Driving widespread deployment of GaN electronics

Saving energy is a top priority for many nations. To succeed in this endeavour, they can make greater use of clean technologies that enable efficiency gains in the production of energy, the way it is transported and how it is used. In this context, one area demanding the utmost attention is that of power conversion. There is still a tremendous opportunity to deliver substantial reductions in energy losses associated with AC-to-DC conversion and DC-to-DC conversion. If a new generation of electronic devices can be unleashed that combine the capability to handle higher powers with lower switching loss and higher operating frequencies, that could boost the efficiency of power inverters, while trimming their size and weight. Benefits that would follow include better power supplies for computers and more efficient power conversion in solar converters and hybrid electrical vehicles. The power electronics market that could underpin this revolution in efficiency savings is growing at more than 8 percent per year, and is worth billions and billions of Euros. Silicon devices dominate this sector. However, the room left for improvement in these devices is vanishing fast, because the best ones are now operating close to limits imposed by their intrinsic characteristics. It is widely accepted that silicon IGBTs deliver a relatively poor performance at high frequencies and MOSFETs struggle to handle high powers. Fortunately, there is a superior class of materials that can come to the rescue: Wide bandgap semiconductors. This includes GaN-on-silicon devices, which excel at combining affordability with great performance. One of the biggest attributes of these nitride electronic devices is their high breakdown voltage, which stems from a field strength that is an order of magnitude higher than that of silicon. Thanks to the high mobility and large carrier concentration associated with the twodimensional electron gas (2DEG) of the AlGaN/GaN heterostructure, nitride devices built for switching applications can also combine a low on-resistance with a high switching speed. What’s more, the wide bandgap properties associated with GaN-based devices enable them to operate at high temperatures. Development of nitride power devices has been underway for more than a decade and progress in this direction has enabled today’s switching devices to outperform their silicon rivals. In the performance stakes, SiC is a tougher opponent, but GaN more than holds its own. Making the nitrides affordable Great performance is no guarantee of market success: Products must also be affordable. GaN-on-silicon excels in this regard, because it is the most cost-efficient wide bandgap technology. It is already possible to deposit advanced heterostructures on silicon substrates up to 150 mm in diameter, and in the near future this growth process will be extended to 200 mm silicon. There is also an opportunity to develop process compatibility with standard CMOS technology, which would open the door to further cost reduction by enabling these wafers to be put through lines at under-utilized 200 mm silicon fabs that are dotted around the globe. Given it’s potential, it is of no surprise that GaN power electronics technology is attracting more and more interest. But no one is yet to deliver the first big commercial breakthrough – a reliable device operating at 600 V. One of the biggest challenges is to drive the first significant penetration for a compound semiconductor technology of a domain where silicon dominates, and many potential customers have been scarred by the experience of SiC. Although the performance of SiC diodes is attractive for power converter manufacturers, these chips are too pricey. In addition, until recently these diodes couldn’t be paired with a SiC transistor, which is detrimental to the uptake of this first-on-the-market wide bandgap solution. Another reason behind the lack of a commercially attractive, reliable 600 V device is that it is tough to manufacture GaN-on-silicon epitaxial structures, which are the starting point for making power electronics. This is precisely the challenge that we have picked up at EpiGaN, a spin-off of imec, a large international independent nanoelectronics centre located in Leuven, Belgium. Our start-up, which we co-founded in May 2010, is built on our expertise developed at imec, where we were involved in GaN research from 2001. Key successes during that time included the world’s first low-sheet-resistivity 6-inch HEMT structures in 2006, and the first GaN-on-silicon 200 mm epiwafers, a feat achieved in partnership with the MOCVD toolmaker Aixtron. Our approach differs from that of several players, which employ SiC as the substrate for their nitride devices. We have always focused our efforts on GaN-on-silicon devices, due to their cost advantages. Initially, we developed material for RF devices (we also offer epiwafers for RF applications). However, given the strengths of GaN-on-silicon for power electronics, it was an obvious move for us to switch our focus and target this lucrative market. During our time at imec advances in device performance governed our development of materials growth. This approach worked effectively and efficiently, thanks to very fast feedback of device performance. Armed with this method of working, we were able to simultaneously improve: Material aspects, such as channel conductivity and defect density; device related aspects, including in-situ passivation, dynamic performance, enhancement-mode devices, high-voltage operation, power density and operating frequency; and last, but by no means least, reliability. During this effort we paid careful attention to the reproducibility of the results. This approach brought us much success, including state-of-the-art device performances and the unveiling of a device concept for an enhancement-mode 600 V HEMT. The commercial promise of these achievements has helped us to attract a strong, balanced consortium of investors. Our vision is shared by Robert Bosch Venture Capital, Capricorn Cleantech fund and LRM. Their funding in 2011 has been used to set-up a new plant for producing GaN epiwafers by MOCVD in Hasselt, Belgium. Taking the strain Our growth process deals with the strain that arises when GaN is deposited on silicon. These two materials have significantly different crystalline properties and thermal expansion coefficients. Left unchecked, this can lead to strain in the epilayer and substrate that can ultimately cause the wafer to bow and even crack. By carefully managing this strain, we can make wafers that are suitable for passing through silicon processing lines. Day-in, day-out, we can manufacture 150 mm epiwafers with a bow well below 50 μm – it is typically 20-30 μm, depending