1. Introduction
Nowadays the AlGaN/GaN high-electron mobility transistor (HEMT) has emerged as an attractive candidate for future microwave power transistor. Owing to the combination of excellent material characteristics and large two-dimensional electron gases (2DEG) density up to 1013 cm-2 induced by polarization charges at AlGaN/GaN interface, AlGaN/GaN HEMT gives great potential for high breakdown voltage, low on-resistance, and high frequency application[1-4]. The ability to block high voltage is a key parameter for power devices. It has been proved that breakdown voltage is strongly restricted by the non-uniform off-state electric-field distribution in conventional AlGaN/GaN HEMT.
Much effort has been made to achieve higher breakdown voltage, for instance, employing field plates (FPs)[5, 6], doping the buffer with Fe or carbon[7], increasing the epitaxial-layer thickness[8, 9], inserting AlGaN buffer layer to the HEMT structure[10, 11], and use of Schottky contacts in drain[12]. But it is still far from the GaN material limitation. Application of a GaN cap is an effective effort to achieve high breakdown voltage, and it has a significant decrease in ohmic contact resistance, an increase in ideality factor, a decrease in gate and drain leakage currents, an increase in gain, and improved reliability in direct current[13]. The GaN cap layer influences the density of 2DEG which is a key issue to breakdown voltage, but currently sufficient investigation is still lacking in this area.
In this paper, the impact of GaN cap layer with different doping is discussed. Firstly, the breakdown voltage of HEMT with GaN cap layer with diverse doping concentration from 2 × 1016 to 5 × 1019 cm-3 of N-type and P-type cap is investigated. Furthermore, we demonstrate a new structure with P-type buried GaN layer which enhances the breakdown voltage. It has been reported that GaN cap layer is crucial to the AlN/GaN heterostructure material characteristics by changing the 2DEG electron density and the 2DEG electron mobility[14].
2. Device structure and breakdown model
With the development of the GaN power device, the structures under investigation are illustrated in Figure 1. Conventional AlGaN/GaN HEMT is shown in Figure 1(a) GaN/AlGaN/GaN HEMT with inserted GaN layer between SiN passivation and AlGaN barrier layer is shown in Figure 1(b). As depicted in Figure 1(b), GaN/AlGaN/GaN HEMT consists of a 0.05-μm SiN passivation, a 3-nm GaN cap layer a 0.02-μm AlGaN barrier, and a 2-μm GaN buffer. The source-to-gate distance is 1.0-μm, gate-to-drain distance is 3.0 μm. The length of drain, gate and source electrode are 0.5, 0.8, and 0.5 μm, respectively. The mole fraction of AlxGa1-xN in the proposed device is kept as 0.25.
Due to the unintentional doping of residual impurities such as oxygen atom[14] or the absence of nitride atom UID GaN typically exhibits some degree of N-type conductivity, so N-type concentration of 5 × 1015 cm-3 is employed in GaN material. Based on the analysis above, donors in an unintentionally N-doped GaN buffer should be compensated by deep acceptor states to obtain high resistivity, therefore acceptor trap density of 1 × 1016 cm-3 with the location of 1.8 eV from conduction band is taken into consideration.
The spontaneous and piezoelectric polarizations of the wurtzite phase of a-polar GaN is taken into account using built-in self-consistent polarization model (Formula=2)[15] of SDEVICE toolbox by Sentaurus. Based on the polarization vector, the piezoelectric charge is computed according to
qPE=−activation∇P. |
(1) |
The amount of the piezoelectric polarization in the direction of the c-axis can be determined by
Pstrain=2strain⋅(e31−e33c13c33), |
(2) |
where e31, e33 are strain-charge piezoelectric coefficients, c33 is stiffness constant. The value of strain is computed as
strain=(1−relax)⋅(a0−a)/a, |
(3) |
where a0 represents the strained lattice constant. a is the unstrained lattice constant, and 'relax' denotes a relaxation parameter[16].
Breakdown is commonly defined as a drain voltage at which a sudden increase is observed at the drain electrode. In this paper, breakdown is determined by the numerical simulator when drain current reaches to 1 mA/mm.
Avalanche generation (impact ionization) produces an electron--hole pair which requires a certain threshold field strength and the possibility of acceleration. If the width of a space charge region is greater than the mean free path between two ionizing impacts, charge multiplication occurs, which can cause electrical breakdown. The reciprocal of the mean free path is called the ionization coefficient α. The exponential expressions for modeling of the impact ionization coefficients (α and b) of electrons and holes in GaN are used in the numerical simulations, with van Overstraeten-de Man Model, with parameters an=2.81 × 108 cm-1, bn=3.43 × 107 V/cm, αp=5.41 × 106 cm-1, and bp=1.96 × 107 V/cm γ=1[1], Fava is the driving force for impact ionization, which is the value of the gradient of the quasi-Fermi level in this paper.
α(Fava)=γaexp(−γbFava). |
(4) |
3. Results and discussions
The two-dimensional model is based on meticulous calibration of a GaN capped lateral GaN HEMT against its experimental output characteristics. The output characteristics (ID-VDS) of the GaN HEMTs is shown in Figure 2. As can be seen, it shows a good agreement.
