GaN-based power devices offer game-changing potential in power-conversion electronics

Over five years of research and development has resulted in a commercially viable GaN-on-Si-based (gallium nitride on silicon) power-device technology platform that offers dramatic improvements in key device figures of merit (FOMs) R_dson and R_on × Q_sw. Combined with improved packaging and drive electronics, GaN-based high electron mobility transistors (HEMTs) will enable power-conversion FOM :efficiency × density/cost figures that are an order of magnitude better than existing silicon and silicon carbide (SiC) FETs.

Historically, paradigms exist until they fail to represent the actual boundary conditions which have become evident. Clearly, the breaking of paradigms is a rare event. Normally, even the most innovative designer keeps some very definitive "rules of thumb" handy with which to evaluate potential new ideas. These rules go out the window when a paradigm is broken.

Within power-conversion electronics, a general rule of thumb is this: it is not possible to achieve high power density through single-stage conversion at frequencies above 10 MHz, while at the same time as achieving high energy-conversion efficiency, without paying a large cost premium through complexity, such as in the use of resonant topologies. This is due to the generally slow, highly variable and lossy switching characteristics of the power devices used in the power conversion stage.

From the advent of rectification, through linear regulation, to the adoption of switch-mode power supplies and resonant converters, power-conversion topologies and the control circuitry that realize these architectures have been developed to take advantage of the inherent capabilities and avoid the deficiencies of the available power-switch technology. When these power-semiconductor technologies vary incrementally, it is advances in the architectures, drivers and control circuits which drive radical advances in power conversion electronics, such as multi-phase or sine wave conversion.

However, on the rare occasion when the nature of the power-switch itself changes radically, usually only once a decade or so, it is the nature of the power-switch which drives radical, potentially revolutionary, changes in power electronics. Such a change is clearly approaching widespread introduction in the near future. This is the general availability of commercially viable, GaN-based power devices which provide the opportunity for more than an order-of-magnitude in performance improvement over state-of-the-art, silicon-based devices within the next five to ten years.

For the last three decades, silicon MOSFETs have been the power device of choice for the majority of applications in the power-electronics application market. Ever since the introduction of the first commercially viable silicon MOSFET (trademarked HEXFET) by International Rectifier in 1978, the silicon power device has continued its onward march both in terms of volume and market size, especially by enabling rapid adoption of switch-mode power supplies (SMPS) over then-dominant linear supplies and bipolar devices.. From planar HEXFETs to TrenchFETs to superjunction FETs, silicon MOSFETs have continued to evolve for the last 30 years to satisfactorily serve many markets. During this time, there has been improvement of approximately two orders of magnitude in the device-performance figures of merit.

Currently, the silicon-based power FET is approaching a performance plateau. Therefore, further enhancements are incremental and costs of advancements, both in terms of development resources and production complexity, are becoming prohibitively high. At the same time, next generation and emerging applications are continuing to demand further, substantial improvements in power conversion performance.

Thus, to meet new requirements and challenges, novel materials and transistor structures are needed to fill this gap. Even though silicon carbide (SiC) FETs have emerged and have been undergoing refinements for the past 10 years to address these issues, they suffer from significant cost premiums due to limited-quality material supply, as well as the intrinsic cost structure of the material. Additionally, SiC-based technology is not highly scalable in substrate size, epitaxial deposition equipment throughput, material supply, and device-fabrication manufacturing platforms.

Foreseeing such a need, scientists and engineers at International Rectifier Corporation (IR) have developed a revolutionary GaN-based power device technology platform which promises to deliver cost-effective performance that is at least ten times better than existing silicon devices, to enable dramatic reductions in energy consumption in end applications in markets such as computing and communications, consumer appliances, lighting and automotive applications.

In fact, over five years of device R&D has resulted in a proprietary GaN-on-silicon epitaxial process and device design and fabrication technology platform that heralds a new era in power conversion. IR expects the potential impact of the new GaN based device technology platform to be at least as significant as the introduction of HEXFETs some three decades ago. Prototypes have been built to demonstrate some of the capabilities of the GaN-based power device technology.

