Germanium – A Surprise Base For High-quality Nitrides
The high cost and low availability of GaN substrates have driven nitride-chip manufacturers to use other platforms for growth, such as sapphire, SiC and silicon. These three alternatives have provided the basis for the production of countless devices, but in every case intermediate layers have had to be inserted into the structure to combat significant lattice and thermal mismatches (see table). Although the additional buffer layers are only an irritation for LED manufacturers, they are the show-stopper for the fabrication of heterostructure devices such as GaN HBTs that require contacts to be made directly to the substrate. This is because they prohibit the formation of a heterojunction between the upper, high-quality, GaN layer and the substrates.
This particular limitation has prompted our team at the Interuniversity Micro Electronics Center (IMEC) in Belgium to investigate alternatives that could ultimately be used for our development of an InGaN-based photo-electrolysis cell, which will use solar radiation to split water. This type of device would really benefit from a conductive substrate in contact with the active InGaN layer, because back contacting can be applied to the design.
Our pursuit of a new nitride platform has focused on germanium (111) material. A theoretical lattice mismatch of −20% with GaN suggests that it would be very challenging to produce high-quality films. However, if we could overcome this obstacle, we could reap the rewards of germanium s otherwise excellent properties. These include a relatively low bandgap that will lead to more uniform heating in MBE growth chambers; a native oxide that is much easier to remove prior to growth than the oxide associated with silicon; and a very small thermal mismatch with GaN, which prevents the formation of large thermal-induced stresses during sample cooling after the growth stage.
It turned out that GaN can actually form high-quality films by direct growth on germanium, probably because there is a very favorable relationship between the sizes of the germanium and GaN lattices (see figure 1). The mesh ratio between the two hexagonal structures is close to 5:4, and the lattice mismatch between the two materials is just +0.4%.
We produced our films by MBE using a Riber tool fitted with an Oxford Applied Research nitrogen plasma source (see figure 2). The common alternative technique, MOCVD, is not suited to this material system because high-quality GaN demands a reactor temperature of 1000–1100 °C and germanium melts at 940 °C. With MBE this is not an issue, because the epitaxy is carried out below 800 °C.
We have characterized our GaN-on-germanium epiwafers using various techniques. Reflection high-energy electron diffraction (RHEED) images taken in the reactor reveal an abrupt transition at the onset of growth, followed by the formation of a smooth surface. Post-growth X-ray diffraction measurements on a 38 nm-thick film confirm that this material has a good crystal quality. X-ray rocking curve measurements on the same sample produced a peak from the GaN epilayer with a full-width half-maximum of 371 arcsec, indicating that the crystal quality is comparable to that of GaN grown on sapphire and silicon. Additional "ω-2θ" scans on the same sample demonstrated that the GaN–germanium interface is also of high quality. Current–voltage measurements on our material also show rectifying behavior for n-type GaN on p-type germanium (see figure 3).
This series of studies shows that high-quality GaN films are grown directly on germanium (111) without the need for intermediate layers. When formed in this way, GaN has a small compressive stress as the thermal mismatch between these layers is only 5.5%. Because GaN can undergo "surface wetting" on germanium, and no chemical reaction occurs between gallium and germanium, the interface between the two materials is of high quality.
These results are very encouraging, but two issues still need to be addressed. The first is that the GaN layer always behaves as an n-type material, which is probably due to some germanium diffusion. The exact cause of the n-type doping is unknown, but it probably results from etching of defects in the germanium surface during the growth, which creates voids in the substrate. This is consistent with secondary ion mass spectrometry measurements, which show that any diffusion between the gallium and germanium atoms is limited. However, even a limited diffusion of germanium into GaN can create a considerable carrier concentration.
The other drawback of our current GaN-on-germanium samples is that they contain two phases that have a 4° twist with respect to one another, according to transmission electron microscopy images and X-ray diffraction curves performed at Cambridge University. However, resolving this issue should only require a switch to off-axis germanium (111) substrates.
It could therefore be possible to make nitride-based structures on germanium substrates that can be used as the back contact in vertical devices. Junctions can be formed between n-type GaN and p-type germanium, or p-type GaN and n-type germanium, and used in devices such as HBTs, LEDs, heterojunction solar cells, and heterojunction diodes.
HBTs could be formed by growing n-doped GaN on p-type and n-type germanium (see figure 4). This would create a device that would have a bandgap difference at the GaN–Ge heterojunction of 2.7 eV, far larger than that of Si/SiGe, InP/InGaAs, and AlGaAs/GaAs, which have bandgap differences of just 0.2, 0.7 and 0.3 eV respectively.
The enormous bandgap difference between GaN and germanium could provide a very large DC current gain, or be traded to invert the base and emitter doping concentrations. A higher base doping concentration would deliver two major benefits: a cut in base resistance, leading to lower power consumption and faster switching; and an increase in pinch-off voltage, which can also boost the switching speed through a shortening of the base width.
There are more ways in which the switching speeds and power consumptions can be improved. The large bandgap difference suppresses the emitter transit time, which increases switching speeds, while the very small bandgap of germanium cuts the operating voltage of the transistor s base-emitter junction and lowers its power consumption.
GaN-on-germanium HBTs should also benefit from good intrinsic characteristics. Electron mobility for germanium, a key parameter for NPN HBT performance, is 3900 cm2/Vs, and the surface recombination speed in this material is lower than that of GaAs and InP. This means that GaN-on-germanium HBTs would not suffer from poor low-current gain and non-ideal base characteristics that can plague other forms of this device.
All of this sounds very promising. However, before we can unlock the great potential of this material combination, we must minimize the HBT s depletion recombination current by fabricating a heterojunction with very few defects. To do this we must improve our fabrication processes, as the one used today appears to create defects through unwanted etching into the germanium substrate.
The alignment of the conduction band edges of GaN and germanium is also under investigation. If they are not aligned, conduction band spikes occur at the interface that can restrict transistor gain (although this reduction is only small owing to the large differences in bandgap between the two materials). The band structure is also influenced by GaN s polarity on germanium, which is currently unknown and under investigation.
Although there are unresolved issues, the GaN-on-germanium material system shows great promise for the fabrication of very fast transistors operating with lower power consumptions than current offerings. At IMEC we will continue to refine this structure before developing an initial device. We will then assess whether this transistor merits further investigation.