AlGaN/GaN Schottky barrier diode (SBD) features excellent performance for high-power, high-frequency, and high-temperature applications, attributed to the outstanding properties of GaN material^[1-5]. Recently, some researchers have demonstrated that AlGaN/GaN SBD has great potential for future wireless power transfer (WPT) applications^[6-7]. The SBD converts the received RF signal into DC power, which determines the efficiency of the WPT system. To increase the conversion efficiency, AlGaN/GaN SBD requires a low turn-on voltage (V_T) while maintaining a low reverse leakage current (I_R).

The Schottky metal is well recognized to have a considerable impact on the V_T and I_R of Schottky junction (SJ). For AlGaN/GaN SBD, the most common electrode used as the anode is Ni/Au and TiN metal^{[2-5, 8-10]}. Some researchers have compared the performance of AlGaN/GaN SJ based on Ni/Au and TiN Schottky metal. The related results show that the electrical characteristics of TiN-based SJ are better than Ni/Au-based SJ^[11-13], but the mechanism is not well understood. A common explanation for this result is that the interaction between Ni and AlGaN(GaN) causes many defects or interface states, thus increasing the I_R^[12-14]. To confirm this hypothesis, a clear understanding of the current transport mechanism is necessary to determine whether and how the related defects act on the electrical characteristics of SJ. However, the comparative analysis on the current transport mechanism of AlGaN/GaN SBDs with Ni/Au and TiN anodes is still lacking, and it would help obtain deeper physical insights into the impact of Schottky metal on the electrical properties of SBDs.

The recess-free technique can improve device uniformity and reliability by eliminating the plasma damage to the AlGaN barrier layer from the etching process^[15-17]. In previous work, we have demonstrated high-performance recess-free AlGaN/GaN SBDs based on Ni/Au anode^{[16, 17]}. However, recess-free AlGaN/GaN SBDs based on TiN anode have never been reported. The TiN anode can be compatible with CMOS processes^[18], thus significantly reducing the fabrication cost of recess-free AlGaN/GaN SBDs.

In this paper, we have achieved the simultaneous improvement of V_T and I_R for recess-free AlGaN/GaN SBDs by substituting the Ni/Au anode with the TiN anode. The impact of Schottky metal on the electrical properties of SBDs is revealed by analyzing the current transport mechanism.

The epitaxial wafer is a commercial product from the Enkris Semiconductor, grown by MOCVD on a 3-inch sapphire <0001> substrate, consisting of a 1.5 μm C-doped GaN buffer layer, a 400 nm UID GaN channel layer, and a 7 nm Al_0.25Ga_0.75N barrier layer. With a 24 nm LPCVD-SiN_x passivation (780 °C, with RCA pretreatment), the 2DEG channel in the access region was effectively preserved^[19]. After passivation, the sheet resistance is reduced to 303 Ω/sq.

The schematic cross-section of SBD is shown in Fig. 1, and the device fabrication flow is the same except for the anode metal. First, the LPCVD SiN_x cap in anode and cathode region was opened with a fluorine-based ICP etch. Then, the metal layer of Ti/Al/Ni/Au (20/150/45/55 nm) was deposited, followed by rapid thermal annealing (RTA) at 870 °C for 50 s in N₂ ambient to form the ohmic cathode. Finally, Ni/Au (50 nm/150 nm) was deposited by evaporation to form the anode electrodes for device A, and TiN metal was deposited by reactive sputtering in Ar:N₂ (7 : 3 sccm) mixed gas atmosphere to form the anode electrodes for device B. The devices in this work have the same dimensions of L_SC/L_AC/L_FP/W = 2/6/0.75/100 μm. The electrical characteristics of the device were measured with Keysight B1500A.

The I–V characteristics at room temperature (RT) on the log scale and linear scales are shown in Figs. 1(b) and 1(c), respectively. The SBD with TiN anode exhibits a low V_T of 0.94 V (defined as the voltage corresponding to current = 1 mA/mm), a low differential specific on-resistance (R_ON,diff) of 0.52 mΩ·cm² and a low I_R of ~ 0.5 μA/mm. Compared with SBD with TiN diode, the Ni/Au SBD exhibits similar R_ON,diff, while presenting a higher V_T of 1.16 V and a higher I_R of ~1 μA/mm. I–V curves of 8 devices are also plotted in Fig. 1(d), suggesting a good device uniformity attributed to the recess-free technique. The lower V_T in TiN SBD can be attributed to the difference between the work function of Ni (~5.15 eV) and of TiN (~4.7 eV)^[13]. However, the lower I_R in TiN SBD is difficult to understand. To further explain the phenomenon, the temperature-related I–V measurement was carried out.

The I–V characteristics of devices A and B measured from 223 to 473 K are shown in Figs. 2(a) and 2(b) and used to analyze the current transport mechanisms. The I–V characteristics of both devices can be divided into three regions, namely, region-I (low forward bias), region-II (high forward bias), and region-III (reverse bias). In region-I, the current show weak temperature and voltage dependence on the logarithmic axis, indicating that the tunneling-assisted transport mechanism dominates. Obviously, the tunneling effect in this region is more significant for Ni/Au SBD, although the metal work function of Ni is higher than that of TiN. The strong tunneling effect in Ni/Au SBD may be attributed to the interface states existing at the Ni/AlGaN interface caused by the interaction^{[12, 20]}. In region-II, the current show strong temperature and voltage dependence, indicating that the thermionic emission (TE) is a possible transport mechanism. The equation of TE model can be simplified as^[21]

$ I = S{A^*}{T^2}\exp \left( { - \frac{{q{\varphi _{\rm b}}}}{{kT}}} \right)\exp \left( {\frac{{qV}}{{nkT}}} \right), $  (1)

$ \ln I = \frac{{qV}}{{nkT}} + {\rm{ln}} \left( {S{A^*}{T^2}} \right) - \frac{{q{\varphi _{\rm b}}}}{{kT}}, $  (2)

where S is the SJ area, A^* is the effective Richardson constant of 30 A/(cm·K)² for Al_0.25Ga_0.75N^[22], T is the absolute temperature, q is the fundamental electronic charge, V is the forward-bias voltage, k is the Boltzmann constant, qφ_b is the apparent Schottky barrier height, and n is the ideality factor. Then, the experimental values of qφ_b and the n can be extracted from intercepts and slopes of the forward lnI versus V plots in Figs. 2(a) and 2(b).

Fig. 2(c) shows the qφ_b and n as a function of temperature. It can be observed that both qφ_b and n show a strong temperature dependence for our devices. This observation is inconsistent with the ideal TE mechanism, in which the qφ_b and n should be almost constant with temperature. Besides, the calculated values of qφ_b and n are also far from the ideal value (n = 1, qφ_b = qφ_M – χ_AlGaN), especially at low temperatures. Several studies have suggested that such behavior is attributed to the spatial non-homogeneous distribution of the Schottky barrier height^{[23, 24]}. At low temperature, the current is dominated by the low barrier regions because the electrons do not have enough energy to cross over the high barrier regions and thus lead to the low extracted qφ_b at low temperatures. According to Schmitsdorf et al.^[25] and Tung^[26], the presence of inhomogeneity of the Schottky barrier height can be proved by the linear relationship between qφ_b and n. Fig. 2(d) shows a linear reduction of qφ_b with increasing n, which suggests the presence of inhomogeneity of the Schottky barrier height. By the extrapolation of qφ_b to n = 1 in Fig. 2(d), the mean barrier height of 1.14 eV for Ni/Au anode and of 1.0 eV for TiN anode was determined. The extrapolated values agree with the experimental value of 1.10–1.30 eV for Ni/AlGaN SBD^[27] and of 0.88–1.13 eV for TiN/AlGaN^[13], thus validating our analysis. Moreover, TiN SBD exhibited lower n and higher qφ_b compared with Ni/Au at low temperature (Fig. 2(c)), indicating the homogeneity of the Schottky barrier height of TiN/AlGaN contact is better than that of Ni/AlGaN contact. The poor homogeneity of Ni/AlGaN may be attributed to the defects induced by the interaction between Ni and AlGaN.

In region-III, it can be roughly seen that the temperature dependence of I_R varied with temperature for both devices, which indicated I_R could be dominated by different transport mechanisms at different ranges of temperature. To identify how many transport mechanisms work, the Arrhenius plots of I_R at V_A = –0.5 V are drawn in Fig. 3(a). Then, the activation energy (E_A) was extracted from the Arrhenius plots, as shown in Fig. 3(b).

The physical meaning of E_A is the energy barrier for electrons at the metal fermi-level to pass through the AlGaN barrier. Therefore, different E_A corresponds to different reverse leakage mechanisms. According to the extracted E_A in Fig. 3(b), the reverse leakage behavior can be roughly divided into two regions. At high temperatures, as temperature increases, the leakage current shows the apparent temperature dependence, indicating that the thermal emission process could dominate the leakage current in this region. The extracted E_A (< 0.6 eV) in this region (Fig. 3(b)) is much smaller than the Schottky barrier height reported in the literature (~ 1 eV)^[11], suggesting that the emission of electrons is associated with the trap. At low temperatures, the temperature dependence of I_R is weaker, and E_A is almost zero (Fig. 3(b)), which indicates that the tunneling process could dominate the leakage current. A detailed analysis of reverse leakage mechanisms is presented in the next section.

Electric field (E) estimation is necessary to analyze I_R quantitatively. E across the AlGaN barrier layer can be calculated using the equation E(V) = q[σ_p– n_s(V)]/ε_sε₀^[28]. The fixed polarization charge density (σ_p) at the hetero-interface is estimated to be 1.35 × 10¹³ cm^–2[29]. ε_s = 8.9^[30] is the static dielectric constant of Al_0.25Ga_0.75N, and ε₀ is the vacuum dielectric constant. The 2DEG concentration (n_s) at the hetero-interface can be extracted from 1 MHz capacitance–voltage (C–V) curves in Fig. 4(a). The calculated E as a function of reverse bias is shown in Fig. 4(b), which will be used to analyze the leakage mechanism. Moreover, it can be seen in Fig. 4(a) that the pinch-off voltage (V_pin-off) in Ni SBD is higher than that in TiN SBD, which proves that the actual barrier height of Ni is higher than TiN.

Considering the strong temperature dependence at high temperatures, Poole–Frenkel emission (PFE) is the most probable carrier transport mechanism. The PFE refers to electric-field enhanced thermal emission from a trap state into a continuum of electronic, and the current density of PFE can be described as^[31]:

$ {J_{{\text{PF}}}} = CE\exp \left[ { - \frac{{q({\varphi _{\text{t}}} - \sqrt {{{qE} \mathord{\left/ {\vphantom {{qE} {\pi {\varepsilon _{{\text{s(h)}}}}{\varepsilon _0}}}} \right. } {\pi {\varepsilon _{{\text{s(h)}}}}{\varepsilon _0}}}} )}}{{kT}}} \right], $  (3)

where C is a constant associated with trap concentration, E is the electric field across the barrier, qφ_t is the barrier height for the electron emission from the trap state, ε_s(h) = 5.1^[32] is the relative dielectric constant at high frequency for Al_0.25Ga_0.75N. Eq. (3) can be rearranged as given below

$ $$\begin{array}{c}\mathrm{l}\mathrm{n}\left(J_{\text{PF}}/E\right)=\frac{q\sqrt{qE/\phantom{qE\pi {\epsilon }_{\text{s(h)}}{\epsilon }_0}\pi {\epsilon }_{\text{s(h)}}{\epsilon }_0}}{kT}\sqrt{E}-\frac{q{\phi }_{\text{t}}}{kT}+\mathrm{l}\mathrm{n}C\\ =m(T)\sqrt{E}-\frac{q{\phi }_{\text{t}}}{kT}+\mathrm{l}\mathrm{n}C,\end{array}$$ $  (4)

or

$ $$\begin{array}{c}\ln \left(J_{\text{PF}}/E\right)=-\frac{q({\phi }_{\text{t}}-\sqrt{qE/\phantom{qE\pi {\epsilon }_{\text{s(h)}}{\epsilon }_0}\pi {\epsilon }_{\text{s(h)}}{\epsilon }_0})}{kT}+\mathrm{l}\mathrm{n}C\\ =-\frac{q{\phi }_{\text{eff}}(E)}{kT}+\mathrm{l}\mathrm{n}C.\end{array}$$ $  (5)

From Eq. (4), ln(J_PF/E) should be a linear function of E^0.5. Figs. 5(a) and 5(b) show that the experimental plots of ln(J/E) versus E^0.5 fit well with the linear dependence at different temperatures for both devices, which proves the possibility of PFE. To further confirm the PFE, the ε_s(h) in Eq. (4) were extracted from the m(T) at different temperatures, as shown in Fig. 5(c). It is found that the ε_s(h) extracted at high temperature (T > 400 K) is consistent with the theoretical value of Al _0.25Ga_0.75N, which strongly indicates that PFE is the dominant mechanism of I_R at high temperature. However, at medium and low temperatures (T < 400 K), the extracted ε_s(h) is much higher than the theoretical value, indicating that different transport mechanisms dominate in this case.

To identify the defect origin of the trap-involved PFE, the extraction of qφ_t is necessary. According to Eq. (5), ln(J_PF/E) should be a linear function of 1/T, and qφ_eff(E) can be extracted from the slope. The ln(J_PF/E) versus 1000/T plots were shown in Figs. 6(a) and 6(b), it can be seen that the data fit well with straight lines at high temperatures (T > 400 K) with various E. At medium and low temperatures (T < 400 K), the dependence of ln( J_PF/E) on temperature is weak, suggesting that the tunneling mechanism dominates in this case, although the data also follows a linear relationship. Then, the extracted E-related qφ_eff(E) are shown in Figs. 6(c) and 6(d), and the qφ_t obtained by extrapolating qφ_eff(E) to zero electric field is 0.65 and 1.02 eV for Ni SBD and TiN SBD, respectively. The value of qφ_t in Ni SBD and TiN SBD is respectively consistent with the energy level of threading dislocation (TD)^{[33, 34]} and of N vacancy (V_N)^{[35, 36]} in AlGaN. Thus, the PFE from the V_N to conduction band and the TD to conduction band may dominate the I_R of Ni SBD and TiN SBD at high temperatures, respectively.

The difference in electrically active PFE defects between the two devices can be explained by the N diffusion mechanism based on Fick’s law^[12]. A significant difference in concentration or chemical potential between two materials facilitates the diffusion process from the region of high concentration to low concentration. The TiN anode with a high N concentration effectively suppresses the N diffusion process from the AlGaN layer to the Schottky metal, which reduces the V_N density in AlGaN, so PFE was dominated by natural defects such as threading dislocations. In contrast, the Ni anode with low N concentration facilitates the N diffusion process from the AlGaN layer to the Schottky metal, increasing the V_N density in AlGaN, so the V_N dominates the PFE. The N diffusion at Ni/AlGaN interface has been reported in as-fabricated HEMT^[37], supporting our finding. Besides, the formation of Ni-nitrides has also been reported by annealing at the temperature of 200 °C^[38]. In short, it can be considered that the PFE induced by the high-density V_N is responsible for the higher I_R in Ni SBD at high temperatures.

At low temperatures, the weak temperature dependence of I_R indicates that the tunneling process dominates. A possibility is Fowler–Nordheim (FN) tunneling, which is weakly related to temperature but strongly related to barrier height and electric field. The FN current density is given as^[32]:

$ {J_{{\text{FN}}}} = A{E^2}\exp \left( { - \frac{{8\pi \sqrt {2m_{}^*q} }}{{3hE}}\varphi _{\rm b}^{3/2}} \right), $  (6)

where A = ((q²(m₀/m^*))/(8πhφ_b)) is constant, h is Planck’s constant, m₀ is the free-electron mass, m^* ~ 0.27 is the conduction band effective mass in the barrier layer, estimated by linear interpolation of the values in AlN and GaN^[39]. Rearranging Eq. (7), we get

$ {\rm{ln}} ({J_{{\text{FN}}}}/{E^2}) \propto - \frac{{8\pi \sqrt {2m_{}^*q} \varphi _{\rm b}^{3/2}/3h}}{E}, $  (7)

which indicates that the ln(J/E²) versus 1/E plot should follow a linear dependency if FN tunneling process is the dominating mechanism, and the slope should be a weak function of T. Figs 7(a) and 7(b) show that the calculated ln(J/E²) versus 1/E plot is very consistent with FN tunneling characteristics, indicating that FN tunneling dominates the leakage current at low temperatures for both devices. However, it is found that the qφ_b of Ni SBD extracted from the slope is much lower than the qφ_b of TiN SBD (Fig. 7(c)), which is inconsistent with the difference between the work function of Ni (5.15 eV) and TiN (4.7 eV). Moreover, the C–V characteristics (Fig. 4(a)) and the forward I–V characteristics (Figs. 2(c) and 2(d)) show that the qφ_b of Ni SBD is higher than that of TiN SBD, which is also contradictory to Fig. 7(c).

