Slashing GaN Costs With SILICON
Production costs for GaN-based devices will plummet when epilayers are formed on 200 mm silicon
BY DENIS MARCON AND YOGA SARIPALLI FROM IMEC
Devices based on GaN can serve a vast number of applications. Transistors constructed from this wide bandgap semiconductor can increase the efficiency of power supplies, solar invertors, and base station transmitters; LEDs made from GaN are already backlight billions of screens and illuminate numerous homes and offices; and sensors fabricated from this material can detect gases such as nitrogen dioxide, a source of air pollution. With all these GaN devices, if the cost of their manufacture falls, their deployment will rise.
One way to drive down costs is to switch the growth of the GaN-based epitaxial structures to a cheaper foundation âˆ' the best in this regard being silicon with a diameter of 200 mm. Epiwafers formed in this manner can be processed at very low costs, by shipping material to under-utilized, fully depreciated 8-inch silicon fabs.
Producing GaN devices in this manner is appealing, but challenging. It is far from easy to obtain wafers that have a small, controlled bow, are free of cracking or surface pits, and are made up of layers with a low density of defects. That's partly because the thermal expansion mismatch between silicon and GaN can cause the build-up of excessive tensile stress, which is alleviated by film cracking; and it is partly because the lattice mismatch during growth, plus the thermal mismatch during cool down from higher temperatures, can give rise to a high density of defects.
These lattice and thermal mismatches can also cause the wafer to bow, and if exceeds just 50 mm for a 200 mm wafer, this can prevent it from being processed in a silicon line. Note that wafer bow tends to increase with increasing wafer size, making up-scaling difficult.
At imec, an internationally renowned microelectronics research centre based in Leuven, Belgium, we are addressing these challenges. We are developing an epitaxial process for the growth of GaN on 200 mm silicon, while many vendors active in this area are focusing on the smaller 150 mm platform. Our expertise in GaN is considerable, as we have spent more than a decade developing GaN devices: depletion-mode and enhancement-mode power transistors, power diodes, and other types of devices, such as air sensors and LEDs (see Figure 1). We share the technology that we have developed with today's and tomorrow's partners, so that they can use this to design and produce their first devices in-house or at imec.
Figure 1: imec's 200 mm GaN-on-silicon platform, which can be used to manufacture a wide variety of devices.
By carefully optimising each of the layers in our GaN-on-silicon structures, including the key buffer layer, we have been able to form epiwafers with a low defect density and smooth surface. According to atomic force microscopy measurements on areas of 5 mm by 5 mm, the surface has a root-mean-square roughness of just 0.4 nm.
From a manufacturing perspective, producing a few good-quality 200 mm GaN-on-silicon wafers is promising, but this will not amount to much unless it is followed up with a demonstration of long-term reproducibility. We have done just that, using various measurements to determine the quality and reproducibility of our buffer layer in structures grown on a MaxBright Veeco reactor. Crystal quality of the buffer layer shows small variation from wafer to wafer, according to X-ray diffraction measurements with a Jordan Valley QC3 tool (see Figure 2a). What's more, the average of the full-width-at-half-maximum (FWHM) for the AlGaN buffer is among the best reported in literature, which includes smaller wafer sizes. In addition to the high crystal quality of the buffer, it is sufficiently flat to allow our epiwafers to be run through a silicon line. Measurements on 90 wafers show that all of them are suitable for processing (see Figure 2 b).
Figure 2: The 200 mm GaN-on-silicon wafers produced at imec have a buffer with excellent material quality, according to X-ray diffraction measurements (a), and a very low level of wafer bow (b). Bow must be below 50 Âµm bow for processing in a silicon line.
Another essential characteristic for all GaN-on-silicon epiwafers, if dedicated to power switching devices, is a low leakage current at high voltages. We determine whether our wafers meet this criterion by applying a voltage between two isolated metal pads. At 1100 V "“ the limit of our system âˆ' leakage measured over 20 structures across the wafer is uniform, and well below a typical spec of 1 mA/mm.
Silicon foundry suitability
Flat wafers are not the only consideration when processing GaN-on-silicon wafers in silicon lines. Engineers working in these fabs are very concerned over the use of gallium, which can contaminate the lines, because this element is a p-type dopant in silicon.
At imec, we have faced this issue, with the device processing team that is keen to to push GaN in the line working together with the contamination team. Together, we have found ways to prevent gallium contamination, which opens the door to putting GaN-on-silicon wafers through the lines after standard silicon lots.
A second challenge that we have faced in introducing our GaN technology in a CMOS line is associated with the contacts: they must be as good as those that include gold, but are free from this element. We have succeeded in this endeavour, producing gold-free ohmic modules with very little variation in contact resistance and a high level of reproducibility from wafer to wafer (see Figure 3). These contacts are reliable under high current density and high temperature conditions, and with an annealing temperature for the module below 600Â°C, they offer flexibility in process integration.
Figure 3: Gold cannot be employed as a contact metal in silicon lines, so imec's engineers have developed a gold-free alternative that has a low contact resistance and a high level of reproducibility (a). There is a small degree of variation in contact resistance between within six wafers within one lot (b).
