By Eric Higham
Shortly after the conclusion of World War II, research began on a solid-state replacement for vacuum tube-based devices. The goal was to develop devices that would be more robust and more reliable than the tube devices in use at the time. Bell Labs started a group, led by William Shockley, to develop this solid-state alternative for amplification purposes. In the late 1940s, this group announced the invention of a point-contact transistor with gold contacts to a sliver of germanium (1). This was a start, but this device was also very fragile, relying on a spring to ensure contact of the gold probes to the germanium surface. Shockley was not satisfied with this solution and pushed on in his research. This work culminated in the theory of p-n junctions and minority carrier injection and what Shockley called the junction transistor (2). In 1951, Bell Labs fabricated a working germanium transistor (3).
The size, portability, performance and reliability of the germanium-based transistor fueled the growth of both military and commercial applications. Computers became one of the biggest early users of the germanium transistor, but the material had issues with temperature range and most notably, reverse leakage. The temperature range was a problem for military applications and the reverse leakage created serious issues for computer manufacturers. In 1954, the chemist who was instrumental in the germanium transistor fabrication process at Bell Labs announced that a team he was running at Texas Instruments had fabricated a silicon-based transistor (4).
Silicon technology has become the preeminent high-volume process technology, but the US Department of Defense started funding efforts to develop the capabilities of III-V semiconductor technologies in the 1980s. These efforts started with the GaAs Pilot Line Program that sought to develop GaAs digital integrated circuits to compete with silicon (5). The program was successful, but the advantages of silicon were undeniable and the government shifted their funding to refining GaAs MESFET technology and developing high frequency GaAs amplifiers with the Microwave/Millimeter Wave Monolithic Integrated Circuits (MIMIC) program. (6)
While the initial focus of the MIMIC program was on defense applications, the funding also helped develop ancillary developments in test and measurement, assembly and manufacturing applications. Because of this and other funding, GaAs devices have seen performance, reliability, manufacturability and market share increase appreciably. From the groundbreaking transistor work in the 1940s, the RF semiconductor industry has grown to become a large, vibrant segment of the broader electronics market.
To predict where the compound semiconductor market will head in 2015 and beyond, it is useful to review the historical performance and the factors that have influenced the growth profile to date. Germanium transistors quickly transitioned to silicon and this technology has enjoyed a tremendous ramp in volume, first with discrete transistors and now with integrated circuits going to process nodes below 20nm. The performance of compound semiconductor-based devices has been superior to silicon and GaAs has become a very mature, high volume technology. Even though other competitive technologies will factor into the prospects for the future of the RF semiconductor market, GaAs is currently the dominant technology.
Figure 1 shows the historical performance of GaAs device revenue from 1999 to a forecast for 2014. It tells an interesting story and illustrates market drivers. There was fast growth in the 1999/2000 timeframe as the .com era hit full stride. The working theory was “build it and they will come”, but the unfortunate reality was there was very little demand for the increased capacity and speed of the new networks. The .com bubble burst; companies built networks, but no one came! As a result, GaAs revenue dropped as quickly as it rose and floundered around at this level for several years.
Figure 1 • Historical Performance of GaAs RF Device Revenue.
In 2004, the effects of the “wireless revolution” and mobile communications become apparent on GaAs revenue and the revenue trajectory has been steadily upward. The initial attraction of mobile communications was the “anywhere, anytime” aspect of staying in touch. As analog communications evolved into digital and data rates started to increase, the clunky bag phone evolved into a much more sophisticated terminal that is fueling the tidal wave of data consumption. Smartphone usage has grown dramatically and Strategy Analytics estimates that nearly two-thirds of all phones sold in 2014 will be smartphones. The CAGR for GaAs device revenue since 2004 has been almost 11% and this will push 2014 results to an estimated $6.6B.
The evidence for how much the GaAs and the RF compound semiconductor industries rely on the cellular segment should be clear in Figure 2. This block diagram, courtesy of TriQuint, shows the front end for a representative smartphone. It includes eleven amplification functions, many of them accommodating multiple transmission bands and eight switching functions, with most of them being multi-throw.