Plans are underway to offer a broad range of commercially viable products (for applications requiring 20 to 1200 V device ratings) supporting discrete as well as circuit solutions (modules and chipsets) for a variety of DC-DC and AC-DC converters, lighting, Class D audio, and motor drives. First commercial products are scheduled for production release by the end of 2009. It is reasonable to expect that 1-2% of all applications in power electronics will use GaN-based power devices within three to five years of production release, Figure 1.

Figure 1: As silicon-based power-device technology reaches maturation, GaN-based power devices appear ready to continue to provide the rapid and dramatic improvement in performance enjoyed by the electronics industry of the past 30 years.
(Click on image to enlarge)

GaN-on-Silicon
As bulk GaN substrates have been prohibitively expensive, with limited availability and at small sizes (under 2-inch diameter), developers were prompted to explore the use of hetero-epitaxial films for GaN-based power devices. Subsequently, most work has been performed using SiC or sapphire substrates for hetero-epitaxial growth of GaN films. Both have been relatively expensive and provide limited opportunity for a commercially viable GaN-based power-device technology platform, having limited volume supply (compared to the 10 million 150-mm wafer equivalents used in today's power device fabrication) as well as relatively small substrate sizes (commercially 6 inches for sapphire and 4 inches for SiC), which make the device fabrication processing non-economical compared to silicon-based platforms.

Though SiC substrates have many attractive properties to lend to power device designs (reasonably small thermal expansion and lattice constant mismatches to AlN/GaN, as well as excellent thermal conductivity), its cost structure is roughly 100 times more than that of silicon. Though sapphire allows very thick epitaxial film growth, due to an excellent match to the thermal expansion of AlN/GaN, it suffers from very poor thermal conductivity, a critical issue for power devices. While silicon is clearly a very attractive, low-cost alternative substrate, GaN hetero-epitaxy on silicon is not a trivial development to achieve, because of defects and deformations due to significant intrinsic mismatches in lattice constants and thermal expansion coefficients of substrate/epitaxial films.

Achieving the necessary uniformity of both epitaxial properties and device characteristics across large volumes of wafers requires the application of rigorous process design and control methodologies. In addition, significant engineering efforts were made to improve device performance FOMs, as well as device initial quality and long term reliability. Silicon wafers also provide much larger-diameter substrates (6-, 8- and 12-inch) and in higher available volumes than are available with either sapphire or SiC substrates.

Designing the device fabrication process to be CMOS compatible also required significant engineering. Generally, periodic-table group III-V semiconductor-device processing has used relatively low throughput, expensive techniques such as gold interconnects, liftoff and e-beam lithography. Through the use of standard CMOS processing involving Si-compatible interconnects, photolithography and plasma etching, IR has created a platform that is compatible with silicon manufacturing facilities to offer a high-volume, commercially viable manufacturing platform for GaN-based power devices. IR's GaN-on-Si technology platform, referred to as GaNpowIR, provides the power-conversion circuit-designer community with a commercially viable manufacturing platform for GaN-based power devices.

To take full advantage of the intrinsic features and capabilities of these new GaN-on-Si based power devices, the GaN technology platform includes the development of complementary gate-driver circuitry, advanced packaging solutions, as well as controller ICs and novel circuit topologies.

As shown in Figure 2, the basic present GaN-on-Si structure is a high electron mobility transistor (HEMT), based on the presence of a two-dimensional electron gas (2DEG) spontaneously formed by the intimacy of a thin layer of AlGaN on a high-quality GaN surface.

Figure 2:Basic GaN-based device structure is a high electron mobility transistor (HEMT), formed by the intimated presence of a thin AlGaN layer on a high-quality GaN layer.
(Click on image to enlarge)

Ohmic contacts are made to the 2DEG, typically using Ti/Al-based metallurgy. An insulated or rectifying metal-gate structure is formed between the ohmic contacts and provides for the field-induced modulation of the 2DEG.

It is clear, then, that the native nature of this device structure is a FET with a buried high electron mobility channel, which conducts in the absence of applied voltage (normally-on). There are several techniques which have been developed to provide a built-in modification of the 2DEG under the gated region, providing for normally off behavior. Internal studies show that GaN-based power devices can offer performance that is comparable to that of SiC devices, at much lower cost.

