Scrutinizing AlN-GaN heterostructures atom by atom
GaN has many attributes. It can handle very high electric fields, electrons can zip through this material at very high speeds, and when this wide bandgap semiconductor is paired with Al(Ga)N, it is possible to form a two-dimensional electron gas (2DEG) with a high charge density.
Thanks to all of these beneficial characteristics, it is possible to produce a range of high-performance HEMTs that can deliver incredibly high output power from the S-band through to the W-band with high efficiency. These transistors also sport a high drain current density and a high cut-off frequency, making them suitable for integrated digital or control functions. What’s more, monolithic integration of miniaturized enhancement- and depletion-mode (E/D) GaN HEMTs can deliver an unprecedented combination of high frequency and high-breakdown characteristics. This promises the feasibility of high-density, GaN-based, mixed-signal circuits.
Table I. Hall measurements including 2DEG sheet density (ns), mobility (µ), and sheet resistance (Rs) for AlGaN/AlN/GaN heterostructures grown by plasma-assisted MBE, ammonia MBE, and MOCVD
Like other classes of transistor, shrinking device dimensions has spurred GaN HEMTs to higher speeds. However, continuing in this vein is getting more challenging as parasitics will limit device performance, and it now appears that the best way to quash parasitics as these transistors move to millimetre- and sub-millimetre-wave frequencies is to turn to a disruptive device technology based on the N-polar (0001) orientation of GaN.
The reason why this switch is beneficial is that there is a lack of inversion symmetry in wurtzite III-nitride materials: The polarization of N-polar crystals is opposite to that of the Ga-polar (0001) crystals. This means that polarization-induced electric fields in N-polar heterostructures are opposite to those in the Ga-polar counterpart, inducing a 2DEG above the wide-bandgap barrier layer, instead of below it (see Figure 1). Thanks to this, N-polar GaN HEMTs with an inverted structure – a GaN channel, an Al(Ga)N barrier and a GaN-buffer – possess an inherent back-barrier that confines electrons and diminishes short-channel effects. The new architecture allows contact to the 2DEG through the channel layer, which has a narrower bandgap and lower surface barrier to electrons, compared with the wide-bandgap Al(Ga)N barrier. Ultimately, this means that N-polar devices have much lower ohmic contact resistances than conventional, Ga-polar HEMTs.
Figure 1. Equilibrium band diagrams of (a) generic Ga-polar (0001) and (b) N-polar (0001) heterostructures
For both Ga-polar and N-polar HEMTs, the idea of employing AlN as the charge-inducing barrier is attracting much attention, because of the large conduction band offset and high polarization charge density at the AlN/GaN interface. AlN can also fulfill other roles: Inserting it between an AlGaN barrier and GaN channel layer improves the electron mobility by reducing alloy scattering.
In both classes of GaN HEMTs, the 2DEG is confined at the AlN/GaN interface with a net positive fixed polarization charge. The transport properties of this electron gas, which impact HEMT performance, are directly related to the roughness and abruptness of the interface, as well as the chemical purity of the AlN. Meanwhile, at the opposite AlN/GaN interface – which has a net negative polarization charge – deep donor trap states form that are uncovered by deep level transient spectroscopy. Extensive electrical characterization has been used to try and uncover the physical nature of this trap, but it remains a mystery.
If a characterisation tool is to help to uncover these physical properties of AlN/GaN heterostructures, it will have to reveal compositional and interfacial profiles of polar AlN/GaN interfaces, and expose the incorporation of unintentional impurities into the HEMT’s mobility-enhancing AlN interlayers.
Probing atomic structures
Unfortunately, the widely used techniques for scrutinizing semiconductor structures are of limited assistance. Although high-resolution microscopes – namely Transmission Electron Microscopy (TEM) and Scanning TEM (STEM) – provide high lateral spatial resolution imaging and detailed structural information, they are limited to two-dimensional atomic resolution, and they fail to address the critical and fundamental capability of generating three-dimensional atomic resolution. This restriction to just two dimensions also hampers techniques for extracting chemical information, such as electron energy loss spectrometry (EELS), energy-dispersive X-ray emission spectrometry (EDS) and secondary ion mass spectrometry (SIMS). What is needed is a reliable technique that delivers precise, three-dimensional characterization at sub-nanometre levels. Such a technique can also be useful for determining dopant diffusion profiles around the drain/source, providing interface profiles between junctions, and uncovering the roughness of a multilayer contact.
The power of atom probe tomography is associated with its capability to generate three-dimensional maps of material with atomic resolution
Atom probe tomography is capable of doing all of this, and has the unique capability to combine detailed composition – with better resolution than EELS or SIMS – with the determination of structural features near the atomic scale. Three-dimensional maps can be constructed of detected atoms or molecules with equal efficiency (detection efficiency is greater than 50 percent) and without prior knowledge of the composition. This technique has a spatial resolution better than 0.3 nm, an analytical sensitivity of 1 atom per million and it can be used to study volumes greater than 106 nm3.
Atom probe tomography is based on a combination of field ion microscopy and mass spectrometry (see Figure 2). Like the former technique, a needle shaped specimen with an end radius of less than 100 nm must be prepared before the measurement can begin. This specimen is cooled to cryogenic temperature under an ultra-high vacuum of around 10-11 mbar and subjected to an electric field, which creates an incredibly high DC field at the apex of the tip – it is tens of volts per nanometre. In addition to this field, voltage or laser pulses are applied to the specimen, triggering ionization of surface atoms that are subsequently field evaporated. The extracted ions are projected onto a position-sensitive detector that records their locations, which are linked to the former positions of the atoms on the tip. The isotopic identity of these atoms is revealed with a time-of-flight mass spectrometer that determines the mass-to-charge ratio of the ion. Armed with this information, it is possible to generate a three-dimensional reconstruction of the locations of tens of millions of neighbouring atoms.
Figure 2. Atom probe tomography can provide a three-dimensional map of a structure on the atomic scale
Our team at the University of California, Santa Barbara, has used atom probe tomography to scrutinise nitride-based heterostructures. This effort kicked-off with a detailed study of a series of MBE-grown, GaN/AlN/GaN heterostructures, which were deposited on Ga-polar and N-polar GaN substrates using two different nitrogen sources: plasma nitrogen and ammonia.
The sample grown with a plasma nitrogen source has two AlN layers separated by a 20 nm-thick GaN layer. To study any effect of a gallium adlayer on the AlN/GaN interfaces, the first AlN layer was grown under metal-rich conditions, while the second was deposited under a slightly nitrogen-rich condition (prior to growth of the second AlN layer, the gallium adlayer was desorbed during a growth interruption, and no gallium was supplied during AlN growth). Note that when we switched to an ammonia source of nitrogen, we had intrinsically nitrogen-rich deposition conditions.
Atom probe tomography allowed us to identify differences in the atomic structure of both these samples (see Figure 3). A compositional profile based on a cylindrical volume with a 20 nm diameter reveals aluminium compositions of the N-polar and Al-polar AlN layers of 0.98 and 1, respectively, which are in close agreement with the target compositions.
Figure 3. Three-dimensional atom distributions of plasma-assisted, MBE-grown (a) N-polar (b) Ga-polar AlxGa1-xN/GaN heterostructures, where the red and blue dots represent aluminium and gallium atoms, respectively. A cylindrical volume with a 20 nm diameter was selected for the composition measurement from the centre data set of both samples, in order to make the interfaces perpendicular to the cylinder. (c) and (d) show chemical profiles of aluminium and gallium in the pure AlN layer indicated in red rectangle in (a) and (b)
Our approach also exposes the asymmetric nature of the AlN/GaN interfaces in the metal-rich GaN/AlN/GaN heterostructure. For N-polar, chemical abruptness is observed at the top interface between AlN and GaN, while the bottom interface between this pair of materials is diffuse. Meanwhile, the opposite situation occurs in our Ga-polar structure: The top interface is chemically diffuse, and the bottom one is abrupt.
What about artefacts?
Artefacts can plague atom probe measurements. We took this into account when interpreting our results, investigating the possibility that the tip shape changed during the measurement, due to differences in the evaporation field of AlN and GaN. This evaporation field is higher for the former material, ruling out any possibility of an artefact causing our results, which would require an opposite state of affairs for differences in evaporation fields.
Our findings are strengthened by atom probe analyses on N-rich GaN/AlN/GaN heterostructures grown by both MBE-based techniques. Measurements confirmed observed asymmetry did not originate from the metal-rich, plasma-assisted MBE process. What we have concluded is that all GaN/AlN/GaN interfaces are asymmetric, and that inversion of crystal polarity reverses the characteristics observed in these regions, regardless of growth conditions (Figure 4a).
Figure 4. (a) Chemical profiles of aluminium and gallium of the pure AlN layer grown by ammonia MBE (b) High angle annular dark-field STEM image of the AlN layer shown in (a). The step-terrace features at the interfaces were due to a 4º miscut. The enclosed region in (b) is enlarged in (c) along with the corresponding line intensity profile in (d)
To provide an independent validation of our atom probe results, we employed STEM, high-angle annular dark-field imaging to scrutinise our N-polar samples grown by MBE with ammonia as the nitrogen source. A cross-sectional micrograph obtained with this technique reconfirms the asymmetry of the GaN/AlN/GaN interfaces, with the top interface appearing abrupt, while the bottom one looks diffuse. A line intensity profile across the GaN/AlN/GaN interfaces clearly shows this asymmetry. In agreement with the results from atom probe tomography, TEM images and line intensity profiles indicate a diffuse interface of about one unit cell.
