Impurity diffusion in ion implanted AlN layers on sapphire substrates by thermal annealing

We report on impurity diffusion in ion implanted AlN layers after thermal annealing. Silicon, tin, germanium, and magnesium ions were implanted into single-crystal Al-polar AlN layers grown on sapphire substrates. By annealing at 1600 °C, silicon and magnesium atoms were diffused in the AlN layer, while less change was observed in the distribution of germanium atoms. Silicon implantation introduced vacancy-related defects. By annealing at temperatures over 1300 °C, the vacancy-related defects were reduced, while oxygen atoms were diffused from the substrate due to sapphire decomposition. We reproducibly achieved silicon-implanted AlN layers with electrical conductance by controlling the annealing temperature and distribution of silicon and oxygen concentrations.


Introduction
Aluminum nitride (AlN) is an attractive material for deep ultra-violet light-emitting diodes and high-power devices due to its high critical electric field and band-gap energy of 6.1 eV. 1,2) AlN epitaxial layers are usually grown on different substrates due to the limited availability of bulk AlN. Highquality AlN layers are grown on sapphire and SiC substrates using metal-organic chemical vapor deposition (MOCVD) and molecular-beam epitaxy. 3,4) The doping concentrations in AlN layers are controlled by dopant incorporation during crystal growth. Silicon and germanium are available for n-type dopants, 5,6) while magnesium is used for a p-type dopant. 1) These crystal-growth technologies have significantly contributed to optical and electrical AlN devices. 7,8) However, the dopant incorporation during growth is limited to the vertical control of dopant profile. The establishment of an ion-implantation technology, which enables both lateral and vertical controls of dopant profile, could further expand the application spaces into the AlN devices with laterally different doping regions, such as metal-oxide-semiconductor field-effect transistors.
Ion implantation into AlN requires high-dose and highenergy ions because of the high atom density of AlN and the high ionization energy of dopants. 5) Bombardment with highenergy ions often kicks out atoms from lattice sites, residing the atom on an interstitial site while leaving vacancies behind. Some point defects and impurities with localized levels in bandgap act as carrier compensation centers, reducing carrier concentration. The interstitials and vacancies migrate during post-implantation annealing, causing selfinterstitial-vacancy recombination. The defects introduced by silicon implantation can be reduced by annealing at temperatures over 1200°C in a nitrogen ambient, resulting in n-type conductance. 9,10) Still, ion implantation into AlN suffers from low reproducibility of electrically conductive films. Recently, we found that silicon and oxygen atoms in the AlN layer diffuse at ∼1600°C. 11) Impurities commonly diffuse in the lattice by exchanging places with interstitials and vacancies in a process, called a point-defect mediated process, in which the diffusion rate of impurities depends on the point-defect concentration and the migration length of point defects. 12,13) The post-implantation annealing not only enhances the diffusion of impurity-defect complexes but also promotes unintentional incorporation of impurities from surface and substrate. Thus, precise diffusion control of point defects and impurities at high temperature is required to improve the reproducibility of electrically conductive AlN layers. In this paper, we report on the impurity diffusion in ion implanted AlN layers by thermal annealing.
2. Ion implantation of donor and acceptor impurities Into AlN layers 2.1. Experimental procedure We used 1.5 μm thick unintentionally-doped (UID) Al-polar AlN layers grown on c-plane sapphire substrates by MOCVD, which were supplied by DOWA Electronics Materials. [14][15][16] Silicon, germanium, tin, and magnesium ions were implanted into the as-grown AlN layers at room temperature by Ion Technology Center. In AlN, silicon and germanium atoms act as shallow donors on the aluminum site, 5,6) while a magnesium atom acts as a shallow acceptor on an aluminum site. 1) The amount and penetration depth of implanted ions was controlled by implantation dose and ionbeam energy, respectively. The desired box profile of dopants, i.e. the concentrations of 2 ± 1 × 10 19 cm −3 in 200 ± 100 nm deep box profiles, was formed by implanting the ions at several levels of ion-beam energy. All ions were implanted at the incident angle of 7°from [0001] to suppress the channel effect. After the ion implantation, the AlN layers were annealed at 1600°C for 30 min in a nitrogen ambient under atmospheric pressure, which was conducted by Taiyo Nippon Sanso (STA1800B), 17) to reduce the concentration of defects introduced by the ion implantation. The impurity concentrations in the ion-implanted AlN layers were experimentally determined using secondary-ion mass spectrometry (SIMS) performed by MST Foundation. The 15 keV cesium and 5 keV oxygen primary beams were used for negative and positive ion emissions, respectively. The detection limits for the concentration of hydrogen, carbon, and oxygen were 2 × 10 17 , 1 × 10 17 , and 1 × 10 18 cm −3 , respectively. For the silicon, germanium, and tin implanted AlN layers, Ti (20 nm)/Al (100 nm)/Ni (25 nm)/Au (50 nm) contacts were formed on 10 × 10 mm 2 samples by electron-beam (EB) deposition, followed by rapid thermal annealing (RTA) at 800°C for 30 s in a nitrogen ambient. For the magnesium implanted AlN layer, Ni (5 nm)/Au (5 nm) contacts were formed, followed by RTA at 500°C for 30 s in an oxygen ambient. The 200 nm deep mesa isolation was obtained by chlorine-based inductivity coupled plasma (ICP) reactive-ion etching (RIE) (SAMCO RIE-400iPS). 18) Electrical properties of these materials were determined at room temperature by a semiconductor parameter analyzer for the contact with 2 μm spacing.

