Charging Effect by Fluorine-Treatment and Recess Gate for Enhancement-Mode on AlGaN/GaN High Electron Mobility Transistors

An enhancement-mode AlGaN/GaN metal-insulator-semiconductor high-electron- mobility-transistor was fabricated using a recess gate and CF4 plasma treatment to investigate its reliable applicability to high-power devices and circuits. The fluorinated-gate device showed hysteresis during the DC current-voltage measurement, and the polarity and magnitude of hysteresis depend on the drain voltage. The hysteresis phenomenon is due to the electron trapping at the Al2O3/AlGaN interface and charging times longer than milliseconds were obtained by pulse I-V measurement. In addition, the subthreshold slope of the fluorinated-gate device was increased after the positive gate bias stress because of the two-dimensional electron gas reduction by ionized fluorine. Our systematic observation revealed that the effect of fluorine ions should be considered for the design of AlGaN/GaN power circuits.


Introduction
AlGaN/GaN high electron mobility transistors (HEMTs) have recently demonstrated to be excellent devices for high-frequency and high-power electronics, thanks to the high breakdown voltage [1] and the low on-state resistance [2] and gate leakage [3]. The AlGaN/GaN HEMTs show normally on operation, since the two-dimensional electron gas (2DEG) channel is generated by the spontaneous and piezoelectric polarization at the AlGaN/GaN hetero-interface [4]. Various methods for enhancement mode (E-mode) device manufacturing through modulating the threshold voltage (V T ) have been studied for the reduction in power consumption and circuit producing, such as an inverter using a depletion mode (D-mode) and an E-mode device [5][6][7][8][9]. The AlGaN barrier recess and fluorine ion implantation methods had been widely studied to shift the V T in the positive direction.
Although these methods modulate the V T , the gate modifications introduce side effects resulting from the surface damage during the device fabrication process [10][11][12], and the electrical and frequency characteristics are degraded due to the poor interface conditions [13]. Various methods have been studied to improve the recessed interface condition [14,15]. However, hysteresis and V T instability remain critical issues so far. In this perspective, it is necessary to better understand the degradation mechanisms and the trends of changing electrical properties related to charge trapping for reliable application.
In this study, the V T was modulated by the gate recess and fluorine treatment employing CF 4 plasma. The fluorinated-gate device (7-nm thick gate recess + fluorine plasma treatment) exhibited more positive-shifted V T compared to the recess only gate device (7-nm thick gate recess). However, hysteresis of the fluorinated-gate device occurred during the DC current-voltage (I-V) measurement and the hysteresis depended on drain voltage (V D ). We demonstrated that the cause of hysteresis was the charge trapping/detrapping effect introduced at the dielectric (Al 2 O 3 ) and fluorinated AlGaN interface and gate leakage characteristics corresponding to V D . The increase in the subthreshold slope (SS) of the fluorinated-gate device after positive gate bias stress indicated the degradation in the AlGaN/GaN interface and the decrease in 2DEG density. As a result, we have revealed the effect of fluorine treatment on the dielectric/AlGaN and AlGaN/GaN interfaces through experimental evidence.

Device Fabrication and Electrical Characterization
The AlGaN/GaN heterostructure was grown by metal-organic chemical vapor deposition (MOCVD) on a sapphire substrate, which consists of a 21-nm AlGaN (26% Al) barrier with a 2-nm GaN cap layer. The Ti/Al/Ni/Au ohmic contact was formed by an E-beam evaporator and annealed at 900 • C for 150 s. After device isolation via phosphorus implantation, a 50-nm SiN passivation layer was deposited by plasma-enhanced chemical vapor deposition (PECVD).
