Smartphones control the semiconductor business

20 February 2012 The highly absorbent smartphone is gaining a controlling interest in how the electronics industry is evolving, as technology developed for it seeps increasingly into other markets. The electronics industry is all about economies of scale, and has…

20 February 2012

The highly absorbent smartphone is gaining a controlling interest in how the electronics industry is evolving, as technology developed for it seeps increasingly into other markets.

The electronics industry is all about economies of scale, and has for much of its existence been organised around major-volume markets. This pattern has been renewed with the advent of the smartphone, and the way in which the electronics industry evolved for smartphone platforms has encouraged innovation in other, often quite dissociated, technologies.

In the beginning, the key customer was the military. When the ‘traitorous eight’ left transistor pioneer Bill Shockley’s Shockley Semiconductor Laboratory to found Fairchild’s semiconductor arm, they sold the first silicon transistors at $200 a shot to the IBM military research team in Owego, which was developing the B-52 bomber’s navigational computers. Ten years later, four of the eight left Fairchild to form Intel and the military began to lose its hold on the industry.

The mainframe computer had become a big user of transistors and one of Intel’s early projects was the microprocessor for a desk calculator. The big shift in power would come with the introduction of the personal computer in the early 1980s.

Intel became the biggest beneficiary of the personal computer revolution. In the mid-1980s, the company was the eighth-largest chipmaker in the world with less than half the revenue of the biggest: Japan’s NEC. By 1993, Intel had overtaken NEC, and by 2000 had achieved sales three times larger than the new number two, Toshiba. The PC, and Intel, seemed unstoppable. In the past decade another shift has taken place and had, in fact, begun as soon as Intel became the dominant player in semiconductors.

The reversal in the PC’s influence over the semiconductor industry began in the mid-1990s with the establishment of what Greg Linden, Clair Brown and Melissa Appleyard tagged the ‘Net World Order’ in a chapter for the 2003 book ‘Locating Global Advantage: Industry Dynamics in the Global Economy’. At that time, the Internet seemed primed to take over: delivering an environment where thin clients would access services in the cloud for anything too big to sit on a stripped-down computer.

Mobile ascendancy

That vision is becoming a reality, but not quite in the way Linden and colleagues posited. The machine that delivered was the mobile phone and its position is demonstrated by the latest league table from IC Insights. Some 400 million PCs are likely to be shipped in 2012. They are less expensive, but smartphone shipments now exceed that number. By 2015, Gartner expects the industry to ship close to a billion smartphones each year. And tablets could be in the hands of almost 100 million more users by the end of the year, according to JP Morgan.

Thanks to the rise of the mobile phone, Samsung has gradually been encroaching on Intel’s dominant position, and the top ten now includes two companies that don’t even own fabs, but have used the communications market to achieve combined revenues of more than $15bn; and it is now driving the technology of semiconductors for most of the big players, with the exception of Intel, which still captures much of the value of a PC sale due to its near-monopoly position as a processor vendor.

Mary Olsen, senior analyst for market intelligence company Gary Smith EDA, reckons that products such as mobiles and tablets are driving the development of new 3D packaging because of the need for high density; if cost trends remain on their current trajectory, however, such high-volume devices may be the only complex digital chips that get designed.

Gary Smith EDA helped develop a cost model for the International Technology Roadmap for Semiconductors produced by research consortium Sematech; and it has shown the cost of chip design spiralling out of control, without much prospect of coming back to levels considered realistic in 2000 for close to another ten years. In Smith’s view, design automation has failed to keep pace with the ability of engineers in the fab to make multibillion-transistor chips. This is having a knock-on effect on the number of chip designs, and has caused some speculation around how this will play out.

Tempting back the VCs

UK-based Icera Semiconductor in Bristol collected more than $250m in venture capital (VC) funding. Sold to graphics chipmaker Nvidia for $367m, the deal was not a disaster for the VC funds but neither was it a raging success that would encourage them to find more of the same, even though Icera’s primary target market lay in mobile handsets and dongles. The main complaint from VCs is that the cost of designing the chips, and of writing the support software, is too high. “If we can keep the cost of designing a SoC to below $25m, the VCs will start funding semiconductor start-ups again,” declared Smith at the 2011 Design Automation Conference. “Above $50m, even the IDMs have a problem.”

