The fabrication processes of p-type regions for vertical GaN power devices are investigated. A p-type body layer in a trench gate metal-oxide-semiconductor field-effect transistor requires precise control of the effective acceptor concentration, which is equal to the difference between the Mg acceptor concentration (Na) and the compensating donor concentration (Nd). The carbon atoms incorporated during growth via metalorganic vapor phase epitaxy substitute nitrogen sites (CN) and function as donor sources in a p-type GaN layer. Since interstitial H atoms also compensate holes, their removal from an Mg-doped layer is crucial. Extended anneals to release H atoms cause the formation of extra hole traps. The p+ capping layer allows effective and rapid removal of H atoms from a p-type body layer owing to the electric field across the p+/p– junction. On the other hand, selective area p-type doping via Mg ion implantation is needed to control the electrical field distribution at the device edge. Ultrahigh-pressure annealing (UHPA) under a nitrogen pressure of 1 GPa enables post-implantation annealing up to 1753 K without thermal decomposition. Cathodoluminescence spectra and Hall-effect measurements suggest that the acceptor activation ratio improves dramatically by annealing above 1673 K as compared to annealing at up to 1573 K. High-temperature UHPA also induces Mg atom diffusion. We demonstrate that vacancy diffusion and the introduction of H atoms from the UHPA ambient play a key role in the redistribution of Mg atoms.
I. INTRODUCTION
The great success of GaN-based light-emitting devices owes much to the invention of p-type conduction.1–3 Although magnesium (Mg) is the only shallow acceptor for GaN, it is difficult to achieve a high free hole concentration owing to two major limitations. One is the relatively large ionization energy of Mg acceptors, resulting in a much lower free hole concentration than the acceptor concentration at room temperature.4 The other limitation is that the high Mg doping needed to reduce the ionization energy leads to a significant reduction in free hole concentration,5 which stems from Mg segregation in the polarity inversion domains (PIDs).6–8 In the case of light-emitting diodes (LEDs), in addition to the engineering of a device structure for enhancing hole injections, improved growth technology to avoid the formation of PIDs has also contributed to their high efficiency.9
p-type formation is key for GaN-based power devices, which have attracted attention because of their high figure-of-merit (FOM).10 The p-type gates in AlGaN/GaN heterostructure field-effect transistors (HFETs) have exhibited a normally off operation and a low channel resistance by injecting holes from the gate.11,12 Even in a unipolar vertical power device, the p-type regions function as critical components. Furthermore, a p-type formation process that meets the demands for device design is a significant challenge in a vertical power device.
Figure 1 illustrates a trench-gate metal-oxide-semiconductor field-effect transistor (T-MOSFET) and the challenges related to p-type formation. The threshold voltage of a MOSFET is determined by the effective acceptor concentration Na (not free hole concentration) in the p-type body layer. To obtain the optimal threshold voltage (3–10 V), precise control of the effective Na in the 1016–1018 cm−3 range is needed.13 Moreover, Mg atoms in the body layer have a preferred uniform distribution in the depth direction to avoid the punch-through under a high drain bias. Such a lightly Mg-doped layer with uniform doping is typically not used in optical devices and therefore needs to be developed for a vertical power device.
Cross-sectional schematic of a T-MOSFET. A p-type body layer uniformly doped with Mg below 1018 cm−3 can be formed by the MOVPE method. Selective-area doping for the edge termination will be achieved by Mg ion implantation. The p+ body contact can be formed by either MOVPE growth or Mg ion implantation.
Cross-sectional schematic of a T-MOSFET. A p-type body layer uniformly doped with Mg below 1018 cm−3 can be formed by the MOVPE method. Selective-area doping for the edge termination will be achieved by Mg ion implantation. The p+ body contact can be formed by either MOVPE growth or Mg ion implantation.
The effective Na is also reduced by donor-like deep levels, which can lead to variations in threshold voltage. Thus, such compensation sources should be identified and minimized. Hydrogen (H) passivation of Mg acceptors also reduces the effective Na. However, the process of H atom release is quite different from that in the surface structures of devices. An n+ capping layer such as a tunnel junction makes it difficult to remove H atoms from a buried p-type layer.14 The process flow of device fabrication needs to be designed by considering how to remove H atoms. As seen in Fig. 1, T-MOSFETs require both an n+ source contact and a p+ body contact to stabilize the threshold voltage and remove holes generated by the avalanche event.15 If the n+ source is formed by epitaxial growth and the p+ region is fabricated by ion implantation, which forms the buried p-type layer like a tunnel junction, the H atoms will be removed through a pass such as a trench sidewall. H atom removal from the trench sidewall requires a relatively long annealing time,16 which can introduce trap states in a p-type body layer.
Selective area p-type doping is also crucial for the fabrication process of T-MOSFETs and can be applied to edge termination, such as a junction termination extension (JTE) structure.17 Without edge termination, the advantage of a high critical electric field for GaN would be reduced. However, p-type formation via Mg ion implantation is a major challenge in the fabrication of a vertical GaN power device. Post-implantation annealing for removing implantation damage and activating Mg acceptors leads to surface decomposition of GaN, which imposes an upper limit on the annealing temperature resulting in insufficient activation. Ultrahigh-pressure annealing (UHPA)18 above the equilibrium N2 partial pressure in the GaN-Ga-N2 system is one approach to removing lattice distortion and substituting Ga sites by Mg atoms without decomposition at high temperature. On the other hand, the high temperature process can cause atomic diffusion. The implantation profile and annealing process must be designed on the basis of the redistribution of impurities.
