WO2007062590A1 - Low density drain hemts - Google Patents

Abstract

Methods and devices for fabricating AlGaN/GaN normally-off high electron mobility transistors (HEMTs). A fluorine-based (electronegative ions-based) plasma treatment or low-energy ion implantation is used to modify the drain-side surface field distribution without the use of a field plate electrode. The off-state breakdown voltage can be improved and current collapse can be completely suppressed in LDD-HEMTs with no significant degradation in gains and cutoff frequencies.

Description

Low Density Drain HEMTs

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority from U.S. Provisional Patent Application 60/740,256 filed on November 29, 2005, and also from U.S. Provisional Patent Application 60/748,339 filed on December 8, 2005, both of which are hereby incorporated by reference.

BACKGROUND AND SUMMARY

The present application relates to a method for breakdown voltage enhancement and current collapse suppression in normally-off high electron mobility transistors ("HEMTs"), and in particular, to fabrication of aluminum- gallium nitride/gallium nitride ("AlGaN/GaN") HEMTs using electrode-less drain- side surface field engineering, resulting in a "Low Density Drain" HEMT.

Excessive electric field can create problems in semiconductor devices. (One type of problem is hot carriers, where sufficiently energetic electrons or holes become able to travel through dielectrics; another type of problem is avalanching, where conduction becomes uncontrolled.) Even in devices designed to operate at minimal logic voltages, it is important to ensure that the voltage does not change too sharply at the drain boundary; and in devices which are intended to switch higher voltages, it becomes increasingly necessary to minimize the peak electric field.

Drain engineering has been one of the longest-running sub-areas of integrated device development, going back to the original LDD proposal of 1974. See Blanchard, "High Voltage Simultaneous Diffusion Silicon-Gate CMOS," 9 IEEE J.S.S.C. 103 (1974). Many techniques have been used to control peak electric field in high-voltage devices, often including various configurations of field plates and non-current-carrying diffusions.

This long-standing development challenge has particular relevance to the relatively new area of enhancement-mode ("E-mode") III-N HEMTs. Normally- off AlGaN/GaN HEMTs are desirable for microwave power amplifier and power electronics applications because they offer simplified circuit configurations and favorable operating conditions for device safety. However, the normally-off AlGaN/GaN HEMTs usually exhibit lower maximum drain current compared to their normally-on counterparts, especially when the threshold voltage is increased to about +1 V to assure the complete turn-off of the 2DEG channel at zero gate bias and provide additional operating safety. To compensate the reduction in maximum current and achieve the same power handling capability, the breakdown voltage (V_BK) needs to be further improved, but preferably not at the cost of increased gate-to-drain distance (which inevitably increases the device size). The use of a field-plate, connected to the gate or source electrodes, can effectively enhance V_BK by modifying the surface field distribution. The gate-terminated field plate, however, can introduce additional gate capacitances (C_Gs and C_GD)_> which reduce the devices' gain and cutoff frequencies. A source-terminated field plate has been used for achieving enhanced V_BK and mitigating the gain reduction, but this requires a thick dielectric layer between the gate and field plate.

A problem in GaN devices has been the current collapse phenomenon: when the source-drain voltage reaches a level at which impact ionization can occur, the maximum current carried by the device can actually decrease. It has been suggested that this undesirable effect is due to a trapping phenomenon in which mid-gap states are populated by hot electrons. Low Density Drain HEMTs

The present application discloses new approaches to the control of an electric field in a field effect transistor. Methods and devices are disclosed for fabrication of an HEMT modifying the surface field distribution between the gate and drain of a normally-off HEMT. Part or all of the region between gate and drain can be transformed into a region with low density of 2DEG using a CF₄ plasma treatment, forming a Low-Density Drain ("LDD") HEMT. The off-state breakdown voltage can be improved and current collapse can be completely suppressed in LDD-HEMTs with no significant degradation in gains and cutoff frequencies.

The disclosed innovations, in various embodiments, provide one or more of at least the following advantages:

• Allows for modification of the surface field distribution in normally-off HEMTs without the use of a field plate electrode.

• Asymmetric modification to drain side only is easily achieved.

• No change in topology is required: the additional fixed charge introduced into the wide-bandgap barrier layer does not affect physical topology.

• Provides for enhanced breakdown voltage and suppressed current collapse without degradation in gains or cutoff frequencies.

• Does not add any additional process steps to those already required for enhancement+depletion M-N fabrication.

• Surface state trapping and de-trapping is reduced or prevented.

• Current collapse is reduced or prevented. BRIEF DESCRIPTION OF THE DRAWINGS

The disclosed innovations will be described with reference to the accompanying drawings, which show important sample embodiments of the present innovations and which are incorporated in the specification hereof by- reference, wherein:

Figure 1 shows a prior art E-mode HFET.

Figure IA shows a DCFL circuit schematic for an E/D inverter.

Figure IB shows a DCFL circuit for a ring oscillator.

Figure 1C shows a photomicrograph of an inverter as one embodiment of the present innovations.

Figure ID shows a photomicrograph of a ring oscillator as one embodiment of the present innovations.

Figure 2 shows transfer characteristics of a conventional D-mode HEMT, an E-mode HEMT without the benefit of the present innovations, and one embodiment of the present innovations.

Figures 3A through 3F show one embodiment of a process of fabricating an E-mode AlGaN/GaN HFET.

Figure 4A shows I-V output characteristics for one embodiment of an E- mode AlGaN/GaN HFET.

Figure 4B shows I_g-V_gs characteristics for one embodiment of an E-mode AlGaN/GaN HFET.

Figure 5 shows fluorine ion concentration profiles as measured by "SIMS" for one embodiment of an E-mode AlGaN/GaN HFET.

Figure 6 shows the cross section of one embodiment of the present innovations prior to implantation of fluorine ions. Figure 7 shows fluorine ion concentration profiles as measured by "SIMS" for various embodiments.

Figures 7A and 7B show fluorine ion concentration profiles as measured by "SIMS" for various embodiments.

Figure 8A shows the I_d versus V_gs transfer characteristics of E-mode AlGaN/GaN HFETs after different CF₄ plasma-treatment conditions.

Figure 8B shows the g_m versus V_gs transfer characteristics of E-mode AlGaN/GaN HFETs after different CF₄ plasma-treatment conditions.

Figure 9 shows the extracted barrier heights and ideality factors of gate Schottky diodes with different CF₄ plasma treatments.

Figure 10 shows the V_& dependence on plasma power and treatment time for various E-mode AlGaN/GaN HFETs.

Figure 11 shows an AFM image depicting the insignificant etching effect of the CF₄ plasma treatment on the AlGaN layer.

Figure 12A shows the DC Id versus V_gs transfer characteristics for various E-mode AlGaN/GaN HFET embodiments.

Figure 12B shows the DC g_m versus V_gs transfer characteristics for various E-mode AlGaN/GaN HFET embodiments.

Figure 13 shows the DC output characteristics for one E-mode AlGaN/GaN HFET embodiment.

Figure 14A shows both reverse and forward gate currents with different CF₄ plasma treatments for various embodiments of an E-mode AlGaN/GaN HFET.

Figure 14B shows enlarged and forward gate currents with different CF₄ plasma treatments for various embodiments of an E-mode AlGaN/GaN HFET.

Figure 15 shows dependencies of f_t and f_max on gate bias, where V_dS is fixed at 12V. Figure 16 shows on-wafer measured f_t and f_max with different CF₄ plasma treatments.

Figures 17A through 17F show a sample process of fabricating an E-mode Si₃N₄AlGaN/GaN MISHFET.

Figure 18 shows sample DC output characteristics.

Figure 19A shows the transfer characteristics.

Figure 19B shows gate leakage currents.

Figure 20 shows pulse measurements.

Figure 21 shows small signal RF characteristics.

Figure 22 shows simulated conduction-band diagrams of conventional D- mode AlGaN/GaN HEMT without CF₄ plasma treatment.

Figure 23 shows simulated conduction-band diagrams of an E-mode AlGaN/GaN HEMT with a CF₄ plasma treatment.

Figure 24 shows the electron concentration of a conventional D-mode AlGaN/GaN HEMT without CF₄ plasma treatment and of an E-mode AlGaN/GaN HEMT with CF₄ plasma treatment.

Figure 25 shows one embodiment for a process flow of monolithic integration of E-mode and D-mode HEMTs for an inverter according to the present innovations.

Figures 26A through 26F show a sample process flow for monolithic integration of E-mode and D-mode HFETs.

Figure 27 shows a planar process flow for monolithic integration.

Figure 28 shows another sample process flow for E/D-mode HEMTs.

Figure 29 shows DC output characteristics of an D-HEMT and an E-HEMT fabricated by a planar process.

Figure 30 compares transfer characteristics of the planar process with those of a conventional process. Figure 31 shows static voltage transfer characteristics of an E/D HEMT inverter fabricated by a planar fabrication process.

Figure 32 show an epitaxial structure for the HEMTs used in a sample embodiment.

Figure 33 show the integrated process flow of monolithic integration of E- mode and D-mode HEMTs for a monolithic inverter.

Figure 34 show sample geometry parameters for inverters and ring oscillators.

Figure 35 show DC I-V transfer characteristics and output characteristics of sample D-mode and E-mode AlGaN/GaN HEMTs as disclosed.

Figure 36 shows performances of fabricated E- and D-mode AlGaN/GaN HEMTs.

Figure 37 show I_g— V_g characteristics of both D- and E-mode HEMTs and simulated conduction-edge band diagrams under the gate electrode for a D-mode HEMT and an E-mode HEMT.

Figure 38 shows static voltage transfer characteristics for a conventional E/D HEMT inverter.

Figure 39 shows static voltage transfer characteristics of E/D HEMT inverters with β = 6.7, 10, 25, and 50 according to various disclosed embodiments.