on wafer specs. Uniformity, in terms of standard deviation of either layer thickness or electrical characteristics, is typically better than 3 percent. These epiwafers are optimized for high voltage/high frequency operation. This requires the formation of a buffer layer that withstands a very high voltage. It is possible to realise this: We have observed that in devices formed with high-quality GaN layers, the upper breakdown voltage is actually limited by the silicon substrate. The epiwafers that we produce for highvoltage devices have a buffer leakage current well below μA/mm at 600 V. Operating frequencies in excess of 100 GHz are also possible with our material, thanks to a reduction of buffer traps. While stress engineering may be the most challenging aspect of forming GaN-on-silicon devices, it is arguably not the hottest topic. That accolade goes to passivation of surface states. GaN is a piezoelectric material that features an excellent high-electron-concentration – associated with high electron mobility – that is obtained ‘for free’, without the need for any extra doping. But there is a significant price to pay: An extreme sensitivity governing device characteristics, such as current density and threshold voltage, on the filling of those surface states, which have a density comparable to that found in the channel. If passivation is poor, the device’s dynamic behaviour suffers. To combat this so-called dispersion problem, devices must be processed in a carefully controlled manner using high-quality epiwafers, because this leads to optimised buffers and controlled surface states. Fail to do this, and uncontrolled charging or discharging of these surface states – which can be modified during processing and device operation – can severely degrade the dynamic properties of the device. To prevent this from happening we deposit a unique in-situ SiN capping layer, which is grown by MOCVD as part of the epitaxy process, on top of HEMT epiwafers. The interface between this capping layer and the top nitride surface is incredibly smooth, and it enables perfect passivation of surface states (see Figure 1). Figure 1. In-situ SiN passivation, a unique concept for controlling surface states filling, enhancing device stability during operation as well as reliability tests We have shown that this capping layer can properly control the filling of the surface states during device operation. In fact, it is believed that SiN can provide enough charge to neutralize the surface charge of the AlGaN barrier layer so that its surface potential no longer contributes to 2DEG depletion. In addition, we have found that the SiN layer aids device stability at elevated temperatures. Specifically, it significantly enhances device reliability in high-temperature, accelerated-lifetime tests. These in-situ deposited SiN films can also lead to lower channel resistance. This equips engineers with another route to adjusting the top part of the FET so that it can hit particular device specifications. GaN FETs are lateral devices, and optimising their performance demands a trimming of conduction losses. This means that for switching applications, aluminium-rich barriers are preferred in a typical AlGaN/GaN structure, because this leads to a higher piezoelectric field, a higher current density and a lower specific on-resistance. Figure 2. Electrical uniformity mapping of 150 mm GaN-on-silicon wafer, measured contact-less sheet resistivity We have shown that one of the major benefits of the SiN cap layer is that it enables increases in aluminium concentration, which are possible without any significant material degradation. This is not the case in transistor structures with an uncapped or GaN-capped AlGaN/GaN 2DEG, where relaxation of the strained top AlGaN layer typically prevents high aluminium content in the top layer. It seems that in-situ SiN capping layers postpone relaxation of the strained AlGaN layer grown on GaN. This delay of relaxation has enabled us to demonstrate a SiN/Al0.35Ga0.65N/GaN/AlGaN FET with sheet resistance (Rsh) well below 300 Ω/ _  on 100 mm and 150 mm silicon (111) substrates. As one would expect, our SiN passivation process improves the performance of nitride FETs grown on other types of substrate too, such as SiC. For our SiN/AlN/AlGaN design that is detailed in Figure 3, sheet resistance falls to 235 Ω/ _ with our passivation technology. In this structure, Hall measurements indicate that the electron sheet concentration is 2.15 x 1013 cm-2 and electron mobility is 1250 cm2/Vs. These are very promising values and they enable the fabrication of devices with a high transconductance, even when gate length is relatively large. What’s more, they highlight the potential of this device for high-frequency operation.   Figure 3. SiN/AlN/GaN heterostructure   The neutralization of surface charges provided by the SiN layer also unlocks the door to an innovative approach for making enhancement-mode devices. This form of transistor, which is required for power converters, can be made by combining a thin AlGaN barrier layer with local removal of SiN under the gate. From 600 V to 1.2 kV Today we can reproducibly manufacture GaN-on-silicon wafers with a breakdown voltage above 600 V and a very low leakage current. This is by no means the upper limit for the breakdown voltage of our devices – recent work has yielded FETs with a breakdown above 2 kV. In every industry, conversion of hero results in the lab to production processes takes time. However, it is clear from our efforts that GaN can already be used today to make products in the 30 V-200 V and 600 V range, and it will not be long before variants operating at 1200 V can be added to the list. This will pave the way for the replacement of two silicon MOSFETs with a single GaN HEMT, a move that will trim the cost and weight of a power converter. To make this happen, we are focussing our development on 1200 V epiwafers on 150 mm silicon.  When we release this product, it will be added to our existing range of 4-inch and 150 mm epiwafers for high voltage and/or high-frequency applications. The production capacity for these products is being ramped up at our new site, while in parallel we are developing manufacturing processes for 200 mm GaN epiwafers operating at 600 V and 1200 V. Although today the demand for these larger epiwafers is weaker than that for those with diameters of 150 mm or less, larger sizes will spur a cost reduction and enable GaN to deliver further success in a field where, up until now, no compound semiconductor has seriously challenged silicon.  © 2012 Angel Business Communications. Permission required. Further reading J. Derluyn et al. IEDM Tech. Dig. 1 (2009) F. Medjdoub et al. Electron. Lett. 47 1345 (2011) J. Derluyn et al. J. Appl. Phys. 98 54501 (2005)