Diffusion of dopant atoms in AlN layers after thermal annealing
Depth profiles of dopant concentrations in the silicon, germanium, tin, and magnesium implanted AlN layers are shown in Fig. 1. For silicon implantation, ion-beam energies of 90, 40, and 10 keV were used at a ratio of 64%, 24%, and 12%, respectively, in the total-implantation dose of 5 × 10 14 cm −2 . The as-implanted AlN layers exhibited a uniform silicon concentration in a 100 nm deep box profile, as shown in Fig. 1(a), indicating that the channeling effect was suppressed at this implantation angle. The measured average silicon concentration of 3 × 10 19 cm −3 corresponds well with the value calculated by SRIM Monte-Carlo simulations. After annealing at 1600°C, the silicon atoms were uniformly diffused in the AlN layer without clustering, forming a 500 nm deep box profile with a silicon concentration of 8 × 10 18 cm −3 , which is lower than the solidsolubility limit of silicon in AlN. 19,20) Before annealing, the concentrations of hydrogen, carbon, and oxygen were under detection limits. The oxygen concentration after annealing at 1250°C was under the detection limit, while that after annealing at 1600°C increased to 1 × 10 19 cm −3 . Singlecrystal Al 2 O 3 is decomposed at 1500°C in a helium ambient at the rate of ∼2 nm min −1 . 21) We consider that oxygen atoms diffused from the decomposed Al 2 O 3 substrate into the AlN layer during annealing at 1600°C. The silicon implanted AlN layer after annealing at 1250°C was electrically conductive at the 2 × 2 mm 2 area, while that after annealing at 1600°C was electrically insulative in the whole area.
For germanium implantation, ion-beam energies of 200, 90, and 20 keV were used at a ratio of 56%, 32%, and 12%, respectively, in the total implantation dose of 5 × 10 14 cm −2 . Even after annealing at 1600°C, the average germanium concentration was 3 × 10 19 cm −3 in an 80 nm deep box profile, which is close to the SRIM simulation of 3 × 10 19 cm −3 in a 200 nm deep box profile, as shown in Fig. 1(b). This suggests that the diffusion of germanium atoms in the AlN layers is slower than that of silicon atoms. After annealing at 1600°C, the concentrations of hydrogen, carbon, and oxygen were 1 × 10 18 , 2× 10 18 , and 2 × 10 19 cm −3 , respectively. These high impurity concentrations may be attributed to the high-energy ion bombardment of germanium. The germanium implanted AlN layer after annealing at 1600°C was electrically insulating.
For tin implantation, the ion-beam energies of 250, 120, and 30 keV were used at a ratio of 56%, 32%, and 12%, respectively, in the total implantation dose of 5 × 10 14 cm −2 . After annealing at 1600°C, the tin concentration showed a peak value of 2 × 10 18 cm −3 at a 50 nm depth from the surface, as shown in Fig. 1(c). The tin concentration is smaller than that of the SRIM simulation of 3 × 10 19 cm −3 in a 130 nm deep box profile. After annealing at 1600°C, the concentrations of hydrogen and carbon atoms were under detection limits, while the oxygen concentration was 5 × 10 18 cm −3 . The tin implanted AlN layer after annealing at 1600°C was electrically insulating.
For magnesium implantation, the ion-beam energies of 90, 40, and 10 keV were used at a ratio of 56%, 32%, and 12%, respectively, in the total implantation dose of 5 × 10 14 cm −2 . After annealing at 1600°C, the magnesium concentration showed the peak value of 6 × 10 18 cm −3 at a 100 nm depth from the surface, as shown in Fig. 1(d). This depth profile is significantly different from the SRIM simulation of 1 × 10 19 cm −3 in a 230 nm deep box profile. Magnesium atoms with a high concentration of 4 × 10 18 cm −3 were accumulated at the AlN/Al 2 O 3 interface. We anticipate that the fast diffusion of magnesium atoms in the AlN layer results from the interstitial mechanism. After annealing at 1600°C, the concentrations of hydrogen, carbon, and oxygen atoms were under detection limits. The magnesium implanted AlN layer after annealing at 1600°C was electrically insulating.
Assuming that all silicon donors are electrically activated, the electron concentration in the AlN layer with the silicon concentration of ∼10 19 cm −3 is estimated to be ∼10 17 cm −3 , 5) which is high enough to show electrical conduction. The silicon implanted AlN layers showed limited electrical conduction after annealing between 1200°C and 1400°C. We expected that higher electrical conductivity would be obtained by annealing at the increased temperature due to the improvement of crystalline quality 17) and the enhancement of selfinterstitial-vacancy recombination. 2,9,10) However, the silicon and germanium implanted AlN layers were electrically insulative after annealing at 1600°C. These results indicate that dopant species and dopant concentrations are not the dominant reason to inhibit the electrical conduction for these ionimplanted AlN layers.

