obvious. BCl₃/Cl₂ plasmas have shown encouraging results in the etching of GaN films. The addition of BCl₃ to a Cl₂ plasma can improve sputter desorption due to higher mass ions and reduce surface oxidation by gettering H₂O from the chamber. In Figure 15, GaN etch rates are shown as a function of % Cl₂ in a BCl₃/Cl₂/Ar ICP plasma. As the % Cl₂ increased, GaN etch rates increased up to 80% due to higher concentrations of reactive Cl. OES showed the Cl emission intensity increased and the BCl emission intensity decreased as the % Cl₂ increased.

Slower GaN etch rates in a pure Cl₂ plasma were attributed to less efficient sputter desorption of etch products in the absence of BCl₃. The fastest GaN etch rates were observed at 10% BCl₃ where the ion current density and Cl radical density were the greatest as measured by OES and a Langmuir probe. In general, GaN:AlN and GaN:InN etch selectivities are <10:1 as a function of plasma chemistry for Cl₂- or BCl₃-based plasmas. GaN:AlN and GaN:InN etch selectivities were higher for Cl₂-based ICP plasmas as compared to BCl₃-based ICP plasma due to higher concentration of reactive Cl produced in the Cl₂-based plasmas thus resulting in faster GaN etch rates. Alternatively, InN and AlN etch rates showed much less dependence on plasma chemistry and were fairly comparable in Cl₂- and BCl₃-based plasmas. An example of etch selectivity dependence on plasma chemistry is shown in Figure 16. GaN, AlN and InN etch rates and etch selectivities are plotted as a function of % SF₆ for an ICP Cl₂/SF₆/Ar plasma. GaN and InN etch rates decreased as SF₆ was added to the plasma due to the consumption of Cl by S and therefore lower concentrations of reactive Cl. The AlN etch rates increased with the addition of SF₆ and reached a maximum at 20% SF₆. As SF₆ was added to the Cl₂ plasma, slower AlN etch rates were expected due to the formation of low volatility AlF₃ etch products. However, due to the high ion flux in the ICP, the sputter desorption of the AlF₃ may occur prior to passivation of the surface. Therefore, the GaN:AlN selectivity decreased rapidly from ~6:1 to <1:1 with the addition of SF₆. The GaN:InN selectivity reached a maximum of 4:1 at 20% SF₆.

The simple Cl₂/Ar chemistry works very well for most device fabrication processes, providing controllable etch rates. Even at biases <90 V, the GaN etch rate is still typically ~1000Åámin^-1.

3.2. I₂ and Br₂ Based.

Figure 17 shows etch rates for the binary nitrides and selectivities for InN over both GaN and AlN as a function of the boron halide percentage by flow in the gas load. The dc chuck self-bias decreases as the BI₃ content increases, suggesting that the ion density in the plasma is increasing. The InN etch rate is proportional to the BI₃ content, indicating the presence of a strong chemical component in its etching. In comparison, AlN and GaN show very low rates until ~50% BI₃ (~500 Å/min^-1 for AlN and ~1700 Å/min^-1 for GaN). An increase in the BI₃ content in the discharges actually produces a falloff in the etch rate for both AlN and GaN. We expect there are several possible mechanisms by which to explain these data. First, the decrease in chuck self-bias and hence ion energy under these conditions may more than compensate for the higher active iodine neutral flux. Second, the formation of the less volatile GaI_x and AlI_x etch products may create a selvege layer which suppresses the etch rate. This mechanism occurs in the Cl₂ reactive ion etching of InP. In this system, etching does not occur unless elevated sample temperatures or higher dc biases are used to facilitate removal of the InCl₃ etch product. In InN etch selectivity to both materials initially increases but also goes through a minimum. Note, however, that selectivities of >100 can be achieved for both InN/AlN and InN/GaN.

Data for BBr₃/Ar discharges are also shown in Figure 17 for fixed source power (750 W) and rf chuck power (350 W). Higher rf powers were required to initiate etching with BBr₃ compared to BI₃ and the dc self-bias increased with the BBr₃ content. The etch rate of InN is again a strong function of the boron halide content, while GaN shows significant rates (~1800 Å/min^-1) only for pure BBr₃ discharges, AlN shows very low etch rates over the whole range of conditions investigated. Maximum selectivities of ~100:1 for InN/AlN and ~7.5:1 for InN/GaN are obtained.

Based on the results in Figure 17, we chose fixed plasma compositions, and varied the ion energy and flux through control of the source and chuck powers. Figure 18 shows that source power had a significant effect only on the InN etch rate for both 4BI₃/6Ar and 4BBr₃/6Ar discharges at fixed rf power (150 W). The etch rate of InN continues to increase with source power, which controls the ion flux and dissociation of the discharge, whereas the GaN and AlN rates are low for both plasma chemistries. The InN etch rates are approximately a factor of 2 faster in BI₃/Ar than in BBr₃/Ar even for lower rf chuck powers. This is expected from taking into consideration the relative stabilities of the respective In etch products (the InI₃ melting point is 210¡C; InBr₃ sublimes at <600¡C). The resultant selectivities are shown at the bottom of Figure 18; once again a value of ~100:1 for InN over GaN is achieved with BI₃, whereas BBr₃ produced somewhat lower values.

The dependence of the etch rate and InN/AlN and InN/GaN selectivities on rf chuck power for both plasma chemistries at fixed source power (750 W) is shown in Figure 19. While the GaN and AlN etch rates (top left) increase only at the highest chuck powers investigated for 4BI₃/6Ar discharges, the InN etch rate increases rapidly to 250 W. This is consistent with a strong ion-assisted component for the latter under these conditions. The subsequent decrease in the etch rate at higher power produces corresponding maxima (³100) in etch selectivity for chuck powers in the range of 150-250 W. This type of behavior is quite common to high density plasma etching of III-V materials, where the etching is predominantly ion-assisted desorption of somewhat volatile products, with insignificant rates under ion-free conditions. In this scenario, at very high ion energies, the active etching species (iodine neutral in this case) can be removed by sputtering before they have a chance to complete the reaction with substrate atoms. Similar data for BBr₃/Ar mixtures are also shown in Figure 19. For this chemistry the InN etch rate saturates and we did not observe any reduction in etch rate, although this might be expected to occur if higher powers could be applied (over power supply is limited to 450 W). GaN does show an etch rate maximum at ~350 W, producing a minimum in the resultant InN/GaN selectivity. The etch selectivity of InN over the other two nitrides for BI₃/Ar is again much higher than for BBr₃/Ar.

The effect of plasma composition on etch rates and selectivities if GaN, AlN and InN in ICl/Ar and IBr/Ar discharges at 750 W source power, 250 W rf chuck power and 5 mTorr is shown in Figure 20. The etch rates of InN and AlN are relatively independent of the plasma composition for both chemistries over a broad composition range, indicating the etch mechanism is dominated by physical sputtering. The dc bias voltage increased with increasing interhalogen concentrations. The decrease in ion flux also implies an increase in the concentrations of neutral species such as Cl, Br and I. The etch rate of GaN steadily increased with increasing ICl concentration. By contrast the etch rate of GaN saturated beyond 66.7% IBr. These results indicate that etching of GaN in both chemistries can be attributed more to chemical etching by increased concentrations of reactive neutrals than to ion-assisted sputtering. The effect of plasma composition showed an overall trend of decrease in selectivities for InN over both AlN and GaN as the concentration of ICl and IBr increased.

