Correlating elemental compositions and charge carrier profiles in ultra-pure GaN/AlGaN stacks grown by molecular beam epitaxy

Attributed to the 3.4 eV bandgap of gallium nitride (GaN) and the high electron saturation velocity of two-dimensional electron gases (2DEGs) confined in GaN/AlGaN heterostructures, GaN-based devices are perfect candidates for commercial high-frequency and high-voltage applications.^1,2 In principal, lateral field-effect transistors (FETs) fabricated from GaN/AlGaN stacks can be used to perform capacitance versus voltage (C(V)) measurements, which, in turn, allow for the calculation of free carrier background levels.³ 

Over decades, the extracted levels of free background charges from C(V) data were commonly interpreted as a metric for the concentration of background impurities—especially donors—in GaN/AlGaN stacks grown by molecular beam epitaxy (MBE).^4,5 Levels of free charges below 10¹⁵ cm⁻³ indicated ultra-high purity of the grown GaN material, in particular, the wide absence of common donor impurities such as silicon (Si) and oxygen (O).

Free background carrier concentrations below 10¹⁵ cm⁻³ were also extracted from C(V) data from one of our GaN/AlGaN heterostructures (sample A) grown by MBE.⁶ As a matter of fact, these levels are by more than one order of magnitude lower than the oxygen impurity levels obtained from secondary ion mass spectroscopy (SIMS) analysis (Fig. 1). It turns out that these low free carrier concentrations are artefacts resulting from an undefined plate capacitor geometry after the 2DEG is depleted and, subsequently, the capacitor bottom electrode disappears. It is demonstrated that the levels of free charges and unintentional donor impurities extracted from C(V) and SIMS, respectively, match very well once a defined plate capacitor geometry with a nonvanishing bottom electrode is employed in acquiring the C(V) data.

Sections II A and II B serve as an overview on the sample growth and device preparation as well as on the measurement routines.

The two discussed GaN/AlGaN heterostructures (samples A and B) were grown in a VG V80H MBE system at growth rates of 240 nm/h on free-standing 2-in, hexagonal, metal-polar GaN substrates at a growth temperature of ∼ 700 °C (measured by optical pyrometry) under slightly Ga-rich conditions. The layer stacks consist of a thick GaN buffer (900 and 600 nm for samples A and B, respectively) followed by an Al_xGa_1−xN barrier (x ∼ 0.06 and 0.05, 16, and 12 nm for samples A and B, respectively) capped with GaN (3 and 30 nm for samples A and B, respectively). The resulting 2DEG is located at the GaN/AlGaN interface. The stack architectures vary slightly to realize different 2DEG densities. The elements Ga and Al were evaporated from single-filament effusion sources. Active nitrogen is provided by an inductively coupled radio-frequency plasma source. The growth conditions result in materials with an unintentional oxygen background concentration of 2 × 10¹⁶ cm⁻³, which is the only donor impurity detectable by SIMS above the detection limits. Here, it is important to mention that the incorporation of oxygen depends on the growth temperature, as already previously reported.^7,8 The employed growth temperature of ∼ 700 °C was found to be slightly below the Ga desorption point (or congruent GaN sublimation temperature in ultra-high vacuum) and thus sets a practical limit not only for the maximum growth temperature, but also for the minimum background of unintentional oxygen incorporated in GaN during MBE growth. The behavior follows an observed trend in a reduction of unintentionally incorporated oxygen by one order of magnitude upon increasing the substrate temperature during growth by ∼ 60 °C (Fig. 2). In other words, GaN with a residual background oxygen concentration of 2 × 10¹⁶ cm⁻³ is practically the purest material which can be grown by MBE, which is a trend previously observed by other groups when growing under similar conditions.⁹ 

The growth conditions for samples A and B are nominally identical, though the main and most important difference between the two samples is the substrate compensation. A heavily compensated, Mg-doped substrate was used for sample A and an uncompensated one for sample B. The consequences for lateral conductivity and C(V) measurements will be discussed in Sec. III.

Microstructures were either defined lithographically or patterned in a reactive ion etch step or by shadow mask patterning during deposition. Two different types of contacts are required to serve as the top and bottom electrodes in C(V) measurements. For the ohmic contact, an annealed Ti/Al/Ni/Au stack is used, which later connects to the 2DEG or deeper conductive layers to serve as a bottom electrode. An as-deposited Ti/Au stack serves as the Schottky counterpart and represents the gate or top electrode. Before the deposition of the Schottky (gate) metal stack, a 27 nm thick aluminum oxide layer is deposited by atomic layer deposition to prevent gate current leakage.

Lateral FETs fulfill all requirements needed for C(V) measurements and are thus used as well (Fig. 3).

Elemental concentrations of oxygen and silicon were obtained from SIMS measurements with element-specific standards for GaN.