Experimental procedures
We used ∼1 μm thick Al-polar AlN layers grown on c-plane sapphire substrates under nitrogen-rich conditions using MOCVD. The silicon implantation with the total dose of 5 × 10 14 cm −2 was performed for the AlN layers at room temperature for the incident angle of 7°from [0001]. The ionbeam energies of 90, 40, and 10 keV were used at a ratio of 64%, 24%, and 12%, respectively. After the ion implantation, the AlN layers were annealed through the face-to-face proximity configuration without any encapsulation caps by Taiyo Nippon Sanso. 17) The thermal annealing was carried out in flowing nitrogen gas at atmospheric pressure for 30 min at temperatures between 1100°C and 1700°C, which means the susceptor temperature was monitored using a pyrometer. We used the UID AlN layer with concentrations of carbon, oxygen, and silicon below 3 × 10 17 cm −3 as a reference sample. 22) The depth distributions of point defects in the AlN layers were determined by positron-annihilation spectroscopy (PAS). A positron mainly locates at the interstitial sites because of Coulomb repulsion from ion cores. For AlN, a positron is repelled from an isolated nitrogen vacancy V N , while a positron is localized in cation vacancies, such as an aluminum vacancy V Al and its complexes with V N . According to theoretical calculations, the triply negatively charged aluminum vacancy V Al 3− has the lowest formation energy in n-type AlN layers under nitrogen-rich conditions and can act as an effective compensating center for shallow donors. [23][24][25] The major species of point defects detected by positrons are neutral and/or negatively charged defects, such as V Al and vacancy clusters. We used monoenergetic positron beams to study the annealing behaviors and interaction of these vacancy-type defects. The PAS spectra were characterized by the S parameter, defined as the fraction of annihilation events over the energy range of 510.24-511.76 keV. Doppler broadening spectra of the annihilation radiation were measured with a germanium detector as a function of the incident positron energy E p .

Positron annihilation spectroscopy of Si implanted AlN
The S values of the silicon implanted AlN layers before and after annealing at various temperatures are shown as a function of E p in Fig. 2(a). The mean penetration depth of positrons for the AlN layers is shown on the upper horizontal axis. The total-implantation dose of silicon was 5 × 10 14 cm −2 . The increased S value at low E p < 1 keV is associated with the annihilation of positron and positronium at the surface, while the constant S value for 2 keV < E p < 10 keV corresponds to the annihilation of positrons in the AlN layer and the almost constant S value for E p > 25 keV corresponds to the annihilation of positrons in the sapphire substrate without diffusing back to the subsurface region. The S values of the as-implanted AlN layers at E p = 2 keV was 0.484, which is higher than that (∼0.47) of the as-grown AlN layers, indicating that V Al -related defects were introduced by the ion implantation. The S value of the AlN layer annealed at 1100°C had a peak of 0.521 near E p = 2 keV, which was higher than that of the as-implanted AlN layers. We consider that the annealing temperature of 1100°C is too low to reduce the V Al -related defects and/or allows to generate new V Al -related defects, causing the electrically insulative AlN layer. The S values at E p = 2 keV dramatically decreased and saturated at 0.459 by annealing at temperatures over 1200°C. 26,27) We consider that the concentration of V Al -related defects was reduced by annealing at temperatures over 1200°C due to self-interstitial-vacancy recombination, reducing the annihilation of positron trapped by V Al -related defects. After annealing at temperatures between 1300°C and 1500°C, the silicon implanted AlN layers reduced the S values and were electrically conducting. The low S values after annealing between 1300°C and 1500°C are attributed to the substitution of silicon for V Al and the decrease of the V Al -related defects due to self-interstitial-vacancy recombination. The silicon implanted AlN layer annealing at 1700°C was electrically insulating.
The observed S-E curves were fitted using s i The region exposed to positrons was divided into four blocks. The solid curves are well fitted to the experimental data, as shown in Fig. 2(a). The derived depth distributions of the S values for the silicon implanted AlN layers before and after annealing at various temperatures are shown in Fig. 2(b). For as-implanted and 1100°C-annealed AlN layers, the position of the first and second blocks with high S values, or less than ∼100 nm depth, agreed with the box profile of silicon, resulting from the ion bombardment and the high concentration of silicon. For all samples, the region with the S value over 0.459 had a depth of ∼900 nm, which corresponds to the AlN thickness determined by SIMS. The S value of 0.459 is slightly higher than that (=0.451) of the reference sample. These indicate that the defects were introduced in the whole AlN layers during crystal growth and/or by silicon implantation.