3.3. CH₄/H₂/Ar.

Vartuli and co-workers compared etch selectivities in CH₄/H₂/Ar and Cl₂/Ar plasmas in both RIE- and ECR-generated plasmas [5]. For CH₄/H₂/Ar plasmas, InN:GaN and InGaN:GaN etch selectivities ranged from 1:1 to 6:1 whereas etch selectivities of 1:1 or favoring GaN over the In-containing films was reported for Cl₂/Ar plasmas.

4. Etch Profile And Etched Surface Morphology

In several studies atomic force microscopy (AFM) has been used to quantify the etched surface morphology as root-mean-square (rms) roughness. Rough etch morphology often indicates a nonstoichiometric surface due to preferential removal of either the group-III or group-V species.

A summary of etch rate results for the nitrides with different chemistries and different techniques is shown in Table II.

5. Plasma-Induced Damage

Ren and co-workers measured electrical characteristics for InAlN and GaN FET structures to study plasma-induced damage for ECR BCl₃, BCl₃/N₂ and CH₄/H₂ plasmas [10]. They reported: (a) doping passivation in the channel layer in the presence of hydrogen; (b) high ion bombardment energies can create deep acceptor states that compensate the material and (c) preferential loss of N can produce rectifying gate characteristics. Ping and co-workers studied Schottky diodes for Ar and SiCl₄ RIE plasmas [6]. More damage was observed in pure Ar plasmas and under high dc-bias conditions. Plasma-induced damage of GaN was also evaluated in ICP and ECR Ar plasma using photoluminescence (PL) measurements as a function of cathode rf power and source power [12]. The peak PL intensity decreased with increasing ion energy independent of etch technique. As a function of source power or plasma density the results were less consistent. The PL intensity showed virtually no change at low ICP source power and then decreased as the plasma density increased. In the ECR plasma, the PL intensity increased by ~115% at low ECR source power and improved at higher ECR source powers but at a lower rate. The effect of postetch annealing in Ar varied depending on initial film conditions, however, annealing at temperatures above 440¡C resulted in a reduction in the PL intensity.

Surface stoichiometry can also be used to evaluate plasma-induced damage. Nonstoichiometric surfaces can be created by preferential loss of one of the lattice constituents. This may be attributed to higher volatility of the respective etch products, leading to enrichment of the less volatile species, or preferential sputtering of the lighter element. Auger electron spectroscopy (AES) can be used to measure surface stoichiometry. Figure 22 shows characteristic Auger spectra for (a) as-grown GaN samples and samples exposed to an ECR plasma t 850 W applied microwave power and cathode rf powers of (b) 65 and (c) 275 W. For the as-grown sample, the Auger spectrum showed a Ga:N ratio of 1.5 with normal amounts of adventitious carbon and native oxide on the GaN surface. Following plasma exposure, the Ga:N ratio increased as the cathode rf power increased with some residual atomic Cl from the plasma. Under high ion energy conditions, preferential removal of the lighter N atoms as observed resulting in Ga-rich surfaces.

5.1. n-GaN.

The layer structure and contact metals are shown schematically in Figure 23. The GaN was grown by rf plasma-assisted molecular beam epitaxy on c-plane Al₂O₃ substrates. The Ti/Au ohmic contracts were patterned by lift-off and annealed at 750¡C, producing contact resistances in the 10^-5 Wcm^-2 range. Samples were exposed to either pure N₂ or H₂ discharges in a Plasma Therm 790 ICP system at a fixed pressure of 5 mTorr. The gases were injected into the ICP source at a flow rate of 15 standard cubic centimeters per minute (sccm). The experimentally varied parameters were source power (300-1000 W) and rf chuck power (40-250 W), which control ion flux and ion energy, respectively. In some cases the samples were either annealed in N₂ for 30 s at 300-850¡C, or photoelectrochemically etched in 0.2 M KOH solutions at 25¡C after plasma exposure. The Pt/Au Schottky metallization was then deposited through a stencil mass by e-beam evaporation. Current-voltage characteristics were recorded on a HP4145A parameter analyzer, and we defined the reverse breakdown voltage (V_B) as the voltage at which the leakage current was 10^-3 A. We found in all cases that plasma exposure caused significant increases in forward and reverse current, with ideality factors increasing from typical values of 1.4-1.7 on control samples to >2. For this reason we were unable to extract meaningful values of either ideality factor or barrier height. Figure 24 shows a series of I-V characteristics from the GaN diodes fabricated on samples exposed to either H₂ or N₂ discharges at different source powers. It is clear that N₂ plasma exposure creates more degradation of the diode characteristics than does H₂ exposure. This implicates the ion mass (²⁸N₂⁺, ²H₂⁺ for the main positive ion species) as being more important in influencing the electrical properties of the GaN surface than a chemical effect, since H₂ would be likely to preferentially remove nitrogen from the GaN as NH₃.

The variations of V_B of the diodes with the source power during plasma exposure are shown in Figure 25. For any exposure to the N₂ discharges V_B is severely reduced. By contrast there is less degradation with the H₂ plasma to the lower average ion energy at those conditions, as shown at the bottom of Figure 25. The average ion energy is approximately equal to the sum of dc self-bias and plasma potential, with the later being in the range Ð22 to Ð28 V as determined by Langmuir probe measurements. Ion induced damage in GaN displays n-type conductivity, and in addition the heavy N₂⁺ ions are also more effective in preferential sputtering of the N relative to Ga, compared to the H₂⁺ ions of similar energy.

Similar conclusions can be drawn from the data on the effect of increasing rf chuck power. There were once again very severe decreases in breakdown voltage and increase in leakage current. The V_B values fall by more than a factor of two even for very low self-biases, and emphasize how sensitive the GaN surface is to degradation by energetic ion bombardment. The degradation saturates beyond ~100 W chuck power, corresponding to ion energies of ~175 eV. We assume that once the immediate surface becomes sufficiently damaged, the contact properties basically cannot be made any worse and the issue is then whether the damage depth increases with the different plasma parameters. Since ion energy appears to be a critical factor in creating the near-surface damage, we would expect damage depth to increase with ion energy in a nonetching process. In the case of simultaneous etching and damage creation (e.g. in Cl₂/Ar etch processing), higher etch rates would lead to lower amounts of residual damage because the disordered region would be partially removed.

The damage depth was established by photoelectrochemically wet etching different amounts of the plasma-exposed GaN surfaces, and then depositing the Pt/Au metal. Figure 26 (top) show the effect on the I-V characteristics of this removal of different depth of GaN. There is a gradual restoration of the reverse breakdown voltage, as shown at the bottom of the figure. Note that the forward part of the characteristics worsens for removal of 260 Å of GaN, and shows signs of high series resistance. This would be consistent with the presence of a highly resistive region underneath the conducting near-surface layer, created by point defect diffusion from the surface. A similar model applies to ion-damaged InP, i.e. a nonstoichiometric near-surface region (deficient in P in that case), followed by a transition to a stoichiometric but point defect-compensated region, and finally to unperturbed InP.