Here, we present our approaches to overcoming these challenges related to p-type formation. In Sec. II, we discuss the growth of a lightly Mg-doped GaN layer by metalorganic vapor phase epitaxy (MOVPE) and demonstrate that carbon atoms compensate holes in a p-type GaN layer like the electron compensation in n-type GaN. Then, the effective removal of H atoms, which compensate holes, using the p+ capping layer is discussed on the basis of the desorption mechanism. In the case of removing from buried p-type layers without a p+ capping layer, the long anneals are needed and form the extra deep levels. In Sec. III, we summarize the current level in the Mg ion implantation process involving UHPA, where post-implantation annealing above 1673 K plays a key role in obtaining a high activation ratio. We also focus on Mg atom diffusion in relation to the introduction of H atoms from the UHPA chamber.
II. FORMATION OF p-TYPE GaN BY MOVPE
A. Light Mg doping
Figure 2 illustrates the Mg supply system of our lateral-flow MOVPE equipment. Bis(cyclopentadienyl) magnesium (Cp2Mg) is used as a dopant gas. A vertical power device requires a p-type body layer and a p+ contact layer doped with Mg at concentrations of around 1017 and 1020 cm−3, respectively. Such a wide doping range cannot be covered by the dynamic range of a single mass flow controller (MFC), which is two orders of magnitude. Thus, we prepared two gas lines for Cp2Mg to cover the desired dynamic range. In addition, the lower doping gas line was equipped with a double-dilution system, as shown in Fig. 2. The 1st MFC supplied the carrier gas with a sufficient flow rate (typically 100 sccm), while the Cp2Mg concentration and flow rate were controlled by the 2nd MFC for dilution and the 3rd MFC connected to the reactor, respectively. This system allows precise control of the effective flow rate of the Mg source gas in the lower doping range. The dilute concentration was measured by an in situ Fourier-transform infrared (FTIR) monitor. Within 5 min of starting the gas supply, the FTIR signal reached a constant level, as seen in Fig. 3(a). The linear relationship between the FTIR signal and the dilute ratio in Fig. 3(b) indicates that the effective flow rate of Cp2Mg is stably controlled.
Schematic of Cp2Mg supply lines in our MOVPE system. The gas line for lower Mg doping, which tentatively covers the doping range of 1016–1018 cm−3, is equipped with a double dilution unit of Mg source gas, where the flow rate of “MFC 1” is set to stably vaporize the Cp2Mg source at a constant level and the effective supply rate is controlled by the flow rates of “MFC 2” and “MFC 3.” The gas line for higher Mg doping allows the fabrication of a p-type contact layer.
Schematic of Cp2Mg supply lines in our MOVPE system. The gas line for lower Mg doping, which tentatively covers the doping range of 1016–1018 cm−3, is equipped with a double dilution unit of Mg source gas, where the flow rate of “MFC 1” is set to stably vaporize the Cp2Mg source at a constant level and the effective supply rate is controlled by the flow rates of “MFC 2” and “MFC 3.” The gas line for higher Mg doping allows the fabrication of a p-type contact layer.
Cp2Mg partial pressure monitored by the FT-IR signal intensity on the low-concentration side of the Mg supply system. (a) Cp2Mg partial pressure as a function of growth time, where the corresponding dilution ratio is 0.33. (b) Averaged Cp2Mg partial pressure as a function of the dilution ratio, where the circle corresponds to the dilute concentration of (a).
Cp2Mg partial pressure monitored by the FT-IR signal intensity on the low-concentration side of the Mg supply system. (a) Cp2Mg partial pressure as a function of growth time, where the corresponding dilution ratio is 0.33. (b) Averaged Cp2Mg partial pressure as a function of the dilution ratio, where the circle corresponds to the dilute concentration of (a).
Figure 4(a) shows the depth profiles of Mg atoms obtained by secondary ion mass spectrometry (SIMS) for the doped layer having a thickness of around 2 μm. The Mg concentrations were perfectly uniform in the depth direction. The doping profiles show abrupt onsets with widths of around 0.1 μm. Uniform and abrupt doping profiles are often prevented by the so-called memory effect, which results from the adsorption of Mg precursors to the reactor parts.19–22 It has been reported that the preflow of Cp2Mg to the reactor reduces the delay in Mg incorporation relative to the onset of Mg source supply and improves the doping profile.19 However, this approach makes it difficult to precisely control the Mg concentration with reproducibility. We realized the uniform and abrupt doping profiles without the use of a preflow process, by designing the heat zone in the reactor and the growth conditions so as to minimize the memory effect. A stable supply of a Mg source and reduction of the memory effect allow linear control with respect to the effective flow rate of Cp2Mg as shown in Fig. 4(b).23 By combining higher and lower doping gas lines, a wide Mg doping range (1016–1020 cm−3) was achieved.
(a) Mg concentration in GaN as a function of depth estimated by SIMS measurement, where the gas line of low-concentration side was used. (b) Averaged Mg concentration as a function of Cp2Mg/TMGa ratio. Linear doping control was achieved in the 1016–1020 cm−3 range via Cp2Mg supply. Modified with permission from Narita et al., J. Appl. Phys. 124, 165706 (2018). Copyright 2018 AIP Publishing LLC.
(a) Mg concentration in GaN as a function of depth estimated by SIMS measurement, where the gas line of low-concentration side was used. (b) Averaged Mg concentration as a function of Cp2Mg/TMGa ratio. Linear doping control was achieved in the 1016–1020 cm−3 range via Cp2Mg supply. Modified with permission from Narita et al., J. Appl. Phys. 124, 165706 (2018). Copyright 2018 AIP Publishing LLC.