The breakdown character is simulated for both of the structures (with and without GaN cap layer). Equal potential lines distribution of (a) conventional GaN HEMT and (b) GaN/AlGaN/GaN HEMTs at breakdown voltage is shown in Figure 3. High electric field crowding at the drain side of the gate electrode corner can be seen (Figure 3(a)), hence the breakdown voltage is limited to 72 V. The GaN/AlGaN/GaN HEMT producing more uniform equal potential lines distribution and its breakdown voltage is 569 V.
Figure 4 shows the mechanism of RESURF effect in GaN/AlGaN/GaN HEMT. The RESURF principle is to introduce negative charges to balance the positive space charges in the drift region. While the amount of negative and positive charges is closer, the effect is better. For the structure discussed, due to the polarization effect of GaN cap, extra -Qp_up is induced at the top GaN/AlGaN interface. The electric field lines go from +Qp_down to -Qp_up that suggests a good RESURF effect (Figure 4). As a result, electric field crowding at the drain side of the gate electrode corner decreases sharply so the breakdown increased to 569 V.
Because of the excellent performance of GaN cap layer, a detailed discussion is made. Figure 5 shows the influence of impurity type and concentration of cap layer on the breakdown characteristic. The thickness of GaN cap is kept constant at 3 nm and doping concentration is varied from 2 × 1016 to 5 × 1019 cm-3. The simulation result indicates obviously that the breakdown voltages bear little difference at low doping concentration while P-GaN cap layer shows more potential to achieve higher breakdown voltage than N-GaN cap layer at high doping concentration. It shows the same tendency against the Reference [17] that the breakdown voltage of HEMT with P-GaN cap layer is higher than the structure with N-GaN cap layer. In the experiment, the breakdown voltage is 184 V of P-GaN cap layer in contrast to 114 V of N-GaN cap layer with the same structure.
During off-state (VG=-5 V), the GaN cap layer is depleted completely. P-type impurity induces net negative space charges region (N-type is just on the contrary). When the doping concentration is low, the order of magnitude of space charges in GaN cap layer is about 1016 cm-3 which is much less than the -Qp_up induced by polarization on the up interface of GaN/AlGaN. The electric field lines mainly go from +Qp_up to Qp_up. With the increasing doping concentration the effect of ionized net negative space charge cannot be ignored, the net negative space charges will disperse the electric lines coming from +Qp_up, the RESURF effect will become better, so the breakdown voltage is higher. According to the RESURF principle, only when the amount of negative charges and positive charges is the same, the breakdown voltage can achieve the largest value. It can be assumed that when the doping concentration is 5 × 1018 cm-3 for P-cap, two kinds of charges are balanced. If the doping concentration keeps increasing, the balance state is broken and breakdown voltage starts to decrease.
It is known that HEMT achieve high breakdown voltage based on the high-resistance region generated after 2DEG is depleted in channel. One way to enhance the breakdown voltage is to widen the depleted region. In this paper, a new structure with P-type buried GaN layer emerges, and Figure 6 shows the schematic cross-section of the new structure. The partly P-buried GaN layer can be fabricated by As+ ion implantation and proton implantation[18-20].
A P-doped layer with concentration of 2 × 1017 cm-3 is embedded in the GaN buffer under drain side. The distance from P-layer to the bottom of AlGaN barrier is H which must be larger than channel thickness (>10 nm), so it has little ability to affect the forward turn-on the forward conduction. W and L are the length and thickness for the buried layer. After inserting the buried layer an electric field maximum value is induced at the left edge of the buried layer. Based on the research on LDMOS[21, 22] electric field optimization, when the peak value is in the middle of the gate and drain electrode, the breakdown voltage can reach the maximum value. So, the length of W and L are 1.8-μm and 0.45-μm discussed in this paper.
Other parameters are the same as GaN/AlGaN/GaN HEMT with UID-cap layer before. The breakdown voltage (in Figure 7) is 640 V, which is increased in comparison to conventional HEMT with UID-GaN cap (569 V).
Figure 8 shows the depletion situation in the channel with different cap layer and HEMT with P-buried layer in off-state (Vg=-5 V). Deplete length in this paper is defined as the electron density is low enough that the region can be regard as highly resistive.
It can be seen that 2DEG concentration near the drain electrode is decreased so P-layer embedded in GaN buffer produces a highly resistive drift region between drain and gate. As can be seen, the electron density of HEMT with P-cap is lower than N-cap, and the deplete length increases to 2.32 μm which is 0.8, 2.07, 2.44 μm for UID-cap, N-cap (2 × 1016 cm-3) and P-cap (2 × 1016cm-3), respectively. As a result, the breakdown voltage of the proposed structure is 640 V which is increased by 12% in comparison to UID-GaN/AlGaN/GaN HEMT.
4. Conclusion
In this paper, the dependence of breakdown voltage on concentration of the cap layer is discussed by analyzing the effect of the GaN cap layer. Simulation results indicate that P-type doping could achieve higher breakdown voltage than N-type GaN cap because the negative fixed charge induced by spontaneous and piezoelectric polarizations and the net ionized negative space charge (off-state) in cap layer disperse the electric lines coming from +Qp_up. A novel structure is proposed by applying the buried P-type layer in GaN buffer which widens the depleted region. The BV of the proposed structure is 640 V which is increased by 12% in comparison to UID-GaN/AlGaN/GaN HEMT.