The fact that plasma exposure severely degraded the surface was clear from the atomic force microscopy (AFM) data. Exposure to the 500 W source power, 150 W rf chuck power (dc self-bias Ð 221 V), 5 mTorr N₂ discharge increased the root-mean-square (rms) surface roughness form 0.8 to 4.2 nm. Subsequent photoelectrochemical etching restored the initial morphology. However, we observed the onset of increasingly rough surfaces for deeper etch depths [4], reducing a relatively inaccurate measure of how much of the surface had to be removed in order to restore the diode breakdown voltage to its original value. We were able to estimate this depth as ~600 ± 150 Å for the N₂ plasma conditions mentioned above. Another method for trying to restore the electrical properties of the plasma-exposed surface is annealing. I-V data from annealed samples are shown in Figure 27. At the top are characteristics from samples, which were plasma exposed (N₂, 500 W source power, 150 W rf chuck power, 5 mTorr), then annealed and the contact deposited. These samples show that increasing annealing temperature up to 750¡C brings a substantial improvement in V_B (Figure 27, bottom). However for annealing at 850¡C the diode began to degrade and this is consistent with the temperature at which N₂ begins to be lost from the surface. In the case where the samples were exposed to the N₂ plasma, and then the Pt/Au contact was deposited prior to annealing, the I-V characteristics show continued worsening upon annealing (Figure 27, center). In this case, the Pt/Au contact is stable to 700¡C on unetched samples. The poorer stability in etched samples could be related to the surface damage enhancing interfacial reaction between the Pt and GaN.

The main findings of this study can be summarized as follows:

There is a severe degradation in the electrical quality of GaN surfaces after ICP H₂ or N₂ discharge exposure. Under all conditions there is a strong reduction of V_B in diode structures to the point at which the Schottky contacts show almost ohmic-like behavior. These observations are consistent with the creation of a conducting n-type surface layer resulting from energetic ion bombardment. Heavier ions (N₂⁺) create more damage than lighter ions (H₂⁺) in this situation, where damage is accumulating without any concurrent etching of the surface.

The depth of the damage is approximately 600 Å, as judged by the return of the diode characteristics to their control values.

Annealing at 750¡C is also effective in helping remove the effects of plasma exposure. Higher temperatures lead to degradation in GaN diode properties for uncapped anneals.

5.2. p-GaN.

Figure 28 shows the I-V characteristics from samples exposed to either H₂ (top) or Ar (bottom) ICP discharges (150 W rf chuck, 2 mTorr) as a function of source power. In both cases there is an increase in both the reverse breakdown voltage and the forward turn-on voltage, with these parameters increasing monotonically with the source power during plasma exposure.

Figure 29 shows this increase in breakdown voltage as a function of source power, and also the variation of the chuck dc self-bias. As the source power increases, the ion density also increases and the higher plasma conductivity suppresses the developed dc bias. Note that the breakdown voltage of the diodes continues to increase even as this bias (and hence ion energy, which is the sum of this bias and the plasma potential) decreases. These results show that ion flux lays an important role in the change of diode electrical properties. The other key result is that Ar leads to consistently more of an increase in breakdown voltage, indicating that ion mass is important rather than any chemical effect related to removal of N₂ or NH₃ in the H₂ discharges.

The increase in breakdown voltage on the p-GaN is due to a decrease in hole concentration in the near-surface region through the creation of shallow donor states. The key question is whether there is actually conversion to an n-type surface under any of the plasma condition. Figure 30 shows the forward turn-on characteristics of the p-GaN diodes exposed to different source power Ar discharge at low source power (300 W), the turn-on remains close to that of the unexposed control sample. However there is a clear increase in the turn-on voltage at higher source powers, and in fact at ³ 750 W the characteristics are those of an n-p junction. Under these conditions the concentration of plasma-induced shallow donors exceeds the hole concentration and there is surface conversion. In other words the metal-p GaN diode has become a metal-n GaN-p GaN junction. We always find that plasma exposed GaN surfaces are N₂-deficient relative to their unexposed state, and therefore the obvious conclusion is nitrogen vacancies create shallow donor levels. This is consistent with thermal annealing experiments in which N₂ loss from the surface produced increased n-type conduction.

An important question is the depth of the plasma-induced damage. We found we were able to etch p-GaN very slowly in boiling NaOH solutions, at rates that depended on the solution molarity (Figure 31) even without any plasma exposure of the material. This enabled us to directly measure the damage depth in plasma exposed samples in two different ways. The first method involved measuring the etch rate as a function of depth from the surface. Defective GaN resulting from plasma, thermal or implant damage can be wet chemically etched at rates much faster than undamaged material because the acid or base solutions are able to attack the broken or strained bonds present. Figure 32 shows the GaN etch rate as a function of depth in samples exposed to a 750 W source power, 150 W rf chuck power Ar discharge. The etch rate is a strong function of the depth from the surface and saturates between ~425-550 Å. Within this depth range the etch rate is returned to the ÒbulkÓ value characteristic of undamaged p-GaN.

The other method of removing plasma-induced damage is annealing. In these experiments we exposed the samples to the same type of plasma (Ar, 750 W source power, 150 W rf chuck power) and then annealed under N₂ at different temperatures. Figure 34 (top) shows the I-V characteristics of these different samples, while Figure 34 (bottom) shows the resulting breakdown voltages as a function of annealing temperature. On this wafer, Plasma exposure caused an increase in breakdown voltage from ~2.5 to ~18 V. Subsequent annealing at 400¡C initially decreased the breakdown voltage, but higher temperature produced a large increase. At temperatures above 700¡C, the diodes characteristics returned toward their initial values and were back to the control values by 900¡C. This behavior is similar to that observed in implant-isolated compound semiconductors where ion damage compensates the initial doping in the material, producing higher sheet resistance. In many instances the damage site density is larger than that needed to trap all of the free carriers, and trapped electrons or holes may move by hopping conduction. Annealing at higher temperatures removes some of the damage sites, but there are still enough to trap all the conduction electrons/holes. Under these conditions the hopping conduction is reduced and the sample sheet resistance actually increases. At still higher annealing temperatures, the trap density falls below the conduction electron or hole concentration and the latter are returned to their respective bands. Under these conditions the sample sheet resistance returns to its pre-implanted value. The difference in the plasma exposed samples is that the incident ion energy is a few hundred eV compared to a few hundred keV in implant-isolated material. In the former case the main electrically active defects produced are nitrogen vacancies near the surface, whereas in the latter case there will be vacancy and interstitial complexes produced in far greater numbers to far greater depths. In our previous work on plasma damage in n-GaN we found that annealing at ~750¡C almost returned the electrical properties to their initial values. If the same defects are present in both n- and p-type material after plasma exposure, this difference in annealing temperature may be a result of a Fermi level dependence to the annealing mechanism. The main conclusions of this study may be summarized as follow:

The effect of either H₂ or Ar plasma exposure on p-GaN surfaces is to decrease the net acceptor concentration through creation of shallow donor levels, most likely N_V. At high ion fluxes or ion energies there can be type conversion of the initially p-type surface. The change in electrical properties is more pronounced with Ar than with H₂ plasmas under the same conditions.