Figure 40 shows noise margins for inverters with different beta values.

Figure 41 shows static voltage transfer characteristics of E/D HEMT inverters with β = 10 measured at different supply voltages.

Figure 42 show noise margins measured at different V_DD'S for an inverter with β = 10.

Figure 43 shows load and input current of an inverter with β = 10 at V_DD ⁼ 2.5 V, according to a sample embodiment. Figure 44 shows a frequency spectrum, and Figure 45 shows time-domain characteristics, of a 17-stage ring oscillator with β = 10 biased at V_DD = 3.5 V.

Figure 46 shows dependences of propagation delay and power-delay product on the supply voltage for one circuit embodiment.

Figure 47 shows the process flow for fabrication of an LDD-HEMT according to one embodiment of the present innovations.

Figure 47A shows the 2DEG for an LDD-HEMT according to one embodiment of the present innovations.

Figure 48 shows off-state breakdown voltages of one embodiment of the present innovations.

Figure 49 shows breakdown voltage dependence on the length of the LDD region for a fixed gate-drain spacing L_GD ⁼ 3 μm for multiple embodiments of the present innovations.

Figure 50 shows DC transfer characteristics for one embodiment of the present innovations.

Figure 51 shows cutoff frequencies for one embodiment of the present innovations.

Figure 52 shows H₂₁ and MSG/MAG for one embodiment of the present innovations.

Figure 53 shows the DC output curve of one embodiment of the present innovations.

Figure 54 shows on-resistance and knee voltages fro multiple embodiments of the present innovations.

Figure 55 shows gate-drain diode I-V characteristics fro one embodiment of the present innovations.

Figure 56 shows DC and pulsed I-V characteristics for multiple embodiments of the present innovations. Figure 57 shows large-signal power characteristics without the benefit of the present innovations.

Figure 58 shows large-signal power characteristics for one embodiment of the present innovations.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

The numerous innovative teachings of the present application will be described with particular reference to the presently preferred embodiment (by way of example, and not of limitation).

The present application presents a simple way to modify the surface field distribution between the gate and drain without using a field plate electrode. The field modification is achieved by turning part or all of the region between gate and drain into a region with low density of 2DEG, effectively forming Low-Density Drain ("LDD"). With the same device dimensions, the off-state breakdown voltage V_BK improves from 60 V in an HEMT without LDD to over 90 V in a device with LDD. No degradation in f_t and slight improvement of power gain and f_max were observed in the LDD-HEMT. In addition, current collapse can be completely suppressed in the LDD-HEMTs.

Example

An AlGaN/GaN LDD-HEMT was fabricated according to the process shown in Figures 47A through 47F. The epitaxial structure and device fabrication flow of the LDD-HEMTs on sapphire substrate are similar to the process used for Figures 3, except one additional step shown in Figure 47D, that defines the LDD region at the end of the device process. After the windows of low-density drain region are defined, a CF₄ plasma treatment under an RF source power of 150W is applied for 45 seconds. The sample was then annealed at 400 ⁰C for 10 minutes. Figure 47E shows the cross-section of a finished LDD-HEMT. The gate length (L_G) at lμm and the gate-source spacing (LQS) at lμm. The gate-drain spacing (L_GD) was chosen to be either lμm or 3μm. The length of the low-density drain region (L_LDD) is 0.5μm and lμm for devices with lμm L_GD, and 0.5μm, lμm, 1.5μm, 2μm and 3μm for devices with 3μm L_GD- A schematic sketch of 2DEG density distribution in different regions of an LDD-HEMT is shown in Figure 47F. For comparison testing, conventional HEMT devices with L_LDD = 0 were also fabricated on the same wafer.

Fluorine ions incorporated in the AlGaN layer of the LDD region provide negative fixed charge which can modulate the surface electric-field and the 2DEG density, enabling the redistribution of the E-field and reduction of the peak field. The function of the LDD region is similar to a metal field plate in terms of improving the breakdown voltage, but without introducing any additional capacitances. Secondly, the fluorine ions incorporated in the AlGaN layer can effectively raise the energy band, effectively hindering the trapping and de- trapping process. The incorporation of fluorine ions in the AlGaN layer is believed to be substitutional and the fluorine atoms can fill out the nitrogen vacancies in the AlGaN layer. Thus, the current collapse associated with the surface states and traps can also be suppressed by implementing the low-density drain.

Figure 48 shows the breakdown voltage enhancement in an LDD-HEMT. By converting half of the gate-to-drain region into a low-density region, a 50% increase is realized in V_Bκ- As shown in Figure 48, the device with L_GD = lμm and L_LDD = 0.5μm exhibits a breakdown voltage similar to that achieved in a bulkier device with L_GD = 3μm and L_LDD = Oμm (without the LDD). The dependence of V_BK on L_LDD is shown in Figure 49. For devices with the same LGD=3μm and different L_LDD (0 or 3 μm), the same V_th=0.75V, IMA_X = 300mA/mm and G_M =150mS/mm are obtained. As shown in Figures 50, 51, and 52, the LDD-HEMT shows no degradations in current gain and f_t, slight improvement in power gain (MSG/MAG) and f_max — a result of increased output resistance (RDS). The only penalty imposed on the LDD-HEMT is the increased on-resistance that results in a knee voltage increase of 2 V at most as shown in Figures 53 and 54, which is much smaller than the enhancement in V_Bκ (> 30 V). Figure 55 shows that the reverse gate-drain leakage current at a bias of 20 V can be reduced from 25μA/mm to 15μA/mm in an LDD-HEMT, compared to a conventional device. Figure 56 shows the DC and pulsed I-V characteristics and shows the passivation effect of fluorine ions with different L_LDD- With L_LDD increasing, the current collapse is reduced, reaching a complete suppression when the gate-to-drain region is fully treated by the fluorine ions (L_LDD ⁼ 3μm). The comparison of large signal measurement is shown in Figure 57 and 58. Due to the current collapse, the maximum power of the conventional HEMT saturates at around 0.69 W/mm. The LDD-HEMT₅ with suppressed current collapse, exhibits a maximum power of 1.58 W/mm.

The LDD-HEMT structures and process described above can be integrated into an enhancement + depletion process which uses patterned fluorine treatment of a wide-bandgap layer, as will now be described.

Figures 3A through 3F illustrate the process of fabricating an enhancement- mode Ill-nitride HFET according to a first embodiment of the present innovations. Figure 3A illustrates a preferred epitaxial structure of the present innovations, where the reference numerals 110, 120, 130 and 140 denote substrate (e.g. sapphire, silicon or SiC), nucleation layer (low temperature grown GaN nucleation layer, AlGaN or AlN), high temperature-grown GaN buffer layer, and Al_xGa_1-xN barrier layer including the modulation doped carrier supply layer. The manufacturing method of enhancement mode Ill-nitride HFET of one embodiment is described below. The mesa isolation is formed using Cl₂/He plasma dry etching followed by the source/drain ohmic contact formation 160 with Ti, Al, Ni and Au annealed at 850° C for 45 seconds as shown in Figure 3B. Next, photoresist 170 is patterned with the gate windows exposed. Then, the fluorine ions are incorporated into Al_xGa_1-xN barrier layer by, for examples, either fluorine plasma treatment or fluorine ions implantation as shown in Figure 3C The gate electrode 180 is formed on the barrier layer 140 by depositing and lift-off Ni and Au as shown in Figure 3D. Thereafter, post-gate RTA is conducted at 400-450⁰C for 10 minutes. A passivation layer 190 is grown on the top of the wafer as shown in Figure 3E. Finally, the contact pads are opened by removing portions of the passivation layer on the contact pads as shown in Figure 3F.

Example

An AlGaN/GaN HEMT structure was grown on a (0001) sapphire substrate in an Aixtron AIX 2000 HT metal-organic chemical vapor deposition (MOCVD) system. The HEMT structure consists of a low-temperature GaN nucleation layer, a 2.5- m-thick unintentionally doped GaN buffer layer and an AlGaN barrier layer with nominal 30% Al composition. The barrier layer consists of a 3-nm undoped spacer, a 15-nm carrier supplier layer doped at 2.5 x 10¹⁸ cm^"3, and a 2-nm undoped cap layer. Room temperature Hall measurements of the structure yield an electron sheet density of 1.3xlO¹³ cm^"2 and an electron mobility of 1000 cmVVs. The device mesa was formed using Cl₂/He plasma dry etching in an STS ICP-RIE system followed by the source/drain ohmic contact formation with Ti/Al/Ni/Au annealed at 85O⁰C for 45 seconds. The ohmic contact resistance was typically measured to be 0.8 ohm-mm.

After gate windows with 1 nm length were opened by contact photolithography, the sample was treated by CF₄ plasma in an RIE system at an RF plasma power of 150 W for 150 seconds. Pressure of the treatment is typically 50 mTorr. The typical depth distribution profile of the fluorine ions thus incorporated via the treatment is Gaussian, and the typical depth when the fluorine concentration drops from the peak by one order of magnitude is 20 nm. Note that ion implantation is another method for incorporating the fluorine ions, and it is estimated that an energy of about 10 KeV would be required.

Ni/Au electron-beam evaporation and liftoff were carried out subsequently to form the gate electrodes. The plasma treated gate region and the gate electrode were self-aligned. Post-gate RTA was conducted at 400 ⁰C for 10 minutes. This RTA temperature was chosen because RTA at temperatures higher than 500⁰C can degrade both the gate Schottky contact and the source/drain ohmic contacts. The devices have a source-gate spacing of L_sg=l μm and a gate-drain spacing of L_gd =2μm. D-mode HEMTs were also fabricated on the same sample without plasma treatment to the gate regions.

