Electronics Weekly Electronics Design & Components Tech News
Is the ARM – Globalfoundries tie-up significant? Some people like the Wall Street Journal this morning, seem to think so.
The proposition is that a 28nm low-power process in the second half of the year at Globalfoundries will deliver Cortex-A9-based mobile phone chip-sets which will bring computing-like performance to mobile phones and head off Intel’s moves, based on 32nm process technology, into the mobile phone business.
Well, like so many theoretical scenarios in the tech business, it’s possible.
But is it likely?
Intel has had 32nm processors out for almost half a year. By the second half of the year, Intel’s 32nm process will have had a year of volume production behind it.
That’s a tidy lead on Globalfoundries which says it will begin production in the second half – which could mean December.
And in getting a new generation of process up and running, there’s many a slip t’wixt cup and lip.
A year into a new ramp delivers substantial improvements in yield, and therefore cost, so Globalfoundries will be up against it trying to produce mobile phone chip-sets as cheaply as Intel.
OK, so Globalfoundries is going to be on 28nm in Q2, while Intel is still, supposedly, on 32nm.
But surely Intel will be working on the half node process – which 28nm was assumed to be before TSMC announced that 28nm it would be supported as fully as if it were a full node process.
It would be naive to assume that Intel will stick around on 32nm waiting for Globalfoundries to overtake it with its 28nm process.
So, although it seems a nice theoretical proposition that ARM-Globalfoundries can match Intel on process this year, or even next year, it seems unlikely.
TOMORROW MORNING; THE TEN BIGGEST NEWS STORIES IN 2009
Well, Mike, I think there’s no doubt about Intel’s commitment to winning in the smartphone market, but I do wonder about Nokia’s commitment to letting x86 into the mobile phone market. Nokia’s old boss Jorma Ollila made public statements about keeping Wintel out of the mobile Internet. Whether any of this attitude still remains in Nokia’s thinking, I do not know. But somewhere, I would imagine, there’s a residual fear of letting the Americans do to the mobile phone industry what they did to the PC industry.
I think Intel do want to win the smart phone market.
“Intel and Nokia merge software to create MeeGo”
http://news.bbc.co.uk/go/em/fr/-/1/hi/technology/8516368.stm
But whether they have the tools is another matter of course.
Netbooks are one thing (for which Atom is just fine and dandy), smart mobiles are another thing entirely.
With netbooks there are many things apart from the CPU which take a lot of power (big LCD display, HDD/SSD, PC motherboard chipset and components). Replacing a 5W CPU with a 1W-2W CPU doesn’t make a dramatic difference when the rest of the components take 5W-10W and you have a 60WH battery, so the case for ARM in netbooks isn’t so compelling — especially given people’s attachment to Windows because “it’s a computer”…
In a smartphone where the total power is only a fraction that of a laptop the CPU power has a much bigger effect — and especially in standby or power-saving mode, since people expect days of standby time, as well as many hours of active time (more than a netbook).
Also Windows (any flavour!) is often seen as an impediment not an advantage, even Windows Mobile is not well suited to such devices since it’s a cut-down version of Windows in the same way that Atome is cut-down x86 CPU — it spends a lot of time and power doing background housekeeping compared to a built-from-scratch mobile-optimised OS.
So if people really do want mobile-computing-like performance and software (internet etc) in something very portable with phone-like battery life, this is much easier starting with an ARM and proper mobile OS than an Atom and x86 OS like Windows.
Remember also that ARM is never a standalone CPU, it’s always SOC with all the low-power peripherals needed to build a handset. Atom can’t really address this market yet, and having peripherals off-chip automatically increases power consumption.
The other point is that I suspect Intel don’t really *want* to win in the smart phone market, given that the margins on these ARM-SOC are *much* smaller than on standalone Atom CPUs 🙂
I hope you’re right Ian, but Intel have been doing a pretty good job on improving power efficiency lately, and I expect they’re putting enormous effort into it, and that the effort will continue. We were told by many vendors of ARM-based netbook chip-sets that netbooks, or smartbooks, would appear in Q4 and they have yet to appear. The Atom-based netbooks have pretty good battery life – some around eight hours. I really hope ARM can do a substantially better job, and get into some big volume netbook runners, and start to gain traction in the computer market, but I don’t think the Globalfoundries’ deal, of itself, will get them there.
Surely the issue here is not the raw silicon technology, but the power efficiency of the CPU design?
The difference between 32nm and 28nm in power is very small — indeed, 28nm is not even that much lower power than 40nm, because voltage scaling and leakage have hit a brick wall.
Chip size is smaller in 32/28nm but cost is not necessarily that much lower since the cost per mm2 is considerably higher. Intel don’t really care about cost for their big CPUs since the margins are huge and so are the chips, and 32/28nm will have higher yield — this isn’t the case for a mobile CPU, the chip area is so small that yield will be very high in any process.
The difference in power efficiency (MIPS/W) between Atom (and its derivatives) and Cortex-A9 is much bigger than any process difference, especially when standby power is taken into account.
ARM cores (including Cortex) have been designed from day one for power and area efficiency, whereas Intel is trying to squeeze down the power of their x86 CPUs which were designed for maximum performance — regardless of what Intel PR says, this is a huge hill to climb, and ARM have been concentrating on low power for many years.
Basically, x86 is a lousy architecture for low power applications — it’s like Windows Mobile, it’s just not the right tool for the job.