EETimes.com – ESC: Confab to host another round of Intel versus ARM
(04/12/2010 3:48 $� EDT)
|Intel is king of the PC processor and ARM is the supplier of the leading processor architecture for mobile phones.The two companies have been squaring up to each other across the netbook and smartphone sectors for a couple of years now. The ARM-versus-Intel battle is set to continue on a number of fronts at the Embedded Systems Conference (ESC) in Silicon Valley, coming up April 26 to 29, at the McEenery Convention Center in San Jose, California.
It is arguable that Intel and ARM see the embedded market differently. As in the netbook contest, they tend to view things from their own side of the battlefield and with their own opinions as to what is important. Intel looks for high performance requirements and similarities with the PC and motherboard businesses while ARM, either leading or surrounded by its partners, looks for low-power opportunities that play to its strengths.
“Intel may have 90 percent of the PC processor market but they only ship about 2 percent of all processor chips each year. ARM and its licensing partners provide about five times the volume that Intel does,” said Jim Turley, founder and principal analyst with consulting firm Silicon Insider (Pacific Grove, Calif.).
Even though Intel is not exhibiting at ESC, it will be there in silicon — on boards from Kontron and Advantech amongst others. And the company is also represented by Wind River Systems Inc. (Alameda, Calif.), a provider of software and software development tools, which Intel acquired in July 2009.
Jonathon Walsh, general manager for software within Intel’s embedded group, said that the reason Intel bought Wind River was to help protect customers’ software investment as they move through the generations in silicon and because those customers are looking for hardware-software solutions, not just a chip.
“We track 35 [embedded] market segments and see incredible diversity,” said Walsh insisting that increased use of networking and rich graphics are two embedded megatrends.
“Intel tends to see the embedded world as PC-lite. Intel is leveraging the success of its x86 architecture into as many applications as possible. There are advantages because there is a wealth of compatible software,” said Turley. “The downside is that the Intel architecture is complex and uses a lot of energy, and increasingly embedded applications are going wireless and sometimes even battery-less.”
But Intel does have the ability to produce lower power versions of established architectures such as Pentium and Core, as can be experienced at ESC.
This year Intel is helping continue ESC’s long-standing tradition of offering a Build Your Own Embedded System conference track. Intel is providing a development board featuring the Dual-Core D510 Atom processor. The processor supports up to four execution threads and the board come with and 8.4-inch SVGA LCD module and runs Microsoft’s Windows Embedded Standard 7.
ARM, which is exhibiting, will have partners close at hand from its ARM Connected Community program as well as many manufacturing licensees, such as Freescale Semiconductors, Marvell, NXP, Renesas, STMicroelectronics, Texas Instruments, Toshiba.
“ARM’s business model could not be more different to that of Intel. It has pioneered the intellectual property licensing business model and avoids the costs of manufacturing while working with IP licensees. ARM are the architects, they don’t swing the hammer,” said Turley.
“ARM’s business model also has the benefit of isolating it from the day-to-day problems of selling and marketing chips. But it means it constantly has to be re-inventing its road-map, helping its licensees succeed, and providing a reason for licensees to pay again to stay with the program.”
Another disadvantage is that the IP licensing business model only lets ARM capture a portion of the value from the supply chain, which partly explains the company’s relatively slow — albeit steady — growth over its 20 year history. In contrast Intel incurs manufacturing costs but gets control and all the revenue from its chip sales, said Turley.
A difference between Intel and ARM also shows up in terms of manufacturing options. Intel tends to sell packaged die along with market-specific reference designs and software. Meanwhile ARM’s cores are available not only to IDM licensees but also to hundreds of fabless chip companies through deals with foundries such as Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC).
Intel has failed to become adept at providing hardware in the form of intellectual property that can be integrated into a system-on-chip (Soc) IC, said Turley. “Just having a separate chip for the microprocessor alienates a lot of customers and consumes more power just from wiggling the I/O pins,” Turley added.
Intel announced it was joining the SoC and IP licensing game in March 2009, when it said it would port Atom processor cores to TSMC’s technology platform.
However, that attempt to take on ARM in the SoC space has yet to prove successful. It was revealed in February 2010 that there are no immediate plans for any TSMC-manufactured Atom-based chips, apparently due to a lack of customer demand. Intel’s Walsh confirmed that the collaboration with TSMC is in a holding pattern. “The technology development has proceeded well. We haven’t found an opportunity, from a technology perspective, for a customer fit,” he added.
“The problem is Intel is new to this kind of business. You cannot build up the necessary infrastructure overnight. But it is in ARM’s blood. ARM has been doing it for twenty years,” commented Turley.