Figure 2 shows the transfer characteristics of both D-mode and E-mode (before and after post-gate annealing) AlGaN/GaN HEMTs. Defining Va₁ as the gate bias intercept of the linear extrapolation of drain current at the point of peak transconductance (g-_m), the Va₁ of the E-mode device was determined to be 0.9 V, while the Va₁ of the D-mode device is -4.0 V. More than 4 V of Va₁ shift was achieved by the plasma treatment. At V_gs=0, the transconductance reaches zero, indicating a true E-mode operation. The drain current is well pinched-off and shows a leakage of 28 μA/mm at Vd_s = 6 V, the smallest value reported up to date for E-mode AlGaN/GaN HEMTs. The peak g_m is 151 mS/mm for the D-mode HEMT and 148 mS/mm for the E-mode HEMT, respectively. The maximum drain current (I_max) reaches 313 mA/mm at the gate bias (V_gs) of 3 V for the E-mode HEMT. Comparison of the current- voltage (I-V) characteristics of E-mode device before and after RTA suggests that RTA at 400⁰C for 10 minutes plays an important role in recovering the damages induced during the plasma treatment and achieving high current density and transconductance. Figure 4 A shows the output curves of the E-mode device before and after the RTA process. No change in threshold voltage was observed after the RTA. At a V_gs of 2.5 V, the saturation drain current (247 mA/mm) of E-mode device after RTA at 400 ⁰C is 85% higher than that (133 mA/mm) before RTA, and the knee voltage of the E-mode device with RTA is 2.2 V, where the drain current is 95% saturation drain current. The off-state drain breakdown voltage at V_gs = OV is larger than 80 V₅ showing no degradation compared to that observed in the D-mode HEMTs. Figure 4B shows Ig/Vg_S curves of these three devices. Lower gate leakage currents were achieved for E-mode HEMT, especially after RTA.

In order to investigate the mechanisms of the V_th shift by CF₄ plasma treatment, secondary ion mass spectrum (SIMS) measurements were carried out on accompanying samples to monitor the atomic composition changes of the CF₄ plasma treated AlGaN/GaN materials. In addition to Al, Ga, and N, significant amount of fluorine atoms were detected in the plasma treated sample. Figure 5 shows the fluorine atom concentration profile of the sample treated at a CF₄ plasma power of 150W for 2.5 minutes. The concentration of fluorine atoms is the highest near the AlGaN surface and drops by one order of magnitude in the channel. It can be deduced that the fluorine ions produced by the CF₄ plasma were incorporated into the sample surface, similar to the effects of plasma immersion ion implantation ("PIII"), a technique developed to realize ultra-shallow junctions in advanced silicon technology. Because of the strong electro-negativity of the fluorine ions, the incorporated fluorine ions can provide immobile negative charges in the AlGaN barrier and effectively deplete the electrons in the channel. With enough fluorine ions incorporated in the AlGaN barrier, the D-mode HEMT can be converted to an E-mode HEMT. The CF₄ plasma treatment can result in a threshold voltage shift as large as 4.9 V. After RTA at 400⁰C for 10 minutes, the peak fluorine atom concentration near the AlGaN surface is unchanged while that around the AlGaN/GaN interface experiences more significant reduction. It should be noted, however, SIMS measurement results from different runs do not offer accurate quantitative comparison because of the lack of reference criterion. Nevertheless, the minute change in Y^ before and after RTA indicates that the total number of fluorine ions incorporated into the AlGaN barrier is near constant before and after RTA, while the plasma damages are significantly recovered by the RTA. The lower gate reverse leakage currents of an E-mode HEMT can be attributed to an upward band bending of the AlGaN layer as a result of fluorine ion incorporation. After the RTA process, the defects at the interface of metal and AlGaN induced by CF₄ were recovered, leading to further suppression of gate leakage current. From the atomic force microscopy ("AFM") measurement conducted on a patterned sample, it was observed that the plasma treatment only results in a 0.8 nm reduction in the overall AlGaN barrier layer (20 nm thick).

On-wafer small-signal RF characteristics of D-mode and E-mode AlGaN/GaN HEMTs were measured from 0.1 to 39.1 GHz. The current gain and maximum stable gain/maximum available gain (MSG/MAG) of both types of devices with 1 μm-long gate were derived from measured S-parameters as a function of frequency, as shown in Figure 5. At V_ds=12 and V_gs=1.9 V, a current gain cutoff frequency (fr) of 10.1 GHz and a power gain cutoff frequency (/_MAX) of 34.3 GHz were obtained for the E-mode AlGaN/GaN HEMT, a little lower than that of its D-mode counterpart, whose and were measured at the drain bias of 12 V and gate bias of- 3 V to be 13.1 and 37.1 GHz, respectively.

One advantage of the present innovations is that the E-mode HFET with fluorine ions incorporated in barrier layer can stand a larger gate bias (> 3V) corresponding to a larger input voltage swing.

Also, thermal reliability testing has shown that the fluorine ion incorporation in the AlGaN barrier is stable up to 700⁰C. However, the Schottky contact, made of nickel, is only stable up to 5OO⁰C. Therefore, the application temperature range is up to 500⁰C unless another Schottky contact technique is used. Tungsten gate is one possible candidate.

In Figure 7, the effect of different post-gate RTA' s on the fluorine atoms' distributions in AlGaN/GaN heterostructures, as measured by SIMS, is shown. The untreated device is used as a reference.

It was found that the fluorine ions, which were incorporated into the AlGaN barrier layer by CF₄ plasma treatment, could effectively shift the threshold voltage positively. The fluorine ions' incorporation in the AlGaN layer was confirmed by secondary-ion-mass-spectrum (SIMS) measurements, as shown in Figure 7. During CF₄ plasma treatment, fluorine ions are implanted into AlGaN/GaN heterostructure in a self-built electrical field stimulated by the RF power.

It is also concluded from the results shown in Figure 7 that the implanted fluorine ions have a good thermal stability in the AlGaN layer up to 700⁰C. It should be noted that, although the presence of the fluorine ions are confirmed as the cause of the threshold-voltage shift, it is not clear what sites, either interstitial or substitutional, the fluorine ions occupy. Deep-level transient spectroscopy ("DLTS") has been conducted on the HEMT samples treated by CF₄ plasma. The fluorine ions incorporated in the AlGaN barrier appear to introduce a deep-level state that is at least 1.8 eV below the conduction-band mhiimum. As a result, the fluorine ions are believed to introduce a negatively charged acceptor-like deep level in the AlGaN.

Note that in SIMS plots such as Figure 7, it is difficult to make accurate calculation of concentration from SIMS measurement because the beam size is not known. However, based on the bandstructure and threshold voltage calculation, the peak value of the F concentration can be as high as about 1 x 20 cm^"3. In Figure 7 A, the effect of different plasma power levels without RTA on the fluorine atoms' distributions in AlGaN/GaN heterostructures, as measured by SIMS, is shown.

Note that the 200W and 400W lines show a "bump" at the interface between the AlGaN/GaN interface. During an incorporation process, the fluorine ions can fill up surface or interface states (or "traps"), producing "anomalous stopping". Therefore, this indicates there are more traps at the interface. Further, the 600W and 800W lines do not show the bump most likely because of the greater penetration depth and overall concentration.

The untreated device is used as a reference. In Figure 7B, the effect of different post-gate treatment temperatures at a fixed power of 600 W for RTA on the fluorine atoms' distributions in AlGaN/GaN heterostructures, as measured by SIMS, is shown. The untreated device is used as a reference. Note that the distributions in the AlGaN for 700⁰C and below show a normal effect of root Dt, but the distribution in the AlGaN layer seems to reflect a very different diffusivity (or perhaps some other activation energy effect). Thus, the data indicates that fluorine ions are more stable in AlGaN than in GaN. Further, the binding energy can be higher, and the fluorine-related energy states are deeper below the conduction band in AlGaN than in GaN.

Sensitivity to plasma treatment parameters was also investigated. Devices were fabricated with different Vu₁ values by applying different CF₄ plasma power and treatment times. Five different combinations were used: 100 W for 60 seconds, 150 W for 20 seconds, 150 W for 60 seconds, 150 W for 150 seconds, and 200W for 60 seconds. For comparison, an HEMT without CF₄ treatment was also fabricated on the same sample and in the same processing run. All the devices were unpassivated in order to avoid any confusion caused by the passivation layer, which may change the stress in the AlGaN layer and alter the piezoelectric polarization. All the HEMT devices have a gate length of 1 μm, a source-gate spacing of L_sg = 1 μm and a gate-drain spacing of L_gd = 2 μm. DC current-voltage (I-V) characteristics of the fabricated devices were measured using an HP4156A parameter analyzer. Transfer characteristics and transconductance (g_m) characteristics are shown in Figures 8A and 8B, respectively. Taking the conventional HEMT (i.e., without CF₄ plasma treatment) as the baseline devices, the threshold voltage of all the other CF₄ plasma-treated HEMTs are shifted to the positive direction. Defining V₁₁₁ as the gate-bias intercept of the linear extrapolation of the drain-current at the point of peak transconductance (g_m), the Vu₁ of all the devices were extracted and listed in Figure 9. For the conventional HEMT, Vu₁ is — 4 V. For the HEMT treated by CF₄ plasma at 150 W for 150 seconds, the V_Λ is 0.9 V, which corresponds to the E-mode HEMT. A maximum Y_th shift of 4.9 V was achieved. In order to further reveal the effects of CF₄ plasma treatment, the dependencies of Vu₁ on both CF₄ plasma treatment time and RF power are plotted in Figure 10. As the plasma power is increased and as longer treatment time are utilized, larger shifts in Vu₁ are effected. With the increase in the plasma treatment time, more fluorine ions were implanted into AlGaN layer. The increased fluorine ion concentration leads to a reduced electron density in the channel, and causes the positive shift of Vu₁. When the plasma power increases, fluorine ions obtain a higher energy and fluorine ion flux increases due to the enhanced ionization rate of CF₄. With higher energy, fluorine ions can reach at a deeper depth closer to the channel. The closer the fluorine ions are to the channel, the more effective they at depleting 2DEG, and a larger shift in Vu₁ is achieved. The increased fluorine ions flux has the same effect on Vu₁ as the increase of the plasma treatment time by raising the fluorine atoms concentration in AlGaN layer. It should be noted that the nearly linear Vu₁ versus time and Vu₁ versus power relationships imply the possibilities of a precise control OfVu₁ of AlGaN/GaN HEMTs. Although the Vu₁ is shifted by CF₄ plasma treatment, the g_m is not degraded. As shown in Figure 8B, all the devices' maximum g_m are in the range of 149-166 mS/mm, except for that treated at 150W for 60 seconds, which has a higher peak g_m of 186 mS/mm. It is suspected that this singularity point was caused by the non-uniformity in epitaxial growth. Confirmed by an AFM measurement conducted on a CF₄-treated patterned sample (with part of the sample treated and other parts protected from the plasma treatment), the CF₄ plasma treatment only results in an AlGaN-thickness reduction of less than 1 nm, as shown in Figure 11. Thus, the almost constant transconductance indicates that the 2DEG mobility in the channel is maintained in the device fabrication according to the present innovations. A key step in maintaining the transconductance is the post-gate annealing process.