Subsequently, the gate recess area was defined by electron-beam (E-beam) lithography and then the SiN was removed via an inductively-coupled plasma (ICP) etch. In order to obtain the etch rate of the gate region using a digital etch process, O 2 plasma ashing (3 min) and 1:10 hydrochloric acid (HCl) dipping (1 min) were repeated 15 times in the test sample. An etch rate of 6.6 Å/cycle was obtained (in Figure 1a). The digital etches of 10 and 21 cycles were conducted with the same etch condition and each device had an etch depth of about 7 and 14 nm. For the fabrication of fluorinated-gate devices, CF 4 plasma using reactive ion etching (RIE) was carried out for 30 s on the 7-nm recessed device. After that, Al 2 O 3 with the thickness of 10 nm was deposited at 300 • C by atomic layer deposition (ALD). A T-shaped Ni/Au gate electrode was deposited by an E-beam evaporator. D-mode metal-insulator-semiconductor HEMTs (MIS-HEMTs) without gate recess and fluorine treatment were also prepared as a reference device. Figure 1b illustrates the cross-sectional view of the fabricated E-mode Al 2 O 3 /AlGaN/GaN MIS-HEMTs with a recessed gate. The fabricated device geometries are L G = 0.25 µm, W G = 50 µm, L GD = 3.5 µm and L GS = 1 µm. The DC I-V, pulse I-V measurement and stress application (for repetition of stress and measurement) were performed using a Keithley 4200-SCS parameter analyzer. The delay between the stress and measurement was less than tens of milliseconds. Capacitance of the device was measured using an Agilent 4294A impedance analyzer. The voltage was applied to the gate while the source and drain were grounded.  Figure 2 shows the I-V transfer curves and transconductance (gm) of AlGaN/GaN MIS-HEMTs with various gate condition. The VT moves in the positive direction from the VT of D-MIS-HEMT after the 7-nm gate recess and fluorine treatment, and the VT further shifted after the 14-nm gate recess. The VT depends on the barrier thickness, since the 2DEG concentration and electron mobility reduce with the decrease in AlGaN barrier thickness [16]. The fluorine treatment modulates the VT by incorporating negatively charged fluorine ions into the AlGaN barrier. As a result, the VT of D-MIS-HEMT, 7-nm recess, 7-nm recess + CF4 and 14-nm recess were −6.5, −5.2, −2.5 and +0.5 V, respectively. The VT can be expressed as an Equation (1) [17]. The tb denotes the barrier thickness, φb is the barrier height, q is the electron charge, σp is the 2DEG sheet charge density, NF is the density of ionized fluorine atoms, XF is the triangular fluorine ion distribution peaked at the surface and extending to a given depth and ε and ΔEc are the dielectric constant and conduction band offset, respectively.  Figure 2 shows the I-V transfer curves and transconductance (g m ) of AlGaN/GaN MIS-HEMTs with various gate condition. The V T moves in the positive direction from the V T of D-MIS-HEMT after the 7-nm gate recess and fluorine treatment, and the V T further shifted after the 14-nm gate recess. The V T depends on the barrier thickness, since the 2DEG concentration and electron mobility reduce with the decrease in AlGaN barrier thickness [16]. The fluorine treatment modulates the V T by incorporating negatively charged fluorine ions into the AlGaN barrier. As a result, the V T of D-MIS-HEMT, 7-nm recess, 7-nm recess + CF 4 and 14-nm recess were −6.5, −5.2, −2.5 and +0.5 V, respectively. The V T can be expressed as an Equation (1) [17]. The t b denotes the barrier thickness, ϕ b is the barrier height, q is the electron charge, σ p is the 2DEG sheet charge density, N F is the density of ionized fluorine atoms, X F is the triangular fluorine ion distribution peaked at the surface and extending to a given depth and ε and ∆E c are the dielectric constant and conduction band offset, respectively.