Figure 2 • Smartphone Block Diagram.
The complication in the front-end is rooted in technical and business considerations. A simplistic analysis shows that data capacity increases as the spectral efficiency increases (more bits/sec/Hz) or the amount of bandwidth increases (more Hz). Wireless operators use both of these techniques by purchasing additional spectrum and developing more sophisticated modulation schemes that allow for wider channel bandwidths. The term “4G” has really become synonymous with faster data speeds. To support higher data rates, network operators have embraced the W-CDMA/UMTS standard that serves as an evolution path for GSM and the newly developed LTE standard. Both use linear modulation schemes that increase spectral efficiency and incorporate flatter network architectures to reduce cost.
The second part of the technical consideration is spectrum and that is a thornier issue. Spectrum is a scarce resource. Since more spectrum cannot be created, the best option is to repurpose it. Governments around the globe are doing this with auctions of reclaimed or underutilized spectrum. This creates some additional challenges for wireless operators, because this additional spectrum may not be close to existing frequency bands, it may not exist over a large geographical footprint, the channel bandwidth may be less than desired and it is expensive!
The final dimension addresses the operator business model. Ideally, operators would like a single handset that covers their entire service footprint and allows users to roam on different networks. This is currently not possible, but operators are pushing to minimize the number of phones they must maintain. To enable this, manufacturers use architectures that incorporate the latest generation of linear PAs, while still accommodating older standards that use saturated PAs. These architectures must accommodate frequency bands that are likely not contiguous and range from 450 MHz to 3.8 GHz. A report from Strategy Analytics identifies 45 E-UTRA WCDMA/LTE bands, with another eight that have been proposed, but not approved. In response to these divergent business requirements, cellular front-end architectures are making more use of multi-mode, multi-band PAs, along with saturated PAs shown as “2G” in the block diagram of Figure 2.
Strategy Analytics estimates that roughly 1.2 billion phones will have block diagrams similar to, or perhaps even more complicated than the one shown in Figure 2. Our research indicates these phones currently handle an average of 4.6 linear bands, in addition to four saturated bands and we expect the number of linear bands to exceed six shortly. Add in approximately 1.4 billion “other” cellular devices (feature phones, tablets, PCs, notebooks, E-readers, etc.) and it is easy to see why GaAs and compound semiconductor revenue is so high in this segment.
In the early days of the wireless revolution, GaAs was the only technology that could provide the performance, frequency coverage, cost and reliability for high volume applications. As volumes and device complexity continue to increase, competitive technologies are beginning to capture market share from GaAs. The best example of this is with the handset switches shown in Figure 2. This application has largely shifted to a Silicon-on-Insulator (SoI) technology that makes use of the high volume processing capabilities of silicon CMOS foundries. Silicon also provides a better opportunity to integrate additional low frequency control circuitry and offers better ESD performance. These and other performance-related features have allowed SoI switches to displace GaAs devices in many of these applications.
While silicon switches stand as the largest volume RF application to capture share from GaAs, power amplifiers represent the largest revenue opportunity for competitive technologies. CMOS-based PAs have steadily been capturing share in entry level, lower tier handsets. Since these applications represent a slowly shrinking opportunity, GaAs PA manufacturers have been willing to cede some market share. The big shock to the GaAs community was the announcement of CMOS PAs that target emerging LTE opportunities. Currently, only Qualcomm and Peregrine Semiconductor have competitive offerings for these applications and they are enduring the usual growing pains, but these devices will take share away from GaAs. In addition to CMOS, SiGe for low power applications and GaN and LDMOS for power applications serve as the main competitive threats to GaAs in RF applications.
Combining these thoughts and accounting for all the technologies that address applications in the RF segment, Figure 3 shows a snapshot of the segmentation of the estimated 2014 RF market revenue.