Such a precise and wide doping range can be applied to not only the formation of a channel region in a T-MOSFET but also the edge termination of a p-n junction. Ohta et al. demonstrated p–n diodes with an avalanche capability by utilizing the punch-through breakdown of p-type GaN layers, which essentially requires precise and uniform Mg doping of around 3 × 1017 cm−3.24 Maeda et al. proposed the negative beveled-mesa termination, where the electrical distribution can be controlled by the bevel angle and the ratio of acceptor and donor concentrations (Na/Nd) in p-type and n-type layers, respectively.25,26 For example, if we form a bevel angle of 10° by thermal reflow of the patterned photoresist, the electric field crowding at the device edge can be reduced to less than 5% by applying Na = 1017 cm−3 and Nd = 2.5 × 1016 cm−3,26 which allows the fabrication of the edge termination of 600 V-class devices. Such designs of the edge terminations require precise control of the effective Na, which is determined by not only the Mg concentration but also the net concentration of donor-like defects that compensate holes. Next, we will discuss the hole compensation sources in MOVPE-grown p-type layers.
B. Hole compensation due to carbon
Hole concentration p at 300 K (symbols) as a function of Mg concentration estimated by Hall-effect measurements on Mg-doped layers with the various carbon concentrations [C]. The dashed and solid lines were calculated on the basis of Eq. (2). Hole compensation due to carbon is clearly observed.
Hole concentration p at 300 K (symbols) as a function of Mg concentration estimated by Hall-effect measurements on Mg-doped layers with the various carbon concentrations [C]. The dashed and solid lines were calculated on the basis of Eq. (2). Hole compensation due to carbon is clearly observed.
Then, was used, which corresponds to the hole effective mass value of 2.07m0.37 The degeneracy factor g of valence band was 4. Based on Eq. (2), the hole concentrations p at 300 K were calculated for different Nd values (see dashed and solid lines in Fig. 5). The experimental hole concentrations are almost located on lines Nd = 0 or for samples with , whereas the p values are significantly reduced by carbon incorporation in the lower Mg doping range for the higher C-doped samples. Moreover, the experimental data lie on the line Nd ∼ [C]. This is indicative of the hole compensation due to C atoms in p-type GaN layers. A comparison of samples with close [Mg] values showed that carbon incorporation reduced not only the free hole concentration but also the mobility, as indicated in Table I. Mobility reduction due to carbon incorporation is significant at lower temperatures, where ionized impurity scattering strongly limits mobility.35 Therefore, incorporated carbon atoms mostly function as ionized donors in a p-type GaN layer.
Summary of representative p-type Hall devices with various [C] values, where [Mg] was within (7.8–8.5) × 1016 cm−3. Carbon incorporation reduced the free hole concentration p and mobility μ.
Sample ID . | [Mg] × 1016 (cm−3) . | [C] × 1016 (cm−3) . | p at 300 K × 1016 (cm−3) . | μ at 300 K (cm2/V s) . | μ at 170 K (cm2/V s) . |
---|---|---|---|---|---|
#2917 | 8.1 | 0.20 | 1.75 | 28.1 | 124.7 |
#3248 | 7.8 | 2.9 | 0.49 | 25.3 | 71.4 |
#3162 | 8.5 | 4.8 | 0.38 | 24.2 | 64.8 |
#3247 | 7.8 | 5.5 | 0.15 | 23.6 | 16.2 |
Sample ID . | [Mg] × 1016 (cm−3) . | [C] × 1016 (cm−3) . | p at 300 K × 1016 (cm−3) . | μ at 300 K (cm2/V s) . | μ at 170 K (cm2/V s) . |
---|---|---|---|---|---|
#2917 | 8.1 | 0.20 | 1.75 | 28.1 | 124.7 |
#3248 | 7.8 | 2.9 | 0.49 | 25.3 | 71.4 |
#3162 | 8.5 | 4.8 | 0.38 | 24.2 | 64.8 |
#3247 | 7.8 | 5.5 | 0.15 | 23.6 | 16.2 |
We found the CN 0/–1 level at an energy of EV + 0.88 eV in p-type GaN layers, i.e., the Hd hole trap, through deep-level transient spectroscopy (DLTS) measurements,35 which was identical to the H1 hole trap in n-type GaN.38–44 This indicates that C atoms incorporated during MOVPE growth most likely substitute nitrogen sites. The Ha hole trap at EV + 0.29 eV was discovered coincidentally, and its concentration exhibited a nearly one-to-one relationship with the Hd concentration.35 This can be explained by the fact that CN has two different charged states. Recent first-principles calculations based on the hybrid functional approach have predicted that CN has not only a 0/–1 state at EV + (0.90–0.93) eV but also a + /0 state with a donor-like nature at EV + (0.25–0.35) eV.32–34 Since the Fermi level is located near the valence band in p-type GaN, a hole becomes trapped at the CN + /0 level, resulting in forming a positive space charge. That is, CN serves as a carrier compensation center in both n- and p-type GaN layers. In the case of a CN acceptor in n-type GaN, the energy level is far from the conduction band minimum, and thus it can act as a fixed charge. The CN donor-like state, however, has a relatively shallow energy of EV + 0.29 eV. The trap parameter of this level45 can give an emission time constant of around 20 ns at room temperature based on the relation of , where and g′ were the thermal velocity of a hole and the degeneracy factor of unity for a hole trap, respectively. This may affect the switching operation by alternating between capture and emission of holes.
C. Impact of capping layer in hydrogen removal from p-type GaN
H atoms can substitute interstitial sites in GaN and function as compensation donors by trapping holes.46 Moreover, MOVPE growth in a hydrogen ambient most likely yields Mg–H complexes owing to their low formation energy,46 and the removal of H atoms after growth is crucial for activating Mg acceptors.2,3 Kuwano et al. reported the difficulty of H atom removal in an LED with a tunnel junction whose surface was capped by a n+-GaN layer.14 Since vertical GaN power devices have various surface structures such as n+-GaN in the source region, p+-GaN in the body contact, and p–-GaN in the channel region, the impact of capping layers on H atom removal needs to be fully understood.
