backtop


Print 40 comment(s) - last by Hector2.. on Jan 5 at 2:57 AM

Qualcomm and other Samsung clients could get a boost from novel power-saving gate design

At the Intel Develop Forum, the world's largest maker of traditional PC central processing units and server chips, Intel Corp. (INTC) is fond of expressing how it's crushing its rivals in process technology.  But by the looks of it the rest of the pack may not be as far behind as Intel would like to have you believe.

On Friday, Samsung Electronics Comp., Ltd. (KSC:005930announced that it had successfully taped out a 14 nm FinFET transistor chip.  For those who don't avidly follow the CPU design industry, most traditional transistors are flat multi-layer designs (think a sandwich).  By contrast the FinFET uses a tall wall-like gate, which towers over the nanoscopic surface like a fin.  The novel 3D design allows leakage to be significantly reduced -- a key source of loss in reliability and power at sub 40-nm designs.

Intel was the first player to get FinFETs, with its 22 nm Ivy Bridgelaunched earlier this year.  It plans to roll over the technology into its mobile offerings next year.  But by the sound of it Samsung won't be far behind.

Samsung worked with a chip-optimizing firm called Synopsys to perfect the difficult process of producing the delicate FinFETs at such a small node.

MOSFET FinFET
A MOSFET FinFET [Image Source: Brews Ohare/Wikipedia]

Samsung LSI vice president Dr. Kyu-Myung Choi cheers, "FinFET transistors can deliver lower power consumption and higher device performance, but they also bring tough challenge.  We chose Synopsys as our FinFET collaboration partner to solve these challenges, because of our successful history together at 20 nanometer and other nodes. We continue to pool our expertise to deliver innovative FinFET solutions."

Synopsys's expertise helped Samsung characterize the prototype chips and to remove so-called "parasitic" impurities.

The chipmaking duo did not announce a time frame for commercial rollout.  But when Samsung does achieve commercial mass-production of 14 nm FinFETs, one interested party will surely be Qualcomm, Inc. (QCOM) whose SnapDragon 4 system-on-a-chip (SoC) design Samsung recently began producing at its monolithic Texas chipmaking facilities.

Source: Samsung [press release]



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Intel taped-out 14nm six months ago...
By CharonPDX on 12/21/2012 2:09:41 PM , Rating: 3
Although this is impressive, only a 6-month lag. Intel often has a MUCH longer lead on new processes. And had functional CPUs on 14nm at least in September - this looks like Samsung has just taped out the process itself, and haven't gotten a functional chip on it yet. (Although that could be fast behind.)




RE: Intel taped-out 14nm six months ago...
By ltcommanderdata on 12/21/2012 2:18:45 PM , Rating: 2
Samsung's currently shipping chips on their 32nm process. 28nm chips come in 2013. After that is their 20nm process. As such, I really don't expect any Samsung 14nm chips to be in consumer hands until at least 2015. Intel should be shipping 14nm chips in 2014. The 1 year gap between Intel and the rest of the industry is pretty standard and it doesn't really seem to be closing despite what the article says.


RE: Intel taped-out 14nm six months ago...
By Master Kenobi (blog) on 12/21/2012 4:30:42 PM , Rating: 2
That is largely because Intel's R&D budget is massive. Their microarchitecture is always pretty good but they back it up with always having the best process in the industry.


RE: Intel taped-out 14nm six months ago...
By dark matter on 12/22/2012 9:46:30 AM , Rating: 1
Samsung isn't no small fry.

And I think IBM wants a word with you about your last statement.


RE: Intel taped-out 14nm six months ago...
By Samus on 12/23/2012 2:11:51 AM , Rating: 4
I agree, IBM is far ahead of anybody in manufacturing concepts (carbon nanotubes, anyone?) but when talking scale, Intel's manufacturing process matures very quickly and they produce TONS of wafers at very high yields.

I also wouldn't call Intel R&D lax. I still have a Core i7-950 that's almost 5 years old, and is still faster than 95% of the computers on the market. 5 years old and faster than 95%. I wouldn't call that bad R&D.


RE: Intel taped-out 14nm six months ago...
By ShieTar on 12/24/2012 5:44:32 AM , Rating: 1
Your own research is a little questionable though. How do you figure that less than 5% of computers on the market include a Ivy Bridge or Sandy Bridge Quad-Core; nevermind any Intel Hexa-Cores? And hey, even AMDs FX-8350 will be faster then the i7-950 in almost every situation.

So what, did you include all the netbooks and tablets in your 95%-Analysis?


By maxxcool on 12/26/2012 2:28:46 PM , Rating: 2
Please... my 4ghz i-750 would bend any amd fake 8 core cpu over a barrel.


By NellyFromMA on 12/26/2012 1:23:10 PM , Rating: 2
Double negative, nooooooooooooooooooooooooooo! lol


By maxxcool on 12/26/2012 2:32:48 PM , Rating: 2
IBM really ? the people who could not make a simple G5 run at 3ghz reliably...


RE: Intel taped-out 14nm six months ago...
By Shadowself on 12/21/2012 4:36:28 PM , Rating: 5
Yes, there is a huge difference in taping out a chip and mass producing it. In one case it is a first cut at the completed design (tape out) and the other is actually mass producing devices in quantities of millions. This article seems to make it feel like going from one stage to the other is trivial. It is not. (The perfect example is the Motorola G5 chip. It taped out then took many, many months to get from test batches to pre production. When the pre production chips were tested in the field errors were found that would have caused a vast chip redesign -- some of which were process based -- so Motorola killed it. For the past decade or more for CPUs/GPUs process and architecture are closely integrated -- hence Intel's tic-toc approach to try to deal with this by decoupling it as much as possible.)

Additionally, the article does not say the number of gates or transistors in the design. Was this design just for a 1,000 transistor proof of concept? If so, then don't expect to see 14 nm FinFET CPUs from Samsung for a few more years!


By Gondor on 12/23/2012 4:45:00 PM , Rating: 4
I believe I saw other tech news sites state that Samsung has used Cortex-A7 as its test vehicle and has produced a functional chip. It's nowhere near as complex as current top of the line x86 CPUs but it's no 1000 transistor chip either.


By SAnderson on 12/24/2012 12:42:57 PM , Rating: 2
Yeah, time between tapout and qual is going to be 1-2 years. Tapout, eng samples, etc, etc and finally qual before any of this gets sold to Tier1 customers.


By Jeffk464 on 12/21/2012 6:57:12 PM , Rating: 2
Yes, but samsung will be competing in the ARM market so being way ahead on the 14nm should be a big boost for them. Samsung hasn't directly competed with intel as of yet.


"DailyTech is the best kept secret on the Internet." -- Larry Barber














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki