NXP Goes With GaN
Following several years of development with UMS and Fraunhofer Freiburg Institute, NXP is starting to launch a family of high-performance GaN-on-SiC power transistors that will complement its hugely successful silicon LDMOS products, says the company’s Director of Marketing for RF Power, Mark Murphy.
At NXP we have recently unveiled our activities in bringing RF power GaN products to the market, which are a significant strand in our portfolio development for the coming years. We believe this demonstrates to the electronics industry something of a breakthrough in the maturity of the technology, as it progresses from being a boutique-only affair to part of a broad range of products from a leading high performance RF supplier like ourselves.
NXP GaN - the story so far
Our firm is not a latecomer to GaN – we are firm believersin the many benefits this wide bandgap semiconductorbrings and our involvement stems back through manyyears of research and development. We are investing inGaN for different market segments, which will inevitablyresult in more than one GaN process to meet thedifferent product-application requirements.
Significantly, for the RF power activity, we have had a great partnership with United Monolithic Semiconductors (UMS) and the Fraunhofer Freiburg Institute that has helped the technology mature by leaps and bounds to the point today, where we can prepare for the first product releases. In-house testing has gone well these past months and we have been boosted by some glowing endorsements from selected customer sampling. As of today, we are well underway with a full release of the wafer process and the first phase of product designs.
The first product in NXP’s GaN power product portfolio is the CLF1G0530-50, a broadband amplifier spanning the 500-3000 MHz range that has a nominal P1dB of 50W
The products that we will release will compete in the market for RF power transistors, which will break the $1 billion mark in 2011, according to independent market research. Products built from GaN will play an increasing role in this sector, and are tipped to take as much as 30 percent of this market by 2014. If true, this would lead to a GaN RF power transistors market of at least $300 million.
For such numbers to come to fruition, GaN technology needs to be supported by mainstream vendors. In the world of RF power, we are the first to offer such an extensive portfolio that covers both silicon LDMOS and GaN products – combined these two technologies will account for nearly 90 percent of the RF power market.
This leads to the obvious question: what makes GaN a winner for RF power applications? Simply put, this material makes a step increase in efficiency and power density performance over silicon LDMOS in most applications. The suitability of semiconductors for RF power transistors is captured in the Johnson’s Figure of Merit (FoM) – a combination of material constants that starts with 1 for silicon as a reference and ends with 324 for GaN. To put this into some context, GaAs, another commonly used compound material in RF, has a FoM of 1.44. Suffice to say, GaN truly represents a breakthrough technology.
To make the most of the benefits of GaN we will use devices fabricated on SiC substrates and package them using the latest low-thermal-resistance materials. The excellent thermal properties of SiC mean that we can exploit the higher operating temperature capabilities of GaN versus LDMOS by keeping the transient thermal impedance down to a minimum – an important requirement for an RF power transistor.
The products that we will release are HEMTs, a class of transistor that exploits one of the intrinsic benefits of GaN: the high electron drift velocity. These transistors will be depletion mode devices, that is, devices that are normally on, without the need for applying a gate bias. A negative gate bias will be needed to switch the transistors off. This biasing is not straightforward, but we like to offer solutions rather than just components, so we already have a tried-and-tested bias circuit available and will provide application support through the life of the product.
In a few applications we will see GaN replace silicon LDMOS. However, ideally this should be the exception rather than the rule, because GaN should be seen as extending the scope for RF power coverage.
A tale of two compounds
From a compound semiconductor perspective it isimportant to realise that the products discussed hereexploit the advantages of two compound materials: SiCand GaN. SiC is used as the substrate owing to itsexcellent thermal conductivity and the junctions areformed by GaN epilayers to improve the efficiency andpower density and also extend the frequency rangecompared to silicon LDMOS devices.
Although the channel is structured laterally in RF power devices the current also flows through the substrate as the principal source contact. For this reason we need low Ohmic substrates for LDMOS devices. Silicon, of course, is available from multiple suppliers and for RF power we use 8-inch (200mm) wafers. In contrast SiC has much a smaller supply base and is in the process of moving from 3-inch to 4-inch substrates. This makes the real-estate on SiC more expensive, but that cost adder is more than compensated for by the incredible performance improvements: a five-fold hike in thermal conductivity and a factor of nine gain in electrical breakdown.
Table1: Comparison of silicon & SiC substrate material
Table 2: Comparison of Si LDMOS & GaN HEMT material
For GaN devices a rather complex heterostructure ends with GaN epi in which the active junctions are formed. But the benefits are to be found in the electron drift velocity and the electrical field breakdown. The peak electron drift velocity (another way of expressing the electron mobility) for GaN is nearly three times that of silicon, leading to devices with lower specific Rds(on) and smaller gate length, in turn yielding devices that can work at much higher power densities.