The reverse recovery (Q_rr) characteristics for a high-voltage GaN diode function is essentially the same as for commercially available SiC diodes, both being significantly better than state-of-the-art silicon fast-recovery diodes (FRED) (Figure 3).

Figure 3: Reverse recovery (Q_rr) performance for a 600 V GaN-based diode function is essentially the same as that of a SiC diode.
(Click on image to enlarge)

This is due to the essential absence of holes in the GaN HEMT structure, eliminating the minority carrier effect in reverse recovery charge Q_rr.

This results in quiet switching behavior, eliminating the otherwise needed filtering (snubber) circuitry. In turn, this reduces system size, cost and weight, and has been used extensively in power factor correction (PFC) circuitry in AC-DC converters using expensive SiC diodes. This naturally fast and quiet switching behavior also provides for significantly reduced switching losses when anti-parallel diode functions are required, such as with insulated-gate bipolar transistors (IGBTs) in motor drive circuitry.

A combination of high electron mobility and higher bandgap provides GaN-based HEMT devices with a significant reduction in device-specific on-resistance R_DS(on) for a given reverse hold-off voltage capability compared with both SiC and silicon devices, as shown in the calculated material-limit curves for (non-highly compensated) unipolar devices in Figure 4.

Figure 4: Comparing specific on-resistance of IR's GaN-on-Si-based HEMTs with silicon and SiC power FETs
(Click on image to enlarge)

Also shown are representative, best case, published measured results for FETs using the three materials, as well as for highly compensated superjunction (SJ) and bipolar (IGBT) device structures in Si.

Results from the early stage development of the GaNpowIR technology platform at IR are also shown (IR GaN). It is clear that an order-of-magnitude improvement in specific on-resistance can be achieved for GaN-based devices over silicon counterparts, even at the early stages of power-device development of GaN technology (less than10 years), as compared to the mature Si (greater than 30 years) or SiC (over 20 years) technologies.

Whereas both the Si- and SiC-based unipolar devices have essentially reached their theoretical limits, the GaN-based HEMT devices are still several orders of magnitude higher than the calculated potential performance. It is expected that engineering efforts over the next 10 to 20 years will result in a significant reduction in this performance gap. Though the absence of a comparable 2DEG device for SiC-based devices, together with higher ohmic contact resistance and lower electron mobility, makes low-voltage SiC devices non-competitive with GaN HEMTs, bulk SiC substrate-enabled homo-epitaxy provides for the very thick films required for voltages above about 1500 V.

In contrast, it is much more difficult to achieve these higher voltages for GaN hetero-epitaxy on silicon substrates. Thick GaN hetero-epitaxial films are easier on sapphire substrates, since the mismatch in thermal coefficient of expansion is small, but insulating substrates restrict the power-handling capability of the device, due to self heating constraints. It is therefore expected that SiC will remain an attractive choice for high voltage switches (above 1500 V) in the future.

Since GaN-based power devices achieve a combination of low gate capacitance and low on-resistance, it permits switching converters operating at much higher frequencies, and therefore more efficiently, than competing silicon transistors. Results based on device modeling, extrapolated from early measured results, indicate that R_on × Q_g FOM for first-generation GaNpowIR HEMTs, to be introduced in 2009, is 33% lower than that of the latest silicon MOSFETs.

Ongoing engineering efforts are expected to provide further significant improvements in the next few years. Figure 5 shows that R_on × Qg for GaNpowIR devices is expected to be as low as 13 mΩ-nC by 2011, representing more than a 50% improvement over GaN-based devices introduced in 2009.

Figure 5: Projected evolution of R_on × Q_g FOM for low-voltage GaNpowIR HEMTs
(Click on image to enlarge)

By 2014, the R_on × Qg FOM for GaNpowIR is expected to be less than 5 mΩ-nC, an order of magnitude improvement over the best Si-based devices available in 2009.

Figure 6 depicts the expected effect of the improvements in R_on × Q_sw FOM of the power switch on the size and efficiency of a DC-DC converter, including the output filter.

