Control of threshold voltage in E-mode and D-mode GaN-on-Si metal-insulator-semiconductor heterostructure field effect transistors by in-situ fluorine doping of atomic layer deposition Al2O3 gate dielectrics

We report the modiﬁcation and control of threshold voltage in enhancement and depletion mode AlGaN/GaN metal-insulator-semiconductor heterostructure ﬁeld effect transistors through the use of in-situ ﬂuorine doping of atomic layer deposition Al 2 O 3 . Uniform distribution of F ions throughout the oxide thickness are achievable, with a doping level of up to 5.5 (cid:2) 10 19 cm (cid:3) 3 as quantiﬁed by secondary ion mass spectrometry. This ﬂuorine doping level reduces capacitive hysteretic effects when exploited in GaN metal-oxide-semiconductor capacitors. The ﬂuorine doping and forming gas anneal also induces an average positive threshold voltage shift of between 0.75 and 1.36 V in both enhancement mode and depletion mode GaN-based transistors compared with the undoped gate oxide via a reduction of positive ﬁxed charge in the gate oxide from þ 4.67 (cid:2) 10 12 cm (cid:3) 2 to (cid:3) 6.60 (cid:2) 10 12 cm (cid:3) 2 . The application of this process in GaN based power transistors advances the realisation of normally off, high power, high speed devices. V C 2016

The development of AlGaN/GaN based power transistors is the focus of widespread research as they offer the potential for large efficiency savings when used in power switching applications compared to Si and SiC based devices. 1,2 The basis of these devices is the utilisation of a high mobility 2-dimensional electron gas (2DEG) underneath the AlGaN/GaN interface which allows rapid switching and low on-resistance transistors to be fabricated for power control applications.
For power transistor applications, it is desirable to have normally off devices. 3 One way to achieve this is to form a negative charge in the gate region of the device by doping or ion implantation with fluorine (F) ions in the gate region of the device. [3][4][5][6][7] Alternative fluorine treatments have been investigated elsewhere, as a means of ameliorating the relatively high defect densities associated with atomic layer deposition (ALD) of amorphous Al 2 O 3 -nitride device structures. 8 A fluorine-based plasma treatment has been previously described as a method of incorporating negatively charged F ions into the AlGaN barrier of an E-mode high electron mobility transistor (HEMT) and resulted in a positively shifted threshold voltage. 9 Further development of normally off AlGaN/GaN MISHEMT devices was achieved by exploiting fluorinated Al 2 O 3 thin film gate dielectrics, prepared using the fluorinebased plasma treatment. The E-mode MISHEMTs exhibited high transconductance (153 mS/mm) and large saturated drain currents (547 mA/mm) when the plasma treatment was performed on the dielectric surface. 10 It was reported that the F-distribution was confined to the top 2 nm of the dielectric and furthermore avoided plasma-induced damage at the interface with the III-nitride. Zhang et al. 4 have reported the compensation of the intrinsic positive charges in Al 2 O 3 gate dielectric by fluorine ions incorporated into the Al 0.26 Ga 0.74 N barrier of GaN metal-oxide-semiconductor high-electron-mobility transistors (MOSHEMTs) by CF 4 plasma treatment. The incorporated fluorine redistributed by diffusion back into the Al 2 O 3 during ALD at 250 C, although a further post-deposition anneal at 400 C left the F-distribution almost unchanged. Here, we report an in-situ process for the uniform F doping of alumina throughout the gate oxide thickness deposited by atomic layer deposition. This approach avoids exposure of the III-nitride to detrimental effects from plasma treatment and reduces the overall thermal budget during fabrication.
