Breaking News
News & Analysis

Samueli Sees Life After Moore's Law

Long life for 28 nm node
12/12/2013 07:50 AM EST
14 comments
< Previous Page 2 / 2
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
TarraTarra!
User Rank
Author
Re: Dealing with the slowdown
TarraTarra!   12/15/2013 6:02:01 AM
NO RATINGS
@rick

"In the future they may not be able to charge the premiumjs for 28nm, 16nm they charge today, but then they will not have the enormous expenditures of new multibillion dollar fabs every 18 monthys either. The biz dynamics will shift radically.

Imagine what that might  be like!"

 

Yes - what egalitarian dreams! I can only imagine the large swaths of un-washed masses storming the bastile. Cake (FinFet) for everyone! 

Susan Rambo
User Rank
Author
Re: Moore's Law
Susan Rambo   12/13/2013 2:34:48 PM
NO RATINGS
Now people are fighting over who predicted Moore's Law would end? It would be good to have a timeline of predictions, just for fun.

zewde yeraswork
User Rank
Author
Moore's Law
zewde yeraswork   12/13/2013 1:29:54 PM
NO RATINGS
I just spoke with a technologist at a leading OEM and server design company who says Moore's Law ended about ten years and claims to have predicted the end about twelve years ago...whether or not that's true, it seems everyone is finally accepting that the days of packing more transistors and getting a denser and denser product are nearing an end--albeit slowly....

moloned
User Rank
Author
Re: Life After Moore's Law
moloned   12/13/2013 7:47:19 AM
NO RATINGS
3D is interesting but even TSVs are too expensive for many applications

Most volume applications use die-stacking in package and wire-bonding for stacked DDR

Going 3D rather than TSV or wire-bonding means you have a huge heat problem and this may prove as much of a limitation as the number of Si atoms in the channel

Samueli who is a brilliant engineer is quite correct that 28nm will be the most cost effective node and few applications will be able to justify going below

For those that do the volumes will have to be huge and the platforms will be programmable so essentially you'll be talking about a commodity SoC business a bit like DRAM is today

Bill Dally's work on ELM at Stanford and now on Echelon at Nvidia charts how this will be possible and programmable solutions will get to between 1.5 and 3x ASIC

The big issue will be how to program these chips efficiently to approach these limits

As proebsting's law notes you can't rely on compilers to deliver the benefits

So SW engineers will have to work for a living instead of relying on Moore's law for a free ride

chipmonk0
User Rank
Author
Re: Dealing with the slowdown
chipmonk0   12/13/2013 12:16:17 AM
as the R&D cost incurred by the leading Foundry to keep up with Intel shoots up, that cost is bound to get xferred to their Fabless customers, cutting into their huge margin up to now. this is what I had meant as "the free lunch is over for the fabless wonders"

Or_Bach
User Rank
Author
Life After Moore's Law
Or_Bach   12/12/2013 6:10:36 PM
NO RATINGS
1 saves
Yes, it is clear thar cost per transistor is not going down with dimension scaling, but new type of scaling - monolithic 3D - would keep Moore's Law in the near future. Samsung is already doing so with 3D NAND and other are moving toward monolithic 3D just as well. The first adaption of monolithic 3D is in the memory segment, and other segment will follow, as we just recently learned whith Qualcomm sign up with CEA Leti. In a recent Blog we articulated why Scaling makes monolithic 3D IC practical http://electroiq.com/blog/2013/10/scaling-makes-monolithic-3d-ic-practical/, and in our site we present the cost and othe benifits the monolithic 3d technology provides - http://www.monolithic3d.com/3d-ic-edge1.html

rick merritt
User Rank
Author
Re: Dealing with the slowdown
rick merritt   12/12/2013 5:55:14 PM
@Tarra: Fabs like TSMC are just starting to ramp up really expensive (profitable) 3-D stacking processes.

In the future they may not be able to charge the premiumjs for 28nm, 16nm they charge today, but then they will not have the enormous expenditures of new multibillion dollar fabs every 18 monthys either. The biz dynamics will shift radically.

Imagine what that might  be like!

junko.yoshida
User Rank
Author
Re: Dealing with the slowdown
junko.yoshida   12/12/2013 3:41:50 PM
NO RATINGS
@TarraTarra! I think you're right. It will be problematic for foundries who need new nodes to keep differentiating their services and charge premiums..

TarraTarra!
User Rank
Author
Re: design
TarraTarra!   12/12/2013 3:05:42 PM
NO RATINGS
This does level the playing field a bit. Design innovation is where companies will have to distinguish themselves. The best, most efficient design shall survive rather than the design on the latest process node.

TarraTarra!
User Rank
Author
Re: Dealing with the slowdown
TarraTarra!   12/12/2013 3:02:45 PM
NO RATINGS
Isnt it more of a problem for the companies with fabs? Process node advantage will not provide the benefits as before and the cost of sustaining new fabs increasing.

Page 1 / 2   >   >>
Most Recent Comments
michigan0
 
SteveHarris0
 
realjjj
 
SteveHarris0
 
SteveHarris0
 
VicVat
 
Les_Slater
 
SSDWEM
 
witeken
Most Recent Messages
9/25/2016
4:48:30 PM
michigan0 Sang Kim First, 28nm bulk is in volume manufacturing for several years by the major semiconductor companies but not 28nm FDSOI today yet. Why not? Simply because unlike 28nm bulk the LDD(Lightly Doped Drain) to minimize hot carrier generation can't be implemented in 28nm FDSOI. Furthermore, hot carrier reliability becomes worse with scaling, That is the major reason why 28nm FDSOI is not manufacturable today and will not be. Second, how can you suppress the leakage currents from such ultra short 7nm due to the short channel effects? How thin SOI thickness is required to prevent punch-through of un-dopped 7nm FDSOI? Possibly less than 4nm. Depositing such an ultra thin film less then 4nm filum uniformly and reliably over 12" wafers at the manufacturing line is extremely difficult or not even manufacturable. If not manufacturable, the 7nm FDSOI debate is over!Third, what happens when hot carriers are generated near the drain at normal operation of 7nm FDSOI? Electrons go to the positively biased drain with no harm but where the holes to go? The holes can't go to the substrate because of the thin BOX layer. Some holes may become trapped at the BOX layer causing Vt shift. However, the vast majority of holes drift through the the un-dopped SOI channel toward the N+Source,...

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed