Silicon Alternatives

[Maxima1]

So how much of a jump is this going likely to be? I read that Intel was looking to switch at 7-10nm.

http://www.pcworld.com/article/2033...are-pushing-pcs-to-blistering-new-levels.html

"OPEL only recently exited the R&D stage and hasn't tried to make itty-bitty transistors at Ivy Bridge's 20nm size, but the company claims that at 800nm, gallium arsenide processors are faster than today's silicon and use roughly half as much voltage.

"If you wanted to match the speed of today's silicon processors, at roughly a 3GHz clock rate, you wouldn't have to go all the way down to 20 or 30 nanometers," says OPEL chief scientist Dr. Geoffrey Taylor. "Heck, you could probably hit that at 200nm." And that's using planar technology, not 3D transistors."

 

[jhu]

I'll believe it when they have a mass produced product.

 

[lehtv]

"[...] the company claims that at 800nm, gallium arsenide processors are faster than today's silicon and use roughly half as much voltage."

On what metric? It has to be something very theoretical and irrelevant to actual processor performance. It just doesn't mean anything.

 

[jdubs03]

From what I have read and my estimates 10nm could be the point in which Intel uses a III-V material (InGaAs, InSb, etc.) in conjunction with silicon-germanium or just germanium. I linked a good article that shows the potential routes for future materials and transistor structures. The graphics are from Applied Materials so that should be a good indicator of where the future lies.

I don't think 14nm for Intel will have a change to the manufacturing process, just a 2nd-gen FinFet, the new materials should enable much higher performing transistors (higher drive current) at lower voltages.

http://www.extremetech.com/computing/162376-7nm-5nm-3nm-the-new-materials-and-transistors-that-will-take-us-to-the-limits-of-moores-law

 

[Homeles]

I'll believe it when they have a mass produced product.

From what I can tell, SiGe or Ge is pretty much a done deal and should land before the end of the decade.

From what I have read and my estimates 10nm could be the point in which Intel uses a III-V material (InGaAs, InSb, etc.) in conjunction with silicon-germanium or just germanium.

They've only just been able to to fabricate III-V materials on a silicon wafer:
http://phys.org/news/2013-11-imec-world-iii-v-finfet-devices.html

10nm seems too early. But anyway, there have been some serious breakthroughs in the past few years, and it's just a matter of time before we see the industry move past silicon.

 

[jhu]

From what I can tell, SiGe or Ge is pretty much a done deal and should land before the end of the decade.

Actually SiGe is being used now and has been for a while (for varying definitions of "mass production").

 

[jdubs03]

10nm does seem rather early, but thats why i I said could Based on the articles out there about finally getting InGaAs on a Si wafer, it seemed pretty positive. It could go out to 7nm, but then it may be GAA or tFets, as it isn't certain that FinFets can scale below 7nm.

Regardless I'm excited to see what happens, especially in seeing the magnitude of improvements of new materials with another change in structure.

If anyone can make it happen its Intel, they'll probably have the same lead time in announcing and manufacturing as they did with FinFets.

 

[Xpage]

so if you use larger size transistors the new designs would make HUGE chips, guess it might be time for AMD to brush off the K8 architecture again though the P4 does go higher in ghz we can relive 2003-2005 again!!!! (obviously joking).

 

[ShintaiDK]

The key is not what is better than silicon. The key is what can be mass produced cheaply.

 

[NTMBK]

Bacon is better than silicon.

 

[podspi]

so if oyu use larger size transistors the new designs would make HUGE chips, guess it might eb time for AMD to brush off the K8 architecture again though the P4 does go higher in ghz we can relive 2003-2005 again!!!! (obviously joking).

Yeah... larger transistors means a lower transistor budget.

If we were actually fabbing things at 200nm again expect a resurgence of speedracer designs :biggrin:

 

[jhu]

Bacon is better than silicon.

/thread

 

[Homeles]

Actually SiGe is being used now and has been for a while (for varying definitions of "mass production").

SiGe is used for straining in the source and drain, and the concentration of Ge has been increasing with every node.

I'm talking about SiGe and Ge as a channel material. Currently, fins/channels are made of silicon.

 

[Phynaz]

Would you believe Tin?

http://www.eetimes.com/document.asp?doc_id=1320283

 

[Homeles]

InGaAs is everywhere at IEDM this year, goodness:
http://www.his.com/~iedm/program/13advprg.pdf

Intel's been playing with TFETs:

The Tunneling Field Effect Transistor (TFET) is of interest for future low-power technologies due to its steep subthreshold-slope (SS). In addition to understanding TFET prospects for future technology nodes, we also need to assess if it enables continued scaling required for increasing transistor density. GaSb/InAs heterojunction TFET (Het-j TFET) is one of the leading TFET options due to its high drive-current. In this paper, double-gate (DG) and nanowire (NW) Het-j TFET is atomisticly modeled and compared to MOSFET down to Lg=9nm, i.e. ITRS 2022 node. To achieve TFET characteristics superior to MOSFET, DG body has to be extremely thin, so NW TFET is preferred due to more relaxed thickness and better characteristics. Scaling roadmap shows Het-j TFET requires a 3nm-NW at Lg=9nm with an Idsat advantage over MOSFET for Vdd<0.4V. A new device &#8211; the Resonant-TFET (R-TFET), is proposed, with average SS~25mV/dec and 100x Idsat advantage over MOSFET at Vdd=0.27V, thus enabling the scaling of tunneling transistors to sub-9nm gate-lengths.