Figure 6 shows the structures of trial samples with various capping structures. Sample #A is composed of a single Mg-doped GaN layer grown by MOVPE on a freestanding GaN substrate. Sample #B has a 0.46-μm-thick n+ capping layer doped with silicon (Si) at a concentration of 3 × 1018 cm−3 on a Mg-doped layer, while sample #C has a heavily Mg-doped capping layer with a Mg concentration of 3 × 1019 cm−3. The concentrations of single and underlying Mg-doped layers are within the (0.6–3) × 1018 cm−3 range. All samples were annealed at 1123 K in nitrogen gas under atmospheric pressure for 1–30 min. The H concentrations before and after annealing were measured by SIMS in the depth direction (see Fig. 7).
Schematics of sample structures for hydrogen removal trials. MOVPE growth of Mg-doped layers and capping layers was performed on freestanding GaN substrates. Concentrations of single and underlying Mg-doped layers were within (0.6–3) × 1018 cm−3.
Schematics of sample structures for hydrogen removal trials. MOVPE growth of Mg-doped layers and capping layers was performed on freestanding GaN substrates. Concentrations of single and underlying Mg-doped layers were within (0.6–3) × 1018 cm−3.
Depth profiles of H atoms estimated by SIMS measurements for samples (a) #A, (b) #B, and (c) #C. The annealing time at 1123 K in nitrogen ambient was varied between 1 and 30 min. H atom profiles in as-grown states corresponded to Mg profiles owing to the formation of Mg–H complexes.
Depth profiles of H atoms estimated by SIMS measurements for samples (a) #A, (b) #B, and (c) #C. The annealing time at 1123 K in nitrogen ambient was varied between 1 and 30 min. H atom profiles in as-grown states corresponded to Mg profiles owing to the formation of Mg–H complexes.
The H concentrations in all Mg-doped layers before annealing were almost consistent with the Mg concentrations. For sample #A without a capping layer, H concentrations gradually decreased with increasing annealing time, as indicated in Figs. 7(a) and 8(a). Moreover, increased linearly with annealing time, as seen in Fig. 8(b). This behavior indicates a desorption process limited by the surface energy barrier, as suggested by Myers et al.47 The surface energy barrier presumably arises from the formation reaction of molecular H2 and the potential difference between the solid and gas phases. Note that the concentration of H atoms decreases uniformly in the depth direction in sample #A in Fig. 7(a). If the desorption process was diffusion-limited, the H atom concentration near the surface would be significantly reduced and the concentration at large depths would be much higher. A sufficiently large diffusion length corresponds to the low diffusion barrier (0.94 eV) of positively charged as predicted by the first-principles calculations using the local density approximation (LDA).48
(a) Normalized H atom concentration [H] as a function of annealing time, where the concentration in the as-grown state is defined as unity. (b) The inverse of [H] as a function of annealing time for sample #A. The linear relation indicates that the desorption process of H atoms in sample #A is limited by the surface energy barrier. (c) Schematic of H atom transport during the initial annealing stage in sample #C. The built-in electric field of p+/p– junction can transport atoms from the p– to the p+ side during the initial annealing stage.
(a) Normalized H atom concentration [H] as a function of annealing time, where the concentration in the as-grown state is defined as unity. (b) The inverse of [H] as a function of annealing time for sample #A. The linear relation indicates that the desorption process of H atoms in sample #A is limited by the surface energy barrier. (c) Schematic of H atom transport during the initial annealing stage in sample #C. The built-in electric field of p+/p– junction can transport atoms from the p– to the p+ side during the initial annealing stage.
For sample #B with n+ capping layers, hardly any H atoms were removed by annealing up to 90 min.16 This can be explained by the high diffusion barrier (1.99 eV) of negatively charged in an n-type GaN layer.48 We previously estimated a very small diffusion coefficient for at 1123 K.16 In our previous study, we observed a small effective Na of sample #B after annealing, which implies hole compensation due to the donor.16 Thus, we tried H atom removal from the etched sidewall. However, this process requires a relatively long annealing time owing to the diffusion limit because the width of the p-type region in actual devices is larger than its thickness.16 When a mesa with a diameter of around 50 μm is formed for the n+/p–/n+ structure, H atoms persisted at a concentration of around 15% even after annealing at 1123 K for 90 min.16
For sample #C with heavily Mg-doped capping layers, the H concentrations rapidly decreased during the first 5 min of the anneal, as seen in Fig. 8(a). This can be explained by the diffusion of atoms from the p– layer to the p+ layer as shown in Fig. 8(c). The p+/p– junction forms an electric field due to a potential difference within a few Debye lengths, which enhances the transport of H atoms in the initial annealing stage. After the redistribution of atoms, the built-in potential can be canceled by the electric field formed through the redistribution of atoms, resulting in a gradual decrease of H atoms in the subsequent stage, as seen in Fig. 8(a).
As discussed in Sec. II D, a long anneal for the n+/p–/n+ structure with mesa leads to the introduction of extra deep levels in the p-type GaN layer. Therefore, the formation of an n+ capping layer in the hydrogen gas ambient should be avoided during device fabrication. This means that it would be more preferable to fabricate the n+ source contacts of the T-MOSFETs in Fig. 1 by a method other than MOVPE growth. Furthermore, the p+ capping layer enhances the release of H atoms and utilizes the fabrication process of T-MOSFETs. Thus, from the standpoint of effective removal of H atoms, it is more appropriate to anneal the p-type body layer after fabricating the p+ body contact.