Results and Discussion
However, the g m of the 7-nm recess + CF 4 device decreased sharply after maximum value in comparison with other devices. It is attributed to the electron trap in the AlGaN barrier under the gate region by the fluorine treatment [18]. The 7-nm recess + CF 4 device shows hysteresis and it depends on the V D , as shown in Figure 3a. The characteristics of recessed gate (7-nm recess) and fluorinated-gate devices (7 nm-recess + CF 4 ) were discussed to compare the fluorine treatment effect. For the hysteresis and V D effect observation, the V G increased in the forward direction (from the off-state to the on-state) and then decreased again in the reverse direction (from the on-state to the off-state) during the DC I-V measurement for various V D . When the V D is increased from 1 V to 10 V and returned to 1 V, the hysteresis magnitudes (V Hysteresis ) are almost symmetric. The V Hysteresis is the V G difference at a specific I D (= 1 × 10 −7 A/mm). It shows that the DC I-V measurement does not cause permanent charging. However, when the V D is higher than 5 V, the direction of hysteresis becomes negative and the V Hysteresis enlarges with increasing V D , as shown in Figure 3c. On the contrary, when the V D is lower than 5 V, the direction of hysteresis is positive and the V Hysteresis is relatively small. The negligible hysteresis of recessed gate device without fluorine treatment was achieved (Figure 3b,d). However, the gm of the 7-nm recess + CF4 device decreased sharply after maximum value in comparison with other devices. It is attributed to the electron trap in the AlGaN barrier under the gate region by the fluorine treatment [18]. The 7-nm recess + CF4 device shows hysteresis and it depends on the VD, as shown in Figure 3a. The characteristics of recessed gate (7-nm recess) and fluorinated-gate devices (7 nm-recess + CF4) were discussed to compare the fluorine treatment effect. For the hysteresis and VD effect observation, the VG increased in the forward direction (from the offstate to the on-state) and then decreased again in the reverse direction (from the on-state to the offstate) during the DC I-V measurement for various VD. When the VD is increased from 1 V to 10 V and returned to 1 V, the hysteresis magnitudes (VHysteresis) are almost symmetric. The VHysteresis is the VG difference at a specific ID (= 1 × 10 −7 A/mm). It shows that the DC I-V measurement does not cause permanent charging. However, when the VD is higher than 5 V, the direction of hysteresis becomes negative and the VHysteresis enlarges with increasing VD, as shown in Figure 3c. On the contrary, when the VD is lower than 5 V, the direction of hysteresis is positive and the VHysteresis is relatively small. The negligible hysteresis of recessed gate device without fluorine treatment was achieved (Figure 3b,d). However, the gm of the 7-nm recess + CF4 device decreased sharply after maximum value in comparison with other devices. It is attributed to the electron trap in the AlGaN barrier under the gate region by the fluorine treatment [18]. The 7-nm recess + CF4 device shows hysteresis and it depends on the VD, as shown in Figure 3a. The characteristics of recessed gate (7-nm recess) and fluorinated-gate devices (7 nm-recess + CF4) were discussed to compare the fluorine treatment effect. For the hysteresis and VD effect observation, the VG increased in the forward direction (from the offstate to the on-state) and then decreased again in the reverse direction (from the on-state to the offstate) during the DC I-V measurement for various VD. When the VD is increased from 1 V to 10 V and returned to 1 V, the hysteresis magnitudes (VHysteresis) are almost symmetric. The VHysteresis is the VG difference at a specific ID (= 1 × 10 −7 A/mm). It shows that the DC I-V measurement does not cause permanent charging. However, when the VD is higher than 5 V, the direction of hysteresis becomes negative and the VHysteresis enlarges with increasing VD, as shown in Figure 3c. On the contrary, when the VD is lower than 5 V, the direction of hysteresis is positive and the VHysteresis is relatively small. The negligible hysteresis of recessed gate device without fluorine treatment was achieved (Figure 3b,d). The V D dependence of hysteresis in the fluorinated-gate device was investigated in more detail by comparing the gate leakage current (I G ) as a function of V D in Figure 4a, and the band diagram and I G flow direction are represented in Figure 4b. The I G also depends on the V D , and when the V D is larger than 6 V, the polarity of the I G is negative. This means that the electrons are trapped in the Al 2 O 3 /fluorinated-AlGaN barrier as they move from the gate metal to the channel along the I G , and the V T is negatively shifted. However, when the V D is in between −2 V and 6 V, the polarity of I G is positive, showing that some of electrons are detrapped from the Al 2 O 3 /fluorinated-AlGaN barrier to the gate metal and the V T is slightly shifted to the positive direction.
