Native substrates spar with the established technology

Capturing these opportunities requires further improvement in device operating performance, reliability and cost - all of which are limited by the currently available substrate technology. Simple materials arguments suggest that all III-nitride devices could benefit from a native substrate (e.g. AlN or GaN), as history has demonstrated for silicon, GaAs, InP and, more recently, SiC (homoepitaxial) devices. However, there is no current supply of large-diameter, uniform, low-cost native III-nitride substrates.

Until recently, device developers and fabricators had been forced to rely primarily on foreign substrates like sapphire and SiC. These materials have supported blue- and green-LED commercialization and have enabled the demonstration of new devices such as UV LEDs and RF FETs with record-breaking physical characteristics.

Progress in blue-laser development has been limited by dislocations in the active region, and was slow until quasibulk GaN became available [S Tomiya 2001]. To meet future market requirements, the performance, reliability and cost of blue lasers and GaN-based RF FETs must improve. The most structurally perfect blue laser, giving rise to record pulsed output, appears to have been fabricated on ultralow defect density, native GaN substrates. Conversely, the performance of UV devices built on sapphire and SiC degrades rapidly at shorter wavelengths, where native AlN substrates may provide a solution. And in spite of the progress made in the performance of RF FETs on SiC, competing efforts on quasibulk and native substrates are beginning to show promise.

The utility of a given substrate in building most devices is shown by several substrate-related properties as well as the quality of the buffer layers grown on it. Table 1 summarizes how the three basic substrate types compare in terms of the materials properties of both the substrate and the buffer layer. The buffer layer is assumed to be GaN for foreign substrates and a lattice-matched (Al,Ga)N composition for the others. Foreign substrates have low defect densities, but, owing to large lattice- and thermal-expansion mismatches, they produce buffer layers with high defect densities.

Quasibulk approaches begin with large, foreign substrates and therefore have a good (large) size and shape. Through cost-additive intermediate processing steps (usually hydride vapor-phase epitaxy (HVPE) or epitaxial lateral overgrowth (ELOG)), their dislocation densities and those of the buffer layers grown on top drop to around 105 to 107 cm-2.

Native III-nitride substrates show excellent structural quality, but suffer in terms of size, shape and uniformity. While native AlN substrate technology can currently achieve 1 inch diameter material and appears highly scalable, truly native GaN substrate technology (i.e. not grown by HVPE or ELOG methods) has not produced significant increases in substrate size over the past few years and only small-diameter material is currently available.

Foreign substrates are, by definition, materials that have already undergone the test of diameter expansion. Quasibulk substrates begin as foreign structures and thus have the potential to be made with a large diameter. Native GaN and AlN substrates lag behind these other materials in terms of size.

The relative thermodynamic instability of GaN demands solution or liquid and high-pressure growth processes that, until now, have been difficult to scale past 1 cm. Native AlN substrates are produced by sublimation recondensation: a slow, yet deliberate mechanism for diameter increase similar to that of SiC. One key difference between AlN and SiC is the relative simplicity of the AlN phase diagram. This means that multiple crystalline phases are not competing during boule growth. As a result, the diameter of native AlN substrates can increase. Limited availability of 2 inch diameter substrates is expected in 2005.

The relative concentration of dislocations in the active region of a typical device structure is expected to vary widely with the substrate used (see figure 1). In practice, growth on native GaN and AlN substrates has often shown defect levels of up to 105 cm-2, primarily due to surface preparation and/or epitaxy initiation issues. The best device results have been obtained almost exclusively by the most experienced research groups.

For device manufacturers, it is device performance, reliability and cost that matter. For a given device performance and reliability, the total manufacturing cost depends on the combined costs of materials, epitaxy and device processing. A low-cost substrate that produces a device that fails basic performance or reliability requirements because of a highly dislocated active region is not a low-cost solution. On the other hand, an expensive yet high-quality substrate (from an isolated single device perspective) that cannot be scaled or reduced in cost is not a long-term solution. The important question is which substrate can provide the lowest cost of manufacturing for devices that meet target performance and reliability specifications.

It is still useful to understand the cost factors that will limit the pricing structure of a given substrate technology. The cost of substrate manufacturing depends strongly on the effective substrate growth rate (EGR), which accounts for three elements of the process speed: actual growth rate during the boule production process; boule growth-process overhead time (i.e. growth initiation and termination); and process equipment downtime.

The EGR depends on many variables, including the length of the boule that is grown (since long boules lead to greater reactor utilization efficiency). The EGR for quasibulk substrates is less than 0.1mm/h and is limited by the relative slowness of the HVPE process or, in the case of ELOG, by additional (epitaxy and lithography) processing steps. In marked contrast, EGRs for native AlN substrates and foreign substrates like SiC are believed to be closer to 1 mm/h. So foreign substrates offer a cost advantage that will be shared by native substrates as the latter mature. Quasibulk substrates may have an inherent cost disadvantage due to process-throughput limitations.

The a-direction lattice constant of AlN and GaN differ at room temperature by 2.4%, which results in a critical thickness of about 12 nm. Hypothetical structurally and otherwise perfect AlN and GaN substrates therefore differ in their ability to lattice-match to any given epitaxial-layer structure. The simplest wisdom, albeit unproven experimentally, is that gallium-rich epitaxial-layer structures are best produced on GaN substrates, while aluminum-rich epitaxial-layer structures are best produced on AlN substrates. Intermediate average-composition epitaxial-layer structures might best be grown on (Al,Ga)N substrates, which are now in development, albeit at dislocation densities characteristic of HVPE.

Gallium-rich epitaxial-layer structures are currently used in visible LEDs and lasers, and for most RF FETs. Gallium-rich designs have led to difficulties in obtaining semi-insulating buffer layers, yet it is difficult to change buffer-layer composition when using foreign substrates since the buffer-layer growth must be re-optimized to minimize cracking and defect generation.

Aluminum-rich FET structures that can increase channel charge and operate at significantly higher voltages should benefit from highly insulating AlN buffer layers. The drive to develop UV LEDs and lasers also requires aluminum-rich structures.

Strain build-up and film-cracking result from differences in thermal expansion coefficients (TECs) between epitaxial films and the underlying substrate as they cool after epitaxial growth, which is typically carried out near 1100 °C. The degree of potential strain build-up is calculated by integrating the change in the TEC between the film and substrate from the film-growth temperature to room temperature.

Figure 2 shows the integrated change in TEC in the a-b plane of GaN grown on AlN, SiC and sapphire. C-plane GaN films grown on a-b plane substrates at 1100 °C that are cooled to room temperature can suffer integrated TEC-mismatch-associated biaxial strains of up to 1200 ppm (tensile) on SiC and 2600 ppm (compressive) on sapphire. Because the TECs of GaN and AlN are fairly close and cross at 475 °C, the potential strain build-up is only 100 ppm (compressive).

In practice these strains may be at least partially relaxed through extended defect generation during cool-down and can be compensated for by building opposing strain elsewhere in the epitaxial structure.

Thermal issues play an important role in the performance and reliability of most III-nitride devices. When pushed hard, the device generates heat due to suboptimal operating efficiencies linked to the quality (traps and scattering centers) of the device s active region and to higher than desired contact and device-layer resistances. The choice of substrate affects not only the quality of the device s active region but also its ability to dissipate heat.

Of the four materials, SiC has the highest thermal conductivity (see table 2; the upper range reflecting the highest-purity variations is quoted). However, the imperfect lattice match and TEC match between SiC and typical III-nitride epitaxial-layer designs results in poor structural quality buffer layers, which are thought to represent major thermal bottlenecks. The thermal conductivity of GaN is known to depend on dislocation density [D Kotchetkov et al. 2001], which is important for understanding thermal bottleneck issues in buffer layers on foreign substrates and quasibulk substrates in general.

The thermal boundary resistance associated with the heterointerface between the buffer layer and a foreign substrate has also been implicated in modeling studies as a general device heat-dissipation bottleneck [K A Filippov et al. 2003]. Native AlN may therefore be the best substrate choice from a thermal-dissipation perspective, thanks to its high thermal conductivity (3 W cm-1K-1) and amenability to producing high-quality AlN buffer layers and active regions.

Since foreign substrates have so far dominated III-nitride device development, and because so few devices have been produced on native and quasibulk substrates, a comparison of device results is of limited usefulness. However, native AlN substrate use has led to increased optical device efficiency at short emission wavelengths, as well as improved RF FET gate characteristics, apparently due to more structurally perfect active regions.

Quasibulk "free-standing" GaN substrates were recently used to build RF FET devices with good power performance [K K Chu et al. 2004], while native GaN substrate based efforts have demonstrated the highest output power (3.9 W for a 30 ns pulse) blue lasers. Piotr Perlin and colleagues at Poland s High Pressure Research Center made these lasers using GaN substrates manufactured by TopGaN in what is likely the lowest defect-density blue-laser structure produced to date.

Cree has made similarly good progress with wide-bandgap electronics on quasibulk substrates. In very recent unpublished work, the US-based company fabricated high-quality HEMT devices grown on its own semi-insulating GaN wafers. Operating at 4 GHz and 68 V, these devices had an output power of 9.2 W/mm a power-added efficiency of 55% and a gain of 16 dB.

K A Filippov et al. 2003 MRS Internet J. Nitride Semicond. Res. 8 4.
P Perlin et al. 2004 MRS Internet J. Nitride Semicond. Res. 9 3.
K K Chu et al. 2004 Electron Device Letters 25 9.
D Kotchetkov et al. 2001 Appl. Phys. Lett. 79 4316.
S Tomiya 2001 Phys. Stat. Sol. (a) 188 (1), 69-72.