Substrates for the electrical properties characterisation were grown by MOCVD on 6 00 Si(111) wafers. For metaloxide-semiconductor capacitor (MOSC) substrates, the layer structure consisted of Si(111)/AlN nucleation layer/graded AlGaN layer/n-GaN doped with 1 Â 10 18 cm À3 Si/n-GaN doped with 1 Â 10 17 cm À3 Si. Planar circular MOSC structures were fabricated utilising Ti/Al/Ni/Au ohmic contacts in contact with the 1 Â 10 18 cm À3 Si doped layer and annealed at 770 C for 30 s in N 2 . Following a cleaning step using acetone and IPA, $10 nm Al 2 O 3 and a range of concentrations of F:Al 2 O 3 were deposited onto different MOSC substrates using the ALD processes described further on in this paper. After ALD, windows through the dielectric layer were opened over the ohmic contacts using a SiCl 4 reactive-ion etch. Ni/Au pads were deposited to form the top contacts of the MOSC structures. Completed MOSC structures were annealed at 430 C in forming gas (90%N 2 /10%H 2 ) for 30 min to improve the interface quality. 11 Substrates for the MISHFET devices consisted of Si(111)/AlN nucleation layer/graded AlGaN layer/C doped GaN at 5 Â 10 18 cm À3 /undoped GaN/1 nm AlN/27 nm 0003-6951/2016/108(7)/072901/5/$30.00 V C 2016 AIP Publishing LLC 108, 072901-1 APPLIED PHYSICS LETTERS 108, 072901 (2016) AlGaN/2 nm GaN. Hall measurements of the substrates yielded sheet resistance of 490 X/ٗ, sheet carrier density of 6.9 Â 10 12 cm À2 , and electron mobility of 1855 cm 2 /V s. The devices were processed by mesa etching, ohmic formation, SiN x passivation, gate foot opening, F ion implantation 12 (for E-mode only), Al 2 O 3 or F:Al 2 O 3 gate deposition, 30 min forming gas anneal (FGA) at 430 C, T-gate formation, and probe pad formation. The gate lengths and gate widths for the MISHFET devices were 1.5 lm and 100 lm, respectively, and each coupon was approximately 10 Â 10 mm.
Atomic layer deposition was used to grow the F:Al 2 O 3 and Al 2 O 3 films presented in this paper. The growth of all ALD films was carried out using an Oxford Instruments Plasma OpAL reactor. Substrate temperatures were set at 200 C for the depositions. The initial depositions were made onto Si(100) substrates for characterisation by spectroscopic ellipsometry (SE) and secondary ion mass spectroscopy (SIMS SE was used for characterisation of the films grown on Si(100) before and after (FGA) at 430 C in 10%H 2 /90%N 2 with a scan range of 500-750 nm. Fitting was performed using a new amorphous model for Al 2 O 3 to determine the film thicknesses and the refractive index of the material. Equation (1) shows the single term Sellmeier dispersion formula 13 The average oscillator strength, S 0 and the average oscillator position, k 0 can be obtained by plotting 1=ðn 2 À 1Þ versus 1=k 2 . The slope of the resultant straight line gives 1=S 0 while the intercept at infinite wavelength yields 1=S 0 k 2 0 . An alternative form of the equation allows the average single oscillator energy gap to be calculated 14 where hx ¼ photon energy, E 0 ¼ the average single oscillator energy gap in eV, and E d ¼ the average oscillator strength in eV. Further characterisation was conducted on 1000 cycle films deposited on Si(100) using SIMS. SRIM modelling and a 35 keV F ion implanted $100 nm Al 2 O 3 on Si(100) were used to enable quantification of the in-situ doped oxide. Figure 1 shows the SE results for the calculation of the single oscillator Sellmeier average oscillator strength, average oscillator position, and average oscillator energy gap. Refractive indices at 632.8 nm were found to be 1.655 for as grown Al 2 O 3 and 1.657 after FGA, within the expected range for thin film Al 2 O 3 grown by other groups. 