Figure 6: Projected evolution of size and effective power-conversion efficiency for a multiphase 100 A, 12 V to 1.2 V converter (including output filter), corresponding to improvements of the power-switch FOM. The per-phase frequency shown on the timeline is chosen to provide a constant converter efficiency of 85%. The added efficiency above 20 MHz corresponds to reduced parasitic power loss downstream from the converter, made possible through the higher-frequency conversion.
(Click on image to enlarge)

Current multi-phase, silicon-based solutions perform 12 V to 1.2 V conversions efficiently up to about 2 MHz per phase. The GaNpowIR technology platform is expected to enable efficient power conversion to greater than 50 MHz per phase in the near future.

The improvements in the power-switch FOM enable a corresponding increase in operating frequency and a corresponding decrease in converter size, without a reduction in power-conversion efficiency. The frequency shown in Figure 6 is chosen to provide a constant conversion efficiency of 85 %. In fact, when the frequency is high enough (20 to 60 MHz) to eliminate the need for significant external components and wasteful distance between the converter and the load, a significant reduction in parasitic-related power loss is achieved.

This is accounted for in the Figure 6, by adding the power-loss savings back into the converter efficiency, since it is the converter properties that enabled the energy savings (admittedly not rigorous but instructive, regardless). This, then, provides for a very significant, simultaneous achievement trio of high density, higher efficiency, and lower system cost.

Consequently, the dominant power conversion application FOM = efficiency × density/cost will be substantially better using GaN-based solutions than alternatives. Due to improvements in the device RQ FOM, together with improved packaging and drive technologies, it is expected that an order-of-magnitude improvement will be achieved in this FOM over the first five years of commercial introduction of the GaNpowIR platform.

To demonstrate the distinct advantages of the new GaN-on-Si based power devices, several prototypes have been built. One such prototype is a low-voltage point-of-load (POL) converter. Designed for a 12 V input to 1.2 V output at 12 A load current, this GaN-based POL converter runs at 6 MHz, delivering efficiency that is comparable to a commercially available silicon solution running at 1 MHz (Figure 7), but at less than one-third the size.

Figure 7: GaN-based POL converter runs at 6 MHz to deliver efficiency that is comparable to the state-of-the-art silicon solution running at 1 MHz, but at less than one-third the size.
(Click on image to enlarge)

Both designs integrate the controller IC and output inductor within the power-stage package.

Summary
IR's GaN-based power-device roadmap anticipates that initial prototypes will switch efficiently for power-conversion ratios of 10 to 20, up to a frequency of 6 MHz, with commercial products introduced over the next few years supporting switching frequencies of 10 to 60 MHz.

Several prototypes of power conversion solution using the new GaN-based power devices were demonstrated at the Electronica trade show in Munich, Germany in November 2008, with products for general availability expected to be released by the end of 2009. High-volume production will begin with industry standard 150-mm silicon wafers. Initial products will include complete POL solutions across a broad range of input and output voltages.

In conclusion, with simultaneous dramatic improvements in GaN power device FOMs R_DS(on) and R_on × Q_sw driving power-conversion FOM (efficiency × density /cost), GaN-on-Si-based power devices should drive a revolution in power electronics. As has been true in the past, it is expected that new conversion architectures and control schemes will be developed to take full advantage of the capabilities of the GaN-based power devices.

About the author
Michael A. Briere is Executive Scientific Consultant, ACOO Enterprises LLC, Woonsocket, RI, under contract to International Rectifier Corp, El Segundo, CA, www.irf.com. He joined IR in 2003 and was Executive Vice President (Research & Development), and chief technology officer. Before joining International Rectifier, Dr. Briere was founder, President, and Chief Executive Officer of Picor Corporation, a developer and marketer of power-IC designs and systems. He has held positions at IBM, the Hahn Meitner Institute of Berlin, Lawrence Livermore National Laboratory, Cherry Semiconductor, and ON Semiconductor.

Dr. Briere has a BSEE and MS in Physics from Worcester Polytechnic Institute, and a Doctorate in Solid State Physics from the Technical University of Berlin. He also served on the IEEE subcommittee on power devices and ICs, the program committee for the International Symposium for Power Semiconductor Devices and ICs, and was a member of the Advisory Board for the Center for Surfaces and Thin Films at the University of Rhode Island, where he also served as an Adjunct Associate Professor of Physics.