MacOSXRumors claims that IBM may bypass 0.09um process PowerPCs and jump straight from 0.13 to 0.06um.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
0.06um PowerPCs?
- Thread starter MacRumors
- Start date
- Sort by reaction score
Longer wait for a PB
I've already waited to get a PB. If things go as this article speculates they might, it would be worth it to wait for a super cool, super fast G5 PB at the end of next year.
Actually, maybe the end of next year is when Apple will roll out the updated 15" Al G4.
I've already waited to get a PB. If things go as this article speculates they might, it would be worth it to wait for a super cool, super fast G5 PB at the end of next year.
Actually, maybe the end of next year is when Apple will roll out the updated 15" Al G4.
Oh, if this were tru...
And the trouble is, we've gotten burned so badly by moto, it's hard to trust another so quickly.
And the trouble is, we've gotten burned so badly by moto, it's hard to trust another so quickly.
Presuming the rumor is in actuallity truth, IBM gould use the .09 hardware for other chips. (If they actually have enough for normal production runs). However, presuming this is true, wouldn't you want to jump to the .06 technology for an even faster processor? This would push IBM into the lead quite possibly in processor speed and capabilities.
If this is true, we would be seeing this technology in Macs next year when I'd be looking at upgrading. I think it's good news if it is true.
Not going to happen. Ever notice that all companies that produce chips use the tried and true chip and downsize it before a new one? The reasoning is to make sure it works properly and it's not the chip. That is a large jump, especially since .09 is available now while .065 is not. The plant that makes the G5 already has .09 hardware for other chips there. It was be easier to move that chip and .09 then .065. IBM would be biting off more then it could chew to pull this one off. Making a batch of bad processors is not cheap; as would the delays it could create. I could see them moving to .09 and then to .065 though.
Intel and IBM are both working on 60nm processes. Of course the design is more advanced than even the 90nm process, but it is possible, and both companies expect to have 60 nm processes up and running next year.
So, no.
I don't work in a chip fab, but I don't think they use a laser (coherent light) for chip fabrication right now. I could be TOTALLY wrong here though...
Making a chip is a LOT like silk screening a t-shirt. You put a reactive chemical on the die and you shine a light through a mask. The light that gets through etches the chip (in the same way that the light in a silk screen sets the dye).
The problem with the current process is that the wave length of the light won't fit through the mask after a certain point. I've heard that .065 would be the limit a while back but apparently .045 is the ultimate limit now (they figured out how to use a few tricks to squeeze some light wave through smaller screens.. coherency and the like).
IBM and AMD are currently working together [at Fishkill] on .065 and .045 micron processes. IBM is currently making chips on a .09 micron process at Fishkill.
I think you'd have to accept that IBM was ahead of schedule for them to skip .09 and go for .065. It's within the realm of possibility but it would be very unusual considering it would be a lot easier to move to .09. I would have to expect that there are some very special circumstances... like they are ahead of schedule on .065 and they are converting large sections of the plant to .065 fab technology.
It would be very cool if IBM could pull it off but it's a huge longshot. I wouldn't bet on it.
There would, however, be other advantages to such a shrink. The die size of the chip would likely become so small that IBM would have trouble packaging it. They would likely make it intentionally bigger to make it easier to package. The easiest way would be to add more L2 cache. I'd expect a .09 micron PPC 970 to ship with 1MB of L2 while a .065 micron 970 would likely ship with 2MB of L2. They could also add more logic, like an on board memory controller (like the Opteron has).
Past .045, it looks like chip designers will have to give up light. The most likely candidate will be x-ray lithography... but it is currently more expensive (and there are other issues with generating x-rays all day.. like radiation).
...but how do they cut the mask pattern? I don't think it's done by some dude with a really steady hand...
Originally posted by idea_hamster
...but how do they cut the mask pattern? I don't think it's done by some dude with a really steady hand...
Fishkill's operations are done entirely by computers (once programmed). Thus you don't need somebody with a steady hand. You just need instruments that are fine tuned enough to do the engraving.
I think you forgot to switch on your irony detector
.Well, duh, but what kind of instruments? Laser?
Ti, Sony and others are as well. Of course Sony doesn't do their own chips, but are working with IBM, who is also working with AMD. All of the top three fabs are working on them. They have to or they will fall behind and lose business to their competitors. Ti has same .09 parts available now.
Late next year at the earliest as .09 is just coming out. I can't see companies buying .09 equipment and doing all the testing for only six months later the .065 process is ready.