15,16 Replacing the H 2 O dose with 40% NH 4 F:H 2 O reduced the refractive index at 632.8 nm to 1.646. This value was further reduced to 1.644 with the application of FGA. The slight increase in n after FGA for the undoped film could be attributed to either the film becoming denser as a result of the heat treatment, 15 the forming gas passivating unsatisfied positive bonds in the oxide 17 or a combination of the two. The reduction in n seen with the addition of 40% NH 4 F:H 2 O indicated that Al-F bonds could be present in the material. Since AlF 3 has a significantly lower n than Al 2 O 3 at 1.2-1.4 (Refs. 18 and 19), a lower refractive index would be expected if a mixed amorphous material consisting of Al-O and Al-F bonds was being produced. However, as the reduction in n was only of the order of 10 À2 , this indicated that the F content of the films was low. Following the FGA, the slight reduction seen in the refractive index for F:Al 2 O 3 could be attributed to displacement of O for F within the oxide 4,20 or by a change in density of the films. Figure 1 shows the results for the average oscillator strength, S 0 , the average oscillator position, k 0 , the average single oscillator energy gap in eV, E 0 , and the average oscillator strength in eV, E d . The average oscillator strength was found to reduce by the addition of F into the oxide from 1.919 Â 10 14 m À2 to 1.862 Â 10 14 m À2 . This was attributed to the replacement of a proportion of the Al-O dipoles with higher mass Al-F dipoles. 21 The average oscillator positions were found to be 94.1 nm for Al 2 O 3 and 94.7 for F:Al 2 O 3 . The resonant frequency x 0 is inversely related to the average oscillator position by where c ¼ velocity of light in free space. Furthermore where k ¼ the spring constant, M, M À , and M þ are the reduced mass and negative and positive ion masses, respectively. 19 In their analysis, M þ represents the Al atomic mass, and M À represents either the negatively charged F or O atomic masses. The increase in k 0 (and decrease in x 0 ) implies that the reduced mass term, M, is increased by the addition of F into the oxide. As the positive Al mass should be unchanged, this increase in M can be attributed to the change from lighter O atoms to heavier F atoms. As ALD Al 2 O 3 grown by thermal methods is known to possess intrinsic positive bulk charge, 4 this Sellmeier analysis indicates that the increase in F, with its additional electron and higher electronegativity compared to O, 7 and the effect of the FGA could compensate for the native positive charge that occurs in ALD Al 2 O 3 . Figure 2 shows the results of SIMS analysis for $100 nm of Al 2 O 3 F ion implanted with a 35 keV ion beam and from the in-situ doped oxide. SIMS results for the in-situ doped oxide show a F concentration of $5.5 Â 10 19 cm À3 , which based on Al 2 O 3 density between 3 g/cm 3 (Ref. 15) and 3.95 g/cm 3 yields a F concentration of between 0.06 at. % and 0.05 at. %, in agreement with that detected using XPS (not shown). The doping profile as observed using SIMS showed a uniform doping density throughout the oxide with slight increases in doping concentration at the upper and lower interfaces.  25 The best value they reported after optimisation of the surface treatment prior to gate deposition was 0.2 V. The reduction in hysteresis seen here was thought to come from negatively charged F atoms within the bulk oxide compensating for residual intrinsic positive traps resulting from the growth process, as shown in Figure 3(b). Figure 3(b) shows the fixed charge within the gate dielectric as calculated from the CV results. 26 Without FGA, the number of fixed charges in the undoped Al 2 O 3 is þ4.67 Â 10 12 cm À2 . With the addition of NH 4 F into the ALD growth process, the positive fixed charges are compensated for. Increasing the percentage of F in the oxide serves to further reduce the positive fixed charge up to a maximum reduction of around one order of magnitude (þ4.67 Â 10 12 cm À2 to þ4.04 Â 10 11 cm À2 ). After FGA, the bulk charge for Al 2 O 3 is halved to þ2.53 Â 10 12 cm À2 . The influence of the F doping process is increased following FGA, with the 100% F:Al 2 O 3 showing negative bulk charge at À6.60 Â 10 12 cm À2 .