The VD dependence of hysteresis in the fluorinated-gate device was investigated in more detail by comparing the gate leakage current (IG) as a function of VD in Figure 4a, and the band diagram and IG flow direction are represented in Figure 4b. The IG also depends on the VD, and when the VD is larger than 6 V, the polarity of the IG is negative. This means that the electrons are trapped in the Al2O3/fluorinated-AlGaN barrier as they move from the gate metal to the channel along the IG, and the VT is negatively shifted. However, when the VD is in between −2 V and 6 V, the polarity of IG is positive, showing that some of electrons are detrapped from the Al2O3/fluorinated-AlGaN barrier to the gate metal and the VT is slightly shifted to the positive direction. The charging time in the Al2O3/fluorinated-AlGaN barrier was studied through a triangular pulse I-V measurement with rising (tR) and falling time (tF) of 1 millisecond (ms). As shown in Figure  5a, the hysteresis and the VD dependence of VT were obtained only in the fluorinated-gate device, whereas insignificant hysteresis was achieved for pulse I-V measurements, and the VT is hardly changed with VD ( Figure 5c). However, the transfer curve of recessed gate devices is almost identical regardless of the pulse I-V measurement (Figure 5b,d). Therefore, it implies that the electron trapping/detrapping in the Al2O3/fluorinated-AlGaN barrier is a very slow charging mechanism and the trapping/detrapping time is longer than ms. The charging time in the Al 2 O 3 /fluorinated-AlGaN barrier was studied through a triangular pulse I-V measurement with rising (t R ) and falling time (t F ) of 1 millisecond (ms). As shown in Figure 5a, the hysteresis and the V D dependence of V T were obtained only in the fluorinated-gate device, whereas insignificant hysteresis was achieved for pulse I-V measurements, and the V T is hardly changed with V D (Figure 5c). However, the transfer curve of recessed gate devices is almost identical regardless of the pulse I-V measurement (Figure 5b,d). Therefore, it implies that the electron trapping/detrapping in the Al 2 O 3 /fluorinated-AlGaN barrier is a very slow charging mechanism and the trapping/detrapping time is longer than ms.
To investigate the effect of fluorine in AlGaN under the gate bias stress, negative (−2 V) and positive (+2 V) gate bias stresses were applied to devices. Although the V T of devices is different, the same stress voltage was introduced to compare the effect of the stress polarity ( Figure 6). The V D of 5 V was used as a condition to monitor the electrical characteristics of the devices and to minimize the charging effects that occurred during DC I-V measurements. The direction of V T shift under the negative gate bias stress is negative; by contrast, the positive direction of V T shift under the positive gate bias stress is positive, and this phenomenon is the same for both devices. It indicates the trapping (or detrapping) of electrons under the negative (or positive) gate bias stress at the AlGaN/GaN barrier. The different directions of V T shift under negative and positive gate bias stress have been analyzed with the same mechanism in many other studies [19][20][21]. However, in the fluorinated-gate device, the V T moved more, even under other stress conditions (V stress = V T + 1.0 V and V stress = +0.5, +1.0 V), as shown in Figure 7a,b, due to the plasma-induced surface damage of the AlGaN barrier during fluorine treatment. In addition, the SS of the fluorinated-gate device is higher than that of the recessed gate device before stress. This indicates a higher interface state and is further evidence of the AlGaN surface damage caused by the CF 4 plasma process. Furthermore, the SS of the fluorinated-gate device increased with stress time under the positive gate bias stress, whereas the SS of the recessed gate device without fluorine hardly changed with stress time, even under negative and positive gate bias stress. To observe in more detail, the change in SS is represented in Figure 7c,d. There are two possible origins of the SS increase after gate bias stress. One is the 2DEG reduction. The accelerated electrons from the channel under the positive gate bias stress induce the ionization of fluorine in the AlGaN barrier [22], and the ionized fluorine atoms are capable of forming bonds with gallium atoms with dangling bonds at the interface. The formation of Ga-F bonds is able to weaken the 2DEG channel and AlGaN/GaN interface [23].