Two different techniques for measuring the damage depth find it to be in the range 400-500 Å under our conditions. After removing this amount of GaN, both the breakdown voltage and wet chemical etch rates are returned to their initial values.

Post-etch annealing in N₂ at 900¡C restores the initial breakdown voltage on plasma exposed p-GaN. Annealing at higher temperatures degraded the electrical properties, again most likely due to N₂ loss from the surface.

5.1. Schottky Diodes.

Dry etching is needed for a range of GaN electronic devices, including mesa diodes rectifiers, thyristors and heterojunction bipolar transistors for high temperature, high power operation. These applications include control of power flow in utility grids, radar and electronic motor drives. It is critical to understand the depth and thermal stability of dry etch damage in both n- and p-type GaN and its effect on the current-voltage (I-V) characteristics of simple diode structures.

In this section we report on a comparison of the effects of Cl₂/Ar and Ar Inductively Coupled Plasma (ICP) exposure on the electrical properties of n- and p-GaN Schottky diodes. In some cases it was found that Cl₂/Ar discharges could produce even more damage than pure Ar, due to the slightly higher ion energies involved. The damage saturates after a short exposure to either Cl₂/Art or Ar discharges and is significant even for low ion energies. Annealing between 700-800¡C restored ³ 70% of the reverse breakdown voltage on n-GaN, while the damage depth was again established to be ~500 Å in p-GaN.

The GaN layers were grown by rf plasma-assisted Molecular Beam Epitaxy on c-plane Al₂O₃ substrates. The Ti/Al (for n-type) and Ni/Au (for p-type) ohmic contacts were patterned by lift-off and annealed at 750¡C. The samples were exposed to either 10Cl₂/5Ar or 15 Ar (where the numbers denote the gas flow rate in standard cubic centimeters per minute) ICP discharges in a Plasma-Therm ICP reactor at a fixed pressure of 3 mTorr. We investigated a range of rf chucks powers (25-250 W) and etch times (4-100 secs), with a fixed source power of 500 W. In some cases, the samples were either annealed in N₂ for 30 secs at 500-800¡C or wet etching in 0.1M NaOH solutions at ~100¡C after plasma exposure. The Schottky metallization (Pt/Au in both cases) was then deposited through a stencil mask (F = 70 or 90µm) by e-beam evaporation. Current-voltage characteristics were recorded on an HP 4145A parameter analyzer, and we defined the reverse breakdown voltage as that at which the leakage current was 10^-3A. The forward on-voltage V_F was defined as the voltage at which the forward current was 100Aácm^-2. In all cases the ideality factors increased from 1.3-1.6 on control samples to >2 after plasma exposure and thus we were unable to extract meaningful values of either barrier height or ideality factor.

Figure 35 shows a series of I-V characteristics from n-type GaN diodes fabricated on samples exposed to either Cl₂/Ar (top) or Ar (bottom) discharges at different rf chuck powers. There is a significant reduction in V_B under all conditions, with Ar producing less damage at low chuck powers. This is probably related to two factors Ð the slightly higher chuck bias with Cl₂/Ar due to the lower positive ion density in the plasma (Cl is more electronegative than Ar) and the heavier mass of the Cl₂⁺ ions compared to Ar⁺. This is consistent with our past data on the relative effects of N₂ and H₂ plasma exposure, in which ion mass was found to be more important in influencing the electrical properties of the GaN surface than any chemical effects.

The variations of V_B and V_F with the rf chuck power during plasma exposure are shown in Figure 36 (top). At powers ²100 W, the Cl₂/Ar creates more degradation of V_B, as discussed above, while at higher powers the damage saturates. The average ion energy is the sum of dc self-bias (shown at the bottom of the figure) and plasma potential (which is about 22-25 eV under these conditions). Thus for ion energies <~150 eV, Ar produces less damage than Cl₂/Ar, even though the etch rate with the latter is much higher. This is also reflected in the variation of _F with rf chuck power.

Figure 37 shows a series of I-V characteristics from n-type GaN diodes fabricated on samples exposed to the two different plasmas for different times at fixed rf chuck power (150 W) and source power (500 W). It is clear that the damage accumulates rapidly, with the I-V characteristics becoming linear at longer times. It should be remembered that this is damage accumulating ahead of the etch front. To examine the thermal stability of the etch damage, n-type samples were exposed to Ar or Cl₂/Ar discharges at a fixed source power (500 W) and rf chuck power 150 W rf), and then annealed at different temperatures prior to deposition of the rectifying contact. The annealing produces a significant recovery of the electrical properties for samples exposed to either type of plasma. The V_B values are shown in Figure 38, as a function of post-plasma exposure annealing temperature. Annealing temperatures between 700-800¡C restore > 70% of the original V_B value, but clearly annealing alone cannot remove all of the dry etch induced damage. Annealing temperatures above 800¡C were found to lead to preferential loss of N₂ from the surface, with a concurrent degradation in V_B.

Figure 40 shows two methods for determining the depth of the damaged region in p-GaN diodes. At top is a plot of the variation of V_F and V_B with the depth of material removed by NaOH etching. The values of both parameters are returned to their control values by depths of 500-600 Å. What is clear from this data is that the immediate surface is not where the p-doping concentration is most affected, since the maximum values peak at depths of 300-400 Å. This suggests that N_V or other compensating defects created at the surface diffuse rapidly into this region even near room temperature. This is consistent with results in other semiconductors, where damage depths are typically found to be many times deeper than the projected range of incident ions. The bottom part of Figure 40 shows the wet etch depth in plasma damaged p-GaN as a function of etching time. The etch depth saturates at depths of 500-600 Å, consistent with the electrical data. It has previously been shown that the wet etch depth on thermally- or ion-damaged GaN was self-limiting. This is most likely a result of the fact that defective or broken bonds in the material are readily attached by the acid or base, whereas in undamaged GaN the etch rate is negligible.

The main finding of our study may be summarized as follows:

(1) Large changes in V_B and V_F of n- and p-GaN Schottky diodes were observed after exposure to both Cl₂/Ar and Ar ICP discharges. In some cases the electrical properties are more degraded with Cl₂/Ar even though this plasma chemistry has a much higher etch rate.

The damage accumulates in the near-surface even for very short exposure times (4 secs). The damage depth was established to be 500-600 Å from both the changes in electrical properties and the depth dependence of wet etch rate.

Annealing in the range 700-800¡C partially restores V_B in n-GaN diodes, but full recover can only be achieved with an additional wet etch step for removal of the damaged material. The combination of annealing and a wet etch clean-up step looks very promising for GaN device fabrication.

5.1. p-n Junctions.