Although it costs several million dollars to make a set of masks that can be used to fabricate a chip, the cost of the hardware design is not the only problem, Smith says: “The cost is also driven by software. It is too high to drive the ecosystem properly.”

For Smith, there is too much reinvention. Chip designs tend to break down into a set of coarse-grained blocks, such as processors, I/O controllers and other virtual devices. Each needs its own set of designers to realise, even if the blocks are bought in: from ARM, for example, in the case of a processor.

“Re-use is the major driver of productivity,” he says. “If you look at the number of engineers needed per block and compare that to the cost of the entire project, the number of blocks needed for an ideal design is five; but there are not a lot of ideal designs. Usually, the number of blocks is more like 25 to 35.”

There is an answer to the problem in Smith’s view: using what is known as platform-based design. This is where you take readymade combinations of processors and I/O and glue them together on design. Companies such as CEVA have put together combinations of hardware and software to produce this kind of readymade block, and are using systems developed for the mobile industry to break into other areas.

The company has taken advantage of the commonality between many of the recently developed communications standards in how they encode and decode signals to reduce its own development load.

“The communications world has been marching towards OFDM,” says Eran Briman, vice-president of marketing at CEVA, referring to the Orthogonal Frequency Division Multiplexing system used by protocols such as the upgrade to 3G, Long-Term Evolution (LTE). OFDM divides the signal among many parallel carriers, each transmitting on its own small part of the frequency spectrum.

OFDM is also a component of the latest wave of digital television standards such as Digital Video Broadcasting – Terrestrial (DVB-T) used in Europe and Integrated Services Digital Broadcasting (ISDB), developed by Japan, but also forming the basis of systems in South America.

The problem for the TV makers is that all of these protocols are subtly different; this is yet another instance of how components vendors from the computing/communications worlds are encountering cultural variance as they move to embrace the TV. Briman says: “If you are a silicon provider who wants to support all the different forms of DVB, you have to have at least ten different products; and, for the OEMs, before their product exits the manufacturing house, they have to say that this product is for the US and this other is for China.”

When chipmakers write software

CEVA developed the XC architecture as a programmable engine that can handle these subtly different changes in protocol through software changes. “We started off with a platform for software-defined radio, and came out with our first offering, which was for LTE. We have added a variety of layers, including HSPA for 3G, and now we have DVB. The technology spans multiple product areas,” says CEVA’s Eran Briman.

To reduce the software development burden, CEVA has worked with partners to develop the software that sits on top of the XC processor. “Often the end customers hook us up together. They say: ‘come up with a complete solution for us’ and then, when it’s ready, it is there for the rest of the market,” Briman explains.

He contrasts the situation with dedicated hardware: “When it comes to hardware-based development you have to start with, say, DVB-T and run it in the labs with some initial algorithms. You then test the first prototype in the field and go back to the lab to do some more tuning. Then you produce a test chip and finally develop the software for it. If you do this with a common software-based platform, you can run the hardware and software development in parallel. If you find out later you have some issues, you can tweak the software.”

Mobile TV drivers

The foray into TV from communications is turning full circle, says Briman, with a TV processor expected to sit alongside one that handles LTE or 3G communications: “We see new players in TV coming from the mobile space, running open platforms such as Android. The TV products will start to look more like an iPhone or an iPad.”

Many of the chip designers have yet to exploit platform-based design fully, but those priced-out of the IC production business have found themselves using the derivatives of smartphone design as their own platforms.

Texas Instruments, for example, has turned the OMAP parts used inside a number of smartphones into the Sitara industrial microcontrollers, replacing some of the phone-specific I/O with more appropriate peripherals but keeping the combination of general-purpose processor and DSP.