The upshot of this study is that we have revealed how the polarity at GaN/AlN interfaces impacts their structural and compositional characteristics. If they are positively polarized, they are abrupt; but if they are negatively polarized, they are diffuse (this conclusion holds whether the heterostructure is Ga-polar or N-polar). It is possible to relate these findings to the well-known electrical properties of GaN/AlN/GaN heterostructures. The highly polar interfaces between GaN and AlN seek to minimize their electrostatic potential energy, through screening of polarization charges by free carriers. At a positively polarized interface, neutralization occurs via the accumulation of a 2DEG – electrons come from intrinsic donors, and for Ga-polar structures these come from the surface; but for N-polar, they come from traps. In contrast, it is not possible to neutralise a negatively polarized interface with a two-dimensional hole gas, because very few holes exist in unintentionally doped III-N materials. This observation tallies with the results of calculations by scientists at Rockwell International in the late 1970s. According to their work, interfaces with a large charge imbalance reconstruct and become chemically diffuse.
Adding AlN interlayers
We have also studied three different AlGaN/AlN/GaN HEMT structures that were grown by either plasma-assisted MBE or MOCVD. They feature an AlN interlayer that reduces the penetration of the electron wavefuction into the AlGaN barrier, and ultimately minimises the impact that alloy scattering can have on 2DEG mobility.
For MOCVD growth, the optimal thickness of the AlN layer is 0.7 nm, while for plasma-assisted MBE it is 2 nm. In addition, we grew a structure with a 2 nm-thick AlN layer by MOCVD: Although we knew that the thicker MOCVD interlayer would degrade material quality, we felt that it would provide a useful comparison with the MBE-grown sample. All samples were scrutinized with atom probe microscopy, and the findings compared with results of Hall electrical data. Atom probe microscopy proved very insightful, revealing a massive difference between the two samples (see Figure 5). A well-defined AlN interlayer exists in the sample grown by MBE, but for both MOCVD-grown samples there is no distinct AlN interlayer, but rather an unintentionally graded AlGaN alloy.
Figure 5. Atom probe microscopy reveals that different growth techniques lead to variations in the composition of a thin interlayer that is nominally made of AlN. Measuremnents were made of the three-dimensional distribution of gallium atoms from AlGaN/AlN/GaN heterostructures grown by plasma-assisted MBE with 2 nm interlayer (a) and MOCVD with 2 nm and 0.7 nm interlayer ((c) and (e), respectively). Group III-site chemical composition profiles (b) (d) and (f) were taken from (a), (c) and (e), respectively
Armed with this knowledge, it is far easier to interpret the Hall data for these samples, which were gathered at room temperature and 77 K and obtained via the van der Pauw Hall technique (see table 1). Our measurements, which show a deviation of below 5 percent between the sheet resistances measured across each sample, reveal that the lower the gallium content in the AlN interlayer, the higher the sheet concentration. For the sample that has a 2 nm-thick interlayer and was grown by plasma-assisted MBE, we recorded higher mobilities due to reduced alloy disorder and interface roughness scattering. Meanwhile, the MOCVD-grown equivalent with an unintentional AlGaN interlayer is plagued by the combination of increased alloy disorder and interface roughness scattering, which both drag down 2DEG mobilities. Interestingly, despite high gallium concentrations in the interlayer, a very high 2DEG mobility was measured in the MOCVD-grown heterostructure that nominally contained a 0.7 nm-thick AlN layer. Note that we cannot infer that unintentional gallium incorporation into AlN will occur in all MOCVD reactors. It is in fact quite probable that our particular gas flow sequence played a role in the unintentional incorporation of residual gallium in the boundary layer.
Values for electron mobility have been put in perspective by performing a measurement at 300K on an MOCVD-grown AlGaN/GaN heterostructure without an AlN interlayer. The mobility in this structure was 1600 cm2/Vs, about 500 cm2/Vs lower than both the sample grown by plasma-assisted MBE and featuring a 2 nm-thick AlN layer, and the MOCVD-grown structure that was designed to have an AlN layer that is 0.7 nm-thick.
Our work showcases the capability and value of site-specific atom probe analysis on electronic devices. At the critical sub-100-nm length scale, the three-dimensional resolution achievable by atom probe tomography offers unique insights into the structural and chemical properties of active device layers and interfaces for growth optimization. Atom probe tomography can thus be a very powerful tool in correlating the electrical properties of a device to material properties. These capabilities, along with the means to provide a wealth of quality information, will strongly impact the processing and analysis of devices and components.