Figure 41 shows the effect of dc chuck bias on the reverse junction leakage current, along with the corresponding GaN etch rates. There is little effect on the current below chuck biases of -250 V. This corresponds to an ion energy of approximately Ð275 eV, since this energy is the sum of chuck bias and plasma potential (about Ð25 eV in this tool under these conditions). The reverse current decreases slightly as the dc self-bias is increased from Ð25 to Ð50 V. This may result from the sharp increase in etch rate which leads to faster removal of near-surface damage. The reverse current increases rapidly above Ð275 V ion energy, which is a clear indication of severe damage accumulating on the sidewall. The damage probably takes the form of point defects such as nitrogen vacancies, which increase the n-type conductivity of the surface. The total reverse current density, J_R is the sum of three components namely diffusion, generation and surface leakage according to

where e is the electronic charge, D_e,_h are the diffusion coefficients of electrons or holes, l_e,_h are the lengths of the n and p regions outside the depletion region in a p-n junction, N_D,_A are the donor/acceptor concentrations on wither side of the junction, n_I is the intrinsic carrier concentration, W the depletion with t_g the thermal generation lifetime of carriers and J_SL is the surface current component which is bias-dependent. The latter component is most affected by the dry etch process, and dominates the reverse leakage in diodes etched in high ion energies.

GaN sidewall profiles and etch morphologies have been evaluated from previous results as a function of dc-bias. The etch becomes more anisotropic as the dc-bias increased from -50 to -150 V dc-bias due to the perpendicular nature of the ion bombardment energies. However, at -300 V dc-bias a tiered etch profile with vertical striations in the sidewall was observed due to erosion of the mask-edge under high ion bombardment energies. The physical degradation (both profile and morphology) of the etched sidewall at -300 V could help explain higher reverse leakage currents above -250 V dc-bias. Under high bias conditions, more energetic ions scattering from the surface could strike the sidewalls with significant momentum thus increasing the likelihood of increased damage and higher reverse leakage currents. Under low bias conditions, the sidewall profile is less anisotropic implying increased lateral etching of the GaN (undercutting of the mask). Under these conditions the etch process becomes dominated by the chemical component of the etch mechanism, which may account for the slightly higher reverse leakage observed at -25 V dc-bias. Figure 42 shows the effect of ICP source power on the junction reverse leakage current. The plasma flux is proportional to source power. In this experiment the ion energy was held constant at Ð100 V dc-bias. There is minimal effect on leakage current for source powers ² 500 W, with severe degradation of the junction characteristics at higher powers even thought the GaN etch rate continues to increase. This is an important result because it shows that the conditions that produce the highest etch rate are not necessarily those that lead to the least damage. Increased sidewall damage under high plasma flux conditions may be due to increased ion scattering as well as more interactions of reactive neutrals with the sidewall of the mesa. SEM micrographs from bulk GaN samples also show a degradation of sidewall profile under high ICP source power conditions. At an ICP source power of 1000 W, the sidewall has a tiered profile with vertical striations possibly due to erosion of the mask edge. However, sidewall profiles at 250 and 500 W looked reasonably anisotropic and smooth. Reverse leakage currents were relatively insensitive to chemistry effects in a Cl₂/BCl₃/Ar ICP discharge. The reverse leakage current ranged between ~10 and 40 nA as the %Cl₂ changed from 0 to 100. This is not too surprising given that BCl₃ ions will be the heaviest ions in the discharge under all these conditions, and we expect ion damage to be dominated at this flue. The reverse leakage currents were measured from a different GaN wafer as compared to other samples used in this study. The surface morphology for the as-grown wafer was significantly higher for this sample and may account for higher reverse leakage currents measured under the standard conditions. Notice the GaN etch rate increased as Cl₂ was added to the BCl₃/Ar plasma up to 80%. In Cl₂/Ar plasma the GaN etch rate decreased due to lower concentrations of reactive Cl neutrals. Etch profiles were relatively anisotropic and smooth except for the Cl₂/Ar plasma where the etch was slightly rough.

Two samples were annealed in this study to determine if the defects caused by plasma-induced-damage to the p-n junction could be removed and low reverse leakage currents recovered. The first sample was initially exposed to the following ICP conditions: 32 sccm Cl₂, 8 sccm BCl₃, 5 sccm Ar, 500 W ICP power, -300 V dc-bias and 2 mTorr pressure. The reverse leakage remained essentially constant up to 600¡C where the reverse leakage increased by more than an order of magnitude (see Figure 43). (Note all reverse leakage data were taken at Ð30 V except for the 600¡C data, which was taken at lower voltages due to breakdown at Ð30 V). A similar trend was observed for the second sample. Although there was much more scatter in the data, which was exposed to the same ICP conditions with the exceptions of 750 W ICP source power and Ð100 V dc-bias. The inability to remove damage from these samples may be due to anneal temperatures which were not high enough. Improved breakdown voltages for dry etched n- and p-GaN Schottky diodes occur after annealing in the range of 400 to 700¡C, however anneal temperatures >800¡C were needed to produce near-complete recovery in breakdown voltage.

In summary there are high-density plasma etching conditions for GaN where there is minimal degradation in the reverse leakage current of p-i-n mesa diodes. Both ion energy and ion flux are important in determining the magnitude of this current, and a high etch rate is not necessarily the best choice for minimizing dry etch damage.

6. Device Processing

6.1. Microdisk Lasers.

6.2. Ridge Waveguide Lasers.

While a number of groups have now reported room-temperature lasers in the InGaN/GaN/AlGaN heterostructure system under pulsed and cw operation [3] [4], the field has been completely dominated by Nakamura et.al. [4]. The growth is performed by MOCVD, generally at atmospheric pressure. Initial structures were grown on c-plane (0001) sapphire, with a low temperature (550¡C) GaN buffer, a thick n⁺ GaN lower contact region, an n⁺InGaN strain-relief layer, and n⁺AlGaN cladding layer, a light-guiding region of GaN, then a multiquantum well region consisting of Si-doped In_0.15Ga_0.85N wells separated by Si-doped In_0.02Ga_0.98N barriers. The p side of the device consisted of sequential layers of p-AlGaN, p⁺GaN light-guiding,, p-Al_0.09Ga_0.92N cladding and p⁺GaN contact. A ridge geometry was fabricated by dry etching in most cases (material removed down to the p-Al_0.08Ga_0.92N layer), followed by dry etching, cleaving or polishing to form a mirror facet. These facets are coated (with TiO₂/SiO₂ in the Nichia case) to reduce laser threshold, while Ni/Au (p type) and Ti/Al (n type) were employed for ohmic metallization.

For this type of structure, threshold current densities are typically ³ 4 kA cm^-2 with an operating voltages of ³ 5 V at the threshold current. The emission mechanism is still the subject of intense study, but may be related to localization of excitons at compositional fluctuations (leading to potential minima in the band structure) in the InGaN wells.. These devices display relatively short lifetimes under cw operation, typically tens to hundreds of hours. The failure mechanism is most commonly short-circuiting of the p-n junction, a result of p-contact metallization punch through. It is not that surprising that in this high defect density material that the metal can migrate down threading dislocations or voids under high drive-current conditions. The threshold carrier density of the laser diodes on sapphire are typically ~10²⁰ cm^-3, well above the theoretical values (~10¹⁹ cm^-3).