D. Deep levels formed through long anneals
We previously employed DLTS measurements in homoepitaxial p-type GaN layers, where six hole traps labeled Ha through Hf and an electron trap labeled E3′ were detected.45,49 On the basis of quantitative relationships with [C], Ha, and Hd at EV + 0.29 and 0.88 eV were assigned to charged states CN + /0 and 0/–1, respectively.35,45 The E3′ trap at an EC –0.57 eV in p-type GaN was identical to the E3 trap in n-type GaN,49 whose origin was recently identified to be an iron atom at the Ga site (FeGa).50 However, the origins of the other four hole traps are still unclear. In addition, Polyakov et al. found hole traps closer to the valence band detected by admittance spectroscopy (G/ω) in heteroepitaxial p-type GaN layers.51 Here, we focused on the shallow hole traps detected by G/ω measurements in our homoepitaxial p-type samples. Then, the annealing times for H atom removal were varied to help identify unknown hole traps. Prolonged annealing is needed to remove H atoms from an n+/p–/n+ structure like a buried Mg-doped body layer beneath the n+-GaN source region formed by MOVPE, which is important for designing the fabrication process of T-MOSFETs.
The p+/p–/n+ diodes used to detect hole traps were fabricated by MOVPE growth, annealing in nitrogen, dry etching for isolation, and formation of ohmic electrodes, in that order. This diode structure reduces junction leakage and series resistance unlike a p– Schottky diode, giving a better signal/noise ratio in deep-level analysis.49,52 Details of the sample fabrication process can be found in Ref. 49. Anneals to release H atoms were performed at 1123 K for 5 min and 300 min. Capacitance–temperature (C–T) and G/ω measurements were conducted over frequency and temperature ranges of 50 Hz–2 MHz and 80–400 K, respectively. MOVPE was used to prepare 2-μm-thick p– layers in the Mg doping range of (0.22–19) × 1017 cm−3, having [C] values within the (2.3–4.0) × 1015 cm−3 range.
Figure 9 shows C–T and G/ω-T curves at 2 and 200 kHz for samples doped with Mg at a concentration of 1.0 × 1018 cm−3. The capacitance drops at 100 and 120 K at frequencies of 2 and 200 kHz, respectively, are due to the un-ionization of Mg acceptors in the neutral region at low temperature, which appears as H0 peaks in the G/ω spectra. A significant decrease in capacitance was observed between 100 and 220 K at 2 kHz in the case of samples annealed for 300 min [see Fig. 9(a)], which was attributed to the SHD trap in the G/ω spectrum. The SHD trap emerged as a result of prolonged annealing and was not observed in the sample annealed for 5 min. We also found SHC traps around 130 and 160 K at frequencies of 2 and 200 kHz, respectively. The SHC trap concentration did not depend on the annealing time, as seen in Table II. Furthermore, the SHE trap was observed at 200 kHz around 250 K and was higher in the sample annealed for 5 min. Figure 10 shows the Arrhenius plots for SHC, SHD, and SHE, whose parameters were extracted (see Table II) assuming the hole effective mass of 0.9m0. Hole traps close to the SHC at EV + 0.106 eV were reported in Mg-doped p-type GaN, which seems to be related to Mg acceptors.51,53 The SHC concentrations were almost 10% of the [Mg] values. The concentration of the SHE trap at EV + 0.051 eV seems to have a correlation with the residual [H] values, as seen in Fig. 11. Lyons et al. theoretically predicted the +/0 transition level of a Mg–H complex at EV + 0.13 eV. Thus, the SHE trap is most likely attributable to the Mg–H + /0 level.
(a) and (b) Zero-bias capacitance as a function of temperature. (c) and (d) G/ω spectra as a function of temperature. The measurement frequencies are 2 kHz for (a) and (c) and 200 kHz for (b) and (d). Mg concentration for the as-grown sample was estimated to be 1.0 × 1018 cm3 by SIMS. Annealing times at 1123 K in nitrogen gas were 5 min (black lines) and 300 min (red lines). The significant decrease of capacitance for samples annealed for 300 min is mainly due to the formation of trap SHD whose concentration is above 1017 cm−3 (see Table II).
(a) and (b) Zero-bias capacitance as a function of temperature. (c) and (d) G/ω spectra as a function of temperature. The measurement frequencies are 2 kHz for (a) and (c) and 200 kHz for (b) and (d). Mg concentration for the as-grown sample was estimated to be 1.0 × 1018 cm3 by SIMS. Annealing times at 1123 K in nitrogen gas were 5 min (black lines) and 300 min (red lines). The significant decrease of capacitance for samples annealed for 300 min is mainly due to the formation of trap SHD whose concentration is above 1017 cm−3 (see Table II).
Arrhenius plots of traps SHC, SHD, and SHE. Black triangles and red circles are data from samples annealed for 5 and 300 min, respectively. The extracted trap parameters are listed in Table II.
Arrhenius plots of traps SHC, SHD, and SHE. Black triangles and red circles are data from samples annealed for 5 and 300 min, respectively. The extracted trap parameters are listed in Table II.
Concentration of hole trap SHE as a function of residual H atom concentration estimated by SIMS.
Concentration of hole trap SHE as a function of residual H atom concentration estimated by SIMS.
Summary of trap parameters extracted from admittance measurements at zero bias for p+/ p−/n+ diodes annealed at 1123 K for 5 and 300 min, where σp and NT are the capture cross section of holes and the trap concentration, respectively. The Mg concentration in the p− layer for the as-grown sample was 1.0 × 1018 cm3. The [H] values for the p− layers were estimated by SIMS measurements.