Recovery of Plasma-Induced Damages by Post-Gate Annealing

As previously discussed, the plasma normally induces damages and creates defects in semiconductor materials, and consequently degrades carriers' mobility. RTA is an effective method to repair these damages and recover the mobility. In the CF₄ plasma-treated AlGaN/GaN HEMTs, drain-current and transconductance degradation occurs just after the plasma treatment. In Figures 12A and 12B, the drain-current and transconductance measured on an untreated device and a treated device (200 W, 60 seconds) before and after RTA (400⁰C for 10 minutes) are plotted. Figure 13 compares the output characteristics of the treated device before and after RTA. The drain-current was 76% and the transconductance was 51% higher after the RTA in the treated device. The RTA process can recover majority of the mobility degradation in the plasma-treated device, while showing an insignificant effect on the conventional untreated device. Therefore, the recovery of Id and g_m in the CF₄ plasma-treated device is the result of the effective recovery of the 2DEG mobility at this RTA condition. Compared to a higher annealing temperature of 700⁰C, which is needed to recover damages induced by chlorine- based ICP-RIE in the case of recessed gate, this lower RTA temperature implies that the CF₄ plasma treatment creates lower damages than the chlorine-based ICP- RIE. It also enables the RTA process to be carried out after the gate deposition, fulfilling the goal of a self-aligned process. If the previous definition of V_th is used, the Vt_h of the CF₄ plasma-treated device seems to be shifted from 0.03 to - 0.29 V after the RTA. When the start point of g_m, as shown in Figure 12B, or the start point of I_d at the logarithm scale, as shown in the inset of Figure 12A, is used as the criteria to evaluate V_th, the Vth of the CF₄ plasma-treated device is not changed after the RTA. The good thermal stability of V^ is consistent with the previously mentioned good thermal stability of fluorine atoms in AlGaN layer.

Suppression ofSchottkv Gate-Leakage Current

AlGaN/GaN HEMTs always show much higher reverse gate leakage currents than the values theoretically predicted by the thermionic-emission ("TE") model. The higher gate currents degrade the device's noise performance and raise the standby power consumption. In particular, forward gate currents limit the gate input voltage swing, hence the maximum drain-current. Other approaches have been attempted to suppress gate currents of AlGaN/GaN HEMTs. These efforts include using the gate metal with higher work function, using copper, modifying the HEMTs structure (such as adding a GaN cap), or diversion to metal-insulator- semiconductor heterostructure field-effect transistors (MISHFETs). In the CF₄ plasma-treated AlGaN/GaN HEMTs of the present innovations, suppressions of gate currents in both reverse and forward bias regions can be achieved. Gate- current suppressions show dependencies on CF₄ plasma-treatment conditions.

Figures 14A and 14B shows gate currents of AlGaN/GaN HEMTs with different CF₄ plasma treatments. Figure 14B is the enlarged plot of the forward gate bias region. In reverse bias region, compared to the conventional HEMT without CF₄ plasma treatment, the gate-leakage currents of all the CF₄ plasma- treated AlGaN/GaN HEMTs decreased. At V_g = -20 V, the gate-leakage current drops by more than four orders of magnitude from 1.2 x 10^~2 A/mm for conventional HEMT to 7 x 10^~7 A/mm for the AlGaN/GaN HEMT plasma treated at 200 W, 60 seconds. In the forward region, the gate currents of all the CF₄ plasma-treated AlGaN/GaN HEMTs also decrease. As a result, the turn-on voltages of the gate Schottky diode are extended, and the gate input voltage swings are increased. Using 1 mA/mm as the criterion, the turn-on voltage of the gate Schottky diode increases from 1 V for conventional HEMT to 1.75 V for the CF₄ plasma-treated AlGaN/GaN HEMT at 200 W for 60 seconds.

The suppression of the gate-leakage current in the CF₄ plasma-treated AlGaN/GaN HEMT can be explained as follows. During CF₄ plasma treatment, fluorine ions are incorporated into the AlGaN layer. These ions with a strong electronegativity act as immobile negative charges that cause the upward conduction-band bending in the AlGaN barrier layer due to the electrostatic induction effect. Thus, an additional barrier height Φ_F , as shown in Figure 23 is formed, and the effective metal-semiconductor barrier height is increased from Φ_B to Φ_B +Φ_F • This enhanced barrier height can effectively suppress the gate Schottky diode current in both reverse and forward bias regions. With higher plasma power and longer treatment time, the fluorine ion concentration in the AlGaN layer increases, and the effective barrier height is raised further, leading to a more significant gate-current suppression. In Figure 9 the effective barrier heights and ideality factors that were extracted from the forward region of the measured gate currents by using the TE model are detailed. The effective barrier height of conventional HEMT is 0.4 eV, while the effective barrier height increases to 0.9 eV for the CF₄ plasma-treated HEMT at 200 W for 60 seconds. The effective barrier heights of the CF₄ plasma-treated HEMT also show a trend of increase with the plasma power and treatment time, except for the HEMT treated at 150 W for 20 seconds, which has a relatively higher effective barrier height. This exception is thought to be due to the process variations. The fact that the extracted effective barrier height is much lower than the theoretically predicted values and very large ideality factors (> 2.4) indicates that the gate currents of fabricated AlGaN/GaN HEMTs are not dominated by the TE mechanism but other mechanisms, such as vertical tunneling, surface barrier thinning, and trap-assisted tunneling. Thus, the barrier heights and ideality factors, which are extracted by using the TE model, are not accurate. Nevertheless, they provide sufficient qualitative information for explaining the mechanism of the gate-current suppression in CF₄ plasma-treated AlGaN/GaN HEMTs.

Dynamic I-V characterizations were conducted by using an Accent DIVA D265 system to investigate the effects of CF₄ plasma treatment on drain-current dispersion. The pulse width is 0.2 μs and the pulse separation is 1 ms. The quiescent point is at VGS slightly (~ 0.5 V) below the pinch-off and V_Ds = 15 V. Compared to static I-V characteristics, the maximum drain-current of conventional D-mode HEMT dropped by 63%, while that of E-mode HEMT with CF₄ plasma treatment at 150 W for 150 seconds dropped by 6%.

The alleviation of drain-current drops for E-mode HEMT is likely due to a raised gate bias of the quiescent point (V_Gs = 0 V for E-mode HEMT, V_Gs ⁼ "4.5 V for D-mode HEMT).

RF Small-Sisnal Characteristics

On-wafer small-signal RF characterization of the fabricated AlGaN/GaN HEMTs were carried out at the frequency range of 0.1-39.1 GHz using Cascade microwave probes and an Agilent 8722ES network analyzer. Open-pad de- embeddings with the S-parameters of dummy pads were carried out to eliminate a parasitic capacitance of the probing pads. The current gain and maximum stable gain/maximum available gain (MSG/MAG) of all devices with 1-μtn long gate were derived from the de-embedded S-parameters as a function of frequency. The current cutoff frequency (J_x ) and maximum oscillation frequency (/_max) were extracted from current gains and MSG/MAGs at unit gain. It has been observed that the intrinsic f_t and f_max are generally 10-15% higher than the extrinsic ones without the de-embeddings process. The dependencies of/ and f_max, on the gate bias are shown in Figure 15 for the E-mode HEMT. Both/ and f_mBX are relatively constant at both low and high gate bias, indicating a good linearity. Figure 16 lists f_t and f_maxof all samples. For the conventional HEMT, f_t and f_mωs_ are 13.1 and 37.1 GHz, while for the CF₄ plasma-treated HEMTs, f_t and f_me^ are approximately 10 and 34 GHz, slightly lower than that of the conventional HEMT, except for the HEMT treated at 150W for 60 seconds. This higher /_t and /_max in the 150W/60 second device are consistent with the higher g_m presented before, and are attributed to a material non-uniformity and process variation. The slightly lower/ and f_meΑ in the CF₄ plasma-treated HEMTs indicate that the post-gate RTA at 400⁰C can effectively recover the 2DEG mobility degraded by the plasma treatment, but the recovery is less than 100%. It suggests that the optimization of the RTA temperature and time is needed to further improve the 2DEG mobility, while not degrading the gate Schottky contact.

