High-density 2D hole gas in p-GaN/AlN/AlGaN on a silicon substrate with polarization-enhanced Mg ionization

Abstract

Improving the hole concentration in p-GaN specimens has posed a major challenge due to the high activation energy of Mg doping in GaN. Recently, a delta doping technique for modulating the valence band and increasing the Mg ionization efficiency in GaN has been proposed. However, the memory effect of Mg in the reaction chamber greatly increases the difficulty of epitaxial growth; additionally, the Mg ionization efficiency remains too low. In this study, we demonstrate a novel p-GaN/AlN/AlGaN structure with an increased Mg ionization efficiency, from 5.16% to 14.22%, relative to the conventional p-GaN/AlGaN structure. The inserted AlN layer effectively modulates the valence band of p-GaN near the p-GaN/AlN interface and decreases the activation energy, as confirmed by density functional theory and the 1D Poisson–Schrodinger equation. Moreover, a high-performance p-channel GaN FinFET with a high output current of 114 mA/mm is fabricated, and the polarization-enhanced epitaxial layer shows great promise for next-generation digital ICs.

Graphical abstract

1. Introduction

GaN, a wide band gap semiconductor, has been rapidly developed in recent years; it is widely used in electronic and optoelectronic devices [1], [2], [3], [4]. Due to the strong spontaneous and piezoelectric polarization effects in wurtzite GaN, the high density and mobility characteristics of the two-dimensional electron gas (2DEG) induced at AlGaN/GaN heterojunction interfaces enable the fabricated GaN high-electron-mobility transistors (HEMTs) to be widely used in radio frequency (RF) and power circuits [5], [6], [7]. Furthermore, the two-dimensional hole gas (2DHG) induced by the polarization effect has attracted the attention of researchers [8], [9], [10], [11]. GaN-based p-channel metal‒oxide‒semiconductor field-effect transistors (MOSFETs) relying on the 2DHG accumulated at the p-GaN/AlGaN heterojunction interface can be integrated with GaN n-channel HEMTs to realize multifunctional GaN-based digital integrated circuits (ICs) [12], [13], [14], [15], [16]. However, the conventional p-type dopant Mg has an activation energy of 170 meV in GaN, making the formation of a high concentration of holes difficult; additionally, the high background carrier concentration induced by N vacancies can further reduce the hole concentration [17,18]. Moreover, the high effective mass of holes in GaN decreases the mobility [19]. Therefore, there is a large gap between GaN HEMTs and GaN PMOSs, hindering the development of GaN digital ICs [20,21].

Improving the p-type doping efficiency is a globally recognized challenge and a popular research topic. Low-energy electron beam irradiation (LEEBI) and thermal annealing techniques have been proven to be effective methods for activating Mg in p-type GaN [22,23]. However, the uniformly doped p-type GaN epitaxially grown by conventional methods has an extremely low activation efficiency, even when treated with thermal annealing in a nitrogen atmosphere [17,24]. A seemingly straightforward approach to increase the hole concentration is to increase the doping concentration of Mg; however, a high doping concentration of Mg has been found to produce a self-compensation effect that severely reduces the quantity and mobility of mobile holes in the GaN epitaxial layer [25,26]. Therefore, improvement of the ionization efficiency of Mg at low doping concentrations is crucial. In recent years, ion implantation and annealing at ultrahigh temperature and pressure have received much attention [27,28]. Meanwhile, some focus has also been on the growth and activation conditions [29], [30], [31], [32]. A delta doping technique has been proposed to substantially improve the ionization efficiency of Mg from 5% to 10%; the epitaxially grown p-GaN shows a high free hole concentration of 10¹⁸ cm⁻³ [33,34]. Additionally, a polarization doping technique has been proven to effectively induce a high-concentration 3D hole gas in GaN; this technique has been successfully applied in GaN-based light-emitting diode (LED) devices [35], [36], [37], [38]. Formation of the AlGaN/GaN superlattice structure is another method for improving the p-type doping efficiency; with the periodically modulated energy band, the average ionization energy of Mg decreases. Furthermore, the polarization-induced holes in the superlattice structure show weaker temperature-dependent characteristics than those of bulk holes formed by thermal activation [39,40]. A high hole concentration of 3 × 18 cm⁻³ has been achieved by combining polarization doping with the superlattice and delta doping techniques [41]. Additionally, due to the strong polarization effect at the GaN/AlN heterojunction interface, a GaN epitaxial layer on an Al₂O₃ substrate with a high 2DHG density reaching 5 × 10¹³ cm⁻² has been achieved by molecular beam epitaxy (MBE), even with an undoped GaN layer [42]. The Mg atoms that diffuse in the AlGaN layer during epitaxial growth severely degenerate the hole mobility through ionized impurity scattering, leading to negative influences on GaN PMOSs. To weaken the influence of ionized impurity scattering, N-surface AlGaN/GaN superlattice structures with steep Mg doping interfaces have been fabricated; additionally, a high 2DHG mobility has been achieved [43,44].