I agree with most of what you said. IBM could go to .09 with the 970 and do no changes, they will when it hits .065 though and lower though. If you look at most systems, the CPU area has stayed rather constant even though the processors have shrunk. Sun will have the US IV that will be produced on a 330mm wafer and will occupy the same space as the current chip does. So while the processors have shrunk, their space has not. The other reason why IBM will probably increase the cache, not only to make the chip larger, but because cache will be cheaper, but because it will also increase the performance. I can't see IBM doing many changes to the 970 though, except the cache. If they add more logic, they will most likely use the Power5 core. Adding in logic is not an easy task like more cache is, and to go through all of that only to use the Power5 later would seem like a waste of resources.
Anyone read the cover story in the last Wired? Look for diamond microprocessors in the near future. Intel isn't researching it because it's "Not needed yet" but I'm hoping that IBM is looking into it. IBM does a lot of speculative research from what I've read of the company, so it's not out of the realm of possibility.
Originally posted by crenz
I think you forgot to switch on your irony detector
You mean a SARCASM detector, right?
I think this is probably not true.
The resources they'd need to go to .06 in that time - I just don't think it's possible.
On the other hand, it may be the case that there's not as much benifit in the .09 process as IBM were expecting. I think Intel just found this out - Prescott is going to start at over 100watts?! Where does it go from there?
Maybe IBM have had similar issues, and are just going to plow on to smaller a process.
The resources they'd need to go to .06 in that time - I just don't think it's possible.
On the other hand, it may be the case that there's not as much benifit in the .09 process as IBM were expecting. I think Intel just found this out - Prescott is going to start at over 100watts?! Where does it go from there?
Maybe IBM have had similar issues, and are just going to plow on to smaller a process.
long story
First, the masks are not cut with light (lasers, UV, x-ray, synchrotron radiation, etc.). The masks are cut with electron beams. They have been (for the most critical masks) for many, many years -- back into the 80s for some masks. Electron beam work is extremely precise, but still very expensive compared to photon work. Also resists (the material put on the wafers and exposed to the light shining through the masks) are not as useable for electron beam work as they are for light (photon) work.
The wavelengths of the light used to expose the wafers through the masks on production systems (as different from the research systems) have been getting shorter and shorter for many years. It is now getting into the UV for some systems. Playing games with the way things are focused, taking advantage of edge effects, and such allows the feature size to get down to about 1/4 of the wavelength of the light when things are really pushed. So a violet (far blue) light at 400nm could do a 100nm feature. Going to near UV can theoretically cut that in half. Thus IBM does not HAVE to go to far UV or X-ray epitaxy to get to the 60nm mentioned.
Now on the research side: I spent some time with the IBM research team at the National Synchrotron Light Source at Brookhaven National Lab back in 1985. They were playing with far UV and soft X-rays to expose masks back then. They used the synchrotron radiation to generate the far UV and soft x-rays. (And optics get really, really different at these wavelengths!) Back then they were playing with feature sizes of 100nm. That's 100nm more than 18 years ago. This is back when feature sizes were measure in microns -- not fractions of a micron or anywhere near 100nm. One of the big issues the team at the NSLS wrestled with was the accuracy the repeititive capabilities of the steppers which held and aligned the masks. Each mask has to be aligned very precisely with the position of the prior mask. To take something from research to production where the masks have to be able to be aligned very precisely on a repetititve basis is a major hurdle. It takes a LONG, LONG time to get these systems from research into production.
That said, IBM's Fishkill plant is one of only 2 or 3 commercial sychrotrons on the planet. They've had it there for quite a while and have been experimenting with it since the 90s. A synchrotron radiation source for far UV could easily get IBM's Fishkill plant down to 60nm or even 40nm.
Another additional point is that it is very, very seldom that chip designs scale by a factor of two or more. It is very unlikely that IBM could take the designs for all the masks (there are several in the process) and just shrink all of them by a factor of two (linearly, not in area) and still have the chip designs work. Capacitances are are all different, internal chip timing is different, leakage currents are different, resistances are different, the relative fraction of the speed of light that electrons travel through the materials are different (based upon such things as the differences in capacitance, resistance, etc.) and such. Thus taking the 970 from 130nm directly to 60nm is a HUGE leap of faith very, very few chip designers would try. Even going from 130nm to 90nm will probably require some adjustments in chip design, but many, many fewer than a jump from 130nm to 60nm.
The question is, "Is IBM to that point yet?" I don't know. Given the rate of progression they were at the last time I talked to them (and the guys at the Thomas J Watson research center) I doubt they are ready to go to 60nm within the next year -- on their production systems that is.
The other question is, "Does it make business sense to jump past 90nm to go to 60nm?" We have to remember that at its core IBM is a very conservative company. If they perceive they can make a lot of money with the 90nm process, they will wring every dime out of it they can before moving on. They've spent lots of money (and time) developing the 90nm process and beginning the move to make the production line move to 90nm. IBM is unlikely to waste this.
Bottom line: Is it possible to move the production line to 60nm in 2003 or even 2004? Yes, it is theoretically possible. Is it probable? Absolutely not.