Microprocessor versus Microcontroller
Ian Ferguson, director of enterprise and embedded solutions certainly sees benefits for ARM as microcontroller applications shift from 8-bit to 32-bits. “Amongst a lot of other applications touch-screen LCD controllers are going to 32-bit, smart meters and a new wave of motor control,” he said.
Intel’s Walsh sees the same trends but Intel tends to produce two chip solutions — I/O chip and processor — rather than integrated microcontrollers. “We’re seeing a lot of consolidation where software that formerly ran on multiple microcontrollers is ported to a single Atom processor. A separate I/O chip provides flexibility. It allows customers to mix and match I/Os with compute power,” argued Walsh.
At ESC ARM is also expected provide further details of the outcome of its partnership with Xilinx, announced in October 2009. “The combination of programmable and ASIC technology seems very powerful. When you think of the ROI [return-on-investment] for an ASSP at 28- and 22-nm then a blend of ASIC and ASSP has to make sense,” said ARM’s Ferguson.
However, important as the design and manufacturing engineering is, it is only half the equation, according to Turley of Silicon Insider. Things have changed drastically with the proliferation and consumerization of electronics.
The technical specifications — energy consumption, chip count and form-factor, reliability and so on — can still be deal-breakers, but often companies sign up for the road-map as much as for individual processors or cores. And that is much more nebulous and comes down to a promise from the vendor and a bet by the customer on who can deliver in two- to five-year’s time.
“In the 1980s processor choices were made purely on technical merits. In the 1990s it also included software support. Now we are in the 2010s decisions include how much free software is provided, time-to-market, reference designs, direct and indirect advertizing support and kudos, which sometimes just comes down to which architecture is cool. In a way it is sad but it’s there,” said Turley. “That’s where Intel has a big advantage. Intel can help make a market happen, but the downside is a single source supplier and the battery life.”
Turley acknowledged that for some customers, those that are perhaps pursuing a low-power eco-friendly agenda, ARM is also cool. “There’s no doubt that ARM has broken through and left a lot of the other processor IP providers behind, right now,” said Turley.
Both Intel and ARM are pushing ahead with increasingly complex multicore versions of their processors. Nonetheless, in terms of annual revenue, Intel Corp. is a $33 billion Goliath and ARM Holdings plc is a $500 million David. That gives Intel advantages in terms of paying for R&D, acquiring technology and creating industry-wide solutions.
“Multicore and virtualization give rise to a spectrum of different use models in embedded,” said Intel’s Walsh. “SMP [symmetric multiprocessing] is being used extensively in networking, in 3G basestations. It’s also there in security and consolidation allowing multiple pieces of legacy software to operate in separate partitions and without crashing the system.”
Intel Ct is a multi-threaded programming API launched in beta form at the Intel Developer Forum in September 2009. Ct is partly a result of an acquisition Intel made of RapidMind Inc. in August 2009. RapidMind was a provider of software tools that allow the development of C++ code for use on multiple processors. At about the same time Intel acquired Cilk Arts Inc., which had licensed Cilk technology from the Massachusetts Institute of Technology.
“If anyone does, Intel has the ability to develop that,” said Turley. “Of course multicore parallel processing has been going on in embedded applications for years, but it has been application-specific,” he added.
ARM’s Ferguson agrees that multicore is starting to impact the deeply embedded space. “The multicore banner has been waving for some time, but more of our new licensees are moving straight to multicore. Our history of providing performance within a certain power budget is helping.”
Multicore progess will be reported at the Multicore Exposition, which is
collocated with ESC. And the ARM-versus-Intel battle is an undercurrent there.
Keynoter Tomas Evensen, CTO of Wind River, is clearly, if not exclusively, in the Intel camp. His presentation is set to illustrate how multicore software can simplify the achievement of multiprocessing advantages, using high-end examples ranging from network equipment to multi-function printers. Keynoter Gene Frantz, Principal Fellow of Texas Instruments, plans to tell attendees what to do with a billion transistors of logic. Although not explicitly an ARM-oriented presentation, TI’s experience is clearly in the heterogeneous multiprocessing mix of multiple DSP and ARM cores.
In addition, Max Domeika, an Intel engineer is due to present an hour-long session on a process to promote the understanding of the characteristics and potential parallelism in application and how to map them to the most suitable programming API.
The challenge for both Intel and ARM is whether the use of multiple and heterogenous resources can be generalized and made simple for programmers. “All that we hear from the academic community indicates that it is really difficult to do,” said Turley. “Progess is likely to be incremental and slow, but whoever can make a difference there, is likely to win out.”