A major breakthrough in laser diode lifetime occurred with two changes to the growth. The first was replacement of the AlGaN cladding layers with AlGaN/GaN strained-layer superlattices, combined with modulation doping. These changes had the effect of reducing formation of cracks that often occurred in the AlGaN, and also to reduce the diode operating voltage [4]. The second was the use of epitaxial lateral overgrowth (ELOG) [3] [4]. In this technique GaN is selectively grown on an SiO₂ masked GaN/Al₂O₃ structure. After ~10µm of GaN is deposited over the SiO₂ stripes, it coalesces to produce a flat surface. For a sufficiently wide stripe width, the dislocation density becomes negligible, compared to ³ 10⁹cm^-2 in the window regions. The laser itself is fabricated slightly off-center from the mask regions, due to gaps that occur there due to imperfect coalescence of the GaN. These devices have lower threshold current density (² 4 kA cm^-2) and operating voltage (4-6 V) and much longer (10,000 h) room-temperature lifetimes. The reduction in threading dislocation density dramatically changes the lifetime, since the p metal no longer has a direct path for shorting out the junction during operation. The carrier density at threshold is also reduced to ~3x10¹⁹ cm^-3, not far above the expected values. Output power > 400 mW, and lifetime > 160 h at 30 mW constant output power has been reported.

Subsequent work from Nichia has focused on growth of the laser diodes on quasi-GaN substrates. The thick (100-200µm) GaN is grown on ELOG structures by either MOCVD or hydride VPE. The sapphire substrate is then removed by polishing, to leave a freestanding GaN substrate. The mirror facet can then be formed by cleaving. The GaN substrate has better thermal conductivity than sapphire.

One of the most important features of the etching of the ridge waveguide is the smoothness of the sidewall. Figure 45 shows SEM micrographs of features etched into pure GaN, using a SiN_x mask and an ICP Cl₂/Ar discharge at moderate powers (500 W source power, 150 W rf chuck power). While the sidewalls are reasonably vertical, one can see striations, which result from roughness on the photoresist mask used to pattern the SiN_x. Another problem than can occur is illustrated in the SEM micrograph at the top of Figure 46. In this a very high ion energy was employed during the etching, leading to roughening of the feature sidewall. This problem is absent when ion energies below approximately 200 eV are employed, as shown in the micrograph at the bottom of Figure 46.

When careful attention is paid to the lithography, the etching of the SiN_x mask and also the etching of the nitride laser structure, then results like those shown in the SEM micrographs of Figure 47 are obtained. The active region of the laser is visible as the horizontal lines along the middle of the sidewall.

6.3. Heterojunction Bipolar Transistors.

In this section we review the fabrication process for GaN/AlGaN HBTs, examine the temperature dependence of the p-ohmic contacts and report measurements of typical background impurity concentrations, determined by Secondary Ion Mass Spectrometry (SIMS).

Structures grown by two different methods were examined. In the first, rf plasma-assisted Molecular Beam Epitaxy (MSE) at a rate of ~0.5µm-hr^-1 was used to grow the HBT structure on top of a 2µm thick undoped GaN buffer that was grown on c-plane (0001) sapphire. An 8000Å thick GaN subcollector (Si ~10¹⁸ cm^-3) was followed by a 5000Å thick GaN collector (Si ~10¹⁷ cm^-3), a 1500Å thick GaN base (Mg acceptor concentration ~10¹⁸ cm^-3), a 1000Å thick Al_0.15Ga_0.85N emitter (Si ~5x10¹⁷ cm^-3), and a 500Å grade to a 2000Å thick GaN contact layer (Si ~8x10¹⁸cm^-3).

The second structure was grown by Metal Organic Chemical Vapor Deposition (MOCVD) on c-plane sapphire, using trimethylgallium, trimethylaluminum and ammonia as the precursors and high purity H₂ as the carrier gas. The basic layer structure is shown in Figure 48.

The process flow for device fabrication is shown schematically in Figure 49. First the emitter metal (Ta/Al/Pt/Au) is patterned by lift-off and used as an etch mask for the fabrication of the emitter mesa. The dry etching was performed in a Plasma-Therm 770 Inductively Coupled Plasma (ICP) system using Cl₂/Ar discharges. The process pressure was 5 mTorr and the source was excited with 300 W of 2 MHz power. This power controlled the ion flux and neutral density, while the incident ion energy was controlled by application of 40 W of 13.56 MHz power to the sample chuck. Base metallization of Ni/Pt/Au was patterned by lift-off, and then the mesa formed by dry etching. The etch rate of GaN under our conditions was ~1100Å-min^-1, and was terminated at the subcollector where Ti/Al/Pt/Au metallization was deposited. The contacts were alloyed at 700-800¡C. It has been firmly established that high specific contact resistivities are a limiting factor in GaN-based device performance, and in particular the p-ohmic contact. We examined the alloying temperature dependence of the current-voltage (I-V) characteristics for several different p-metal schemes. The as-deposited contacts are rectifying. Annealing at progressively higher temperatures produced a significant improvement. But even for 800¡C anneals the contacts were not purely ohmic when measured at room temperature. This is consistent with past data, showing that p-metallization on GaN is often better described as a leaky Schottky contact. As the measurement temperature is increased, the hole concentration in the p-GaN increases through higher ionization efficiency of the Mg acceptors. For example the hole concentration would increase from ~10% of the acceptor density at 25¡C to ~60% at 300¡C, based on Fermi-Dirac statistics. The p-contact becomes truly ohmic at ³ 300¡C. From transmission line measurements, we found r_c ~2x10^-2Wcm^-2 at this temperature. This indicates that the GaN/AlGaN HBT will perform better at elevated temperatures, where the base contact resistivity is lower. The contact barrier is of order 0.5 eV, whereas the Mg acceptor has an ionization level of 0.18 eV.

The device performance of both the MBE and MOCVD grown devices was similar, namely a common-emitter current gain of ² 3 at 25¡C, increasing to ~10 at 300¡C. In both devices the performance was still limited by the base resistance, and methods to increase the base doping and lower the extrinsic resistance in this region will be critical for future efforts in this area. The common base current gain, µ, was in the range 0.75 (25¡C) to 0.9 (300¡C), indicating that the base transport factor is close to unity and that I_B is dominated by re-injection to the emitter.

Another important aspect of the realization of GaN/AlGaN HBTs is confinement of the Mg doping to the base. If the p-type spills over into the relatively lightly doped emitter, then the junction is displaced and the advantage of the heterostructure is lost.

In summary, GaN/AlGaN HBTs have been fabricated both on MBE and MOCVD grown material, and they display similar performance, i.e. a common-emitter current gain of ~10 when operated at ~300¡C. However junction leakage is also higher at elevated temperatures, which is a major drawback in most applications. The fabrication process developed previously for other compound semiconductor systems works well for the GaN/AlGaN system, with the main difference being higher annealing temperatures required for the ohmic contacts. The device performance is still limited by the base doping for both MBE and MOCVD structures.

6.4. Field Effect Transistors.

The exceptional chemical stability of the nitrides has meant that dry etching must be employed for patterning. To date most of the work in this area has concentrated on achievement of higher etch rates with minimal mask erosion, in particular because a key application is formation of dry etched layer facets. In that case etch rate, etch anisotropy and sidewall smoothness are the most important parameters, and little attention has been paid to the effect of dry etching on the stoichiometry and electrical properties of the nitride surface.

