Technical Insight
GaN on anything
Giving GaN freedom from Nb.N
BY DAVID MEYER, BRIAN DOWNEY, D. SCOTT KATZER, MARIO ANCONA, SHAWN MACK AND LAURA RUPPALT FROM THE U.S. NAVAL RESEARCH LABORATORY
GaN devices are transforming our lives. Leading the way is the LED, which is lighting our homes, offices and public spaces, thanks to its competitive pricing, great efficiency and ability to hit full brightness in an instant. But there is also a significant and growing market of GaN electronics, where devices are becoming increasingly attractive for efficient power switching and amplifying RF power. Within this sphere, GaN is enabling the construction of DC-AC inverters with unprecedented efficiencies of above 95 percent that can serve solar photovoltaics and electric/hybrid cars; and it is providing power amplifiers for radar systems and communication networks with RF output power densities that are ten times that of those based on more conventional III-V compound semiconductors, such as GaAs and InP. Impressively, for this application, the level of superiority provided by GaN spans nearly all microwave and millimetre-wave frequencies.
Despite all the material advantages that GaN brings, it would be naive to believe that this wide bandgap technology can provide all the microelectronic functionality desired for today's mixed-signal systems, and do so at low power consumption levels and acceptable costs. Why would anyone choose GaN for high-density digital logic or memory, when silicon CMOS is so far advanced and widely available?
In a circuit designer's ideal world, the optimal solution would involve a seamless mixing of disparate technologies/materials, in a manner that mirrors the construction of objects with coloured LEGO blocks. However, this concept of heterogeneous integration remains far from being fully realized.
In a circuit designer's ideal world, the optimal solution would involve a seamless mixing of disparate technologies/materials, in a manner that mirrors the construction of objects with coloured LEGO blocks. However, this concept of heterogeneous integration remains far from being fully realized.
One major roadblock is that the best substrates for ensuring high-quality growth of electronic materials tend to be incompatible with critical design goals, such as low thermal resistance and cost.
To tackle this issue head on, our group at the Naval Research Laboratory has recently developed a versatile technique that releases III-N devices from their costly SiC substrates. This technology has the potential to form freestanding GaN and AlN.
The pivotal enabler is a thin layer of niobium nitride, Nb2N. It has several important attributes: a close lattice-match to GaN; temperature-compatibility with III-N growth; and a selective etch character over GaN, AlN and SiC. Using Nb2N as a sacrificial layer allows simple processing, based on standard fabrication equipment, for the lift-off and transfer of III-N devices to any desired substrate after completing front-side processing and yield screening.
There are numerous opportunities arising from this new technique. Potential uses range from enhancing discrete device performance via improved thermal management to reducing cost through substrate re-use and achieving superior hybrid circuit performance with device-level heterogeneous integration.
Developing niobium nitride
Our efforts focus on the β phase of Nb2N. This metallic allotrope has a hexagonal crystal structure similar to that of SiC, GaN, and AlN (see Figure 1 for details of how it compares with common wide bandgap, hexagonal materials).
Figure 1: Energy bandgap and lattice constant of several hexagonal crystalline materials.
Aside from their similar crystal structures and common anion, Nb2N and other III-N materials have very little in common. As opposed to GaN, where there is a wide thermodynamic stability window that tolerates large variations in the gallium-to-nitrogen ratio during crystal growth, the Nb-N compound family has as many as nine different phases, making the growth of the desired b-phase very difficult and strongly dependent on stoichiometry. Making matters even more challenging, niobium has to be heated to nearly 2500 °C before it melts. That's substantially higher than the temperature limits of conventional effusion cells used in MBE "“ they typically top out at 2000 °C.
The good news is that the MBE process does not take place in thermodynamic equilibrium. This allows the window for the growth of a thin film of single phase b-Nb2N to be expanded via growth kinetics. This is the route we have taken in our plasma-assisted, nitride MBE system that incorporates an electron-beam evaporator source. Armed with this tool, we can overcome the temperature limitation of conventional effusion cells and generate a sufficient flux of niobium in its metal form to gain control and flexibility over the growth kinetics of the material. Extensive MBE growth development and materials characterization led us to uncover a window of conditions that produce atomically-smooth, single-phase b-Nb2N films on SiC substrates between 775 °C and 850 °C "“ temperatures that are compatible with standard MBE III-N growth.