Another is the degradation of the AlGaN/GaN interfacial layer by accelerated electrons under positive gate bias stress. For metal-oxide-semiconductor field-effect transistors (MOSFETs), the increase in SS primarily reflects the degradation of the interfacial layer [24,25]. However, the recessed gate device without fluorine did not show an SS change, even under the positive gate bias stress. Therefore, the Ga-F bonding mechanism makes more sense than the interface degradation. To investigate the effect of fluorine in AlGaN under the gate bias stress, negative (−2 V) and positive (+2 V) gate bias stresses were applied to devices. Although the VT of devices is different, the same stress voltage was introduced to compare the effect of the stress polarity ( Figure 6). The VD of 5 V was used as a condition to monitor the electrical characteristics of the devices and to minimize the charging effects that occurred during DC I-V measurements. The direction of VT shift under the negative gate bias stress is negative; by contrast, the positive direction of VT shift under the positive gate bias stress is positive, and this phenomenon is the same for both devices. It indicates the trapping (or detrapping) of electrons under the negative (or positive) gate bias stress at the AlGaN/GaN barrier. The different directions of VT shift under negative and positive gate bias stress have been analyzed with the same mechanism in many other studies [19][20][21]. However, in the fluorinated-gate device, the VT moved more, even under other stress conditions (Vstress = VT + 1.0 V and Vstress = +0.5, +1.0 V), as shown in Figure 7a,b, due to the plasma-induced surface damage of the AlGaN barrier during fluorine treatment. In addition, the SS of the fluorinated-gate device is higher than that of the recessed gate device before stress. This indicates a higher interface state and is further evidence of the AlGaN surface damage caused by the CF4 plasma process. Furthermore, the SS of the fluorinatedgate device increased with stress time under the positive gate bias stress, whereas the SS of the recessed gate device without fluorine hardly changed with stress time, even under negative and positive gate bias stress. To observe in more detail, the change in SS is represented in Figure 7c,d. There are two possible origins of the SS increase after gate bias stress. One is the 2DEG reduction. The accelerated electrons from the channel under the positive gate bias stress induce the ionization of fluorine in the AlGaN barrier [22], and the ionized fluorine atoms are capable of forming bonds with gallium atoms with dangling bonds at the interface. The formation of Ga-F bonds is able to weaken the 2DEG channel and AlGaN/GaN interface [23]. Another is the degradation of the AlGaN/GaN interfacial layer by accelerated electrons under positive gate bias stress. For metal-oxidesemiconductor field-effect transistors (MOSFETs), the increase in SS primarily reflects the degradation of the interfacial layer [24,25]. However, the recessed gate device without fluorine did     The capacitance-voltage (C-V) curve was measured before and after positive gate bias stress at 1 MHz, as shown in Figure 8a, and the C-V curve moved to a positive direction and the slope of the C-V curve lowered. AlGaN/GaN interface degradation and reduction in 2DEG after positive gate bias stress is also shown. We represent the schematic degradation model of fluorinated-gate device in Figure 8b. At the Al 2 O 3 /AlGaN interface, the trapped charges in the border trap site and/or interface state induce hysteresis during DC I-V measurement. The fluorine ions ionized by accelerated electrons by positive gate bias, and formation of the Ga-F bonds induces the reduction in 2DEG at the AlGaN/GaN barrier. The capacitance-voltage (C-V) curve was measured before and after positive gate bias stress at 1 MHz, as shown in Figure 8a, and the C-V curve moved to a positive direction and the slope of the C-V curve lowered. AlGaN/GaN interface degradation and reduction in 2DEG after positive gate bias stress is also shown. We represent the schematic degradation model of fluorinated-gate device in Figure 8b. At the Al2O3/AlGaN interface, the trapped charges in the border trap site and/or interface state induce hysteresis during DC I-V measurement. The fluorine ions ionized by accelerated electrons by positive gate bias, and formation of the Ga-F bonds induces the reduction in 2DEG at the AlGaN/GaN barrier.

Conclusions
An E-mode AlGaN/GaN HEMT was fabricated employing a gate recess and CF4 plasma treatment, and the effect of fluorine in the device was investigated using an analysis of electrical characteristics. The fluorinated-gate device showed hysteresis during DC I-V measurement, which depends on the VD. The pulse I-V measurement was used to observe the charge trapping/detrapping

Conclusions
An E-mode AlGaN/GaN HEMT was fabricated employing a gate recess and CF 4 plasma treatment, and the effect of fluorine in the device was investigated using an analysis of electrical characteristics. The fluorinated-gate device showed hysteresis during DC I-V measurement, which depends on the V D . The pulse I-V measurement was used to observe the charge trapping/detrapping effect of fluorine, and the charging time is longer than a few ms. It was also figured out that the SS of the fluorinated-gate device increased due to the 2DEG reduction by fluorine ions and the formation of Ga-F bonds after positive gate bias stress. We highlight that the impact of fluorine ions should be taken into account for circuit design when the E-mode device is employed.