Anneal time (min) . | Trap label . | Level (eV) . | σp (cm2) . | NT (cm−3) . | [H] (cm−3) . |
---|---|---|---|---|---|
5 | SHC | EV + 0.190 | 9.9 × 10−15 | 1.0 × 1017 | … |
SHE | EV + 0.051 | 9.8 × 10−20 | 1.2 × 1017 | 9.7 × 1016 | |
300 | SHC | EV + 0.187 | 1.3 × 10−14 | 1.1 × 1017 | … |
SHD | EV + 0.106 | 4.7 × 10−20 | 2.3 × 1017 | … | |
SHE | EV + 0.051 | 8.8 × 10−20 | 2.5 × 1016 | 8.2 × 1015 |
Anneal time (min) . | Trap label . | Level (eV) . | σp (cm2) . | NT (cm−3) . | [H] (cm−3) . |
---|---|---|---|---|---|
5 | SHC | EV + 0.190 | 9.9 × 10−15 | 1.0 × 1017 | … |
SHE | EV + 0.051 | 9.8 × 10−20 | 1.2 × 1017 | 9.7 × 1016 | |
300 | SHC | EV + 0.187 | 1.3 × 10−14 | 1.1 × 1017 | … |
SHD | EV + 0.106 | 4.7 × 10−20 | 2.3 × 1017 | … | |
SHE | EV + 0.051 | 8.8 × 10−20 | 2.5 × 1016 | 8.2 × 1015 |
The SHD level was evidently associated with prolonged annealing, which resulted in significant temperature and frequency dependences of capacitances, owing to the high trap concentration. This can lead to variations in the threshold voltage of MOSFETs during operation. Accordingly, prolonged annealing to release H atoms from the p-type layer needs to be avoided. As discussed in Sec. II C, p+ capping is effective in removing H atoms through relatively short anneals. These trials reveal important principles for the fabrication of T-MOSFETs.
III. FORMATION OF p-TYPE GaN VIA Mg ION IMPLANTATION
A. Activation of Mg acceptors by UHPA
The main problem of Mg ion implantation (Mg-I/I) is the need to elevate the annealing temperature for activation, which causes surface decomposition in GaN. To prevent decomposition during post-implantation annealing, conventional rapid thermal annealing (RTA) with a SiN or AlN protective layer,54–57 multicycle RTA,58–60 and capless annealing for N-polar surfaces61–65 have been attempted, all of which allow annealing up to 1573 K. Indeed, Mg activation occurred at annealing temperatures above 1473 K, as revealed by the rectification properties of p-n junctions54,61,64 and/or the donor–acceptor pair (DAP) bands in luminescence spectra.55,58,59,61–63 However, the optimal annealing temperature for a sufficient activation ratio in GaN is presumably much higher than 1573 K because the activation temperature is found to be roughly 70% of the melting point for other compound semiconductors.65 Recently, we proposed UHPA as the post-implantation annealing method, which allows annealing at up to 1753 K under 1 GPa. In this section, we discuss the optimal annealing temperature for Mg-ion-implanted GaN samples subjected to UHPA.
Unintentionally doped (UID) GaN grown by hydride-VPE (HVPE) and MOVPE-grown n-type GaN doped with a Si concentration of 2 × 1016 cm−3 were used as host materials for Mg-I/I. Mg ions were implanted with multiple energies to obtain a 300-nm-deep box-shaped profile of 1019 cm−3, as described in Ref. 18. After implantation, UHPA was performed on the MOVPE host materials at 1573 and 1673 K (samples #A and #B, respectively), and on the HVPE hosts at 1673 and 1753 K (samples #C and #D, respectively) for 5 min. As discussed in Sec. III B, UHPA leads to Mg diffusion into deeper positions.18 The effective Mg concentrations after annealing were estimated to be 7.5 × 1018, 2.3 × 1018, 2.2 × 1018, and 2.0 × 1018 cm−3 for samples #A, #B, #C, and #D, respectively.
Figure 12 shows the cathodoluminescence (CL) spectra at 10 K for Mg-I/I GaN samples annealed at various temperatures. For comparison, a p-type GaN sample doped with a Mg concentration of 2 × 1018 cm−3 was prepared by MOVPE growth on a freestanding GaN substrate. The common sharp peak at 3.467–3.468 eV is attributed to the excitonic transition bound to a neutral acceptor (A0X).66 A DAP band with the zero-phonon line (ZPL) at ∼3.28 eV is also commonly observed. Both signals indicate the formation of acceptors in Mg-I/I GaN samples through UHPA. The sharp peaks at 3.39–3.42 eV have been reported in some photoluminescence studies on Mg-doped GaN, although their origin remains unclear.66,67 The intensities of the DAP bands, a signature of acceptor formation, were enhanced on elevating the temperature from 1573 (#A) to 1673 K (#B). This change seems to be larger than the improvement between 1673 (#C) and 1753 K (#D). The intensities of both the A0X peak and DAP band were lower than the intensity of the MOVPE-grown Mg-doped GaN peak, indicating that non-radiative recombination centers persist in high densities in Mg-I/I GaN even after UHPA.
CL spectra at 10 K, where the acceleration voltage and beam current density were 5 kV and 8.3 × 10−5 Acm−2, respectively. A0X peaks and the LO-phonon lines were observed at energies of 3.467–3.468 eV and 3.374–3.476 eV, respectively. Zero-phonon lines (ZPLs) of DAP bands were found at around 3.277–3.278 eV.
CL spectra at 10 K, where the acceleration voltage and beam current density were 5 kV and 8.3 × 10−5 Acm−2, respectively. A0X peaks and the LO-phonon lines were observed at energies of 3.467–3.468 eV and 3.374–3.476 eV, respectively. Zero-phonon lines (ZPLs) of DAP bands were found at around 3.277–3.278 eV.