Experimental procedures
The UID Al-polar AlN layers with the thicknesses of 1-3 μm were grown on sapphire substrates by MOCVD. Silicon ions with a total dose of 5 × 10 14 cm −2 were implanted into the as-grown AlN templates at room temperature for the incident angle of 7°from [0001]. After pumping to a high vacuum of 5 × 10 −4 Pa in the chamber to suppress oxygen incorporation from the surface, the silicon-implanted AlN layers were annealed between 1400°C and 1800°C for 30 min in a nitrogen ambient under a vacuum of 1 × 10 4 Pa. The impurity concentrations in the silicon-implanted AlN layers were experimentally determined using SIMS. The Ti/Al/Ni/ Au contact was formed by EB deposition, followed by RTA at 800°C for 30 s in a nitrogen ambient. The 200 nm deep mesa isolation was obtained by chlorine-based ICP-RIE. The electrical property was determined by a semiconductor parameter analyzer for the contact with 2 μm space.

Dependences of Si diffusion on annealing temperatures
Silicon concentration depth profiles in the silicon implanted AlN layers with 1 μm thickness before and after annealing is shown in Fig. 3(a). The AlN layer annealed at 1200°C and the as-implanted AlN layer had the average silicon concentration of 3 × 10 19 cm −3 in a 100 nm deep box profile. The silicon concentrations in the AlN layers annealed at 1400°C, 1600°C, and 1800°C were 2 × 10 19 , 1 × 10 19 , and 7 × 10 18 cm −3 , respectively. This indicates that silicon diffusion was enhanced by increasing the annealing temperature. The AlN layer annealed at 1800°C formed cracks due to the different thermal expansion coefficient of AlN and Al 2 O 3 . 28) After annealing at 1800°C, the depth from the surface to the where D s is the diffusion coefficient of silicon, Q s is the total amount of silicon atoms per area, and t is the annealing time. As shown in Fig. 3(b), the temperature dependence of D s is described well by the equation where D s0 is the pre-exponential factor of silicon diffusion, E sa is the activation energy of silicon diffusion, k b is Boltzmann's constant, and T is the annealing temperature. The experimental data fitted well in the region of the silicon concentration > 10 19 cm −3 , while there is a discrepancy between experiment data and fitting in the region of silicon concentration < 10 19 cm −3 . Impurity diffusion can depend on the concentration of charged point defects due to the Fermi-level effect. 13,29) The PAS measurement showed that the V Al -related defects are present in the whole AlN layers. D s may depend on the transient gradients of charged point-defects near the silicon-implanted region. Further investigation is necessary to obtain better fitting to the experimental data. The value of D s0 (=4 × 10 −10 cm 2 s −1 ) in the AlN layer is much smaller than that (= 10 −4 -10 −3 cm 2 s −1 ) in GaAs 13,30) and that (= 9 × 10 −8 cm 2 s −1 ) in GaN, 31) indicating the slower diffusion of silicon atoms in AlN.
Silicon atoms in AlN prefer the substitutional position of the aluminum sublattice, forming substitutional defects (Si Al ). There are two major mechanisms in the interstitialsubstitutional diffusion; the dissociative and the kick-out mechanisms. 29,32) In the dissociative mechanism, mobile interstitial silicon atoms (Si i ) occupy vacancies on the aluminum sublattice (V Al ) through the reaction of  + V Si Si .
i Al Al V Al -mediated diffusion of silicon atoms occurs when the exchange rate between Si i and V Al is faster than that between Al i and V Al . In the kick-out mechanism, the interstitial-substitutional exchange of silicon atoms creates an aluminum self-interstitial (Al i ) through the reaction of  + Si Si Al , i i Al causing the Al i -mediated diffusion of silicon atoms. E as = 1.35 eV is much smaller than the theoretical migration-barrier energy (= 2.33 eV) of V Al 3− in AlN. 33) In the PAS measurement, the concentration of V Al -related defects was very low after annealing at temperatures over 1200°C. Therefore, we consider that the silicon atoms in AlN lattices are mainly diffused via the kick-out mechanism.