In these experiments, we used an InAlN and GaN FET structure as a test vehicle for measuring the effect of electron cyclotron resonance (ECR) BCl₃-based dry etching on the surface properties of InAlN and GaN. Preferential loss of N leads to roughened morphologies and creation of a thin n⁺ surface layer which degrades the rectifying properties of subsequently deposited metal contacts.

The InAlN samples were grown by MOMBE on 2Ó diameter GaAs substrates using a WAVEMAT ECR N₂ plasma and metalorganic group III precursors (Trimethylamine alane, triethylindium). A low temperature (~400¡C) AlN nucleation layer was followed by a 500Å thick AlN buffer layer grown at 700¡C. The In_0.3Al_0.7N channel layer (~5x10¹⁷ cm^-2) was 500Å thick, and then an ohmic contact layer was produced by grading to pure InN over a distance of ~500Å.

The GaN layer structure was grown on double side polished c-Al₂O₃ substrates prepared initially by HCl/HNO₃/H₂O cleaning and an in-situ H₂ bake at 1070¡C. A GaN buffer < 300Å thick was grown at 500¡C and crystallized by ramping the temperature to 1040¡C, where trimethylgallium and ammonia were again used to grow ~1.5µm of undoped GaN (n < 3x10¹⁶ cm^-3), a 2000Å channel (n = 2x10¹⁷ cm^-3) and a 1000Å contact layer (n = 1x10¹⁸ cm^-3).

FET surfaces were fabricated by depositing TiPtAu source/drain ohmic contacts, which were protected by photoresists. The gate mesa was formed by dry etching down to the InAlN or n-GaN channel using an ECR BCl₃ or BCl₃/N₂ plasma chemistry. During this process, we noticed that the total conductivity between the ohmic contacts did not decrease under some conditions. CH₄/H₂ etch chemistry was also studied. To simulate the effects of this process, we exposed the FET substrates to D₂ plasma, we saw strong reductions in sample conductivity. The incorporation of D₂ into the InAlN was measured by secondary ion mass spectrometry. Changes to the surface stoichiometry were measured by Auger electron spectroscopy (AES). All plasma processes were carried out in a Plasma Therm SLR 770 System with an Astex 7700 low profile ECR source operating at 500 W. The samples were clamped to an rf-powered, He backside cooled chuck, which was left at floating potential (about Ð30 V) relative to the body of the plasma.

Upon dry etch removal of the InAlN capping layer, a Pt/Ti/Pt/Au gate contact was deposited on the exposed InAlN to complete the FET processing. if pure BCl₃ was employed as the plasma chemistry, we observed ohmic and not rectifying behavior for the gate contact. If BCl₃/N₂ was used, there was some improvement in the gate characteristics. A subsequent attempt at a wet-etch clean-up using either H₂O₂/HCl or H₂O₂//HCl produced a reverse breakdown in excess of 2 V (Figure 50). These results suggest that the InAlN surface becomes nonstoichiometric during the dry etch step, and that addition of N₂ retards some of this effect.

Figure 51 shows the I_DS values obtained as a function of dry etch time in ECR discharges of either BCl₃ or BCl₃/N₂. In the former case the current does not decrease as material is etched away, suggesting that a conducting surface layer is continually being created. By contrast BCl₃/N₂ plasma chemistry does reduce the drain-source current as expected, even though the breakdown characteristics of gate metal deposited on this surface are much poorer than would be expected.

Figure 52 shows the gate current-voltage characteristics when the gate metal is deposited on the as-etched GaN surface. The Schottky contact is extremely leaky, with poor breakdown voltage. We believe this is caused by the presence of a highly conducting N deficient surface, similar to the situation encountered on dry etching InP where preferential loss of P produces a metal-rich surface which precludes achievement of rectifying contacts. Auger electron spectroscopy analysis of the etched GaN surface showed an increasing Ga-to-N ratio (from 1.7 to 2.0 in terms of raw counts) upon etching. However, a 5 min. anneal at 400¡C under N₂ was sufficient to produce excellent rectifying contacts, with a gate breakdown of ~25 V (Figure 53). We believe the presence of the conducting surface layer after etching is a strong contributing factor t the excellent r_c values reported by Lin et. al. [3] for contacts on a reactively ion etched n-GaN.

The drain I-V characteristics of the 1x50µm² MESFET are shown in Figure 54. The drain-source breakdown was Ð20 V, with a threshold voltage of Ð0.3 V. The device displays good pinch-off and no slope to the I-V curves due to gate leakage, indicating that the anneal treatment is sufficient to restore the surface breakdown characteristics. We believe these devices are well suited for high power applications since GaN is a robust material and the contract metallizations employed are also very stable.

III-nitrides FET structures are sensitive to several effects during dry etching of the gate mesa. Firstly, if hydrogen is present in the plasma there can be passivation of the doping in the channel layer. Secondly, the ion bombardment from the plasma can create deep acceptor states that compensate the material. Thirdly, even when these problems are avoided through use of H-free plasma chemistries and low ion energies and fluxes, preferential loss of N can produce poor rectifying gate characteristics for metal deposited on the etched surface. Ping et.al. observed that pure Ar etching produced more damage in Schottky diodes than SiCl₄ RIE. The diodes characteristics were strongly dependent on plasma self-bias, while annealing at 680¡C removed much of the damage.

6.5. UV Detectors.

Nitride-based UV photodetectors which have been reported include p-n photodiode devices with 0.05 mm² junction area and 0.07 A/W peak responsivity [3], with 0.04 mm² junction area devices, with 0.25 mm² junction area 0.1 A/W peak responsivity and 0.59 mm² junction area and 0.195 A/W peak responsivity. Photodetector also reported include semi-transparent Schottky junction devices and metal-semiconductor-metal devices.

The GaN and AlGaN ultraviolet (UV) photodiodes were grown on (0001) basal-plane sapphire substrates by molecular beam epitaxy (MBE) using an RF atomic nitrogen plasma source. µm wide The p-i-n detector epitaxial layers consisted of a 5x10¹⁸ cm^-3 n-GaN layer followed by a 5000Å intrinsic region with unintentional n-type doping in the 10¹⁵ cm^-3 decade. The topmost epitaxial layer consisted of 1000-2000Å 1x10¹⁸ cm^-3 p-GaN. Mesas reaching the n-GaN cathode contact layer were formed by inductively coupled plasma (ICP) plasma etching with chlorine-based chemistry. Ohmic contacts to the n-type and p-type GaN were made by Ti-based and Ni-based metallizations, respectively. All of the GaN p-i-n UV detectors were fabricated with an optical detection area of 0.5 mm² and a p-i-n junction area of 0.59 mm², which is considerably larger (> 12.5 times) than other GaN p-n detectors reported with noise measurements.

In addition to fabricating p-i-n type detectors, shorter UV wavelength MSM photodetectors, operating in a quasi-photoconductive mode, were fabricated from 1.5µm thick silicon doped (~1x10¹⁷ cm^-3) n-AlGaN with a bandgap energy of approximately 320 nm. The MSMs were fabricated by first depositing a Pt-group metallization 1µm wide with a 5µm pitch to form the Schottky contacts. next a dielectric was deposited to act as an insulator between the AlGaN semiconductor and the bond pads. The dielectric process was not optimized to function as an anti-reflection coating. Reported here are results for AlGaN MSMs with active areas of 0.25 mm².