Let us first take a closer look at sample A, which was grown on a heavily-compensated GaN substrate. Here, the 2DEG is the only conduction path and consequently, lateral FETs fabricated from this particular heterostructure show very high on-to-off current ratios of >10⁸ in the transfer characteristics.⁶ In C(V) measurements, the 2DEG serves as the bottom plate capacitor electrode. Once the 2DEG is depleted and all electrons are transferred out of the 2D channel into the source/drain contacts, the bottom electrode is lost and an undefined plate capacitor geometry is left. Equation (1) cannot be applied anymore. Mathematically, of course, the transformation can be done, but since the requirements of a parallel plate capacitor do no longer exist, the low level of free background charges in Fig. 1 in the GaN buffer below the 2DEG has no physical meaning.

The situation dramatically changes when taking a closer look at sample B grown on an uncompensated substrate, where massive parasitic lateral conductivity is observed. This parasitic channel is attributed to silicon adhesion at the wafer surface, as described previously and by other groups and us independently.^11–13 Thus, the conductive parasitic channel is located at the substrate/MBE interface and contacted by the annealed ohmic stack. Since it is only possible to deplete the 2DEG, but not the parasitic channel, it will not be possible to fully switch off lateral FETs fabricated from heterostructure B. On the other hand, this parasitic channel represents a huge advantage, since it persists as a defined plate capacitor bottom electrode even when the 2DEG is entirely depleted. Equation (1) is now fully valid and performing the transformation to calculate the depth profile results in a level of free charges in the GaN buffer of ∼ 2 × 10¹⁶ cm⁻³, consistent with the elemental oxygen concentration extracted from SIMS data (Fig. 4).

One open question from past experiments is the definite assignment of silicon as the dominant contamination at the substrate/MBE interface. There are some more candidates around mass/charge of 28 amu/e, namely, AlH, CO, and CNH₂ complexes, but of course also N₂ escaping from the GaN matrix. High-resolution SIMS data taken at the substrate/MBE interface finally demonstrate that beside ¹⁴N molecules the most counts in the SIMS data result from ²⁸Si (Fig. 5). This ultimately proofs that elemental silicon is the contamination at the substrate/MBE interface, and not other organic or metallic complexes.

C(V) measurements performed with GaN/AlGaN heterostructures can provide valuable information, such as the 2DEG position, layer thicknesses, and charge profiles within the layer stack. Particularly, the latter aspect is of tremendous interest, when it comes to the assessment of unintentional impurity background concentrations. Commonly, the transformation of the C(V) data to depth profiles results in unphysical low concentrations, which are inconsistent with elemental analysis data, e.g., by SIMS. The reason for this discrepancy becomes clear when taking a closer look at the stack architecture: In case the 2DEG is used as a bottom electrode in C(V) measurements and the 2DEG is the only conduction path, this bottom electrode is lost after the 2DEG is depleted and no defined plate capacitor geometry exists anymore. This property of course is very desirable for the fabrication of FETs, since these devices are basically perfect insulators in the off-state but unfortunately not suited for C(V) measurements. On the other hand, deeper conductive paths, which are not electrostatically depleted, can serve as a bottom electrode and guarantee a defined plate capacitor geometry, which, in turn, leads to consistent free carrier and elemental donor concentrations. FETs from these stacks on the other hand will suffer from leakage in their off-state when the 2D channel is fully depleted. It looks like the jack of all trades device for perfect transistor functionality and free carrier profiling beneath the 2DEG does not exist, and one has to choose between the different geometries depending on the intended application.

Charge carrier profiles in GaN/AlGaN heterostructures extracted from C(V) data are often not consistent with doping concentrations determined by elemental analysis. Particularly, when a 2DEG is the only conduction path, charge carrier concentrations in deeper regions often become unphysically low due to the loss of the bottom electrode once the 2DEG is fully depleted. Defined bottom electrodes allow for a consistent extraction of free charge carrier concentrations compared to the elemental composition, but have the disadvantage of altering lateral conductivity and hence, impeding lateral FET functionality.

We are grateful to RTG Mikroanalyse Berlin GmbH for providing the high-resolution SIMS data. The TU Dresden part of the work was financially supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—Project No. 348524434. This work was financially supported by the European Fund for Regional Development EFRD, Europe supports Saxony, and by funds released by the delegates of the Saxon State Parliament (No. 100356328). The NaMLab gGmbH part was funded by the DFG—Project No. 405782347, the German Federal Ministry of Education and Research—BMBF (Project “ZweiGaN,” No. 16ES0145K) and the German Federal Ministry of Economics and Technology—BMWi (Project No. 03ET1398B).

The authors have no conflicts to disclose.

Stefan Schmult: Conceptualization (lead); Data curation (supporting); Formal analysis (supporting); Funding acquisition (equal); Methodology (equal); Project administration (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (lead). Pascal Appelt: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Validation (equal); Visualization (equal); Writing – review & editing (equal). Claudia Silva: Resources (equal). Steffen Wirth: Conceptualization (equal); Formal analysis (equal); Methodology (equal); Validation (equal); Writing – review & editing (equal). Andre Wachowiak: Data curation (equal); Formal analysis (equal); Methodology (equal); Supervision (supporting); Validation (equal); Writing – review & editing (equal). Andreas Großer: Resources (equal); Writing – review & editing (supporting). Thomas Mikolajick: Funding acquisition (equal); Project administration (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal).

The data supporting the findings and conclusions of this study are available from the corresponding author upon reasonable request.