Why can’t the world’s biggest IC company get any market share in the cellular space? I’ve asked myself that question a bunch of times over the years, and find myself asking it again in light of a few recent announcements from the company (another from ISSCC and another from EETimes).
When the cellular market started to take off, Intel was a strong flash memory supplier. They decided to get into the GSM baseband processor business and their approach was to integrate the baseband processor with the flash. Intel had a competitive ARM-based microcontroller core, and acquired a DSP core through a joint development. The resulting chip, code-named Manitoba, was a flop. Why did Manitoba fail? In my view, the flash integration was a bad idea. There was too much on the chip for low-end applications, not enough for high-end applications, and the more-complex flash process made the baseband too expensive.
A few years later, Intel spent $1.6B to acquire Israel-based DSPC, which at the time was the only open-market vendor of CDMA chips aside from Qualcomm. However, as things progressed, DSPC’s only significant customer exited the handset business. There were a few more developments by that group, but none gained any market traction. Eventually the Intel wireless group was sold off (for much less than $1.6B), and that was the end of Intel in cellular. Or was it?
A couple of years ago, Intel bought Infineon’s wireless business for another $1.4B. Infineon was the supplier of RF and baseband chips to Apple for the first couple of iPhone models. In addition, Infineon had just replaced TI as Nokia’s GSM chip supplier on the strength of a fully-integrated RF-plus-baseband chip as TI struggled to get their “LoCosto” single-chip solution using “Digital RF Processing” to production. Thus it looked like Intel had made the acquisition that would put it in highly-visible, leading wireless handset customers.
Then Apple switched to Qualcomm baseband chips. I’m not sure if the single-chip solution is still being used at Nokia. There are lots of alternatives now, and handset manufacturers are more likely to switch suppliers now than in the past.
Intel’s engineers gave several interesting wireless-related papers at this year’s ISSCC. The first, sort of hidden in a processor session, described the integration of an 802.11 transceiver on the same chip as a dual-core Atom processor. They did some smart things to prevent the noise from the processor from trashing the radio. But the die photo shows that the RF transceiver takes more real estate that the two Atom cores.
An Intel executive, commenting on these developments, noted that “Commercial versions of (the Atom+Wi-Fi) chips could emerge by mid-decade” and “We can now build a Wi-Fi radio and hopefully in the not too distant future a cellular radio to make digital RF practical for SoCs.”
The first statement shows that Intel doesn’t yet understand that the mobile market moves too fast to allow research chips to develop into commercial chips over a 3-year time frame. And he needs to look back at TI’s attempt to evolve their DRP technology from Bluetooth to cellular. Part 15 systems like Wi-Fi and Bluetooth are very forgiving of poor sensitivity and transmitter defects. Cellular radios are much much harder to integrate.
Intel has now announced their first design wins in handsets using the Atom-based baseband chip…all with second-tier handset manufacturers. And the multi-year deal announced in January between Intel and Motorola (or perhaps I should say Google) seems to only relate to Atom-based apps processors to run Android, not the “real wireless” stuff. The block diagram for the platform shows Intel (actually Infineon technology) cellular communications, but TI for Wi-Fi/BT.
However, deep pockets allow you to buy a company that can compete. I think Intel will do that within the next few years. I am sure they have the inside track to some companies that will get them in the wireless space.
Intel did have the PXA series of processors which were ARM based but their die sizes were too large and uncompetitive. The parts were sold off to Marvelle and now Intel is trying to get back into the mobile market. The mobile market is seeing roughly 15% growth while the traditional PC market is growing something like less than 5%. If you are going into the mobile market everybody wants ARM.
Assuming that you have a solid plan and sufficiently deep pockets (like Intel!), success favours those who keep at it. If you keep jumping in and out, expecting overnight success in a very conservative market, you are unlikely to succeed in anything other than spending your hard earned. This is something that keeps the wind in Intel's sail on the desktop and in the server space, so they should know better. Perhaps they shouldn't have dumped XScale at the first sign of hardship!
For the last 25 years, Intel has never shown it has better IC designs except in process technologies. Remember Intel's 80286 and 386 were not competitive when others offer the same thing? And through out its history, there is no customer service from Intel.
The PC companies benefited from WinTel monopoly in the 80s and first half of 90's, then they realized they were held hostage by WinTel, but it was too late. So, every company with their right mind understood they should avoid using Intel products whenever possible. Therefore, outside its monopolized PC segment (which should include tablet but Intel is insignificant), Intel has not played any role at all. I don't see this is going to change any time soon.
Looks to be quite dumb as to why Intel engineers were attempting such mammoth Integration tasks. Looks to be much enormous R&D dollars wasted.
TI went in the other direction and adopted PoP technology, wherein their baseband chips have pads on top. DRAM chips can be mounted on these pads right on top of the baseband. This gives flexibilty to the handset manufacturer to delay its memory size decision right upto the last assembly stage. In some cases a manual rework step on finished boards. Much like the Apple in China example from Bert above.
Bert22306 makes a good point. All-in-one integration seems like a good thing to do, but clearly does not allow for rapid changes in feature sets for cell phone differentiation and tiered platform strategies. Furthermore, such complicated chips take longer to design, debug, and validate in an actual phone design. If you miss a key timeline in the customers plans, an alternative solution can get in even if it is not optimized to the nth degree. Best to remain flexible.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.