In this work, a state-of-the-art Mg ionization efficiency has been achieved due to the effect of the AlN spacer between the Al_0.15Ga_0.85N barrier layer and the p-GaN layer on the Si substrate. The valence band (VB) of p-GaN near the AlN spacer is modulated upward; the 2DHG density increases from 1.48 × 10¹³ to 4.08 × 10¹³ cm⁻² as the Mg ionization efficiency increases from 5.16% to 14.22%. The inserted unintentionally doped (UID) GaN layer acting as a hole conduction channel effectively suppresses ionized impurity scattering; a high hole mobility of 49.67 cm²/V·s is obtained for the epitaxially grown p-GaN with a 15-nm UID GaN conduction layer at 90 K. For the fabricated p-channel GaN fin field-effect transistor (FinFET) with a gate nanowire width of 60 nm and a source-to-drain length (L_SD) of 1.9 μm, a high output current of 114 mA/mm and a low effective on-resistance (R_ON) of 88 Ω·mm are obtained. The proposed p-GaN epitaxial layer with the polarization-enhanced AlN spacer and the scattering-suppressed UID GaN conduction layer shows great promise for next-generation GaN-based digital ICs.

2. Material and methods/ experiment

2.1. Epilayer growth

The epitaxial layers of Sample A were grown via metal-organic chemical vapor deposition (MOCVD) on a 6-inch substrate and consisted of a 4.5-μm buffer layer, a 300-nm GaN electron channel layer, a 1-nm AlN spacer, a 15-nm Al_0.15Ga_0.85N barrier layer with 15% Al composition and a 70-nm p-GaN layer, from bottom to top. During growth, trimethylgallium (TMGa), trimethylaluminum (TMA) and ammonia (NH₃) were used as the sources of Ga, Al, and N, respectively. Biscyclopentadienyl-magnesium (CP₂Mg) was used as the Mg dopant source during the growth of the p-GaN layer. The Mg concentration in the p-GaN layer was quantified to be 4.1 × 10¹⁹ cm⁻³ by secondary ion mass spectrometry. Sample B had a similar epitaxial structure, except for a 1.5-nm AlN layer inserted between the Al_0.15Ga_0.85N barrier layer and the p-GaN layer, modulating the VB. The raised valence band maximum (VBM) of p-GaN and the formed deeper and narrower triangular quantum well were conducive to increasing Mg ionization and the hole concentration. Furthermore, additional UID GaN layers with thicknesses of 5, 10, and 15 nm were introduced under the p-GaN layer for Samples C, D, and E, respectively, as shown in Fig. 1a. The UID GaN hole channel with suppressed ionized impurity scattering acted as a hole conduction channel, and the hole mobility increased with increasing UID GaN thickness. The cross-section of Sample B was investigated by high-resolution transmission electron microscopy (HR-TEM) along the (11–20) plane using a Themis Z (Thermo Fisher Scientific, USA) at 300 kV, as depicted in Fig. 1b; the thickness of the sample was thinned to less than 80 nm by a focused ion beam. Energy-dispersive X-ray spectroscopy (EDX) was performed with the Super-X spectrometer attached to the Themis Z to determine the element distribution in the heterojunction, as displayed in Fig. 1c. The 1.5-nm high-quality AlN spacer with a sharp interface effectively modulated the triangular quantum well, which was more conducive to accumulating 3D holes to form a 2DHG. Combined with the fully strained AlN spacer, the increased VBM promoted the Mg ionization efficiency, and a high 2DHG concentration was obtained.