Figure 3 • Segmentation of RF Semiconductor Market Revenue.
The competitive technologies add about $2.4 billion of revenue to the GaAs portion of the market, bringing the total to about $9 billion. It should be very clear how important cellular applications are to the overall RF market, accounting for nearly 66% of the revenue. Adding in other wireless applications like Wi-Fi (the second largest revenue segment), base stations, microwave/millimeter wave backhaul, VSAT, etc. increases the wireless segment to nearly 86% of the total RF semiconductor revenue.
With this snapshot of where the RF compound semiconductor industry stands in 2014 and a good understanding of the developments and trends that got us here, where does the industry go in 2015? The overwhelming, insatiable desire to consume increasing amounts of data will influence every future development and trend. It is not hyperbole to say that every trend in the electronics market ultimately ties to data consumption. Figure 4 shows the Cisco VNI (Visual Networking Index) forecast to 2018, with actual data back to 2009. The CAGR for the data in the chart approaches 28%. To put this into perspective, a petabyte of data is equivalent to about 223,000 DVDs. In 2018, this forecast implies users will generate data equivalent to 29.3 billion DVDs…per month!
The wireless segment shows the fastest growth, with a CAGR slightly greater than 77%. Given the segmentation of the RF semiconductor industry shown in Figure 3, this is a promising trend. As impressive as mobile growth is, it will only represent 12% of the total data consumption in 2018! The rest of the data will reside on wired copper, fiber or coaxial cable networks. Increasingly, a bit of data will travel over several of these networks and there are opportunities for RF semiconductors in all these data segments. The revenue and volume associated with the cellular and Wi-Fi segments shifts the spotlight away from some of the other areas of the industry, but there are a number of interesting developments taking place in these segment.
I think the discussion of the history, trends, drivers and present state of the RF semiconductor industry focuses the view of 2015 and beyond. The last section includes my thoughts on some of the important topics in the industry. Some are obvious, but some will require a bit of faith:
Data Consumption: This engine drives the entire semiconductor market. Any substantial changes to the trajectory presented in Figure 4 would have catastrophic repercussions for the semiconductor industry. Even though the last couple of data forecasts have shown declining growth forecasts, the numbers remain large. There are developments like 4K or UHD TV, the Internet of Things (IoT), increasing HD and UHD video uploading to social networking sites and the ongoing arms race between fiber and coax to provide the highest data rates that will sustain and probably increase data consumption rates.
RF CMOS: This is an easy trend to call. CMOS-based amplifiers and switches will continue to capture market share from GaAs. The trendy discussion in the industry has been “the death of GaAs”. This is unlikely to happen anytime soon. CMOS has undeniable performance, integration and cost benefits, but this technology works best with high volume applications that have stable performance requirements. When volumes are lower, mask costs of CMOS affect the cost competitiveness of the technology. RF CMOS revenue will increase in 2015 and beyond, but it will not be the dominant RF technology in the foreseeable future.
MMMB PAs: To address the rise of LTE bands, carrier aggregation and increasing data consumption, multi-mode, multi-band (MMMB) PAs will continue to capture market share in handset front-end architectures. With the price sensitivity of this market, the price of the MMMB PAs cannot exceed the price of the components they replace. This would seem like a bad development for GaAs device revenue, but manufacturers seem to be making the block diagram even more complicated by including more functionality or expanding the number of bands in the phone. The net effect for the GaAs device market will be neutral to positive. The situation will be a bit different for epitaxial substrate manufacturers, because the MMMB trend will mean less production area and a reduction in the $/mm2 metric.
GaN: There has not been much discussion of GaN here, but the technology has turned the corner and it is seeing significant adoption in commercial applications. Defense applications have driven the development and adoption of GaN and the latest Strategy Analytics forecast shows this will continue with defense applications accounting for more than 50% of GaN revenue in 2018. Commercial adoption is increasing quickly with CATV adoption continuing and base station PA applications increasing quickly. VSAT and point-to-point radio applications are also starting to see growth. The vast majority of these devices will be GaN-on-SiC. There is a chorus to push GaN-on-silicon into lower power, high volume applications and functions. The argument is that the lower cost structure will allow the technology to address more applications. Unless there is a disruptive manufacturing development, the realization of this idea appears to be several years down the road, at best.