Shunt resistance and spectral responsivity data were collected using on-wafer probing. The shunt resistance was determined by the linear trace of the current-voltage (IV) characteristic from Ð10 mV to +10 mV. The spectral responsivities of the UV photodiodes were measured in photovoltaic mode (zero bias) for p-i-n devices and a photoconductive mode (biased) for MSM devices using a 75 W xenon arc lamp chopped at 700 Hz and filtered by a 1/8 meter monochromator set to a 5 nm bandpass. The power of the monochromatic light was measured with a calibrated, NIST traceable, silicon photodiode and then focused onto the GaN wafers resting on a micropositioner stage.

The GaN p-i-n UV photodetector responsivity measurements reported were obtained with the devices operating in the unbiased, photovoltaic mode. Shown in Figure 55 is a 25¡C spectral responsivity curve for a baseline UV photodetector with 0.194 A/W peak responsivity and 4 orders of magnitude visible rejection which has been previously reported. The maximum theoretical peak responsivity at the 360 nm bandgap is 0.28 A/W with no reflection and 0.23 A/W including reflection at the GaN surface. Also included on the plot is a trace for a GaN p-i-n UV photodetector with an improved p-type epitaxial process, which yields a greater visible rejection and more constant deep UV responsivity. The improved GaN p-i-n device was fabricated with a 1000Å p-type cathode layer. The shunt resistances for these improved 0.59 mm² devices ranged from 200 MW to 50 GW depending on the process. The device exhibits a low dark current and excellent forward-biased diode IV characteristics with a built-in potential of approximately 3.1 V.

The AlGaN MSM exhibited a substantial photoconductive gain (> 700 X) at 6 V bias, which yielded a responsivity of 7 A/W at 250 nm. The AlGaN device band-edge response does not decay as rapidly as that seen in devices fabricated from GaN material. The MSM devices, fabricated from AlGaN, which exhibited a luminescent peak at 320 nm, exhibited slightly more than 3 orders of magnitude (1278 X) rejection of 360 nm light over 320 nm light. The rejection ratio for 250 nm light over 360 nm light was more than 5.5 orders of magnitude (5263 X). The increased gain at shorter wavelengths is believed to be the result of greater electron-hole generation near the high electric field regions at the surface of the device. As expected, the MSM devices exhibited excellent shunt resistance (> 100 GW) and dark current characteristics at low bias as shown in Figure 55. GaN p-i-n UV photodetectors with an optically active surface area of 0.5 mm² and a junction area of 0.59 mm² have been fabricated on 3-inch diameter GaN p-i-n epitaxial wafers and characterized. Wafer maps of photodetector peak responsivity (maximum of 0.194 A/W at 359 nm) indicated that more than 60% (±1s) of all the GaN UV photodetectors performed within a ± 12% deviation from the average peak responsivity. Further, the vast majority of GaN UV photodetectors were characterized with shunt resistances that were within one decade of each other.

High temperature testing of the GaN p-i-n photodetectors up to 300¡C indicated no significant increase in visible spectral responsivity or short-term degradation. The room temperature spectral responsivity of the GaN photodetectors was fully recovered after 300¡C testing. The 300¡C GaN photodetector 1/Ä noise power densities were measured to be 6.6x10^-19 and 2.1x10^-21 A²/Hz at 100 Hz and 1 kHz, respectively. The room temperature, 100 Hz and 1 kHz noise power density of the GaN p-i-n photodetectors was extrapolated to be on the order of 10^-30 A²/Hz (~10^-15 A/Hz^1/2 noise current density).

The AlGaN MSM photodetectors, which were fabricated from AlGaN with a near bandgap luminescent peak of 320 nm, exhibited substantial photoconductive gain resulting in 7 A/W responsivity at 250 nm and 1.7 A/W responsivity at 320 nm. The AlGaN MSMs were characterized with 5263 and 1278 responsivity rejection ratios for 250 nm and 320 nm light, respectively, versus 360 nm light.

Acknowledgments

References

[1] Stephen J. Pearton, GaN and Related Materials (Gordon and Breach Science Publishers, Amsterdam, 1997) .

[2] J.I. Pankove, T.D. Moustaka, Gallium Nitride I (Academic Press, San Diego, 1997) .

[3]H. Morkoc, Wide Band Gap Nitrides and Devices (Springer, Berlin, 1998)

[4] Shuji Nakamura, Gerhard Fasol, The Blue Laser Diode - GaN based Light Emitters and Lasers (Springer-Verlag, Heidelberg, 1997) .

[5]S.J. Pearton and R.J. Shul,ÓDry EtchingÓ in Gallium Nitride I, edited by J.I. Pankove and T.D. Moustakas (Academic, San Diego, 1998)

[6]I. Adesida, ÒEtching of GaN and Related Compounds,Ó in Properties, Processing and Applications of GaN and Related Semiconductors, ed. J. Edgar, S. Strite, I. Akasaki, H. Amano and C. Wetzel, EMIS Data Review No. 23 (INSPEC, IEE, London 1999)

[7] R. J. Shul, G. B. McClellan, S. J. Pearton, C. R. Abernathy, C. Constantine, C. Barratt, Electron. Lett. 32, 1408 (1996).

[8]R.J. Shul, ÒICP Etching of III-Nitrides,Ó in GaN and Related Materials II, edited by S.J. Pearton (Gordon and Breach, New York 1998)

[9] H. P. Gillis, D. A. Choutov, K. P. Martin, L. Song, Appl. Phys. Lett. 68, 2255 (1996).

[10] F. Ren, S. J. Pearton, R. J. Shul, J. Han, J. Electron. Mater. 27, 175 (1998).

[11] C. R. Eddy, Jr, B. Molnar, J. Electron. Mater. 28, 314 (1999).

[12] R.J. Shul, R.D. Briggs, J. Han, S.J. Pearton, J.W. Lee, C.B. Vartuli, K.P. Killeen, M.J. Ludowise, Mater. Res. Soc. Symp. Proc. 468, 355 (1997).

[13] J. I. Pankove, M. Leksono, S. S. Chang, C. Walker, B. Van Zeghbroeck, MRS Internet J. Nitride Semicond. Res. 1, 39 (1996).

[14] L. S. McCarthy, P. Kozodoy, S. P. DenBaars, M. Rodwell, U. K. Mishra, IEEE Electron Dev. Lett. EDL-20, 277 (1999).

[15] Fan Ren , Cammy R. Abernathy , J. M. Van Hove, P. P. Chow, R. Hickman, J. J. Klaassen, R. F. Kopf, Hyun Cho, K. B. Jung, J. R. La Roche, R. G. Wilson, J. Han, R. J. Shul, A. G. Baca, S.J. Pearton, MRS Internet J. Nitride Semicond. Res. 3, 41 (1998).

Table I

Figure 1. Schematic diagram of (a) RIE, (b) ECR and (c) ICP etch platforms.

Figure 2. Schematic diagram of high density plasma etching process.

Figure 3. RIBE nitride removal rates as a function of Cl2 percentage in Cl2/Ar beams

Figure 4. GaN etch rates in RIE, ECR, ICP and RIBE Cl2-based plasmas as a function of dc bias.