Characterization of electrically active defects at III-N/dielectric interfaces

 
« PreviousUpNext »Contents
Previous: Acknowledgement    Top: Home    Next: 1.2 Defects in GaN/AlGaN MIS-HEMTs

1 Introduction

This chapter is dedicated to lay the foundation for the present work. We start with a review of the GaN–based devices properties, from the advantages in power applications to the physical working principles of the high electron mobility transistor (HEMT). We also present a brief summary on device fabrication and its challenges. Then we focus on the effect of defects, in particular those arising at the interface with the dielectric in composite metal–insulator–semiconductor (MIS) HEMTs. Finally, we determine the objective of the present work and we give an overview of how the thesis is organized throughout the following chapters.

1.1 Fundamentals

Wide bandgap semiconductors have become one of the main research subjects of the semiconductor industry in the past years. In particular, GaN and SiC are today the best candidates to substitute silicon in certain applications because of their superior material parameters, as we can see from Table 1.1.

Si SiC GaN
\( \mr {E_g} \) in eV 1.1 3.2 3.4
\( \mathcal {E}_\mr {BD} \) in MV/cm 0.3 3 3.3
\( \Theta    \) in W/(cm K) 1.5 3.3 1.3
\( v_\mr {sat} \) in 107 cm/s 1 2 2.5*
\( \mu                   \) in cm/s 1350 600 2000*

Table 1.1: Comparison of some material parameters for silicon, four–layer hexagonal (4H) SiC and GaN (from [1]). The values marked with * are for a composite GaN/AlGaN structure.

In the first place, a large bandgap results in a high breakdown electric field \( \mathcal {E}_\mr {BD} \), thus allowing higher operating voltages than silicon. Together with a good thermal conductivity \( \Theta              \), this is a clear advantage for power applications as power supplies, DC/DC converters and AC/DC adapters [2].

Furthermore, composite GaN/AlGaN devices can achieve high intrinsic speed because of the large saturation velocity \( v_\mr {sat} \) combined with an excellent electron mobility \( \mu                        \). As a consequence, their performance in terms of operating frequency is superior to both silicon and GaAs. This opens a series of opportunities in the field of RF electronics, like telecommunications and radar applications [3].

In the rest of this section we discuss in detail the properties of the material, the advantages of GaN/AlGaN heterostructures and the fundamental characteristics of GaN–based devices.

1.1.1 The GaN buffer: crystal structure, polarization, growth

All group–III nitrides like GaN can crystallize in three structures: zincblende, rocksalt and wurtzite, the latter being thermodynamically stable at ambient temperatures [4, 5]. The GaN wafers used in this thesis have a wurtzite structure. The unit cell is hexagonal and it consists of two interpenetrating hexagonal closed packed sublattices, as shown in Fig. 1.1. The first sublattice is formed by the cations, and the other by the anions. Their spatial separation along the vertical \( c \) axis results in a spontaneous polarization field in this direction.

Figure 1.1: The wurtzite crystal structure of GaN and alloys like AlGaN. The distance between the cation and anion sublattice results in a spontaneous polarization field along the \( c \) axis. (Illustration adapted from [6].)

The GaN buffer is usually grown using metal–organic chemical vapor deposition (MOCVD) or molecular beam epitaxy (MBE) on different kinds of substrates: sapphire, silicon carbide and silicon. All these materials have different lattice constants and thermal expansion coefficients compared to GaN, and this can cause strong strain effects in the GaN bulk. The best match in this sense is 6H–SiC (3.5% lattice mismatch), which furthermore offers an excellent thermal conductivity, possibly allowing an efficient heat dissipation for power devices. However, the very high price limits the use of this substrate, in favor of the cheaper and widely available silicon (17% lattice mismatch). In fact, a 2–inches SiC wafer can cost up to 500$, while a Si wafer with the same dimensions costs between 20$ and 50$, depending on the doping [7]. Silicon wafers are fabricated in a wide range of sizes and are already in use in semiconductor industries. In this way, the implementation of GaN devices on a Si substrate would benefit from all the existing facilities and processing environments, and the integration of GaN structures on silicon chips would be possible as well.

(image)

Figure 1.2: (a) Optical microscope image of the surface of a MOCVD grown sample. A mosaic structure of hexagonal crystals of about 3 µm is clearly visible (from [4]). (b) Schematic illustration of the GaN columnar growth (from [5]).

Still, one of the main challenges for GaN–based device fabrication is to grow a high–quality GaN buffer. In fact, the first epilayer grown on any substrate contains a large concentration of dislocations, clearly visible in transmission electron microscope (TEM) cross-sections [5]. The dislocation density decreases in the layer above this region, where the GaN continues to grow in a columnar fashion. An image of a GaN sample surface is shown in Fig. 1.2a. The columns develop along the \( c \) direction, displaying some relative twist and tilt (illustrated in Fig. 1.2b), until in the top part of the film the structure becomes more ordered. The quality of the GaN bulk depends on the growth parameters as temperature, pressure as well as the thickness of the nucleation layer. This influences the density of threading dislocations and other defects in the buffer, which can be electrically relevant during device operation.

Figure 1.3: Calculated conduction band edge diagram, neglecting or considering the spontaneous and piezoelectric polarizations. The structure consists of an AlGaN layer of 20 nm above a 50 nm thick GaN buffer (from [8]).

