It's hard making money as a chip maker. Your set-up and development costs are enormous. And, in something like 18 months, that chip foundry you spent so much money designing, building, fitting out and getting up to production speed and quality goes from state of the art to just a place for building cheap chips to keep paying the bills, thanks to Moore's law.
Most people forget the second half of Moore's law. You cram in twice as many transistors but the facility costs that much more to build. Paying down those old fab costs is one reason Intel is so big in SSDs. Atom is also there for making use of older fabs as well as trying to keep AMD from eating Intel's market alive.
If you don't have something to differentiate your ARM chips, that leaves you in a difficult situation
In general, Intel's ability to get new fabs online with high production quality in a short time is unmatched and keeps the company ahead even when it can take 10 years to get a new chip production process such as high-K metal gates up and running.
ARM chip builders
If you're an ARM chip builder, you don't build the fab — unless you're Samsung — but you pay to use it. And you also have extra work to do as well. What you get from ARM isn't a blueprint; it's a licence. You have to design your own chip and platform around that, and you have to add something special to compete with the other ARM makers.
Samsung has scale and production experience, Nvidia has graphics know-how, and Ericsson has massive experience in radio and modem development. If you don't have something to differentiate your ARM chips, that leaves you in a difficult situation where it can be hard to make chips that you can make money from.
When your way out of that situation is to cancel products that don't make you money, that leaves your former customers in a difficult place. And that's happened more than once to Texas Instruments (TI) customers over the years, when the company has cancelled various OMAP mobile processors.
Palm had so many other issues with the development of webOS that moving from OMAP in the Pre to Qualcomm Snapdragon in the TouchPad might have been just one more thing to deal with. But changing chipset means rewriting all your drivers and it could have been a factor in slowing down HP's ill-fated tablet.
MeeGo also had more problems as a project that I can easily count, many of them in management and governance. But MeeGo development handsets had been entirely based on OMAP chips, so when TI cancelled development of its OMAP chip with an integrated baseband modem in late 2008, that was pretty much the death knell for the project — especially as Nokia worked with Intel to develop MeeGo, which meant no CDMA MeeGo handsets for Sprint or Verizon and no LTE MeeGo handsets for any market at all.
Toshiba and TI OMAP chips
More recently, we know Toshiba was building its Windows RT devices on TI OMAP chips. We also know it cancelled its initial RT plans because of delayed components, which prevented it getting its planned RT tablet and notebook out for the launch of Windows 8 alongside Surface and tablets from PC makers such as Asus.
Toshiba declines to confirm whether the delayed components were the OMAP chips that TI is no longer making or something entirely unrelated, but the timing is certainly thought-provoking.
We don't think Toshiba has the same reservations about Windows RT that other OEMs such as Acer have expressed. At CES this year, TI was showing off tablet and notebook form factors it said Toshiba would make and it was still demonstrating the tablet prototype at Computex this June.
HP was also expected to release a Windows RT tablet based on OMAP chips from TI. When it announced it would be sticking with Windows 8, the company cited customer interest rather than missing chips, but it would certainly be hard to get customers interested in a tablet you can't build because you haven't got the chips for it.
It's worth noting that although RIM stuck with OMAP as the processor for the LTE version of the PlayBook, it didn't use TI's LTE modem. Instead it has a Qualcomm modem, and we're expecting the BlackBerry 10 handsets that arrive next January to use Qualcomm S4 Snapdragon chipsets with LTE rather than anything from TI.
Poor BlackBerry sales
One of the things that RIM CEO Thorsten Heins blames for poor BlackBerry sales over the past year isn't users getting fed up with the interface — it's the lack of LTE handsets. In 2011, the then co-CEO Mike Lazaridis said he was waiting for a promised ARM chip that had a "highly-integrated, dual-core LTE platform" for BlackBerry 10 handsets.
The dilatory supplier was never named, but like the PlayBook, previous BlackBerry handsets have used OMAP chips from TI. Again, while RIM has plenty of problems it can't blame on suppliers, switching from OMAP to Snapdragon would certainly have given RIM yet more work to do to get its new OS up and running.
So I was struck when I noticed recently that TI is talking up ARM-based servers for the datacentre, launching new KeyStone chips with dual and quad-core Cortex-A15 packages and the option of multiple DSPs.
I'm still unconvinced about the value of ARM for datacentre loads, despite the interesting suggestion of using KeyStone to offload suitable tasks from a more traditional server.
TI certainly has an impressive list of design wins for KeyStone with familiar names such as embedded systems company Wind River. But if I were a server vendor considering ARM chips from TI, I would be acutely aware of the problems it has had providing others with long-term component supplies.
I would also know that these KeyStone chips only use the older 32-bit ARM v7 instruction set rather than ARM's newer 64-bit architecture. Consequently, at the very least I'd want some hefty penalty clauses in the contract.