Fig. 1.

Description of epitaxial layer. (a) Cross-sectional view of the epitaxially grown structure of the samples. (b) High-resolution TEM images, including EDX analysis of sample B; the 1.5-nm AlN spacer with a sharp interface is clearly visible between the p-GaN layer and the Al_0.15Ga_0.85N barrier layer. (c) Quantified atomic ratios of grown Sample B extracted from EDX.

2.2. Device fabrication

p-channel GaN FinFET fabrication commenced with alignment definition via an inductively coupled plasma (ICP) process with a gas mixture of Cl₂ and BCl₃, and device isolation was implemented using multienergy Ar⁺ implantation. The sample was subsequently treated with buffered oxide etchant (BOE) solution for 2 min to remove the native oxide before ohmic metal deposition, which was effective for eliminating the contact barrier and lowering the contact resistance. Then, evaporation of Ni/Au (30/120 nm) and annealing at 550 °C in an O₂ atmosphere for 300 s produced a good ohmic contact with a low barrier height of 0.53 eV. Then, fins with a width of 60 nm were defined by electron beam lithography (EBL), followed by a low-damage ICP process, and the total etching depth was 138 nm, as confirmed by atomic force microscopy (AFM). Afterward, the sample was annealed at 450 °C in a N₂ atmosphere for 300 s, followed by boiling in tetramethylammonium hydroxide (TMAH) at 80 °C for 10 min to eliminate etching damage. A 15-nm Al₂O₃ layer was then deposited via plasma-enhanced atomic layer deposition (PEALD) as the gate dielectric and the passivation layer in the active region. The gate metal was formed by evaporating a metal stack of Ni/Au (50/150 nm), followed by a post-gate annealing treatment at 450 °C d to eliminate the interface states between GaN and Al₂O₃. The fabricated p-channel GaN FinFET possesses a gate length (L_G) of 450 nm, a gate-to-source length (L_GS) of 600 nm, a gate-to-drain length (L_GD) of 850 nm, and a gate nanowire width (W_Fin) of 60 nm.

2.3. Computational methods

Density functional theory (DFT) calculations were carried out with the Vienna Ab initio Simulation Package (VASP). The projector-augmented wave (PAW) method with plane-wave cutoff energies of 530 and 450 eV was adopted for structural relaxation and electronic structure calculations, respectively. Semicore Ga_3d electrons were treated as the valence electrons. The Perdew-Burke–Ernzerhof (PBE) and HSE06 (α = 28%) exchange-correlation functionals were selected for the structural relaxation and electronic structure calculations, respectively. By transferring the wurtzite structure to an orthogonal lattice, the atomic model used for doping included 120, 300, and 420 atoms for Al, Ga and N, respectively, which resulted in a large cell volume of

15.96 × 16.59 × 36.27 ų. All atoms were allowed to move until the Hellmann–Feynman forces were below 0.02 eV/Å. The electronic step convergence criterion of 10⁻⁵ eV and a single Γ were used for the doped model calculations. The activation energy of the acceptor can be obtained by:

$$E_A=E_{tot}^{q_{-1}}-E_{tot}^{q_0}-E_{vbm}-\Delta V$$ (1)

where $E_A$ is the activation energy level referenced to the VBM and $E_{tot}^{q\_-1}$ and $E_{tot}^{q\_0}$ are the total energies of the ionized and deionized Mg-doped configurations, respectively. $E_{vbm}$is the energy of the VBM at the dopant site of the pristine bulk, and $\Delta V$ is the corresponding potential alignment between the ionized and pristine configurations.