MISHFETs

In another embodiment, E-mode Si₃N₄/AlGaN/GaN MISHFET were constructed with a two-step Si₃N₄ process which features a thin layer of Si₃N₄ (15 nm) under the gate and a thick layer of Si₃N₄ (about 125 nm) in the access region. Fluorine-based plasma treatment was used to convert the device from D-mode to E-mode. The E-mode MISHFETs with 1-μm long gate footprint exhibited a threshold voltage of 2 V, a forward turn-on gate bias of 6.8 V (compared to about a 3 V realized in E-mode AlGaN/GaN HEMTs) and a maximum current density of 420 niA/mm.

The AlGaN/GaN HFET structure used in this example was grown on (0001) sapphire substrates in an Aixtron AIX 2000 HT MOCVD system. The HFET structure consists of a 50-nm thick low temperature GaN nucleation layer, a 2.5- μm thick not-intentionally doped GaN buffer layer, and an AlGaN barrier layer with nominal 30% Al composition. The barrier layer is composed of a 3-nm undoped spacer, a 16-nm carrier supplier layer doped at 2 X 10¹⁸ cm^"3, and a 2-nm undoped cap layer. The capacitance-voltage ("C-V") measurement by mercury probe yields an initial threshold voltage of -4 V for this sample. The process flow is illustrated in Figures 17A through 17F. The device mesa is formed using CVHe plasma dry etching in an STS ICP-RJE system followed by the source/drain ohmic contact formation with Ti/Al/Ni/Au (20 nm/150 nm/50 nm/80 nm) annealed at 85O⁰C for 30 seconds, as shown in Figure 17A. Then, the first Si₃N₄ layer (about 125 nm) is deposited on the sample by plasma enhanced chemical vapor deposition (PECVD) as in Figure 17B. After gate windows with 1-μm length are opened by photolithography, the sample was put in an RIE system under CF₄ plasma treatment, which removed the Si₃N₄ and incorporated fluorine ions in the AlGaN. The RF power of the plasma was 150 W, as shown in Figure 17C. The gas flow was controlled to be 150 seem, and the total etching and treatment time is 190 seconds. After removing the photoresist, the second Si₃N₄ film (about 15 nm) was deposited by PECVD to form the insulating layer between gate metal and AlGaN as in Figure 17D. Subsequently, the Si₃N₄ layer was patterned and etched to open windows in the source and drain ohmic contact regions, as shown in Figure 17E. Next, the 2-μm long gate electrodes were defined by photolithography followed by e-beam evaporation of Ni/ Au (-50 nm/300 nm) and liftoff as in Figure 17F. To ensure that the gate electrode covers the entire plasma-treated region, the metal gate length (2 μm) was chosen to be larger than the treated gate area (1 μm), leading to a T-gate configuration. The gate overhang in the source/drain access regions is insulated from the AlGaN layer by the thick Si₃N₄ layer, keeping the gate capacitances at low level. Finally, the whole sample was annealed at 400⁰C for 10 minutes to repair the plasma-induced damage in the AlGaN barrier and channel. Measured from the foot of gate, the gate-source and gate-drain spacings are both 1.5 μm. The E-mode MISHFETs are designed with gate width of 10 μm for dc testing and 100 μm for RF characterizations.

The constructed device was then characterized. The DC output characteristics of the E-mode MISHFETs are plotted in Figure 18. The devices exhibit a peak current density of about 420 mA/mm, an ON-resistance of about 5.67 Ω-mm and a knee voltage of about 3.3 V at V_GS = 7 V. Figure 19A shows the transfer characteristics of the same device with 1 x 10-μm gate dimension. It can be seen that the V_& is about 2 V, indicating a 6-V shift of Vn₁ (compared to a conventional D-mode HFET) achieved by the insertion of the Si₃N₄ insulator and plasma treatment. The peak transconductance gm is about 125 mS/mm. Figure 19B shows the gate leakage current at both the negative bias and forward bias. The forward bias turn-on voltage for the gate is about 6.8 V, providing a much larger gate bias swing compared to the E-mode HFETs. Pulse measurements were taken on the E-mode MISHFETs with 1 x 100-μm gate dimensions with a pulse length of 0.2 μs and a pulse separation of 1 ms. The quiescent bias point is chosen at V_GS = 0 V (below V_th) and V_Ds = 20 V. Figure 20 shows that the pulsed peak current is higher than the static one, indicating no current collapse in the device. The static maximum current density of the large device with a 100-μm gate width is about 330 mA/mm, smaller than the device with 10-μm gate width (about 420 mA/mm). The lower peak current density in the larger device is due to the self- heating effect that lowers the current density. Since little self-heating occurs during pulse measurements, the maximum current for the 100-μm wide device can reach the same level as the 10-μm wide device. On wafer small-signal RF characteristics were performed from 0.1 to 39.1 GHz on the 100-μm wide E-mode MISHFETs at V_DS = 10 V. As shown in Figure 21, the maximum current gain cutoff frequency (f_τ ) and power gain cutoff frequency (f_max) are 13.3 and 23.3 GHz, respectively. When the gate bias is 7 V, the small-signal RF performance does not significantly degrade, with an f_τ of 13.1 GHz and an f_max of 20.7 GHz, indicating that the Si₃N₄ insulator offers an excellent insulation between gate metal and semiconductor.

Models

A theoretical characterization model was developed for some of the present innovations. For a conventional AlGaN/GaN HEMT with silicon modulation doped layer, as shown in Figure 7, the polarization charges need to be taken into account in the calculation of HEMTs threshold voltage. Modified from a generally used formula by taking into account the effects of charge polarization, surface and buffer traps, the threshold voltage of the AlGaN/GaN HEMT can be expressed as:

J-U — ΦB/* ^■- dσ/ε ^■■■■ ΔϋV/e H- E&/&

Where the parameters are defined as follows:

φ_B is the metal-semiconductor Schottky barrier height. σ is the overall net (both spontaneous and piezoelectric) polarization charge at the barrier — AlGaN/GaN interface.

d is the AlGaN barrier-layer thickness.

N_sl(x) is the silicon-doping concentration.

ΔE_cis the conduction-band offset at the AlGaN/GaN heterostructure.

Ep is the difference between the intrinsic Fermi level and the conduction band edge of the GaN channel.

s is the dielectric constant of AlGaN.

iVa is the net-charged surface traps per unit area.

Nt, is the effective net-charged buffer traps per unit area.

C₆ is the effective buffer-to-channel capacitance per unit area.

The last two terms in equation (1) describe the effects of the surface traps and buffer traps, respectively. The AlGaN surface is at x = 0, and the direction pointing to the channel is the positive direction for the integration. To represent the devices described above, immobile negative charges are introduced into the AlGaN barrier layer under the gate. Because of electrostatic induction, these immobile negative charges can deplete 2DEG in the channel, raise the energy band, and hence modulate V&. Including the effect of the negative charges confined in the AlGaN barrier, the modified threshold voltage from equation (1) is given by:

{NJr)~-N_F{x}) dx-~caNy?-~eN_h{C_b, Q)

The positive-charge distribution profile N/pc) is replaced by the net charge distribution N_si(x) — N_F (x), where N^ (x) is the concentration of the negatively charged fluorine ion. The surface-trap density (ΛQ could be modified by the plasma treatment.

By applying Poisson's equation and Fermi-Dirac statistics, a simulation was made of the conduction-band profiles and the electron distributions of AlGaΝ/GaΝ HEMT structures with and without fluorine ions incorporated in AlGaN layer. Both structures have the same epitaxial structure, shown in Figure 7. For the fluorine ions incorporated HEMT structure, the negatively charged fluorine ions' profile was extracted from SIMS measurement results of the fluorine atoms' distribution of an AlGaN/GaN HEMT structure that was treated by CF₄plasma at 150 W for 150 s and converted to an E-mode HEMT. The simulated conduction band diagrams at zero gate bias were plotted in Figures 22 and 23 For the simulated conduction band of E-mode HEMT, as shown in Figure 22 the fluorine concentration is approximated by using a linear distribution that the peak fluorine concentration is 3xl0^iScm^~3 at the AlGaN surface, and the fluorine concentration is assumed to be negligible at the AlGaN/GaN interface. A total fluorine ion sheet concentration of about 3x10¹³ cm^"2 is sufficient to not only compensate the silicon doping (about 3.7xlO¹³ cm^"^ in the AlGaN barrier but also to compensate for the piezoelectric and spontaneous polarization-induced charges (about IxIO¹³ cm^"2). Two significant features can be observed. First, compared to the untreated AlGaN/GaN HEMT structure, the plasma-treated structure has its 2DEG channel's conduction-band minimum above Fermi level, indicating a completely depleted channel and E-mode HEMT. As shown in the electron profiles in Figure 24, there are no electrons in the channel under the zero gate bias in the plasma-treated structure, indicating an E-mode HEMT operation. Second, the immobile negatively charged fluorine ions cause an upward bending of the conduction band, especially in AlGaN barrier, yielding an additional barrier height ΦF, as shown in Figure 23 Such an enhanced barrier can significantly suppress the gate Schottky diode current of AlGaN/GaN HEMT in both the reverse and forward bias regions.

The epitaxial structure for the monolithically-integrated E/D-mode HFET consists of: (a) a semiconductor substrate (sapphire, SiC, silicon, AlN or GaN, etc.); (b) a buffer layer grown on the substrate; (c) a channel layer; (d) a barrier layer including an undoped spacer layer, a modulation doped carrier supply layer and an undoped cap layer. The fabrication process includes: (f) active region isolation; (g) ohmic contacts formation on source and drain terminals; (h) photolithography of the gate regions for the E-mode HFETs; (i) fluoride-based plasma treatment to the exposed barrier layer of the E-mode HFETs; (j) gate metal deposition of the E- mode HFETs; (k) photolithography of the gate regions for the D-mode HFETs; (1) gate metal deposition of the D-mode HFETs; ml) surface passivation of the D- mode and E-mode HFETs; (n) gate annealing at elevated temperatures. A schematic process flow for this monolithic integration is depicted in Figure 25.