1.1.2 The GaN/AlGaN interface: 2DEG creation, surface donor states

In an AlGaN alloy, some of the gallium atoms are substituted by aluminum. Since the sizes of these two atoms are different, the lattice constant \( a \) of the AlGaN crystal structure is smaller than that of GaN. In fact, in GaN \( a= \) 3.19 Å while for AlN \( a= \) 3.11 Å [2]. The AlGaN lattice constant takes a value in between, depending on the aluminum content. For this reason, a thin AlGaN layer grown on a thick GaN layer is subject to a certain amount of tensile strain. As a consequence of the crystal structure stretch, a piezoelectric polarization field is created, which sums up to the spontaneous polarization of the AlGaN layer. The discontinuity of the polarization field at the GaN/AlGaN interfaces results in the creation of a two–dimensional sheet of charge, the two–dimensional electron gas (2DEG). The band diagram of such a heterostructure is shown in Fig. 1.3. The polarization fields cause a large discontinuity at the GaN/AlGaN interface, pushing the conduction edge below the Fermi level. In this way, a triangular quantum well is formed. Here, electrons must have discrete energy values along the \( c \) direction, therefore their motion is confined to a two–dimensional plane parallel to the interface.

The thin AlGaN layer can accommodate the lattice mismatch by tensile strain up to a certain maximum thickness. For example, the critical thickness is 80 nm for an AlGaN layer with 20% of aluminum, and it decreases to 10 nm for 60% aluminum content [9]. Above this value the strain cannot be sustained anymore and the material relaxes, going back to its original lattice constant. This results in the creation of dislocations, compromising the material quality. The critical thickness depends on the aluminum content: the higher the percentage, the smaller the thickness of the layer which can tolerate the lattice mismatch.

(image)

Figure 1.4: The donor state model: the surface states are neutral below the Fermi level (a) and positively charged when above (b). The AlGaN thickness determines the position of the donors with respect to the Fermi level. This explains the critical thickness observed in experimental data (c): solid markers are measured data and the line is calculated from the model (from [10]).

The electron gas at the GaN/AlGaN interface offers very high electron density (about 2 × 1013/cm2) and mobility (2000 cm/s) with no intentional doping, qualities that have attracted the interest of semiconductor device manufacturers and many research groups. Nevertheless, the exact origin of the 2DEG is not yet completely understood. The polarization charges at the GaN/AlGaN interface and at the AlGaN surface constitute a dipole, whose net charge must be zero. Because of the nature of polarization charges, they cannot be the source of the free electrons in the 2DEG, which must therefore be found elsewhere. The currently most accepted model is the surface donor state model, introduced by Ibbetson at al. [10]. In this picture, donor states at the AlGaN surface are considered the source of electrons for the 2DEG. These states would be energetically located inside the AlGaN bandgap, and would contribute to the electron gas when they rise above the Fermi level. On the contrary, if they are below the Fermi level, there are no electrons at the GaN/AlGaN interface. Since the electric field throughout the AlGaN layer remains constant, by changing the AlGaN layer thickness the position of the surface state with respect to the Fermi level changes as well. This is illustrated in Fig. 1.4a and b. This is confirmed by experiments, as shown in Fig. 1.4c: below a certain critical thickness, no electron gas is formed.

In conclusion, the AlGaN layer must be thick enough in order to allow the existence of the electron gas, but beyond a certain thickness the material would relax and develop many defects like dislocations. Therefore a trade–off between these two limits must be found. In this thesis we use structures with an AlGaN layer thickness between 20 and 30 nm.

1.1.3 GaN–based devices: the high electron mobility transistor

The two–dimensional electron gas originated at the GaN/AlGaN interface can be used for the fabrication of field effect transistors (FETs). This type of device is a heterojunction FET (HFET) and it is usually called high electron mobility transistor (HEMT), because of the high mobility achieved by the electrons in the 2DEG.

Two ohmic contacts are in direct connection with the 2DEG, which therefore constitutes the channel between source and drain. The top gate contact can be realized directly above the AlGaN as a Schottky contact (HEMT device), or with a dielectric layer in between (MIS–HEMT device). In both cases this contact regulates the flow of the electrons in the channel, from the full saturation current down to zero, when the 2DEG is depleted. The insertion of the dielectric allows to suppress the parasitic leakage currents through the gate stack [11]. The structure of an insulated gate HEMT (MIS–HEMT) is schematically shown in Fig. 1.5.

Figure 1.5: Schematic illustration of a MIS–HEMT structure.

Since the channel is formed as a consequence of the polarization properties of the heterojunction, these devices are normally on. This means that they must be turned off by the application of a negative gate voltage. Especially in power applications, transistors with normally off characteristics are preferred. The reasons include safety during operation at high voltage or high power densities, lower driving power consumption and compatibility with existing driving circuits. Several modified device design concepts and processes during fabrication allow to push the threshold voltage to positive values, for example fluorine implantation [12], the use of p–type GaN for the gate contact [13] or the recess of the Schottky gate [14]. As an alternative, normally on transistors can be used in cascode configuration [15].

In this work we use normally on MIS–HEMTs in order to investigate the defects at the AlGaN/dielectric interface and their contribution to device degradation through charge trapping. In this way we keep the complexity of the structures to a minimum and we focus on the properties of interface states inherent to the system.

« PreviousUpNext »Contents
Previous: Acknowledgement    Top: Home    Next: 1.2 Defects in GaN/AlGaN MIS-HEMTs