Yet another frontier in this scenario is the payment component: handset manufacturers and service providers alike are getting keen on the idea of more integration between the smartphone as a payment device and the kinds of services users will be prepared to pay to have delivered on the move. *

Further information

Jargon buster

3rd Generation Partnership Project (3GPP) – Collaboration between groups of telecommunications associations, known as the Organisational Partners. The initial scope of 3GPP was to make a globally applicable third-generation (3G) mobile phone system specification based on evolved Global System for Mobile Communications (GSM) specifications within the scope of the International Mobile Telecommunications-2000 project of the International Telecommunication Union (ITU).

3GPP Long-Term Evolution (LTE) – Standard for wireless communication of high-speed data for mobile phones and data terminals; based on the GSM/EDGE and UMTS/HSPA network technologies, boosting the capacity and speed using innovative modulation techniques. Developed by the 3GPP (3rd Generation Partnership Project).

Digital Video Broadcasting – Terrestrial (DVB-T) DVB European-based consortium standard for broadcast transmission of digital terrestrial television. First published in 1997 and first broadcast in the UK the following year. This system transmits compressed digital audio, digital video, and other data in an MPEG transport stream, using coded orthogonal frequency-division multiplexing (COFDM or OFDM) modulation.

Field-programmable gate array (FPGA) – Integrated circuit designed to be configured by the customer or designer after manufacturing – hence ‘field-programmable’. The FPGA configuration is generally specified using a hardware description language (HDL), similar to that used for an application-specific integrated circuit (ASIC).

IDM – Integrated device manufacturer

Integrated Services Digital Broadcasting (ISDB) – Japanese standard for digital television (DTV) and digital radio (DAB) used by radio and television networks. The standard replaced the previously-used MUSE Hi-vision analogue HDTV system. A derivative of ISDB, ISDB-T International, was developed by the Brazilian government, and is being adopted elsewhere in South America.

Orthogonal Frequency Division Multiplexing (OFDM) – Method of encoding digital data on multiple carrier frequencies. OFDM has developed into a popular scheme for wideband digital communication, whether wireless or over copper wires, used in applications such as digital television and audio broadcasting, DSL broadband Internet access, wireless networks, and 4G mobile communications.

‘Deconfiguration’ and selective decommissioning

In the late 1990s, Bob Payne, vice president of engineering at custom-chip developer VLSI Technology (later bought by Philips Electronics) coined the term ‘deconfiguration’ to describe what would happen with platform-based design:

Some of the peripherals in one of the large blocks envisaged by Gary Smith EDA will not be needed. The cost of having them on silicon is outweighed by the greater cost of trying to redesign the part to not have those peripherals. Many of the microcontrollers on the market quietly use this approach so that manufacturers can avoid multiple versions of a product.

The unwanted peripherals are often ‘killed’ by blowing fuses during test, or are made unavailable by the absence of critical firmware. Customers will perform their own deconfiguration by disabling functions they do not want to use in the off-the-shelf chip. As the economies of scale favour high-volume, standard ‘cookie-cutter’ parts, this generally works out cheaper than buying a part that is a better fit for the application but which ships in much lower volume.

Deconfiguration does not satisfy everyone. The big advantage of designing your own silicon comes when you need to move large amounts of data around quickly. The external memory bus is a major bottleneck and one that SoC designers can often avoid by integrating more on-chip. Makers of programmable logic have taken core platforms for smartphones and combined them with programmable logic. Both Altera and Xilinx have alighted on a dual-core ARM Cortex-A9 – the guts of a Samsung Galaxy S II – as the main processing subsystems for Cyclone V and Zynq SoCs.

Stéphane Monboisset, product line manager at Xilinx, explains: “Many of the applications are based around video. For example, driver assistance systems in automotive have to perform a lot of real-time video analytics.” The programmable logic section can act as a dynamically reconfigurable co-processor, working in concert with the two main processors, which have their own multimedia accelerators. Having all the processing on one device makes it easier to share high volumes of video data.

Senior director of embedded products at Altera, Chris Balough agrees that the connection between the programmable-logic and processor sections is vital, claiming that the company has put a wider connection between the two than its prime competitor to maximise the bandwidth between two co-operating elements. “We see this form of processing as a significant macro trend in the industry and an important one for the company,” he says.