3. Results and discussion

The hole concentration and mobility characteristics of these epitaxial layers, explored through Hall effect measurements conducted in the van der Pauw geometry, are shown in Fig. 2a and b, respectively. The sheet hole concentrations of Samples A, B, C, D, and E at 300 K are 1.48 × 10¹³, 4.08 × 10¹³, 3.88 × 10¹³, 2.98 × 10¹³, and 1.75 × 10¹³ cm⁻², respectively, while the hole mobilities are 8.73, 6.17, 7.34, 8.97, and 10.48 cm²/V·s, respectively, as confirmed by contact Hall measurements. Relative to Sample A, the 1.5-nm AlN spacer in Sample B plays a role in dramatically increasing the hole density from 1.48 × 10¹³ to 4.08 × 10¹³ cm⁻², with an increase in the Mg ionization rate from 5.16% to 14.22%. The extremely high hole density results in increased phonon scattering [8,45,46], degenerating the hole mobility from 8.73 to 6.17 cm²/V·s. The inserted UID GaN layer, acting as a hole conduction channel, is conducive to suppression of ionized impurity scattering. The hole mobility increases relative to Sample B to 10.48 cm²/V·s when the UID GaN thickness is 15 nm. However, the inserted UID GaN layer separates the AlN spacer from p-GaN and weakens the modulation of the energy band by the AlN spacer. Therefore, although increasing the UID GaN thickness can improve the hole mobility, the hole concentration limits further increases in the UID GaN thickness.

Fig. 2.

Characteristics and benchmarks of 2DHG. (a) Hole concentration, (b) hole mobility, and (c) sheet resistance values of designed epitaxial layers measured by the contact Hall method. (d) Benchmarking the hole concentration and mobility characteristics of the p-GaN epitaxial layers relative to other works. (e) Statistics of the Mg ionization efficiency obtained via various methods [[17], [24], [25], [27], [28], [30], [31], [32], [33], [47], [48], [49], [50], [51], [52], [53]].

It displays the sheet resistance (R_SH) values of these epitaxial layers in Fig. 2c, which are calculated to be 48.27, 23.86, 21.94, 23.32, and 34.05 kΩ/□. Due to the energy band modulation caused by the 1.5-nm AlN spacer and the inserted 5-nm UID GaN conduction path with suppressed ionized impurity scattering, a low R_SH of 21.94 kΩ/□ is obtained for Sample C; this value corresponds to a more than twofold reduction relative to the traditional epitaxy Sample A at room temperature. The performance levels of our epitaxial layers are benchmarked against some other group results in Fig. 2d, and the novel structure of Sample C with a 1.5-nm AlN spacer and a 5-nm UID GaN interlayer shows great potential to further improve the GaN PMOS performance. The statistics of the Mg ionization efficiency are shown in Fig. 2e. Relative to conventional methods, such as annealing, delta doping, and Mg ion implantation, the proposed polarization-enhanced AlN spacer induces a great improvement in the Mg ionization efficiency and an increase in the Mg ionization rate from 5.16% to 14.22%.

The temperature-dependent sheet hole concentrations of these epitaxial layers are presented in Fig. 2a. The hole concentrations of Samples B and C decrease at high temperatures and increase with decreasing temperature. For epitaxial layers with thick UID GaN interlayers inserted or epitaxial layers without AlN spacers, the hole concentrations decrease at high temperatures and saturate with decreasing temperature. This trend differs from that of the bulk p-GaN epitaxial layer, in which the holes only come from thermal ionization. The sheet hole concentration of Sample B increases from 4.08 × 10¹³ cm⁻² at 300 K to 5.23 × 10¹³ cm⁻² at 78 K; the Mg ionization rate increases from 14.22% to 18.22%. The holes of the multiheterostructure come from thermal and polarization ionization processes; the thermal ionization tends to dominate at high temperatures, whereas the polarization ionization tends dominates at low temperatures. Fig. 2b shows the temperature-dependent hole mobility characteristics of these designed epitaxial layers; the mobility increases with decreasing temperature at low temperatures because of the suppressed scattering effects of ionized impurities. Due to the stable hole concentration and the increased hole mobility at low temperatures, the R_SH of the proposed epitaxial layers with AlN spacers decreases with decreasing temperature; the R_SH of Sample C is only 10.51 kΩ/□, which is less than one-quarter of that of conventional Sample A, as shown in Fig. 2c.