Figure 13 shows the hole concentrations at 300 K for the Mg-I/I samples. The data for the MOVPE-grown p-type GaN samples were plotted as a function of Mg concentration. The lines in Fig. 13 were calculated on the basis of Eq. (2), assuming different compensation ratios and activation ratios. We could not obtain the p value for sample #A annealed at 1573 K owing to insufficient activation and/or a poor ohmic contact. On the other hand, the Hall-effect measurements were successful for the samples annealed at 1673 K (#B and #C) and 1753 K (#D),68 which agrees with the dramatic improvement observed in the CL spectra at the corresponding temperatures in Fig. 12. This indicates that the threshold temperature for the improvement is between 1573 and 1673 K. On the other hand, the p values for samples #B, #C, and #D at room temperature were estimated to be 3.6 × 1016, 4.6 × 1016, and 6.8 × 1016 cm−3, respectively, which indicates that an enhancement in activation and/or reduction in hole-compensating donors occurs on elevating the temperature from 1673 to 1753 K. Even after UHPA at 1753 K, the p value at 300 K was slightly lower than that of the MOVPE-grown p-type GaN. The calculated data in Fig. 13 suggest that the compensation donors and/or inactive Mg atoms partly remain under the present annealing conditions.
Hole concentration p at 300 K (symbols) as a function of Mg concentration estimated by Hall-effect measurements. Closed symbols are from Mg-ion-implanted samples #B, #C, and #D after UHPA, while open circles show data from MOVPE-grown Mg-doped GaN samples. Solid, dotted, and dashed lines are calculated by Eq. (2), assuming different activation ratios and compensation ratios.
Hole concentration p at 300 K (symbols) as a function of Mg concentration estimated by Hall-effect measurements. Closed symbols are from Mg-ion-implanted samples #B, #C, and #D after UHPA, while open circles show data from MOVPE-grown Mg-doped GaN samples. Solid, dotted, and dashed lines are calculated by Eq. (2), assuming different activation ratios and compensation ratios.
The structural change of extended defects in Mg-I/I GaN through UHPA was recently discussed by Iwata et al.69 Their Mg-I/I sample annealed at 1573 K exhibited a number of inversion domains along whose boundaries Mg atoms were segregated. This kind of defect most likely reduces Mg atom substitution at Ga sites and induces an inactive state. On the other hand, no Mg-segregated defects could be detected after UHPA at 1753 K. Moreover, vacancies agglomerated, forming intrinsic dislocation loops during UHPA at 1753 K, resulting in clearly defect-free regions.69 The changes in luminescence and electrical properties suggest that this dramatic structural change presumably occurs between 1573 and 1673 K. Since defect evolution also depends on annealing time,62 systematic investigations involving annealing temperature and time variations are currently under way. In addition, the Mg-ion-dosage dependence of the hole concentration needs to be determined over a wide range as in the case of the MOVPE-grown p-type GaN samples in Fig. 13.
B. Mg diffusion through UHPA
Impurity diffusion needs to be considered in the ion implantation process because the function of a device is based on its impurity profile after post-implantation annealing. In the case of silicon carbide, the diffusion coefficient of aluminum as a p-type dopant is extremely low.70 Thus, the device can essentially be designed on the basis of the implantation profile. In the case of Mg-I/I GaN, the change in the Mg profile due to annealing seems to be small up to 1573 K.57,71 By contrast, we found significant Mg diffusion due to a 5-min UHPA run in the range between 1573 and 1753 K.18 This means that the optimal UHPA temperature (presumably above 1673 K) for obtaining a high activation ratio can lead to the redistribution of the Mg profile. Furthermore, the MOVPE-grown p-type body layer is also exposed to the UHPA ambient during the fabrication of T-MOSFETs, as seen in Fig. 1, which may induce the diffusion of Mg atoms from the body layer to the n-type drift layer. Thus, we investigated the Mg diffusion induced by the UHPA process.
The 0.3-μm-thick layer doped with 1.4 × 1019 cm−3 Mg on a 1.6-μm-thick undoped GaN layer was grown on a freestanding GaN substrate by MOVPE. Since no annealing to remove H atoms was performed, the H atom profile in the as-grown material almost corresponded to the Mg atom profile owing to H passivation as seen in Fig. 14(a).72 We also prepared a 2-μm-thick undoped layer and performed Mg-I/I to form a 0.3-μm-thick box-shaped profile with [Mg] ≈ 1019 cm−3 as shown in Fig. 14(b).72 UHPA was performed at 1673 K under a N2 pressure of 1 GPa and the hold time at 1673 K was varied from 0.5 to 15 min. Figure 14 shows the depth profiles of Mg and H atoms before and after UHPA.72
Depth profiles of Mg atoms (closed symbols) and H atoms (open dots) before and after UHPA for (a) MOVPE-grown Mg-doped GaN and (b) Mg-I/I GaN sample. Both samples were annealed at 1673 K under a N2 pressure of 1 GPa. Diffusion profiles of H atoms (solid lines) were calculated by the electric-field-induced diffusion model, as reported previously.73 From Sakurai et al., Appl. Phys. Express 13, 086501. Copyright 2020 The Japan Society of Applied Physics.
Depth profiles of Mg atoms (closed symbols) and H atoms (open dots) before and after UHPA for (a) MOVPE-grown Mg-doped GaN and (b) Mg-I/I GaN sample. Both samples were annealed at 1673 K under a N2 pressure of 1 GPa. Diffusion profiles of H atoms (solid lines) were calculated by the electric-field-induced diffusion model, as reported previously.73 From Sakurai et al., Appl. Phys. Express 13, 086501. Copyright 2020 The Japan Society of Applied Physics.
We identified several features of the diffusion profiles. First, the H atom profiles were consistent with the Mg atom profile of the MOVPE-grown sample even after UHPA, as shown in Fig. 14(a). If the UHPA ambient were free of H atoms, H atoms were removed.3 We believe that the moisture in the UHPA equipment functioned as a source of H atoms.73 Furthermore, the introduction of H atoms was clearly observed in Mg-I/I samples after UHPA despite the absence of H atoms in the as-implanted sample [see Fig. 14(b)]. Thus, free holes as indicated in Fig. 13 were observed by removing H atoms after post-annealing to release H atoms at 1123 K in N2 ambient under atmospheric pressure.
