Shrink progression of Sony CELL chip

[Idontcare]

I was perusing Goto-san's latest weekly digest which this week is all about the PS3 slim and I thought this graphic nicely displayed the iterative shrinking of the CELL processor over three successive technology nodes (90nm, 65nm, and 45nm).

(one thing to note is that the Area% shrink for 45nm was calculated with 90nm as a reference, why I don't know, but FWIW the 45nm areal shrink over 65nm was 66%, better than the 74% areal shrink of 65nm over 90nm but still no where close to the expected node entitlement shrink factor of 50%)

There's actually quite a bit of interesting data in the article itself relating to performance/watt and GHz vs. Vcc (shmoo plot) for the 45nm chip and comparisons to its more dated 65nm sibling.

 

[palladium]

Didnt know you can read japanese, IDC.

 

[Idontcare]

Ha ha if only, a semester of Japanese long long ago doesn't count, nah google translator is my friend but in this case I relied on the universal language of mathematics.

I wish I had kept the die pics, at TI we had a similar situation with our mobile phone chip - dubbed UPP8M internally - which we successively shrunk thru some 4 or 5 nodes. The chronology of the dieshrinks was quite neat to behold when all the dieshots were laid out in succession and scaled accordingly...the last shrink for that chip was at 65nm which resulted in a 14mm^2 chip, not bad for something that was around 10x the size at 180nm.

Getting a microprocessor shrunk 3 times without any major microarchitecture or ISA changes is pretty rare, speaking of the CELL B.E., which makes the dieshot comparisons all the more "textbook worthy" from an academic example point of view.

 

[Bateluer]

What is the manufacturing process of the Cell and PSX in the currently shipping PS3s?

 

[BenSkywalker]

PS3s have been 65nm for quite a while, Slims are 45nm. It is possible that some of the later 'fats' are 45nm, I haven't seen anything to support that at all mind you, just Sony controls those chips themselves and can change them at will.

RSX has been following the same build process as Cell. I'm not sure if Sony is fabbing RSX themselves for the Slim parts or farming them out to TSMC, I would assume they took them in house if they are continuing with having both chips on the same build process as they have been doing for a while now.

I like how that shot demonstrates that differring portions of Cell responded differently to die shrinks, a point I have always seemed to have trouble getting across to people(while you may be able to ballpark what a die shrink will do for you, even on a given chip some parts 'shrink' better then others).

 

[thilanliyan]

Originally posted by: Bateluer
What is the manufacturing process of the Cell and PSX in the currently shipping PS3s?

I'm not sure but I think the 40gb PS3s started off with 65nm?

 

[Arkaign]

Hmm, hopefully this is a step towards an eventual $199 PS3 price. I'd like to have one as a cheap 1080p media center / linux box.

 

[Soulkeeper]

die shrinks excite me more than most anything else for some reason
nice pics

 

[thilanliyan]

Originally posted by: BenSkywalker
RSX has been following the same build process as Cell.

I think at least the 40GB PS3s still had RSX on 90nm.

 

[BenSkywalker]

I think at least the 40GB PS3s still had RSX on 90nm.

If they did it had to be the early models, before the 80GB non BC launched all the PS3s were using 65nm for both.

Hmm, hopefully this is a step towards an eventual $199 PS3 price. I'd like to have one as a cheap 1080p media center / linux box.

Slim PS3s already have removed the ability to run Linux, so by the time there is a PS3 for that price point they will have long since stopped running Linux. If you are going to use it as a media center, the PS3's OS already handles the task just fine. It will likely be a couple of years at least before we see the PS3 at $199, it already is one of the cheapest high end BluRay players you can get ignoring the gaming capability and media extender abilities. Sony would be killing another division of theirs if they were to drop the price that agressively.

For the record- the PS3s that you can buy in a store today still run Linux just fine, that situation is just going to end in the next week or so.

 

[palladium]

Wow, the 45nm Cell runs at only ~0.8V vcore at 3.2GHz!? That's some awesome engineering feat.

 

[Idontcare]

Originally posted by: palladium
Wow, the 45nm Cell runs at only ~0.8V vcore at 3.2GHz!? That's some awesome engineering feat.

I think it's also impressive that they hit 6GHz with 1.15V at 45nm, and even more impressive to me is that at 90nm they were able to hit 5.6GHz with 1.4V.

Granted for the PS3 purposes that the CELL B.E. was designed for there is no benefit of improving clockspeed with the shrinks, so the trade-offs taken when shrinking CELL were all maximized for reducing cost (die-size) and reducing power-consumption for comparable clockspeeds.

 

[thilanliyan]

Originally posted by: Idontcare
Granted for the PS3 purposes that the CELL B.E. was designed for there is no benefit of improving clockspeed with the shrinks, so the trade-offs taken when shrinking CELL were all maximized for reducing cost (die-size) and reducing power-consumption for comparable clockspeeds.

So games wouldn't run much faster? There are still slowdowns in some games on my PS3. Although RSX is the GPU, I thought Cell did a lot of work as well.

 

[Idontcare]

Originally posted by: thilan29

Originally posted by: Idontcare
Granted for the PS3 purposes that the CELL B.E. was designed for there is no benefit of improving clockspeed with the shrinks, so the trade-offs taken when shrinking CELL were all maximized for reducing cost (die-size) and reducing power-consumption for comparable clockspeeds.

So games wouldn't run much faster? There are still slowdowns in some games on my PS3. Although RSX is the GPU, I thought Cell did a lot of work as well.

I'm not 100% sure but I don't believe the clockspeed for the 45nm-based PS3 is going to be increased over that of the 65nm-based PS3.

Originally posted by: BenSkywalker
I like how that shot demonstrates that differring portions of Cell responded differently to die shrinks, a point I have always seemed to have trouble getting across to people(while you may be able to ballpark what a die shrink will do for you, even on a given chip some parts 'shrink' better then others).

Here's an even nicer graphic highlighting this precisely.

http://pc.watch.impress.co.jp/.../310/373/kaigai-06.jpg

edit: from the table in this link - http://pc.watch.impress.co.jp/...i/20090824_310373.html