First, the masks are not cut with light (lasers, UV, x-ray, synchrotron radiation, etc.). The masks are cut with electron beams. They have been (for the most critical masks) for many, many years -- back into the 80s for some masks. Electron beam work is extremely precise, but still very expensive compared to photon work. Also resists (the material put on the wafers and exposed to the light shining through the masks) are not as useable for electron beam work as they are for light (photon) work.
The wavelengths of the light used to expose the wafers through the masks on production systems (as different from the research systems) have been getting shorter and shorter for many years. It is now getting into the UV for some systems. Playing games with the way things are focused, taking advantage of edge effects, and such allows the feature size to get down to about 1/4 of the wavelength of the light when things are really pushed. So a violet (far blue) light at 400nm could do a 100nm feature. Going to near UV can theoretically cut that in half. Thus IBM does not HAVE to go to far UV or X-ray epitaxy to get to the 60nm mentioned.
Now on the research side: I spent some time with the IBM research team at the National Synchrotron Light Source at Brookhaven National Lab back in 1985. They were playing with far UV and soft X-rays to expose masks back then. They used the synchrotron radiation to generate the far UV and soft x-rays. (And optics get really, really different at these wavelengths!) Back then they were playing with feature sizes of 100nm. That's 100nm more than 18 years ago. This is back when feature sizes were measure in microns -- not fractions of a micron or anywhere near 100nm. One of the big issues the team at the NSLS wrestled with was the accuracy the repeititive capabilities of the steppers which held and aligned the masks. Each mask has to be aligned very precisely with the position of the prior mask. To take something from research to production where the masks have to be able to be aligned very precisely on a repetititve basis is a major hurdle. It takes a LONG, LONG time to get these systems from research into production.
That said, IBM's Fishkill plant is one of only 2 or 3 commercial sychrotrons on the planet. They've had it there for quite a while and have been experimenting with it since the 90s. A synchrotron radiation source for far UV could easily get IBM's Fishkill plant down to 60nm or even 40nm.
Another additional point is that it is very, very seldom that chip designs scale by a factor of two or more. It is very unlikely that IBM could take the designs for all the masks (there are several in the process) and just shrink all of them by a factor of two (linearly, not in area) and still have the chip designs work. Capacitances are are all different, internal chip timing is different, leakage currents are different, resistances are different, the relative fraction of the speed of light that electrons travel through the materials are different (based upon such things as the differences in capacitance, resistance, etc.) and such. Thus taking the 970 from 130nm directly to 60nm is a HUGE leap of faith very, very few chip designers would try. Even going from 130nm to 90nm will probably require some adjustments in chip design, but many, many fewer than a jump from 130nm to 60nm.
The question is, "Is IBM to that point yet?" I don't know. Given the rate of progression they were at the last time I talked to them (and the guys at the Thomas J Watson research center) I doubt they are ready to go to 60nm within the next year -- on their production systems that is.
The other question is, "Does it make business sense to jump past 90nm to go to 60nm?" We have to remember that at its core IBM is a very conservative company. If they perceive they can make a lot of money with the 90nm process, they will wring every dime out of it they can before moving on. They've spent lots of money (and time) developing the 90nm process and beginning the move to make the production line move to 90nm. IBM is unlikely to waste this.
Bottom line: Is it possible to move the production line to 60nm in 2003 or even 2004? Yes, it is theoretically possible. Is it probable? Absolutely not.
Either way, if anyone does it, I know it'd be IBM. They have a serious amount of resources devoted to just theoretical technologies and future processes so I bet there are already computer instructions/test hardware set up somewhere in their plant already working on .065 as we speak. Even if they aren't skipping .09, I still bet they have their hands in both currently.
Maybe that's what's holding up the PBook's.
The 60nm process is nearly developed and Apple's waiting the few extra days to stick the spanking new 60nm G5's into the machine.
Hmmmm
On second thought, you can file that under "Not Even Close!!"
The 60nm process is nearly developed and Apple's waiting the few extra days to stick the spanking new 60nm G5's into the machine.
Hmmmm
On second thought, you can file that under "Not Even Close!!"
Are there even reliable EDA tools for 65nm?
Just not going to happen... I've no doubt that IBM is getting fantastic results from their early tests, but they won't leapfrog a generation. Processes have just shrunk too fast in the last couple years to make it profitable... Companies are having trouble keeping .13µm fabs full.
Nice to dream about though-- cool, fast G5's and Athlons that cut Intel to the bone...
Just not going to happen... I've no doubt that IBM is getting fantastic results from their early tests, but they won't leapfrog a generation. Processes have just shrunk too fast in the last couple years to make it profitable... Companies are having trouble keeping .13µm fabs full.
Nice to dream about though-- cool, fast G5's and Athlons that cut Intel to the bone...