It has been found that the ionization of V_N donor can cause the thin surface barrier (TSB) effect^[40], thus leading to an increase of E in the barrier layer. Therefore, the low value of qφ_b for Ni SBD extracted by the FN model may be attributed to the underestimation of the E. Considering the TSB effect, Eq. (7) should be corrected to:

$ {\rm{ln}} ({J_{{\text{FN}}}}/{(\beta E)^2}) \propto - \frac{{8\pi \sqrt {2m_{}^*q} \varphi _{\rm b}^{3/2}/3h}}{{\beta E}} ,$  (8)

where β is the electric field enhancement coefficient defined as the ratio of the actual electric field to the theoretical electric field. According to Eq. (8), the slope in Figs. 7(a) and 7(b) can be expressed as:

$ {\text{slope}} = - \frac{{8\pi \sqrt {2m_{}^*q} \varphi _{\rm b}^{3/2}/3h}}{\beta }. $  (9)

Using Eq. (9), the relationship between β and qφ_b for Ni SBD can be obtained, as shown in Fig. 7(d).

Fig. 7(d) shows that the extracted qφ_b increases as β increases. When the β is between 1.5 and 2, the qφ_b0 for Ni SBD is close to the value reported in the literature^{[11, 27]}. The large β indicates that the strong TSB effect exists in Ni SBD, which leads to the high I_R for Ni SBD compared to the TiN SBD at low temperatures. The TSB effect in Ni SBD is mostly caused by the donor-like V_N defect, which is introduced by the N diffusion process. For TiN SBD, the N diffusion process is effectively suppressed, and thus the extracted value of qφ_b by the FN model is relatively normal (Fig. 7(c)).

Curiously, for reverse characteristics, the qφ_b of TiN is higher than Ni (Fig. 7(c)), while for the forward characteristics, the qφ_b of TiN is lower than Ni (Fig. 2(d)). This phenomenon can be explained as follows: at high forward bias, the quasi-Fermi level in semiconductor shifts above the level of V_N donor, and the ionized V_N will be neutralized, so the TSB effect in Ni/AlGaN will be weakened. Besides, the E in AlGaN barrier is decreased with forward bias, and the thermal emission related mechanism will gradually dominate the current transport, so the TSB effect has less influence on the forward characteristics.

In short, the carrier transport mechanisms of both devices are shown by the schematic energy band diagram in Fig. 8. According to the leakage mechanism analysis, it can be concluded that the main reason for the high I_R of Ni SBD is that the high-density V_N induced by the N diffusion enhances the PFE and FN tunneling. TiN SBD with high N concentrations is efficient to mitigate the N diffusion, thus obtaining a better trade-off between good V_T and I_R. Therefore, to reduce V_N density and improve SBD performance, conductive metal nitride, such as TiN, TaN, and WN_x, is more suitable as Schottky metal for AlGaN/GaN SBD.

In summary, we have comparatively investigated the reverse leakage mechanisms of recess-free AlGaN/GaN SBD with Ni/Au and TiN anodes. It is found that TiN SBD exhibits lower V_T and I_R than Ni SBD. For forward characteristics, the current of both devices is dominated by the TE mechanism. It is found that both qφ_b and n show a strong temperature dependence. Such temperature dependence can be explained by the presence of inhomogeneity of the Schottky barrier height, which is proved by the linear relationship between qφ_b and n. For reverse characteristics, two types of mechanisms work at different temperature ranges. At high temperatures, the I_R for Ni/Au anode is dominated by the PFE from V_N to the conduction band, while the I_R for TiN anode is dominated by the PFE from TD to the conduction band. The decrease of V_N can explain the observed low I_R in TiN SBD due to the suppression of N diffusion. At low temperatures, the I_R of both devices is dominated by FN tunneling. By modifying the FN model, we concluded that the electric field of the barrier layer in the Ni SBD is higher than in the TiN SBD, thus leading to the relatively higher FN leakage in Ni SBD. The TSB effect caused by V_N may be responsible for the increased electric field of the barrier layer in Ni SBD.

This work was supported in part by Natural Science Foundation of China (Grant No. 61804172), in part by GuangDong Province Key Technologies Research and Development Program (No. 2019B010128001) and in part by the Youth Innovation Promotion Association of CAS.