The preceding topics have addressed a shorter time horizon. The final two topics are longer term, with the potential to change the trajectory of the entire industry.
Internet of Things (IoT): This is one of the hottest discussions in the electronics industry. The premise is that deploying a large number of embedded computing sensors and interconnecting them into wide area networks with the Internet will dramatically improve society. The latest Strategy Analytics forecast anticipates more than 33 billion devices connected to the Internet in 2020. The concept involves smart sensors sending data that enables better decisions. This “intelligence” gives rise to applications involving telemedicine, “smart” cities and utilities, industrial automation, security and a whole host of others. The connected devices and networks will create a vibrant service economy, which will provide substantial revenue. This concept is a superset of M2M communications, Wi-Fi, cellular terminals and the whole host of devices already connected to the Internet, so it is clear that the IoT is already happening. The use cases currently involve low data rate, low power applications, so silicon-based semiconductors manufactured in high volume seem the most likely choice. With the breadth of devices and applications included in the IoT concept, there is little doubt that there will be growth. The challenge will be determining applications for RF compound semiconductors.
Figure 4 • IP Data Consumption.
5G: If IoT is the most discussed topic, then 5G is running a close second. This concept assumes that existing network architectures will not be able to keep up with the anticipated data consumption increases. This effort will revolutionize the RF industry, because the goal is to increase user data rates, capacity, battery life and network devices by orders of magnitude over existing capabilities. Network deployments may not be until 2020, but development work streams are currently underway, under the auspices of Alcatel-Lucent, Fujitsu, NEC, Ericsson, Samsung and Nokia. This is a disruptive opportunity for the RF semiconductor industry because several of the activities involve developing networks at frequency bands of 5 GHz to 86 GHz, with more available bandwidth. Other concepts under development involve the use of antenna beamforming, beam tracking and massive MIMO. These all play into the strengths of compound semiconductor devices and 5G represents an exciting opportunity for the entire RF semiconductor supply chain.
This is a very exciting time for the RF semiconductor industry. High volume applications are growing, new technologies are gaining traction and new applications are in development to handle the tidal wave of data consumption. There will undoubtedly be twists and turns, along with a surprise or two along the way, but the future for the industry looks rosy.
About the Author:
Eric Higham serves as Director, Advanced Semiconductor Applications Service, Strategy Analytics. He has held various positions in engineering, applications, business development and marketing at Raytheon, MicroDynamics and M/A-COM. He received a BSEE from Cornell University with a concentration in solid-state semiconductors and an MSEE from Northeastern University with a concentration in Fields, Waves and Optics.
(1) Bardeen, J. and Brattain, W. “Three-Electrode Circuit Element Utilizing Semiconductor Materials,” U. S. Patent 2,524,035 (Filed June 17, 1948, issued Oct. 3, 1950).
(2) Shockley, W. “Circuit Element Utilizing Semiconductive Material,” U. S. Patent 2,569,347 (Filed June 26, 1948. Issued September 25, 1951)
(3) Teal, Gordon K. “Methods of Producing Semiconductive Bodies,” U. S. Patent 2,727,840 (Filed June 15, 1950. Issued December 20, 1955)
(4) Teal, G. K. “Some recent developments in silicon and germanium materials and devices,” Presented at the National Conference on Airborne Electronics Dayton, Ohio (May 10, 1954)
(5) Zolper, John Z. “A DARPA Perspective on the Future of Electronics”, CSMantech Presentation, 2003.
(6) Cohen, Eliot D. “MIMIC from the Department of Defense Perspective”, IEEE Trans. Microwave Theory and Technique, vol. 38, no. 9, 1171 (1990).