The active device isolation in the above-described monolithic integration process uses the mesa etching, which features the active region removal by etching techniques in the areas without the HFETs. Such an approach imposes limits to the integration density, photolithography resolution. For high frequency circuits, the edges of the mesas also introduce additional discontinuities for wave propagation, which in turn, complicate the circuit design and analysis. Since the fluoride-based plasma treatment is able to deplete the electrons in the channel (providing electrical turn-off of the channel), it can be used for device isolation. With increased plasma power and treatment time, the regions where active devices are not desired can be completely turned off electrically, providing electrical isolation between devices. Such an approach does not involve any material removal, therefore, enables a flat wafer surface for planar process. Example

Figures 26 A through 26F illustrate the process of monolithically integrating the E/D-mode HFETs for integrated circuits according to one embodiment of the present invention. Figure 26A illustrates a preferred epitaxial structure of this invention, where the reference numerals 110, 120, 130 and 140 denote substrate, low temperature grown GaN nucleation layer, high temperature grown GaN buffer layer, and Al_xGa_1-xN barrier layer including the modulation doped carrier supply layer. The manufacturing method of monolithic integration of E/D-mode HFETs for integrated circuits is described below. For both D-mode and E-mode HFETs, the mesa isolations are simultaneously formed using C12/He plasma dry etching followed by the source/drain ohmic contact formation 160 with Ti, Al, Ni and Au annealed at 850⁰C for 45 seconds as shown in Figure 26B. The gates as well as gate-source interconnections of D-mode HFETs are patterned by photoresist 170 as shown in Figure 26C, followed by depositing and lift-off Ni and Au 178. Thereafter, the E-mode HFETs' gates, pads and second interconnections are patterned with photoresist 175 as shown in Figure 26D. Then the fluoride ions are incorporated into Al_xGa_1-3JNf barrier layer beneath E-mode HFETs' gates by, for examples, either fluoride plasma treatment or fluoride ions implantation as shown in Figure 26D. The gate electrode 180 is formed on the barrier layer 140 by depositing and lift-off Ni and Au. Thereafter, post-gate rapid thermal annealing (RTA) is conducted at 400-450⁰C for 10 minutes. A passivation layer 190 is grown on the top of the wafer as shown in Figure 26E. Then the contact pads and via holes are opened by removing portions of the passivation layer on them as shown in Figure 26F. Finally, a third interconnection is formed.

An E/D HFETs inverter and a 17-stage direct-coupled ring oscillator were created on a 20nm Alo.25Gao.75N barrier layer on 2μm GaN buffer layer with typical CF₄ plasma treatment condition of 150W for 150 seconds and typical post-gate RTA condition of 450⁰C for 10 minutes for E-mode HFETs. The inverter has a NM_L of 0.21 V and a NM_n of 0.51V at a supply voltage of 1.5 V. When supply voltage of 3.5 V is applied, the 17-stage ring oscillator shows a maximum oscillation frequency of 225 MHz corresponding to a minimum propagation delay of l30 ps.

Example

This embodiment describes a method for planar monolithic integration of E- mode and D-mode AlGaN/GaN HFETs. As described in the first embodiment, the isolation among active devices can be obtained by creating active device mesa through etching, which creates non-flat wafer surfaces. In integrated circuit fabrications, planar process are always desirable. Following the same principle of channel depletion by the negatively charged fluorine ions in the AlGaN, depletion of the desired inactive (isolated) areas by fluoride-based plasma treatment can be effected. The plasma power and treatment time can both be increased to enhance the carrier depletion. The process flow is illustrated in Figure 27, where: (a) source/drain ohmic contacts formation; (b) D-mode HFET gate definition by photolithography; (c) D-mode HFET gate metallization and part of the interconnects formation; (d) E-mode HFET gate definition by photolithography followed by plasma treatment; (e) E-mode HFET gate metallization and part of the interconnects formation; (f) isolation region definition by photolithography followed by the second fluoride-based plasma treatment; (g) final chip followed by passivation.

Example

The AlGaN/GaN HEMT structure in this example was grown on a (0001) sapphire substrates in an Aixtron AIX 2000 HT MOCVD system. The HEMT structure consists of a low-temperature GaN nucleation layer, a 2.5-μm thick unintentionally doped GaN buffer layer, and an AlGaN barrier layer with a nominal 30% Al composition. The barrier layer is composed of a 3-nm undoped spacer, a 21-nm carrier supplier layer doped at 2 x 1018 cm^"3, and a 2-nm undoped cap layer. Room-temperature hall measurements of the structure yield an electron sheet density of 1.3 x 1013 cm^""2 and an electron mobility of 950 cmVVs.

The integration process flow is illustrated in Figure 28. First, the source/drain ohmic contacts of the E/D-mode devices were formed simultaneously by a deposition of e-beam evaporated Ti/Al/Ni/Au (20 nm/150 nm/50 nm/80 nm) and rapid thermal annealing at 850 °C for 30 seconds, as shown in Figure 28(a). Second, the active regions for both E/D-mode devices were patterned by a photolithography, which is followed by the CF₄ plasma treatment in a reactive ion etching system. The plasma power was 300 W, and the treatment time was 100 seconds. The gas flow was controlled to be 150 seem, and the plasma bias was set to be 0 V. The isolation regions are the locations where a large amount of fluorine ions are incorporated in the AlGaN and GaN layers near the surface, and then deplete the two-dimensional electron gas in the channel, as shown in Figure 28(b). The D-mode HEMTs' gate electrodes were then patterned by the contact photolithography, which is followed by the e-beam evaporation of Ni/Au (50 nm/300 nm) and liftoff as shown in Figure 28(c). Next, E-mode HEMTs' gate electrodes and interconnections were defined. Prior to the e-beam evaporation of Ni/Au, the gate regions of the E-mode HEMTs were treated by the CF₄ plasma (which has a negligible etching to AlGaN) at 170 W for 150 seconds, as shown in Figure 28(d). This plasma treatment performed the function of converting the treated devices from the D-mode to E-mode HEMT. A 200-nm-thick silicon nitride passivation layer was deposited by PECVD, and the probing pads were opened. Then, the sample was annealed at 400 °C for 10 minutes to repair the plasma- induced damage in the AlGaN barrier and channel of the E-mode HEMTs as in Figure 28(e). As a comparison, the D-mode devices were fabricated on another piece of the sample from the same substrate by the standard process, in which inductively coupled plasma reactive ion etching was used to define the mesa as the active region. For the direct-coupled FET logic inverter shown in Figure IA, the E-mode HEMT driver is designed with a gate length, gate-source spacing, gate- drain spacing, and gate width of 1.5, 1.5, 1.5, and 50 μm, respectively; the D-mode HEMT load is designed with a gate length, gate-source spacing, gate-drain spacing, and gate width of 4, 3, 3, and 8 μm, yielding a ratio β = (W_E/L_E)/(W_D/L_D) of 16.7. Discrete E-mode and D-mode HEMTs with 1.5 x 100 μm gate dimension are fabricated for characterizations.

Device and Circuit Characteristics

For the E/D-mode HEMTs fabricated by the planar process, the output characteristics are plotted in Figure 29. The peak current density for D-mode and E-mode HEMTs are about 730 and 190 niA/mm. Figure 30 shows the DC transfer characteristics comparison between the planar and the standard process. It can be seen that the drain leakage current for the planar process is about 0.3 mA/mm, reaching the same level as the devices fabricated by the standard mesa etching. The D-mode HEMTs by the planar process have the comparable drain-current and transconductance characteristics as shown in Figure 30(b), as the ones by the standard process. Also, the leakage current between two pads (400 x 100 μm ) was measured with a spacing of 150 μm. At the DC bias of 10 V, the leakage current by the planar process is about 38 μA, at the same level of the standard mesa etching sample (about 30 μA). Compared with the standard mesa process, the fluoride-based plasma treatment can achieve the same level of the active device isolation, enabling a complete planar-integration process. The E-mode HEMTs exhibit a smaller transconductance ("§"»,") compared to the D-mode devices, which is due to the incomplete recovery of the plasma-induced damage. The fact that the sample has been through a thermal annealing at 400 °C also indicates that a good thermal stability is expected at a temperature at least up to 400 °C. It should be noted that an ion-implantation technique has also been developed for inter-device isolation accomplished by a multiple energy N+ implantation to produce significant lattice damage throughout the thickness of the GaN buffer layer. Compared to the ion-implantation technique, the CF₄ plasma-treatment technique has the advantages of low cost and low damage.

The E/D-mode HEMTs DCFL inverter fabricated by the planar-integration process was characterized. Figure 31 shows the measured static voltage transfer curve of the inverter at a supply voltage V_DD = 3.3 V. High- and low-output logic levels (V_0H and V_OL) are 3.3 and 0.45 V, respectively, with the output swing (V_OH ~~ V_OL) of 2.85 V. The DC voltage gain in the linear region is 2.9. By defining the values of Vn, and Vm at the unit gain points, the low and high noise margins are 0.34 and 1.47 V. The inverter DC current is also shown in Figure 31. The leakage current with the E-mode device pinch-off is about 3 μA, which is consistent with the discrete device results.