The evident increase in the hole density was then evaluated by DFT. As shown in Fig. 3e, three typical doping configurations were designed with variation of the distance between the dopant and the AlN/GaN interface. As shown in Fig. 3a, the insertion of the AlN layer obviously decreases the activation energy of Mg. The highest value is 0.31 eV, corresponding to the GaN bulk region, due to the site being the farthest from the interface. The slightly higher value of 0.31 eV compared to the experimentally reported value of 0.26 eV is attributed to the superlattice-like Al_0.29Ga_0.71N alloy, which is derived from the periodic configuration in the DFT calculations. However, the trend of the activation energy as a function of the doping site remains reasonable. When the dopant approaches the interface, the activation energy decreases from 0.26 to 0.24 eV. The decrease from 0.31 to 0.26 eV is understood from the VBM variations. As shown in Fig. 3b, the VBM increases by 0.25 eV from the GaN bulk to the AlN/GaN interfacial region due to the additional polarization field induced by the inserted AlN. The existence of an electric field is seen in Fig. 3e, which shows that ionized holes tend to accumulate at the AlN/GaN (0001) interface but are depleted at the (000-1) interface. In addition, the decrease from 0.26 to 0.24 eV is not derived from the VBM difference but from the symmetry variation induced by the inserted AlN. The substitution doping of Mg_Ga leads to four Mg-N bonds with a tetrahedral configuration. Due to the wurtzite phase, the bonds are divided into two groups: bonds parallel and orthogonal to the (0001) direction. As shown in Fig. 3c, the four Mg-N bonds of Mg_Ga_1 show less deviation than those of Mg_Ga_3, suggesting the higher symmetry of Mg_Ga. The low symmetry derived from the Jahn-Teller effect leads to a large splitting of energy levels and accordingly an increase in the acceptor energy level. The increase in the VBM and symmetry of Mg_Ga derived from the inserted AlN is the main cause of the increase in the hole density.

Fig. 3.

Mg_Ga calculation characteristics near the interface by DFT. (a) Mg_Ga activation energy as a function of the dopant site. (b) VBMs at different sites obtained from the N_2p orbital projections of the N layer nearest to the corresponding dopant. The Fermi level is at 0 eV. (c) Mg-N bond length of bonds parallel and orthogonal to the (0001) direction. (d) Side view of the atomic model. The red rectangular areas are the layers including Mg_Ga. (e) Isosurfaces of holes of Mg_Ga.

It shows the VB distributions of Samples A and B calculated by the 1D Poisson–Schrodinger equation in Fig. 4a. Due to the polarization-enhanced VB induced by the inserted AlN layer, a deeper and narrower triangular quantum well forms that can attract more holes and prevent holes from escaping from the quantum well. Additionally, the VB of Sample B near the heterostructure interface is modulated toward the Fermi level, which can effectively improve the Mg ionization efficiency and increase the 2DHG density. The extracted VB differences with depth between Samples A and B are displayed in the inset of Fig. 4a. The polarization modulation effect becomes weak with increasing distance from the interface, and the 2DHG density decreases to the same value as that of Sample A when the UID GaN thickness reaches 15 nm. The temperature-dependent VB of Sample B is shown in Fig. 4b, and the relationship between the VB and temperature is displayed in the inset of Fig. 4b. The VB near the heterojunction clearly increases with decreasing temperature, boosting the Mg ionization efficiency and the 2DHG density. Although the holes are frozen under the suppressed thermal ionization effect, the decreased Mg ionization energy near the p-GaN/AlN interface contributes to the increased 2DHG density at low temperatures. However, for Samples D and E, the thick UID GaN channel layer weakens the influence of the heterostructure, leading to a monotonic decrease with temperature.

Fig. 4.

VB distributions calculated by the 1D Poisson–Schrodinger equation. (a) VB distribution and extracted VB difference (inset) characteristics of Samples A and B. (b) Temperature-dependent VB as a function of depth and VB as a function of temperature.