Dependences of O diffusion on annealing temperatures
The depth profile of the oxygen concentration in the silicon implanted AlN layers with 1 μm thickness before and after annealing is shown in Fig. 4(a). The AlN layer annealed at 1400°C had the oxygen concentration less than 1 × 10 18 cm −3 near the surface, while the AlN layer annealed at temperatures over 1600°C had the oxygen concentration over 8 × 10 18 cm −3 . The oxygen concentration increased with increasing depth in the AlN layers, showing the oxygen concentration over 10 20 cm −3 near the AlN/Al 2 O 3 interface. These results indicate that the oxygen atoms are diffused from the decomposed Al 2 O 3 substrate during high temperature annealing.
Using Fick's second law under the assumption of Gaussian distribution and infinite oxygen source from the Al 2 O 3 substrate, the oxygen concentration is given by = where C o is the oxygen concentration at the AlN/Al 2 O 3 interface. The Arrhenius plots for the intrinsic diffusion coefficients are shown in Fig. 4(b). These are well described by the Arrhenius equation where D o0 is the pre-exponential factor of oxygen diffusion and E oa is the activation energy of oxygen. The value of D o0 (=3 × 10 −10 cm 2 s −1 ) in AlN is smaller than that (=2 × 10 −7 cm 2 s −1 ) in GaN, 34) indicating slower diffusion of oxygen atoms in AlN. E oa =1.32 eV is close to E sa . We consider that the oxygen atoms in the AlN lattice are mainly diffused via the kick-out process of The slightly smaller E oa may result from the fast exchange rate between O i and O Al .

Control of impurity distribution using thick AlN layer
Silicon ions were implanted into a 3 μm thick AlN layer, followed by annealing at 1600°C. Depth profiles of the silicon, oxygen, hydrogen, and carbon concentration in the silicon implanted AlN layer after annealing are shown in Fig. 5. The average silicon concentration was 1 × 10 19 cm −3 in a ∼300 nm deep box profile. The concentrations of hydrogen and carbon were under detection limits. The oxygen concentration gradually decreased when moving towards the surface from the substrate side, reaching the detection limit near the surface. We reproducibly achieved electrically conductive AlN layers with silicon implantation after annealing at 1600°C.
The 1 μm thick AlN layers annealed at temperatures over 1500°C were electrically insulating, while the 3 μm thick AlN layer annealed at 1600°C was electrically conducting. Major defects in the MOCVD-grown AlN layers are V Al (V N ) n and V Al (O N ) n . 21) From the results of the V Al -related defects and impurity distribution after annealing, we suggest that the high oxygen concentration is the dominant reason that inhibits the electrical conduction of AlN layers on sapphire substrates. Oxygen atoms substitute nitrogen sites, forming complexes with aluminum vacancies V Al -O N with the stable deep acceptor level. 25,35) The complex of an aluminum vacancy and three oxygen atoms V Al 3− -3O N + is the most favorable oppositely charged defect for n-type AlN. 36) We consider that the V Al -O N defect complexes act as electron compensation centers, reducing the electron concentration. We need further investigation on the optical properties of the silicon-implanted AlN layers to clarify the point-defect species affecting the electrical properties.
The electron mobility μ e and electron concentration n e of the 3 μm thick AlN layer after annealing at 1600°C were determined by Hall-effect measurements. The values of μ e and n e at 500°C were 12 cm 2 V −1 s −1 and 1 × 10 17 cm −3 , respectively, while those at 800°C were 1.0 cm 2 V −1 s −1 and 1 × 10 19 cm −3 , respectively. These carrier concentrations are comparable to the other reports. 10,37) Silicon on aluminum substitutional site Si Al + often forms a DX − centers, reducing the electron concentration due to self-compensation. 35) Breckenridge et al. suggested that the annealing of highly disordered AlN samples at high temperatures allows silicon atoms to relax into the DX − acceptor state. 10) Further