Following optimisation of the general processing modules, such as ohmic contacts and passivation, device engineers working with us, in synergy with processing engineers, can optimise their device technology. Support for this effort comes from intense TCAD and modelling activity.
Our 200 mm GaN-on-silicon platform can be used to form a range of power switching devices: depletion mode (D-mode) MISHEMTs; enhancement mode (E-mode) MISHEMTs and J-HEMTs; and power Schottky diodes, which are standalone devices that are compatible with transistor processing, so allow a high level of integration. The D-mode transistors that are normally on are well established, with 600 V-rated devices exhibiting an on-resistance of just 0.8 mÎ© cm2. However, most research efforts are now focused on the realisation of highly performing E-mode devices, because they are easier to use in circuits "“ they do not have to be paired with a silicon transistor to form the preferred, normally-off mode of operation. It is the E-mode form of the device, which is normally off, that will lead to a ramp in sales of GaN transistors.
One of the common approaches to making an E-mode device is to recess the AlGaN barrier and then deposit a gate dielectric. An alternative, popular process involves growing a p-type layer on top of the AlGaN, and then etching the p-type layer in the active region only. Both approaches, which form MISHEMTs and J-HEMTs respectively, have their pros and cons. This is why our partners "“ whose names cannot disclosed for reasons of confidentiality "“ are pursuing both options to de-risk their future investment on processing development.
MISHEMTs that we have produced exhibit a uniform threshold voltage of around +1.2 V, and a low leakage current that is in the pico-amp range at a gate voltage of 0 V (see Figure 4 a). Under forward gate bias with a 10 V gate voltage, current is below 1 nA/mm, and the breakdown voltage exceeds 15 V (see Figure 4 b).
Figure 4: E-mode MISHEMTs produced at imec combine a uniform threshold voltage of typically 1.2 V (a) with a low leakage current (b) and a breakdown voltage in excess of 600 V (c).
At a gate voltage of 0 V, the breakdown voltage exceeds 600 V, demonstrating the true E-mode nature of these devices. They deliver state-of-the-art performance for an E-mode transistor, with an on-resistance of 1.5 mÎ© cm2, which is far superior to that of a standard silicon equivalent. We are now trying to take these 600 V devices to a higher level of maturity, and to increase the threshold voltage beyond 2 V, while trimming the on-resistance to below 1 mÎ© cm2.
We have also used our 200 mm GaN-on-silicon platform for the development of power rectifiers. The biggest challenge with this class of device is to combine a low turn-on voltage with a low leakage current. We have succeeded in this regard by pioneering an innovative, proprietary device architecture named the gate-edge terminated (GET) diode. This class of diode has a turn-voltage that is below 1 V, and a leakage current less than 1 mA/mm at high reverse voltage (see Figure 5).
Figure 5: The low turn on voltage (a) and low leakage current of imec's novel, proprietary gate-edge terminated diodes make them ideal for use in electrical systems delivering highly efficient switching.
An additional strength of these devices is their incredibly short reverse recovery time, which results from them being majority-carrier Schottky diodes. Short recovery times slash switching losses, enabling these devices to increase the efficiency in electrical systems, such as solar inverters. We expect that once these devices appear on the market, they will replace silicon and SiC diodes, because they offer a better performance than the former, and are inherently cheaper than the latter.
Our 200 mm GaN-on-silicon platform is not restricted to the production of power devices: it is also capable of producing NOx sensors and LEDs, while RF devices offer an opportunity for future collaboration. It is astonishing how the same GaN material and platform can be used for so many different applications!
Detecting levels of NO2 is very important, because emissions of this gas can jeopardize human health and cause ecosystem damage. Consequently, there is a great demand for ultra-compact, portable and even wearable low-cost continuous NO2 monitoring devices. Sensors on the market today do not satisfy all these requirements. They are either too costly, too bulky, or fail to provide the NO2 sensitivity or reversibility required for continuous air quality monitoring, where detection at concentrations of less than 50 parts-per-billion is needed.
It is possible to meet all these requirements with our GaN-based technology that can yield devices with a reaction time below 2 seconds and a detection limit below 1 ppb for NO2 (see Figure 6). With these devices, the level of NO2 concentration is revealed by variations in the resistance of the two-dimensional electron gas in the channel. These devices are showing much promise, having already been used to monitor NO2 inside underground parking garages, where they have produced very good results.
Figure 6: imec's GaN-on-silicon platform can be used to make NO2 sensors (a). Detection of this pollutant oxide is possible downto levels of 10 parts per billion (ppb) and below, with the sensor delivering a quick response to changes in concentration that lead to changes in the resistance of a two-dimensional electron gas.
It is clear that GaN devices are accounting for an ever-larger proportion of the semiconductor market. The 200 mm GaN-on-silicon platform that we have developed with our partners will help to accelerate the increasing deployment of this wide bandgap semiconductor, which can be used to make many different devices. Some companies might find it a formidable challenge to make the transition to manufacturing GaN devices, but this should not be too daunting, as we can help them to gain easy access to this technology, accelerate their internal GaN development and provide them with access to next-generation GaN-based epitaxy and device technologies currently under development.