And another:

Reducing supply voltage (Vdd) while keeping leakage current low is critical for minimizing energy consumption and improving mobile device battery life. A Tunneling Field Effect Transistor (TFET) is not limited by subthreshold thermal tail and may perform better than CMOS at low Vdd. In this paper, a leading N-TFET option -GaSb/InAs heterojunction- is atomistically modeled and circuit simulation models are developed. Het-j TFET logic with symmetric N- and P-TFET transistors is expected to have 64% energy savings compared to CMOS for low-power applications at Lg=13nm (i.e. 2018 ITRS node). Both MOSFET and TFET device variations are dominated by work-function variation and energy savings are slightly lowered when variations are considered. Without a good pull-up transistor option, energy efficiency is reduced greatly, highlighting the need for exploration of a steeper SS P-TFET to realize significantly superior power efficiency from TFET logic circuits.

Would you believe Tin?

http://www.eetimes.com/document.asp?doc_id=1320283

That's incredibly interesting.

 

[Borealis7]

i read today that the cost per 1 million transistors, which the writer estimated to be around 3 cents, is not expected to drop in the transition between 22nm and 14nm (or was it 28nm to 20nm?) which would make 22nm the "cheapest ever node in history" since he estimates the cost will only rise in the future (basically another guy forecasting the death of Moore's law), and that's not including R&D costs of prototype chips for each new node and the cost of producing the lithography masks which skyrocketed by 100% from 32nm to 22nm.
the article was about the mass layoffs in the semiconductor industry around the world, and trying to figure out some of the financial reasons why these companies are struggling.

 

[Ajay]

InGaAs is everywhere at IEDM this year, goodness:

That's incredibly interesting.

Thanks for the link. Stanene - well, at least Stanford's not too up on itself

Hmm, I wonder how much current they will be able to drive along the superconducting edges? Getting rid of parasitic wire delay/resistance in CPUs would be huge!

 

[Ajay]

i read today that the cost per 1 million transistors, which the writer estimated to be around 3 cents, is not expected to drop in the transition between 22nm and 14nm (or was it 28nm to 20nm?) which would make 22nm the "cheapest ever node in history" since he estimates the cost will only rise in the future (basically another guy forecasting the death of Moore's law), and that's not including R&D costs of prototype chips for each new node and the cost of producing the lithography masks which skyrocketed by 100% from 32nm to 22nm.
the article was about the mass layoffs in the semiconductor industry around the world, and trying to figure out some of the financial reasons why these companies are struggling.

Well, the decline of the PC is shaking the tree on all those who lost marketshare - even of peripheral makers. Smart phones just don't generate as much a need for design and fabs are consolidating - not good for employment in the semicon market I'd think.

 

[schmuckley]

Graphene :awe:

http://gizmodo.com/scientists-just-figured-out-how-to-make-lightning-fast-1177727488

 

[Morbus]

Graphene is probably where it's up, but my ignorant mind thinks the future of processors is increasing their size while decreasing their temperature. Flexible CPUs, like a brain, all rolled up in a ball, submerged in liquid. Like a brain...

If it's doable, we'd only need to allow CPUs to dynamically rearrange their logical optimizations and we'd have real-life CPU-only machines (like a brain) that had RAM, data storage, processing, everything in the same CPU... How awesome is that?!

 

[videogames101]

There are many novel materials and structures that enable scaling beyond 10nm, however there are not many which enable that scaling to be economically viable for mass production.

 

[Maxima1]

On what metric? It has to be something very theoretical and irrelevant to actual processor performance. It just doesn't mean anything.

I don't know, but the material in the article has about 6 times more electron mobility. There's factors that end up reducing the difference, but they're not equals.

Look at how long we have had crappy resolution screens, but next year we'll start to see a drastic increase resolutions because of IGZO displays.

 

[Imouto]

Graphene :awe:

http://gizmodo.com/scientists-just-figured-out-how-to-make-lightning-fast-1177727488

I think graphene won't replace silicon for a while but will be included in CPUs as an extra cooling feature in the near future.

 

[monkeydelmagico]

Bacon is better than silicon.

Saline is better than silicon. Feels much more natural.

 

[Homeles]

Saline is better than silicone. Feels much more natural.

FTFY