x-ray lithography

[e-drood]

http://www.tomshardware.com/news/ib...smc-samsung-semiconductor-new-york,13561.html

although enhanced conventional lithography can produce commercial wafer/dice yields at 12nm, there is compelling need to transit to x-ray lithography in foundries --- this study/development center will fleshout commercial yield limits, applied to 450mm wafer production, of fully enhanced conventional lithography or will develop protocols/practices of initial x-ray lithography applied to large scale wafer fab having high yields

3.6B is about right & the heavy hitters are jointly developing this tech --- I ask "Where are PRC reps?"

past time if Moore's Law is to remain applicable...

 

[jhu]

What happens when (if?) we get to atomic scales? Then we're stuck.

 

[Idontcare]

What happens when (if?) we get to atomic scales? Then we're stuck.

Then we use different atoms (whole periodic table of elements for new compounds to be created).

After that we will have to resort to using time-dilation and the theory of relativity to our advantage, slowing down our computers physical velocity relative to ours, such that it appears to be completing computations in less and less time in our reference field.

Thankfully this phenomenon is infinitely scalable, no limit, as we need merely accelerate ourselves closer and closer to the speed of light while keeping our compute resources stationary wrt our reference field.

Its fool-proof I tell you, absolutely fool-proof.

 

[Soccerman06]

Then we use different atoms (whole periodic table of elements for new compounds to be created).

After that we will have to resort to using time-dilation and the theory of relativity to our advantage, slowing down our computers physical velocity relative to ours, such that it appears to be completing computations in less and less time in our reference field.

Thankfully this phenomenon is infinitely scalable, no limit, as we need merely accelerate ourselves closer and closer to the speed of light while keeping our compute resources stationary wrt our reference field.

Its fool-proof I tell you, absolutely fool-proof.

Real man of genius here folks

 

[Voo]

Then we use different atoms (whole periodic table of elements for new compounds to be created).

After that we will have to resort to using time-dilation and the theory of relativity to our advantage, slowing down our computers physical velocity relative to ours, such that it appears to be completing computations in less and less time in our reference field.

Thankfully this phenomenon is infinitely scalable, no limit, as we need merely accelerate ourselves closer and closer to the speed of light while keeping our compute resources stationary wrt our reference field.

Its fool-proof I tell you, absolutely fool-proof.

But I won't have to pay more than at most 200 bucks for one of those chips, yes?

 

[e-drood]

jhu, socman06 & voo

the production of cpu's & gpu's & certain other specialized u-lsi chip arrays use manufacturing techniques functioning at the atomic level of scale, ie: deposition layering presently averages 25 atoms thick (which includes electron reserviors & provison for registration tolerance)- and there are several layers on silicon wafer to create cpu or gpu physical structure...

your concern about cost is well founded - cost (individual cpu dice) will decrease when larger diameter silicon wafer yields larger numbers of validated functional cpu's (or gpu's) - the present 300mm (say 11 7/8inch) wafer does not provide sufficient validated (as example) cpu's to cause meaningful retail price reduction - in fact if you look at intel retail offerings there is smaller cluster of higher performance cpu's and larger cluster of lower medium to low end cpu's --- when the expectation has been better yields of higher end product

wafer yield implies the actual "sweet spot" consistantly produced by the fab (foundry) --- say you are looking down at circle 300mm (11 7/8 inch) diameter and circle 450mm (17 11/16 inch) diameter the "sweet spot" is the smaller center circle and will yield highly accurate multilayer image registration & multilayer thickness accuracy - the home of the very good cpu's when we move to the 1st inner ring, again good cpu's (when wafer is diced) continuing to move outward, the accuracy of registration and thickness begins to degrade and cpu performance becomes measurably different - still quite usable but will not be as fast & will be slightly hotter during operation

the outer bands or rings yield low end cpu's

it's parallax & beam dispersion - smaller silicon wafer, the lower the beam parallax, having small saleable quantities of cpu's with high market prices

with larger silicon wafers scale-of-economy results in lower unit cost BUT parallax significantly increases resulting in yield reduction

more accurate mask registration & layer thickness control is v expensive to achieve - consistant 450mm silicon (w/ proprietary doping) is v much an art heavily supported by cutting edge science

conventional fab technique v good at 32nm & reasonably good at 24nm but there is great market pressure to further increase yields reducing component cost - you the buyer & user are demanding this...

x-ray lithography supports high accuracy mask registration & thickness control - later layer thickness reduced to 5-10 atoms thick & hopefully 6nm with greater complexity...

idontcare i agree w/ you, we all have enjoyed dr. who...

 

[bryanW1995]

Idc has a good idea, but it would be a real pita to keep accelerating ourselves closer and closer to c. Why not just tether our computers to a neutron star at a distance of , say , 1 parsec. Then the angular momentum would be low enough to allow the tether to be made out of carbon nanotubes, and we could remain in the solar system of our choice while the computers do all the moving.

 

[e-drood]

first, this cooperative development center is all about increasing fab yields as viable commercial yield levels for BOTH ibm & intel have been less than stellar (neutron star indeed) - reduced wattage + significant complexity level increases are absolutely mandated - the "usual" blackarts tinkering have not worked...

second, considering the total numbers viewing + registered on anand forums, I will wager my own money that 1-2 end up with tech positions in this dev center...

I'm thinking "Century Rain"... distributed intelligence(s) & consequences...

 

[jhu]

Then we use different atoms (whole periodic table of elements for new compounds to be created).

After that we will have to resort to using time-dilation and the theory of relativity to our advantage, slowing down our computers physical velocity relative to ours, such that it appears to be completing computations in less and less time in our reference field.

Thankfully this phenomenon is infinitely scalable, no limit, as we need merely accelerate ourselves closer and closer to the speed of light while keeping our compute resources stationary wrt our reference field.

Its fool-proof I tell you, absolutely fool-proof.

Well, clearly if we can make time dilation work to our advantage, even a lowly single-core 80386 will seem faster than any Core i7 out there. You may be on to something there...

 

[e-drood]

http://www.anandtech.com/show/4894/amd-conforms-32nm-yield-issues-at-global-foundries

it is not "yield issues" - it's real serious production failure

let's see - global foundries (dresden / amd) and samsung and tsmc just happen to be the three (3) participants who only have observer status until each ponies up the full individual partner investment share

samsung is chaebol + major player + good yields @ 32nm and wants to stay current - possibly cross-licensing some of it's own proprietary tech

what about tsmc? has been struggling for some time & ibm + nippontypes have given tech assistance in hope of continued survival and wow tsmc is also sf's foundry & tsmc has yield issues & sf is experiencing continuing user issues w/ controllers from tsmc

lets hope tsmc can pony up all the bucks to use new fab tech... this world is connected in everyway you can imagine and more...