While demonstrating the growth of a thin film of β-Nb2N on SiC is a significant experimental milestone, the far more important challenge is this: can Nb2N be incorporated within a III-N device heterostructure, so that it adds new functionality, while not impacting the material quality and electrical performance of the subsequent III-N layers? Fortunately, it can, with our work showing that AlN and GaN heterostructures with high crystallinity and electrical quality can be grown on Nb2N/SiC templates, much in the same way these films can be grown directly on SiC.
One example that confirms device-quality epitaxy is possible is our fabrication of high-performance N-polar GaN/Al0.4Ga0.6N HEMTs, grown on Nb2N/6H-SiC (silicon-face). These devices feature a two-dimensional electron gas with a sheet resistance of 350 Ω/sq., a sheet carrier density of 1.30 à— 1013 cm-2, and Hall mobility of 1400 cm2/Vs. The mobility results are within 10 percent of those measured on similar N-polar HEMT structures grown directly on 6H-SiC (carbon-face).
Interestingly, the standard growth conditions that generate metal-polar films on silicon-face SiC have a tendency to produce N-polar films on Nb2N. However, by adjusting the growth parameters of the AlN nucleation layer on Nb2N, it is possible to realise polarity control and obtain the metal-polar orientation.
The capability to produce a thin layer of single-crystal metal adds a new instrument to the III-N device designer's toolbox. Exploring Nb2N films of varying thickness in a series of experiments confirmed that the material was indeed metallic, with a "˜bulk' conductivity of approximately 40 µΩ·cm. This conductivity increased slightly as film thicknesses dropped below 10 nm, but even at 4.4 nm Nb2N has a sheet resistance of just 236 Ω/sq.
The metallic nature of these films is retained after III-N overgrowth, implying that Nb2N layers have the potential to provide integrated ground planes for microstrip transmission, or buried contacts for active or passive vertical devices. Since Nb2N and AlN are predicted to be in thermodynamic equilibrium up to 1500 °C, buried or top-side Nb2N electrodes should be able to tolerate high thermal budgets and exhibit high reliability, due to the limited solid-state reaction. This means that they should be able to support higher junction temperatures than conventional Schottky gate contacts.
Piquing the interest of solid-state physicists, our Nb2N films undergo a superconductivity transition at around 10 K. This opens up interesting possibilities for novel devices that integrate a nanometre-thick superconducting material within an ultra-wide bandgap semiconductor material system that features polarization.
In addition to the potential of Nb2N as an electrode material, it has a very promising attribute that may ultimately deliver a far greater impact: it can be easily and selectively removed from other III-N materials. Etching with the reactive non-plasma gas XeF2, which is commonly used in silicon micro-electro-mechanical system processing, Nb2N can be rapidly removed vertically "“ and laterally, in the case of a buried layer. While this proceeds, there is virtually no etching of GaN, AlN, or most common metals and dielectrics. Alternatively, selective wet chemical etching based on nitric and hydrofluoric acid can remove Nb2N. In either case, by inserting a thin Nb2N layer, a fully processed III-N device can be completely separated from its substrate.
Transfer techniques
Capitalizing on the selective etch capability of Nb2N, we have developed an epitaxial lift-off (ELO) technique that enables the transfer of III-N material, devices, or circuits to an alternative substrate (an overview is provided in Figure 2). Starting with an Nb2N/SiC template, epitaxy creates a III-N heterostructure prior to conventional front-side processing steps to produce devices or circuits. Subsequent electrical test and/or yield screening can then identify known-good devices or circuits for transfer. After this, a masking layer is patterned over the devices intended for ELO, before a deep mesa dry etch exposes the buried Nb2N layer. Prior to the XeF2 release etch step, photoresist anchors and a top-side protective layer can be patterned to temporarily tether devices in place.
Figure 2: The epitaxial lift-off (ELO) technique developed at the Naval Research Laboratory can be applied to a GaN HEMT grown on Nb2N/SiC.