The cross-sectional view of the fabricated p-channel GaN FinFET based on Sample B is shown in Fig. 5a. A post-gate annealing treatment was implemented at 450 °C for 5 min in a N₂ atmosphere to eliminate the interface states, which has been proven to be an effective method for reducing the leakage current and R_ON. Fig. 5b shows the good output characteristics of the fabricated device with a source-to-drain distance of 1.9 μm, in which V_DS was swept from 0 to −10 V with a step of 0.1 V. Benefitting from the optimized epitaxial layer with the polarization-enhanced AlN spacer, the normalized maximum output current at a V_GS of −8 V is 114 mA/mm, with an extracted effective R_ON of 88 Ω·mm. Note that a slight nonlinear phenomenon occurs when V_DS is low, which results from the nonlinear characteristics of ohmic contact. The transfer characteristics on linear and semilog scales are displayed in Fig. 5c, and a threshold voltage (V_TH) of 3.8 V is extracted by linear extrapolation from the transfer I–V curve. In addition, due to the high-quality Al₂O₃ deposited by PEALD as a gate dielectric, the gate leakage current is only 1 nA/mm with a V_GS bias of −8 V. Boiling wafers in TMAH solution to eliminate the etching damage is critical to obtain a high ON−OFF current ratio of 2 × 10⁶ and a low subthreshold slope of 180 mV/dec of devices.

Fig. 5.

The characteristics of fabricated GaN p-channel FinFET. (a) Cross-sectional view of the fabricated p-channel GaN FinFET with a W_Fin of 60 nm and an L_SD of 1.9 μm. (b) Output and (c) transfer characteristics of the fabricated GaN FinFET.

4. Conclusion

A method to obtain a polarization-enhanced Mg ionization efficiency by inserting an AlN spacer between the AlGaN barrier layer and p-GaN was reported. A 2DHG density of 4.08 × 10¹³ cm⁻² with a 14.22% ionization efficiency was obtained, which was more than twofold the density of the traditional sample of 1.48 × 10¹³ cm⁻² with a 5.16% ionization efficiency. A higher symmetry of Mg_Ga and a lower activation energy were observed near the interface of the heterojunction by DFT calculations, increasing the Mg ionization efficiency. Additionally, the VB distribution calculated by the 1D Poisson–Schrodinger equation illustrated that the VB was modulated by the AlN spacer layer near the heterojunction, and a high−density 2DHG was obtained. The temperature-dependent VB distribution, which showed that the VB moved upward near the heterojunction with decreasing temperature, well explained the experimental results of the increase in the 2DHG density at low temperatures observed in the temperature-dependent Hall measurements. Finally, a p-channel FinFET with a high output current of 114 mA/mm at a V_GS of −8 V, a high ON−OFF current ratio of 2 × 10⁶ and a low subthreshold slope of 180 mV/dec was fabricated due to the high 2DHG density obtained by the p-GaN/AlN/AlGaN structure. The polarization-enhanced p-GaN/AlN/AlGaN structure exhibited promising prospects for high-performance p-channel MOSFETs and next-generation GaN-based digital ICs.

Declaration of competing interest

The authors declare that they have no conflicts of interest in this work.

Biographies

Tao Zhang received her Ph.D. degree from Xidian University in 2020 under the supervision of prof. Jincheng Zhang. Currently, he is an associate professor at Xidian University. His research interest focuses on wide-bandgap semiconductors and electronic devices.

Shengrui Xu (BRID:08337.00.52059) received the B.S. and Ph.D. degrees from Xidian University, Xi'an, China, in 2005 and 2010, respectively. He is currently a professor with the School of Microelectronics, Xidian University. His current research interests include GaN-based optoelectronic devices and wide gap-band materials and devices.

Jincheng Zhang (BRID:03137.00.03285) received the M.S. and Ph.D. degrees from Xidian University, Xi'an, China, in 2001 and 2004, respectively. He is currently a professor with Xidian University. His current research interests include wide gap-band semiconductor GaN and diamond materials and devices.