The positive shift in V FB can be attributed to the higher concentration of negative charges near to the n-GaN/dielectric interface, introduced by the F doping as evidenced in Figure 3(b) and implicated by the Sellmeier analysis above. Undoped Al 2 O 3 and 100% F:Al 2 O 3 were subsequently incorporated within E-and D-mode MISHFET fabrication processes to ascertain the effect of the gate oxide in GaN-on-Si MISHFET devices. Figure 4(a) shows drain current (I D ) against gate source voltage (V GS ) and the gate leakage current (I G ) for representative D-mode MISHFET devices with 20 nm Al 2 O 3 and 20 nm F:Al 2 O 3 gate oxides and gate to drain distances of 12 lm. 15 devices of each type were measured for the data reported here. All devices tested were found to work, and no relationship between device location and HEMT properties is observed. The maximum drain current is uneffected by the addition of F into the gate oxide layer with both sets of devices giving I Dmax between 600 and 800 mA/mm. The threshold voltage (V TH ) for the standard undoped gate oxide devices ranges from À9.39 V to À8.86 V, with a mean V TH of À9.23 V. With the addition of F into the gate stack measurements across several devices give mean V TH values of À8.48 V, an increase in þ0.75 V, with values ranging from À8.65 V to À8.10 V. Gate leakage currents are of the order of 1 Â 10 À6 mA/mm for both oxides, although a small proportion of the devices showed higher gate leakage up to  1 Â 10 À3 mA/mm. Devices with the higher gate leakage still showed consistent V TH and I Dmax values. Figures 4(b) and 4(c) show the drain current characteristics for both sets of devices and shows kinks in the drain current between 8 and 10V DS . This is thought to be related to the presence of slow traps in the GaN buffer layer within the substrate. 27 The peak transconductance for both sets range from 120 to 128 mS/mm. The mean, variance, and standard deviation (r) of V TH for 15 of each type of MISHFET are shown in Figure  4(d). The mean shift in V TH is observed to be 0.75 V due to the F-doping of the alumina gate dielectric in the D-mode devices. The F-induced shift is statistical significant with a 99% confidence limit as determined by a t-test. Figure 5(a) shows I D and I G against V GS for a representative E-mode MISHFET (produced by the ion implantation of F into the barrier as described elsewhere in Ref. 12) with both 20 nm Al 2 O 3 and 20 nm F:Al 2 O 3 gate oxides. I Dmax for the standard Al 2 O 3 gate dielectric range from 400 to 500 mA/mm across all devices tested. This is slightly reduced in the F:Al 2 O 3 devices to around 350-400 mA/mm. Mean V TH for the Al 2 O 3 gated device is þ0.99 V. The maximum and minimum V TH values are þ1.83 V and þ0.10 V, respectively. With the addition of the in-situ deposited F, the mean V TH value is found to increase to þ2.35 V, an increase in þ1.36 V, with values ranging from þ1.97 V to þ2.60 V. The addition of F into the gate oxide layer also reduces the off-state drain current for gate-source voltages below V TH compared to the standard Al 2 O 3 gate oxide. Figures 5(b) and 5(c) show the drain current characteristics for E-mode devices with standard ALD Al 2 O 3 and in-situ doped F:Al 2 O 3 . Peak transconductance for E-mode devices with Al 2 O 3 gate dielectric range from 65 to 95mS/mm and 70 to 90mS/mm for E-mode devices with F:Al 2 O 3 gate dielectric. Figure 5(d) shows the mean, variance, and standard deviation (r) of V TH for 15 of each type of MISHFET. Fluorine-doping of alumina gate dielectric causes a shift in the mean V TH of 1.36 V for the E-mode devices. The statistical significance of the F-induced shift has a 99% confidence limit as determined by a t-test. Although direct comparisons between results from different groups are awkward due to differences in device layout, processing steps, etc., recent advances and publications in the field for D-and E-mode GaN-based MISHFET devices show that higher drain currents and peak transconductances are achievable (1550 mA/mm and 330 mS/mm, respectively, for the D-mode, 28 and 1130 mA/ mm (Ref. 29) and 153 mS/mm for E-mode 10 ) through further device optimisation.
In conclusion, positive V FB and V TH shifts in GaN-on-Si MOSCs and GaN-on-Si MISHFETs were demonstrated via atomic layer deposition in-situ doping and growth of Al 2 O 3 gate oxide with F. This was achieved using a 40% NH 4 F:H 2 O precursor in place of the standard H 2 O precursor. Spectroscopic ellipsometry results, single oscillator Sellmeier analysis, and fixed charge calculations from MOSC CV analysis showed that increasing levels of F compensated for positive traps in the oxide. Following gate deposition, annealing in 10%H 2 /90%N 2 increased the effect of the F, resulting in a maximum fixed negative charge of À6.6 Â 10 12 cm À2 . SIMS results showed that the doping of F in Al 2 O 3 gave a more uniform concentration of F atoms through the oxide compared to other methods such as ion implantation, with a doping concentration around 5.5 Â 10 19 cm À3 . In total, 60 MISHFET devices were measured. 30 E-mode and D-mode devices were produced, with half of each device type using Al 2 O 3 gate oxide and the other half using F:Al 2 O 3 gate oxide. The mean D-and E-mode V TH shifts due to F-incorporation were found to be 0.75 V and 1.36 V, respectively, with a confidence limit of 99%. F:Al 2 O 3 gate oxide can be used in conjunction with other E-mode MISHFET methods, such as ion implantation, gate recess, tri-gate to achieve devices with more positive V TH , and low gate leakage currents.