GaN’s suitability as a general purpose power switch exploits the better electron mobility plus a factor of 12 improvement in breakdown field: this opens up the market for a complete new level of very low Ohmic, high-voltage devices. For RF power we will see devices with a typical breakdown voltage more than twice that of silicon LDMOS, enabling higher Vds bias voltages to be used. A further advantage of GaN is that it is a very hard material capable of withstanding very high temperatures. Our GaN transistors will be specified to a maximum temperature of 250 °C, compared to 225 °C for silicon LDMOS.
The potential for GaN as a material for RF Power devices can be stated, without exaggeration, as outstanding. As an illustration of what this means in terms of real device performance, see the efficiency-power density comparison for LDMOS and GaN in figure 1.
Fig.1 GaN clearly outperforms silicon LDMOS at 2.1 GHz
From an industrialisation aspect, we have managed to incorporate silicon LDMOS processing into a standard CMOS fab environment, with just a few minor process variants. This has been essential for a market that consumes thousands of wafers and millions of products per year. With this comes great economy-of-scale advantages for silicon LDMOS. For now, GaN is processed in dedicated fabs, the mask count is lower but the economies of scale are some years away.
Initially we will focus on bringing a range of broadband devices to market. This will give customers the chance to evaluate a GaN device in as many applications as possible. But these devices are far more than evaluation vehicles: we will bring them to full release this year and offer all the product and application support required for customers to use them in volume production.
NXP has built a 2.7 GHz Doherty demo using three small, unmatched GaN devices that is claimed to achieve a decent power density and efficiency performance when compared to silicon LDMOS
The first released type will be the CLF1G0530-50, hardly a catchy name but one with meaning: the ‘C’ is our code for GaN technology; ‘F’ denotes a ceramic package type; ‘1G’ stands for first-generation technology; 0530 describes the optimal frequency range of 500-3000 MHz; and ‘50’ is short-form for a nominal P1dB of 50 W. This is the naming convention we will apply to our portfolio: 100W and 150W, unmatched broadband versions will follow before we start on a few frequency-specific matched types.
The biggest market segment for RF power transistors is telecom base-station infrastructure equipment. Typical operating frequencies range from less than 1000 MHz for various GSM, WCDMA and LTE standards to 2700 MHz for other LTE use and to 3800 MHz for WiMAX. In recent times the power amplifier (PA) architecture of choice has changed from the classic class AB to the more exotic Doherty configuration. A Doherty is a hybrid amplifier, with one portion for the main signal and one for the peak power. The Doherty concept sacrifices linearity in favour of efficiency. Combined with system improvements in the signal handling – digital pre-distortion (DPD) – base-station designs can be made with much higher efficiencies whilst retaining linearity.
As a small step in the right direction, we have produced a neat 2.7 GHz Doherty demo using three small, unmatched GaN devices that achieve a decent power density and efficiency performance when compared to silicon LDMOS – this drew much interest at the recent International Microwave Symposium (IMS/MTTS) show in Baltimore, June 2011. Commercial devices for basestation applications will require the next process version and improved matching designs; these will start to become available during the first half of 2012.
From analogue to digital
As well as being a complement to LDMOS in existinglinear topologies, GaN offers a much more excitingprospect – it is an enabling technology for digital transmission. This is another key research and development project for us, which illustrates how our focus in the high performance RF domain allows us to demonstrate our abilities as an innovator of improved systems. The digital transmitter rationalises the digital signal chain and culminates in a switched mode PA (SMPA) – a concept impossible without GaN. The thinking behind the SMPA is to make a break from the current family of linear designs of PA, with their inherent current and voltage losses, to a true switching design, with near-zero switching losses. The SMPA will follow the class E/F topology, taking the efficiency up to the 70-80 percent mark. The whole concept is aimed at producing far smaller, cheaper, cooler, and hence greener base-stations.
Our application insight for system concepts like the digital transmitter is complemented by a deep appreciation of knowing how to get perfect device partitioning. In this respect, we are not constrained by a single material and adopt the right technology for the required functionality.
Here it is important to mention SiGe, another compound semiconductor receiving much of our attention. The benefits of SiGe are distinct but quite different to GaN. Whereas the benefits of GaN are as the optimum RF power transistor material, SiGe’s benefits are as a very cost-effective technology for mixed signal RF solutions compared to GaAs.
We will continue to adopt an agnostic approach to materials throughout this year and beyond. 2011 could well go down as a year in the evolution of GaN with great strides in becoming a material that is available, workable and reproducible in significant volumes.
NXP has developed its power transistors on 3-inch SiC substrates in partnership with United Monolithic Semiconductors (UMS) and the Fraunhofer Freiburg Institute
© 2011 Angel Business Communications. Permission required.