One promising technique for manufacturable transfer of tethered devices is micro-transfer printing. It is a process that has been pioneered by John Rogers' group, previously at University of Illinois and now at Northwestern University, and commercialized by X-Celeprint, Inc. The key technology is the use of a polydimethylsiloxane (PDMS) elastomer stamp to retrieve devices from a source wafer and place them on a target substrate.
There is a great deal of freedom with this approach, because the PDMS stamp can be moulded into custom shapes and sizes to define the area of contact between the stamp and individual released devices. What's more, it allows the batch transfer of multiple devices.
During retrieval, the stamp is brought into contact with the tethered device intended for transfer. Stamp retraction then preferentially breaks the device tethers, leaving the device adhered to the stamp and completely separated from the source wafer. To transfer to a target substrate "“ this could be another semiconductor wafer, or glass, paper, metal, or just about anything "“ the device-laden elastomer stamp is brought into contact with the new substrate. Once this occurs, the stamp is retracted at a slow speed with a degree of lateral (shear) motion to reduce the adhesion force of the stamp and ultimately leave the device behind on the target substrate.
Bonding may be improved by applying a thin adhesive layer to the target platform prior to transfer. If further processing of the substrate with transfer-printed devices is required, this is possible, thanks to a placement accuracy for the transferred devices that is better than 2 µm.
We have evaluated the electrical performance of our N-polar HEMTs after their transfer to target substrates using ELO. To improve adhesion to the target substrate, an 80 nm-thick, spin-on interlayer dielectric polymer was included in the process.
Electrical measurements have compared the performance of devices bonded to different substrates (see Figure 3, which also includes a scanning electron micrograph of a transferred device on silicon). Reductions from pre-release values are probably caused by device self-heating, which can be traced back to the thermal resistance at the device/substrate bonding interface, as well as thermal conductivity differences across target substrates.
Figure 3: (Left) Scanning electron micrograph of a 2 à— 75 µm N-polar GaN HEMT transferred to an 80 nm interlayer dielectric on silicon and (right) device transfer characteristics after bonding to various substrates with an interlayer dielectric adhesive. The slight curvature in the 1.5 µm-thick bonded device is caused by stress induced by device alloyed metal electrode processing, as released regions of unprocessed III-N heterostructure material are relatively flat and exhibit excellent bonding to target substrates. Mitigation strategies are being investigated for residual stress. Modifications to the released device membrane geometry and/or the application of a temporary external strain compensation photoresist or dielectric layer are expected to help reduce curvature during the bonding step.
In a circuit designer's ideal world, the optimal solution would involve a seamless mixing of disparate technologies/materials, in a manner that mirrors the construction of objects with coloured LEGO blocks. However, this concept of heterogeneous integration remains far from being fully realized.
One major roadblock is that the best substrates for ensuring high-quality growth of electronic materials tend to be incompatible with critical design goals, such as low thermal resistance and cost.
One major roadblock is that the best substrates for ensuring high-quality growth of electronic materials tend to be incompatible with critical design goals, such as low thermal resistance and cost.
Evaluating opportunities
As the idea of using Nb2N as a sacrificial layer for ELO is new, it is difficult to predict the impact it will have on the compound semiconductor community. However, it certainly has great promise in several areas, including: the fabrication of RF power amplifiers with improved heat-sinking; the provision of a new route to heterogeneous integration; the housing of LEDs on flexible substrates; and improved RF filters at higher frequencies, for next-generation wireless communication. All of these opportunities are now discussed in turn.
One of the biggest weaknesses of GaN-based MMICs is that the performance, while surpassing that of conventional III-Vs by a significant margin, is limited by power de-ratings that are imposed to reduce peak junction temperatures and prevent early chip failures. To address this shortcoming, several recent research initiatives have been launched to improve the thermal management of GaN HEMTs, such as the DARPA Near-Junction Thermal Transport (NJTT) and Intrachip/Interchip Enhanced Cooling (ICECool) programmes. One option enabled by our technology "“ and similar to the NJTT approach "“ is to transfer a device or a MMIC membrane from a SiC substrate to a single-crystal diamond wafer. This would situate the device on a substrate with a thermal conductivity that is five times higher, and lead to improved thermal management for the chip. Simulations suggest a significant benefit, even when the interlayer dielectric between the GaN device and diamond is 40 nm-thick (see Figure 4).