Example

Figure 32 shows AlGaN/GaN epitaxial heterostructures during the fabrication of an HEMTs according to the present innovations. They include the following: 2.5 μm GaN buffer layer and channel, 2 run undoped Al₀.₂₅GaO.₇₅N spacer, 15 nm Alo.₂₅Gao_.75N carrier supply layer with Si doping at 1 x 1018 cm^"3, and a 3 nm undoped Alo.25Gao.₇₅N cap layer. The structures were grown on sapphire substrate in an Aixtron 2000 HT MOCVD system. The process flow is shown in Figures 33(a) through 33(f). The mesa and source/drain ohmic contacts were formed simultaneously for both E-mode and D-mode HEMTs, as shown in Figure 33(a) and (b). The D-mode HEMTs' gate electrodes were then formed by photolithography, metal deposition, and liftoff as shown in Figure 33(c) and (d). After defining the patterns of E- mode HEMTs' gates and interconnections, samples were treated by CF₄ plasma at a source power of 150 W for 150 seconds in an STS RIE system as shown in Figure 33(e), followed by gate metallization and lift-off for the E-mode HEMTs. Inspected by atomic force microscope ("AFM") measurements, the AlGaN barrier thickness was reduced by 0.8 nm after the plasma treatment. Next, a post-gate thermal annealing was conducted at 450 °C for 10 minutes as shown in Figure 33(f). The CF₄ plasma treatment converts the treated GaN HEMT from D-mode to E-mode. The magnitude of threshold voltage shift depends on the treatment conditions, e.g., plasma power and treatment time, as described previously. The post-gate annealing is employed to recover the plasma-induced damages in AlGaN barrier and channel. In principle, the higher is the annealing temperature, the more efficient is the damage repair. However, in practice, the post-gate annealing temperature should not exceed the highest temperature (~ 500 °C, in our case) that the gate Schottky contact can endure, as mentioned earlier. It was found that the D- mode HEMTs' characteristics remain the same after the annealing, whereas the E- mode HEMTs' drain current density increases significantly. The post-gate annealing was found to have no effect on the threshold voltage shift introduced by the plasma treatment.

For the E/D inverter and the ring oscillator, the most important physical design parameter is the drive/load ratio, β = (W_g/Lg)E-mode/(W_g/L_g)D-mode. Several E/D inverters and ring oscillators with β varying from 6.7 to 50 were designed and fabricated on the same sample. The geometric parameters of each design are listed in Figure 34. Discrete E-mode and D-mode GaN HEMTs with 1 x 100 μm gate dimension were simultaneously fabricated on the same sample for dc and RF testing.

Characterisήcs ofE/D-Mode HEMTs

DC current-voltage (I-V ) characteristics of the discrete devices were measured using an HP4156A parameter analyzer. The transfer characteristics of the E/D-mode HEMTs are plotted in Figure 35(a). On-wafer small-signal RP characterization of the discrete devices were carried out in the frequency range of 0.1-39.1 GHz using Cascade microwave probes and an Agilent 8722ES network analyzer. The measured parameters of E/D-mode HEMTs are listed in Figure 36. The threshold voltage and peak transconductance (g_m,_maχ) are 0.75 V and 132 mS/mm for the E-mode HEMT and -2.6 V and 142 mS/mm for the D-mode HEMT. The relatively low peak current density of 480 mA/mm for D-mode HEMT is due to relatively low Al composition of 25% and relatively low doping density of 1 x 1018 cm^"3 in AlGaN barrier layer. Different from the AlGaN/GaN HEMTs used for RP/microwave power amplifiers, the digital ICs are less demanding on the current density. As shown in Figure 35(b), a low knee voltage of 2.5 V is obtained for E-mode HEMTs. At a gate bias of 2.5 V, an on-resistance of 7.1 Ω ^• mm was achieved for the E-mode HEMT, which is the same as that for the D-mode HEMT at the same saturation current level. One observation is that the gate current in both the reverse- and forward-bias conditions is significantly reduced in the E-mode HEMT as shown in Figure 37(a) compared to the D-mode HEMT. The mechanism of this gate current suppression is the modulation of the potential in the AlGaN barrier by the negatively charged fluorine ions that are introduced by the plasma treatment. The conduction-edge band diagrams simulated for both D- and E-mode HEMTs by solving Poisson's equation and Fermi-Dirac statistics. For the simulated conduction band of E-mode HEMTs, the profile of fluorine distribution is approximated by a linear function that features a maximum fluorine ion concentration of 3 x 1019 cm^"3 at the AlGaN surface and reaches zero (negligible) at the AlGaN/GaN interface. A total fluorine ion sheet concentration of about 3 x 1013 cm^"2 is sufficient to compensate not only the Si+ donors' concentration of about 3.7 x 1012 cm^"2 but also the piezoelectric and spontaneous polarization-induced charges (about 1 x 1013 cm^"2). It should be noted that the Schottky baπier height at the gate/AlGaN junction is assumed to remain the same in this example. As seen from the simulated conduction bands shown in Figure 37(b) and (c), the potential of the AlGaN barrier can be significantly enhanced by the incorporation of the fluorine ions, resulting in an enhanced Schottky barrier and the subsequent gate current suppression. The gate current suppression in the forward bias is particularly beneficial to digital IC applications. The suppressed gate current allows the E-mode devices' gate bias to be increased up to 2.5 V. Such an increase results in a larger gate voltage swing, larger dynamic range for the input, and higher fan-out. The increased input voltage swing permits higher supply voltage that is an important factor in achieving higher operation speed and higher noise margins for digital ICs. Without the increased gate input swing, a larger supply voltage will lead to an output voltage (at logic "high") that exceeds the turn-on voltage of the following stage's input gate. The wider dynamic range for the input enables direct logic level matching between the input and the output, eliminating the need for level adjustment between adjacent stages.

It should be noted that silicon nitride passivation, which is an important technique generally used for the stable operation of the GaN-based HEMTs, can also affect the threshold voltage to a lesser degree. The deposition of silicon nitride passivation layer on the active region, in general, can alter the stress in the AlGaN and GaN layers. Subsequently, the piezoelectric polarization charge density and the threshold voltage of the device can be slightly modified. In general, the widely used silicon nitride layer deposited by high-frequency PECVD introduces additional tensile stress in the AlGaN layer, resulting in a negative shift of the threshold voltage in the range of a few tenths of a volt. In practice, this effect should be taken into consideration in the process design. The plasma treatment dose can be increased accordingly to compensate the negative shift in threshold voltage by the SiN passivation layer. The stress of the SiN passivation layer can also be reduced by modifying the process parameters of the PECVD deposition so that the negative shift in the threshold voltage is minimized.

Example: DCFL Inverter

The circuit schematic of an E/D HEMT inverter is shown in Figure IA, where the D-mode HEMT is used as load with its gate tied to its source and the E- mode HEMT is used as a driver. Figure IB shows a fabricated photomicrograph of an inverter according to the present innovations. The fabricated inverters were characterized using an HP4156 A parameter analyzer. Figure 38 shows the static voltage transfer characteristics (the solid curve) for a typical E/D HEMT inverter. The rise in the output voltage at the large input voltages (> 2.1 V) is a result of the gate Schottky diode's turn-on. The dashed curve is the same transfer curve with the axis interchanged and represents the input-output characteristics of the next inverter stage. The parameter definitions follow those given for GaAs- and InP- based HEMTs. The static output levels (VO_H and VO_L) are given by the two intersections of the curves in stable equilibrium points, and the difference between the two levels is defined as the output logic voltage swing. The inverter threshold voltage (V_TH) is defined as Vj_n, where Vi_n is equal to V_out- The static noise margins are measured using the method of largest width for both logic-low noise margin (NM_L) and logic-high noise margin (NM_H)- The measured static voltage transfer curves of E/D inverters with β varied from 6.7 to 50 at a supply voltage V_DD = 1.5 V are plotted in Figure 39. High output logic level (VOH) is maintained at 1.5 V₅ indicating that the E-mode HEMTs are well switched off, whereas low output logic level (V_OL) is improved from 0.34 to 0.09 V as a result of β increasing from 6.7 to 50. As a result, the output logic swing defined as V_OH ~ V_OL increases from 1.16 to 1.41 V. As β is increased from 6.7 to 50, V_TH decreases from 0.88 to 0.61 V, the DC voltage gain (G) in the linear region increases from 2 to 4.1. Figure 40 lists the measured values of static noise margins, as well as V₀H, V₀L₅ output logic swing, VT_H, and G. Both NM_L and NM_H are improved as β increases.

The static voltage transfer curves of the inverter with β = 10 were measured at different supply voltages and are plotted in Figure 41. The circuit performance parameters are listed in Figure 42. When supply voltage increases, all the parameters of E/D inverter increase accordingly. This means that the increase of supply voltage improves the static performance of the E/D inverter. As well known, for HEMT and MESFET E/D inverters, the input voltage is always limited by the turn-on voltage of the gate Schottky diode. At a large input voltage, gate conduction causes an increased voltage drop across the parasitic source resistance of the E-mode device that is used as a driver, raising the voltage of the logic low level. The rise in the output voltage can be observed in the static transfer curves as the supply voltage and the required input voltage increase, as shown in Figure 41. The gate current, when increased by the large input voltage, can significantly degrade the inverter's capability of driving multiple stages, reducing the fan-out. Usually, the turn-on voltage of the gate Schottky diode is around 1 V for a normal AlGaN/GaN HEMT. For a gate-recessed E-mode GaN HEMT, the thinned AlGaN barrier further decreases the turn-on voltage due to an enhanced tunneling current. As a result, for the inverter based on a gate-recessed E-mode GaN HEMT, the output voltage rises when the input voltage is beyond 0.8 V. As disclosed earlier, the E-mode GaN HEMT fabricated by CF₄ plasma treatment possesses a suppressed gate current because of the enhanced Schottky barrier in the AlGaN layer, which is induced by the electronegative fluorine ions. Such a gate current suppression enables a larger input voltage swing for the E/D inverter. As can be seen in Figure 41, the rise in output voltage does not occur until the input voltage is beyond 2 V, indicating about 1 V extension of input voltage swing. Figure 43 shows the dependences of the load current and input current on the input voltage. The lower input current (gate current of the E-mode HEMT) implies a larger amount of fan-out. At "ON" state, the input current exceeds 10% load current when the input voltage is larger than 2 V.