The second feature of the diffusion profiles is the agreement between the profiles of Mg and H atoms near the diffusion edge. This is because the Mg–H complex has a lower formation energy than isolated MgGa and Hi atoms.46 Since Hi forms a bond with a Mg atom on a Ga site, the H atom profile can be an indicator of the MgGa profile.72 Indeed, the depth profiles of Mg and H atoms were perfectly consistent in the MOVPE-grown Mg-doped GaN, as seen in Fig. 14(a). By contrast, the H atom concentration in the box-shaped region of the Mg-I/I sample was much lower than the Mg concentrations [see Fig. 14(b)], indicating that a certain portion of the Mg atoms sits at sites other than Ga sites. As discussed in Sec. III B, the electrically inactive Mg atoms are most likely located at the inversion domain boundaries.69 We have previously reported an acceptor activation ratio of 78% for Mg-I/I GaN after UHPA at 1673 K for 5 min, which agreed with the [H]/[Mg] ratio of 75% for the same sample.72
Third, we found that the diffusion profile did not obey Fick's law. An abrupt diffusion profile appeared in the MOVPE-grown Mg-doped sample after UHPA. We previously explained this profile by the drift effect of atoms due to the electric field of the depletion layer,73 as illustrated in Fig. 15. That is, the atom diffusion can act as a driving force for Mg atom diffusion. One possible mechanism is that H atoms break the bonds of Ga atoms like a smart-cut process and create Ga vacancies (VGa), followed by Mg atom substitution of VGa sites. Further understanding of the diffusion mechanism requires confirmation via theoretical calculations.
The expected band diagram for Mg-ion-implanted UID GaN during UHPA at 1673 K, where the depletion electric field F can drift atoms toward the surface. The H atoms are likely derived from residual moisture in the UHPA chamber.
The expected band diagram for Mg-ion-implanted UID GaN during UHPA at 1673 K, where the depletion electric field F can drift atoms toward the surface. The H atoms are likely derived from residual moisture in the UHPA chamber.
The fourth feature is the enhanced diffusion in the Mg-I/I samples due to UHPA as compared with MOVPE-grown Mg-doped GaN. The diffusion coefficient for the Mg-I/I sample was around 30 times higher than that for the Mg-doped GaN grown by MOVPE,72 based on the electric-field diffusion model.73 Moreover, a significant diffusion of Mg and H atoms occurred within 1 min and while increasing or decreasing the temperature followed by slower diffusion in the subsequent annealing stage. This can be explained by vacancy-assisted diffusion.72 That is, VGa complexes created by Mg-I/I such as divacancies (VGa-VN) distribute into the deeper position than the box-shaped region of Mg atoms71 and partly diffuse by annealing, and thus Mg atoms substitute VGa complex sites. MgGa-vacancy complexes such as MgGa-VN likely dissociate because their binding energy is small, resulting in the formation of MgGa sites.74 Annealing also decreases the vacancy concentration,71 reducing the vacancy-assisted diffusion effect with increasing annealing time. This well explains the rapid Mg/H diffusion in the initial annealing stage, where the vacancy concentration is high, and the slower diffusion in the subsequent stage. Thus, vacancy diffusion plays a key role in controlling the Mg atom profile.
IV. SUMMARY
The fabrication of Mg-doped GaN via MOVPE, ion implantation, and post-implantation annealing was investigated in connection with vertical GaN power MOSFETs. Carbon impurities significantly compensate free holes in a p-type GaN layer by forming CN + /0 charged states and therefore should be suppressed at a low enough level, as compared with the Mg doping concentration. Since interstitial H atoms also compensate holes, the release process is crucial. A p+ capping layer effectively removed H atoms from a lightly Mg-doped layer during a short anneal, whereas other surface structures require relatively longer annealing times in an H-free ambient. Long anneals led to the formation of high-density traps, which significantly reduced the zero-bias capacitance. These results indicate that the p+ body contact formed by MOVPE benefits H atom removal from the p-type body layer. Accordingly, the device fabrication process involving MOVPE growth needs to be designed on the basis of dehydrogenation from the p-type body layer.
Vertical power devices also require selective-area p-type doping for the fabrication of edge termination. UHPA at 1 GPa is employed as a post-implantation anneal without thermal decomposition up to 1753 K. Free holes were observed in Mg-ion-implanted GaN after UHPA at temperatures above 1673 K. The CL spectra and Hall-effect measurements indicated that a dramatic improvement in the acceptor activation ratio occurred in the temperature range between 1573 and 1673 K. UHPA at above 1673 K caused Mg atom diffusion along with H atom diffusion. Moreover, the H atoms likely derived from residual moisture in the UHPA ambient and played a key role in the diffusion process. The Mg atom diffusion along the c-axis was enhanced in the Mg-ion-implanted GaN, which is indicative of vacancy-assisted diffusion. These findings are important for controlling the redistribution of Mg atoms through the UHPA process in device fabrication. Since the dosage of Mg ions and annealing time are expected to affect the device performance, in a future study we will investigate the dependency of acceptor activation and Mg diffusion on these conditions.
ACKNOWLEDGMENTS
This work was supported by MEXT “Research and development of next-generation semiconductor to realize energy-saving society” Program under Grant No. JPJ005357. Part of this research was also supported by the Polish National Science Centre through Project No, 2018/29/B/ST5/00338. The authors thank Dr. Iwinska (Unipress), Mr. Hirukawa (Nagoya University), Mr. Iwata (Nagoya University), Mr. Nakashima (Nagoya University), and Mr. Kogiso (Aichi Institute of Technology) for their experimental support. The authors would like to thank Professor Uedono (University of Tsukuba) and Professor Chichibu (Tohoku University) for valuable discussions.
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.