Re: long story
Hi,
thanks for your good post. Its good to see that there are people who really know what they are talking about. What I often have to doubt.
Cheers
Hi,
thanks for your good post. Its good to see that there are people who really know what they are talking about. What I often have to doubt.
Cheers
Hey I got a Idea
Hmm as far as I know IBM owns the latest and best Fabs out there. For example east Fishkill is the newest as I can remember so imagine this, IBM realy can produce this PPC in 0.65 that would be the best thing for Apple even better than the idea going to IBM itself.
Steve said in 12 Months 3GHZ so till June 04 we are on 3 GHz what I would say is that this chip must be produced in 0.09 because the heat on the 0.13 is too big for such Clocks. So if they now jump directly to 0.65 Apple and IBM would have a Rage till at least 3.5 GHz and when he can be cooled and driven to such High Clocks than Apple has much more space in Clock rates in all Apple products like Power Book being really "Power Book with 2GHz G5 or even 2.5 and the iMac could get 1.8. But the best thing is that Apple can switch the hole Apple linup to G5 and that would be perfect for Apple and IBM, the Chips will be cheaper not only because of the 0.65 also of the mass of produced chips for all Macs.
Would be really cool and why it should not be ?
Hmm as far as I know IBM owns the latest and best Fabs out there. For example east Fishkill is the newest as I can remember so imagine this, IBM realy can produce this PPC in 0.65 that would be the best thing for Apple even better than the idea going to IBM itself.
Steve said in 12 Months 3GHZ so till June 04 we are on 3 GHz what I would say is that this chip must be produced in 0.09 because the heat on the 0.13 is too big for such Clocks. So if they now jump directly to 0.65 Apple and IBM would have a Rage till at least 3.5 GHz and when he can be cooled and driven to such High Clocks than Apple has much more space in Clock rates in all Apple products like Power Book being really "Power Book with 2GHz G5 or even 2.5 and the iMac could get 1.8. But the best thing is that Apple can switch the hole Apple linup to G5 and that would be perfect for Apple and IBM, the Chips will be cheaper not only because of the 0.65 also of the mass of produced chips for all Macs.
Would be really cool and why it should not be ?
I think probably not true as well and here ase the factors supporting the original claim:
Steve has been burned so bad by Motorola, including right now today with the deferral of the release of a SPEED BUMP Ti 15 PB, that he will look to IBM for anything and everything by default.
Steve is still the "insanely great" guy. His idea of a desktop computer is the fastest on the planet for a cool $3k. He is not out there making $400 portable pizza boxes and populating every school desk with one linked to a DPG5 server in every room.
So "insanely great" Steve is very likely to accept a 6 month chip delivery delay in exchange for a GENERATIONAL improvement that WILL deliver lower heat AND higher Ghz ratings, Apple's bane for the past decade.
Thinner processes bring us closer to a portable internet appliance in every pocket with Apple supplying all the network, music and media services and outsourcing the voice phone service.
PowerPC has legs and this is a generational leap for the product line that will translate to all price points and power levels and user niches.
But it's not gonna happen . . . .
Rocketman
Steve has been burned so bad by Motorola, including right now today with the deferral of the release of a SPEED BUMP Ti 15 PB, that he will look to IBM for anything and everything by default.
Steve is still the "insanely great" guy. His idea of a desktop computer is the fastest on the planet for a cool $3k. He is not out there making $400 portable pizza boxes and populating every school desk with one linked to a DPG5 server in every room.
So "insanely great" Steve is very likely to accept a 6 month chip delivery delay in exchange for a GENERATIONAL improvement that WILL deliver lower heat AND higher Ghz ratings, Apple's bane for the past decade.
Thinner processes bring us closer to a portable internet appliance in every pocket with Apple supplying all the network, music and media services and outsourcing the voice phone service.
PowerPC has legs and this is a generational leap for the product line that will translate to all price points and power levels and user niches.
But it's not gonna happen . . . .
Rocketman
Re: Hey I got a Idea
Those are details that only IBM knows for sure.
Well, we don't know how far the current process will scale. The 970s are pushing around 45 watts @ 2GHz. Intel and AMD are shipping chips that generate around 80 watts of heat. We probably won't see 3GHz on .13 because I think .09 is around the corner but I don't know how fast they can push the current design/process. The 970 has pretty long pipes, it's got pretty good power characteristics, and IBM is probably constantly fine tuning the process. Intel currently ships .13 micron P4s in speeds ranging from 1.8GHz to 3.2GHz. The wattage in this range scales from 68 watts to 82 watts.
There are other issues related to process shrinks that can limit the potential speed gains. electron migration... the inability to remove heat from such a small area. Going to .065 might not guarantee 3.5 GHz anymore than stick with .13 would prevent the 970 from clocking much higher than initial speeds.
Those are details that only IBM knows for sure.