Figure 4: Thermal simulation of 20 à— 150 µm GaN HEMT on SiC (left) and 40 nm of interlayer dielectric on single-crystal diamond (right).
Our ELO process for heterogeneous circuitry could aid the construction of circuits that employ the best material for each function. DARPA is backing efforts in this area via its Diverse Accessible Heterogeneous Integration (DAHI) programme, which involves the integration of GaN power amplifiers with other semiconductor technologies, such as silicon CMOS, InP and SiGe.
Transferring GaN circuits or devices directly to silicon with ELO could enable additional front-side processing to cost-effectively create hybrid circuits with very short signal routing. If through-device vias were processed after transfer, this could facilitate vertical connections within this heterogeneous technology.
Wearable displays, a sector continuing to gain popularity, could also benefit from our technology. Here, the incentive to produce micro-LED displays on flexible substrates is increasing. This market could be served by growing and fabricating high-quality III-N LEDs on SiC, and then using our ELO process to transfer these devices to glass or plastic. Since the released LEDs are lithographically defined, they can combine a similar pitch with a very small pixel size "“ on the order of a few microns. Possibly the greatest benefit could come in the deep UV, as our ELO process would allow the removal and recycling of the bulk AlN substrate, which can compromise efficiency by absorbing UV emission from the LED.
Another opportunity for our technology exists in the wireless sector, which is facing ever-greater crowding of the allocated frequency bands within the electromagnetic spectrum. This is being addressed with smarter bandwidth management, and a key technology for accomplishing this is the reconfigurable, miniaturized, high-Q filter operating above the S-band. Such filters are in high demand, but are challenging to produce due to the poor material quality of the sputtered AlN films used in modern cell phone duplexers. The quality of these films, which are only a few tenths of a micron thick, are so poor that it is difficult to increase the centre frequency in bulk acoustic wave (BAW) and contour-mode resonator (CMR) filters much above a few gigahertz. Turning to our Nb2N template technology can address this: it enables the growth of high crystalline quality AlN, or improved piezoelectrics like ScAlN, at thicknesses of less than 200 nm. These films can then be suspended via partial removal of the underlying Nb2N layer. To assess the potential of such devices, we have simulated the performance of a 200 nm-thick AlN CMR. Results are summarized in Figure 5.
Figure 5: Simulation of a single-crystal AlN contour mode resonator, which could provide a filter for wireless communication at several GHz.
In addition to all of these promising opportunities for our Nb2N technology, it could also have a role in bringing together III-Vs with vastly different materials, such as paper and plastic. As the ELO technique is compatible with micro-transfer printing, one could imagine ways to populate a sheet of paper with an LED display or to integrate a power amplifier and antenna into the glass windshield of a car.
New energy and medical applications could also emerge, based on the micro-assembly of multi-junction photovoltaics on flexible plastic or the formation of high data-rate wireless communication circuitry on bio-resorbable polymers. To showcase the potential of the transfer of devices to different foundations, we have transferred one of our N-polar GaN HEMTs to a wooden ruler (see Figure 6 for static and pulsed open-channel, drain-current curves).
Figure 6: Static and pulsed current-voltage open-channel characteristics for an ungated N-polar GaN HEMT that has been transferred to a wooden ruler.
We are very excited to be exploring the impact that Nb2N can have on III-N applications and technology. When applied to ELO and transfer, Nb2N offers a number of important advantages over competing lift-off methods involving ion implantation or photo-electro-chemical etching. Aside from producing an atomically-smooth, bonding-ready released device or circuit backside, our ELO process can recycle substrate material indefinitely, potentially leading to significant cost savings.
With a predicted thermodynamic stability for Nb2N and AlN of up to 1500 °C, typical III-N growth methods should be capable of producing standard device heterostructures that tolerate a high thermal budget during subsequent processing. Although epitaxial conductors are a relatively unexplored frontier in microelectronics, it's easy to envisage many applications.