Example: DCFL Ring Oscillator

Figure IB shows a schematic circuit diagram of a DCFL ring oscillator, which is formed with an odd-numbered E/D inverter chain. Seventeen-stage ring oscillators were fabricated with inverters' β = 6.7, 10, and 25. For each ring oscillator, 36 transistors were used including an output buffer. Figure ID shows a photomicrograph of a fabricated ring oscillator according to the present innovations. The ring oscillators were characterized on-wafer using an Agilent E4404B spectrum analyzer and an HP 54522A oscilloscope. The DC power consumption was also measured during the ring oscillators' operation. Figures 44 and 45 show the frequency- and time-domain characteristics of the 17 -stage ring oscillator with β = 10 biased at V_DD = 3.5 V. The fundamental oscillation frequency is 225 MHz. According to the formula of propagation delay per stage τpd = {Inff¹, where the number of stages n is 17 , and τ_pd was calculated to be 130 ps/stage. The dependences of τ_Pd and power-delay product on V_DD were plotted in Figure 46. With the increase of supply voltage, the propagation delay was reduced, whereas power-delay product increases. Compared to τ_pd (234 ps/stage) measured at 1 V, τ_pd measured at 3.5 V is reduced by 45%. The fact that the ring oscillator can operate at such a high V_DD attributes to the larger input voltage swing realized by the CF₄ plasma treatment technique used in the integration process. A minimum power-delay product of 0.113 pJ/stage was found at a V_DD of 1 V. Figure 46 also shows τ_pd and power-delay product characteristics of ring oscillators with β = 6.7 and 25. For the ring oscillator with β = 6.7, the larger τ_pd and power-delay product is due to the larger input capacitance determined by the larger gate length (1.5 μm) of the E-mode HEMT. For the ring oscillator with β = 25, the larger τ_pd is due to the lower charging current determined by the larger gate length (4 μm) of the D- mode HEMT, whereas the power-delay product is at the same level as the one with β = 10. When this integration technology is implemented in the sub-micrometer regime, the gate delay time is expected to be further reduced.

Recently, the discrete E-mode HEMTs and the DCFL ring oscillators have been tested at elevated temperature up to 375 C. No significant shift has been observed in the threshold voltage of the E-mode HEMTs, and the ring oscillator exhibits an oscillation frequency of 70 MHz at 375 C.

According to a disclosed class of innovative embodiments, there is provided: a field-effect transistor, comprising: a source contact and a drain contact; a channel, in a vertically inhomogeneous semiconductor material overlain by a gate, which electrically separates said source contact from said drain contact; said vertically inhomogeneous material having a higher aluminum fraction and a wider bandgap near a surface; and a region of trapped charge, within said semiconductor material, which lies between said gate and said channel, and which also extends laterally toward said drain.

According to a disclosed class of innovative embodiments, there is provided: a method for fabricating semiconductor active devices, comprising the actions of: i) introducing dopants into a first semiconductor material, where exposed by a patterned layer, to populate at least one deep level thereof and thereby introduce trapped charge; and ii) forming heterostructure transistors, which include channel regions in a narrower-bandgap semiconductor which is immediately under a respective portion of said first semiconductor; wherein ones of said transistors also include said trapped charge over portions of said first semiconductor material which connect said channel region to a respective drain region.

Modifications and Variations

As will be recognized by those skilled in the art, the innovative concepts described in the present application can be modified and varied over a tremendous range of applications, and accordingly the scope of patented subject matter is not limited by any of the specific exemplary teachings given.

In the embodiments shown above, two levels of fixed-charge are provided between gate and drain. However, in alternatives, additional intermediate steps can be used if desired, or a continuous increment can be used.

For high-voltage applications, it is also possible to include a fixed-charge component on the source side too, using the same techniques as described above for use on the drain side. This will slightly degrade on-resistance, but can be advantageous where the maximum voltage withstand is required in a HEMT.

The methods and structures described above provide a new tool in electric field control. These concepts are not only applicable to HEMT or MISHFET devices, but also to III-N MESFET (Metal-semiconductor FET) and MOSFET devices. (MESFET devices do not use a gate insulator, but instead provide a Schottky barrier between gate and channel.)

For another example, minor variations in the semiconductor composition, e.g. use of a phosphonitride instead of a pure nitride, or use of an Al_xGa_1-xN over Al_yGai-_yN heterostructure for the basic HEMT structure, are contemplated as alternatives. For another example, in the various device structures shown, a variety of materials can optionally be used for gate electrodes (taking into account any resulting differences in work function).

Similarly, various changes or substitutions can be made in the epitaxial layer doping.

Similarly, as noted above, various materials can optionally be used for the substrate.

Various disclosed embodiments provide field-effect transistors with a new kind of drain engineering, i.e. a new approach to electric field control on the drain side. However, it is also contemplated that the disclosed transistor embodiments can be part of a merged device structure, e.g. with a lateral transistor used to control injection into another device structure.

In another class of alternative embodiments, it is also contemplated that the proposed field-shaping fixed charge can be used for electric field shaping in a high-voltage diode, especially on the anode side.

It is also contemplated that the present invention can be combined with conventional drain-field modification techniques, using e.g. differential diffusion, field plates, and/or sidewall spacers.

Additional general background, which helps to show variations and implementations, may be found in the following publications, all of which are hereby incorporated by reference:

. Y. Cai, et al. IEEE EDL, Vol. 26, pp. 435-437, July 2005;

. W. Saito, et al., IEEE T-ED, vol. 53, pp. 356-362, Feb. 2006;

- Y. Ando, et al., IEEE EDL, vol. 24, pp. 289-291, 2003;

. Y. F. Wu, et al, IEDM 2004, pp. 1078-1079;

• Y. Ando, et al. IEDM 2005, pp. 576-579. None of the description in the present application should be read as implying that any particular element, step, or function is an essential element which must be included in the claim scope: THE SCOPE OF PATENTED SUBJECT MATTER IS DEFINED ONLY BY THE ALLOWED CLAIMS. Moreover, none of these claims are intended to invoke paragraph six of 35 USC section 112 unless the exact words "means for" are followed by a participle.

The claims as filed are intended to be as comprehensive as possible, and NO subject matter is intentionally relinquished, dedicated, or abandoned.

Claims

CLAIMSWhat is claimed is:

1. A field-effect transistor, comprising: a source contact and a drain contact; a channel, in a vertically inhomogeneous semiconductor material overlain by a gate, which electrically separates said source contact from said drain contact; said vertically inhomogeneous material having a higher aluminum fraction and a wider bandgap near a surface; and a region of trapped charge, within said semiconductor material, which lies between said gate and said channel, and which also extends laterally toward said drain.

2. The transistor of Claim 1, wherein said semiconductor material is an

AlGaN/GaN layered structure.

3. The transistor of Claim 1, wherein said semiconductor material is an epitaxial layer supported by a substrate of sapphire, silicon, SiC, AlN, or GaN.

4. The transistor of Claim 1, wherein said semiconductor material is an epitaxial structure comprising a nucleation layer of GaN or AlN, a buffer layer of GaN or AlGaN, a GaN channel, and an AlGaN barrier.

5. The transistor of Claim 1, wherein said source contacts and said drain contacts are formed by depositing multiple metal layers and rapid thermal annealing, wherein said metals are selected from the group consisting of Ti, Al, Ni, and Au.

6. The transistor of Claim 1, wherein said channel is subjected to fluorine-based plasma treatment using a feed gas selected from the group consisting of CF₄, SF₆, BF₃, and mixtures thereof.

7. The transistor of Claim 1, further subjected to the subsequent step of depositing, over said transistor, a passivation material selected from the group consisting of silicon nitride, silicon oxide, polyimide, andbenzocyclobutene.

8. The transistor of Claim 1, further subjected to a thermal annealing at approximately the highest temperature which will not change the Schottky barrier below said gate.

. A method for fabricating semiconductor active devices, comprising the actions of: i) introducing dopants into a first semiconductor material, where exposed by a patterned layer, to populate at least one deep level thereof and thereby introduce trapped charge; and ϋ) forming heterostructure transistors, which include channel regions in a narrower-bandgap semiconductor which is immediately under a respective portion of said first semiconductor; wherein ones of said transistors also include said trapped charge over portions of said first semiconductor material which connect said channel region to a respective drain region.

10. The method of Claim 9, wherein said respective channel region is connected to said respective drain region both by said portion which is overlain by said trapped charge and also, in series therewith, by another portion of said semiconductor material which is not overlain by said trapped charge.

11. The method of Claim 9, wherein ones of said transistors include said trapped charge over portions of said first semiconductor material which connect said channel region to a respective drain region, but not over portions of said first semiconductor material which connect said channel region to a respective source region.

12. The method of Claim 9, wherein said semiconductor material is an

AlGaN/GaN layered structure.

13. The method of Claim 9, wherein said semiconductor material is an epitaxial layer supported by a substrate of sapphire, silicon, SiC, AlN, or GaN.

14. The method of Claim 9, wherein said semiconductor material is an epitaxial structure comprising a nucleation layer of GaN or AlN, a buffer layer of GaN or AlGaN, a GaN channel, and an AlGaN barrier.

15. The method of Claim 9, further comprising the subsequent step of depositing, over said transistors, a passivation material selected from the group consisting of silicon nitride, silicon oxide, polyimide, and benzocyclobutene.

16. The method of Claim 9, further comprising a thermal annealing at approximately the highest temperature which will not change the Schottky barrier below said gate.