S/A: Atom takes the process lead on 14nm

[cbn]

http://semiaccurate.com/2012/04/23/atom-takes-the-process-lead-on-14nm/

During their Q1 conference call, Intel stated that Atom was going to move faster than Moore's law. This will happen not by magic, but by less hamstringing. Intel has traditionally treated Atom like an unwanted houseguest …

Click the link for the rest of the article. (S/A has blocked copy and paste)

According to the article Intel has decided to launch the 14nm atom before the 14nm mainstream CPUs.

Anandtech forum members have pointed out this wasn't done in the past due to the low profit margins of atom.

So what future plans does Intel have that will allow these chips to be manufactured on the most advanced process node?

Maybe Intel needs Atom on the best process to help sell other packaged services?

 

[tynopik]

i was wondering how long it would take them to get serious about mobile

 

[khon]

This is a very logical move from Intel.

From a consumer perspective the main benifit of going to a smaller node is reduced power consumption, and that is obviously far more important for mobile devices, thus Atom will benifit more than Core will.

Also Intel is trying hard to move into the growing market for smartphones and tablets, and for that they need to improve Atom. For desktops and laptops on the other hand, they already have a significant lead over AMD, so they can afford to wait a bit, and let Atom go first.

 

[Mr. Pedantic]

During their Q1 conference call, Intel stated that Atom was going to move faster than Moore's law. This will happen not by magic, but by less hamstringing.

Intel has traditionally treated Atom like an unwanted houseguest, leaving it on 40nm until late in 2011, and removing features with reckless abandon. The lack of attention showed, Atom was about as beloved as uninvited relatives asking for money, and sales reflected this.

Late in 2011, the chips finally were shrunk to the 32nm node just as the mainstream CPUs were going to 22nm. Although the low power process that Atom is fabbed on is distinct from the mainstream CPU process, the nearly three year lag was not warranted for technical reasons. In rectifying this, Atom will indeed progress faster than Moore&'s law, even if Intel's fab technology does not.

With the late 2011 mostly vaporous introduction of 32nm Atoms, the line moved to what is known as a -1 process, basically current generation minus one. As we stated earlier, this is true for the gross geometry, not the exact same process, there are nuance differences. It went from being four years behind the times to a mere two, but that was just the beginning.

Sources tell SemiAccurate that the 22nm Atoms will be announced at CES in 2013, less than a year away. This means that the line is on a current process, and is only lagging mainstream parts by a year. From there, the next step is 14nm, a process that is not due until 2014.

Luckily for Atom, it becomes the lead product on the 14nm node, so you should hear about 14nm Atoms before you hear about 14nm mainstream CPUs. These are set for introduction during CES 2014. In the course of a little over three years, Atom will have gone from a -2 process with a four year lag to the lead vehicle for new nodes. Quite the change, eh?

That's pretty cool. So this will be on the next smartphone I'll be buying?

 

[dkm777]

So, it's the successor to Medfield that will be top priority from now on? And there I was, thinking I'll soon be able to buy a netbook with a 1366x768 screen, adequate CPU and GPU that will last for over 15 hours on battery while at the same time being totally fanless.

 

[ShintaiDK]

Looking at Medfield phone on Anandtech today. And then knowing it will be 22nm next year and 14nm in 2014. It looks really bad for ARM.

Qualcomm told customers to use older chips because they cant deliver enough 28nm for example.

But this isnt news. Maybe for SA...

There is a new 32nm version coming out in Q4 this year too.

 

[T_Yamamoto]

Net books become more powerful sooner or later

 

[VirtualLarry]

Atom is and always has been crap. Even at <1 W.

 

[SetiroN]

Not when compared to ARM.

 

[pelov]

Not when compared to ARM.

But it's far more expensive. Carrying that x86 license means a heftier price tag and people have been complaining about it for ages. Consider that OEMs were asking Intel to drop down the prices of their chips so they can sell more ultrabooks and Intel stuck their collective fingers in their ears and ignored them for months until we finally saw an SB mobile price drop earlier this year. Coincidentally this was also just a couple of months before IB was due out...

The market the Atom is targeted for is another beast entirely and I don't think Intel realizes that just yet. People don't care how fast their phone or tablet is so long as it's cheaper and accompanied by better hardware. Selling an x86 license in that segment means very little to the market if it's price is going to be inflated. This also means competing with ARM licensees who make efficient and cheap chips by the millions.

Unless these new Atoms are going specifically into netbooks/laptops, but that's not exactly a new market for them.

 

[Lonyo]

Leading on a new process with small, simple chips?
Almost sounds like what GPU manufacturers have tended to do when they transition to a new process. Lead with something easy to iron out any issues, before moving onto the big boys.

 

[dealcorn]

22 NM Atom implements Ivy Bridge tick+ graphics. Will Atom 14 NM be used as the introductory vehicle for next generation tock graphics? If so, I would characterize Atom as fully integrated into Intel's technology roadmap.

 

[ShintaiDK]

Leading on a new process with small, simple chips?
Almost sounds like what GPU manufacturers have tended to do when they transition to a new process. Lead with something easy to iron out any issues, before moving onto the big boys.

Atom wont be the first 14nm chip. It will be Broadwell. 14nm Atoms are for Q4 2014. Not Q1/2 2014.

 

[cbn]

Leading on a new process with small, simple chips?
Almost sounds like what GPU manufacturers have tended to do when they transition to a new process. Lead with something easy to iron out any issues, before moving onto the big boys.

Not sure about the current GPUs.....but yes, I remember AMD did that with the HD4770 on TSMC's 40nm.

Intel has also done this before with 32nm by launching the smaller clarkdale chip before the larger die CPUs.

However, for whatever reason (s), the company has left atom on a lagging node.

 

[IntelUser2000]

However, for whatever reason (s), the company has left atom on a lagging node.

They've said though more design sharing will happen among Core and Atom teams in the future so that should change.

Remember that CPU development takes 4-5 years minimum. Decision to develop the next core should have finished in 2008-2009 for a 2013 product like 22nm Silvermont, and 2007-2008 for 2012 product like Medfield. You can't release a CPU that you started working on only say, 2-3 years ago.

The original Atom Silverthorne came in late 2008. That suggests first work started happening in 2003-2004. Back then they might not have thought important to do big core changes too often. Mobile wasn't so important then, and they still had XScale. 2008-2009 seems to be when Intel started considering the phone market really seriously, which is why you see cores coming in 2013.

Another reason I think is due to management changes. Otellini brought the Tick Tock model, where it refreshes CPUs every year, and architecture every other year. He became CEO in 2005. Back in Craig Barrett days were when new designs took 4-5 years to come out. Releasing new designs every 1-2 years require overlapping multiple teams.

 

[pyjujiop]

Atom is and always has been crap. Even at <1 W.

I agree, but in the ultra-mobile segment, everything is crap when you compare it to desktop CPU's. I tried using an Atom-based system once, and wouldn't touch another one with a 10-foot pole. It was sort of like trying to run Vista on an old P4 Willamette with SDRAM.

But Atom does have its uses, and Intel needs a better Atom if it wants to compete in that market segment.

 

[HexiumVII]

Well hopefully Haswell will get us the 8 hours on tablets. Its nice to see intel moving forward while Apple iPad moving backwards in power usage.

 

[Lonbjerg]

I agree, but in the ultra-mobile segment, everything is crap when you compare it to desktop CPU's. I tried using an Atom-based system once, and wouldn't touch another one with a 10-foot pole. It was sort of like trying to run Vista on an old P4 Willamette with SDRAM.

But Atom does have its uses, and Intel needs a better Atom if it wants to compete in that market segment.

I was very suprised over the perfomance of my AT5IONT-I (a gift).
I wouldn't pay for an Atom myself...but I won't toss out my box named "medie luder"...that is my HTPC.

Google translate will tell you the english name...

 

[SetiroN]

But it's far more expensive.

Based on what? You're citing ULV Core processors... a product with 100% market share is hardly a good comparison.

The only Intel powered smartphone is sold for $420. Nobody would enter a new market with uncompetitive prices.

P4 Willamette with SDRAM.

And oh how many of those were sold. :\

 

[khon]

Atom wont be the first 14nm chip. It will be Broadwell. 14nm Atoms are for Q4 2014. Not Q1/2 2014.

Actually Atom will come before Broadwell, that's what Intel stated in their conference call.

 

[blckgrffn]

My crystal ball says that the mobile world is the one that needs to be ready for Intel and MS finally taking it very, very seriously

"seriously" though, I think that Intel flexing it manufacturing might and Microsoft flattening out the app space (Win 8 is going to be close to parity with where Apple has managed to get so far with this, IMHO) between peoples phones, tablets and PCs is going to be disruptive to where some folks had envisioned the market going. That's an only slightly informed opinion, mind you

I view all current Android tablets as hardware prototypes for W8 tablets

MS making W8 on the desktop painful though... that's something that surprised me. I had jokingly talked with coworkers, and we had all agreed that 7 was likely the new XP (in that businesses were moving to it and it would probably "stick" longer than MS wanted) - a year or more ago. It looks like we might have been on to something.

The first Intel phone is fairly impressive. When reading about it, it seemed like some slider W8 variant might make a next nice phone.

 

[Nintendesert]

I always figured Intel's fabs were one of the companies strongest things they could leverage against a competition fight for scraps at the other fabs out there. It's about time Intel actually leverages it against the competition.