IBM and Motorola PPC Yields?

[MacRumors]

MacBidouille claims that IBM's production yields on PowerPC 970's greatly exceed those of Motorola's yields on the PPC 7457.

According to the French site, IBM is able to produce 105 PPC 970's from a 12" wafer with a 80% certification rate of 1.6-2GHz chips. Motorola, however, is reported to produce 98 G4 PPC's with only 1/5th the yield of IBM on (presumably a similar sized?) wafer.

These poor yields are claimed to be part of the delay of the PowerBooks, which are speculatively felt to utilize the 7457 processor from Motorola.

 

[arn]

This is interesting and all... but I don't know what to make of this info.

Where would they get info on both IBM's and Motorola's yields?

I guess someone at Apple may know this information...

arn

 

[NNO-Stephen]

and now we know why Motorola can't ever stick to their plans... and also know why they slipped on the manufacturing list to #11

 

[praetorian_x]

g5 powerbook. now.

Moto needs to die.

Cheers,
prat

 

[danielgrenell]

i hate moto!

 

[stefman]

I think this just shows how much better IBM is at manufacturing chips than Moto.

The fact they have higher yields means their manufacturing technique is better and more efficient than Moto. Higher yield would also mean they can sell cheaper, lower reject rate. This would explain why the 970 is not more expensive than the G4 (I'm sure I read that somewhere).

 

[macnews]

I agree with Arn, where did they get the specs? They did a good job with the G5, not great, but pretty good so who knows. Maybe they do have some good sources. It certainly would expalin why no PB updates recently. Sucks but explains a little.

 

[ZildjianKX]

The industry never wants to give out figures of their yields... that way they can have total price control... since they're the only ones to know their true yield... so this has to be B.S.

 

[RHutch]

Originally posted by ZildjianKX
The industry never wants to give out figures of their yields... that way they can have total price control... since they're the only ones to know their true yield... so this has to be B.S.

Right. And Apple wouldn't post the specs to the new G5's on their site a few days before they were officially announced.

You might be right that they are not publicly available, but these are rumor sites. Bits and pieces of information are gathered from unofficial sources and become stories like this.

I wouldn't be so quick to dismiss it for the reason given.

 

[RoofusPennymore]

As far as I know moto doesnt have a 12" factory, and probably produces them on 8" wafers. No only that the die size of each chip is most likely different. (I suspect the die for the g5 chips is larger, but will get smaller over time.)

Unless these numbers have been factored so they are equivelent for those two issues you cant really compare them. I seriously doubt Moto and IBM share that info with each other.

However yeild rates for usable chips are very telling and are a good example of the work IBM has done with thier new 12" factory. It is day and night between the way IBM makes the chips and Moto makes the chips.

 

[MetallicPenguin]

Also the story below that shows how long and how much Moto has been slowing Apple down for the last 4 years.

 

[DrGonzo]

SO if they can only get 1 good chip out of 5, TWENTY PERCENT SUCCESS RATE, isn't this just gunna drive up the price of the powerbooks? Or has Mots track record always been this bad?

 

[eric_n_dfw]

Wafers

Just curious, can/do they take the bad chips (and the unused portions) from those silicon wafers and melt them down into new wafers? Do any of you semiconducter engineers out there know?

 

[RoofusPennymore]

Re: Wafers

Originally posted by eric_n_dfw
Just curious, can/do they take the bad chips (and the unused portions) from those silicon wafers and melt them down into new wafers? Do any of you semiconducter engineers out there know?

There are layers upon layers of metals and insulators on those chips. That would all have to be stripped off. To expensive to do that so the best thing you can do is make key chains with them.

 

[robg]

You can learn the yield rates from people that work at the company. No one is going to be proud about anything less than 70% yield though. That is definately something the managerial staff gets fired for.

And Roofus is right; Mot isn't running on 12" wafers yet. 12" is relatively new technology in the semi world. I believe the first 12" wafer fab opened up in Taipei 18 months ago. Mot isn't really on the cutting edge of the semi world these days.

 

[zac4mac]

Since the raw material for the chips is sand, it's cheaper to throw 'em out than recycle. So says my friend at Moto - they have a lot of experience in throwing out silicon.

 

[Capt Underpants]

Well isn't that what everyone expected? Motorola is a slacker when it comes to chips, and IBM has one of the most funded R&D's in the world. You would think IBM would have a better yield.

 

[robg]

It isn't just that Cap'n...

Mot was very respectable back in the day but they never looked beyond 6 months. They never worried about updating their fabs or building new fabs. Instead of diverting some profits to R&D/new capital they made the deadly mistake of keeping it all. Now they are trying to use a smaller process on the same equip (it costs about half the price of buying new equip to retool the current ones for a new chip, btw) and it is killing them. They know they are dead in the computer chips industry. They need to cut their losses and abandon the capital and fall back upon other, easier to build, semiconductors.

 

[RoofusPennymore]

Just look at how many factories moto has closed down in recent years. Compare that to Intel who has built fabs in recent years and even has 12" factories. Despite the downturn.

I'm really quite surpised that Moto has found it profitable to stay in the semi buisness. They don't seem like they want to stay in the semi business.

 

[RBR2]

Or has Moto always been this bad?

Originally posted by MetallicPenguin
Also the story below that shows how long and how much Moto has been slowing Apple down for the last 4 years.

Moto's yields have never been all that great, although they have been better than AMD's in years past. When you are making more bad ones than good ones your costs are just too high.

I am out of date on Moto's fabs, but they have not kept up in general. The new fab in Taiwan should be their best hope of getting yields up to the level of everyone else.

Does anyone know if the new Moto chips are to be fabbed in Taiwan?

 

[Orome]

Originally posted by RoofusPennymore
I'm really quite surpised that Moto has found it profitable to stay in the semi buisness. They don't seem like they want to stay in the semi business.

Could be contractual agreements holding them in...

 

[3G4N]

For sale: Motorola's PowerPC 7457 for $5046.96

From the Register:
http://www.theregister.co.uk/content/39/31980.html

It verifies that the 7457 is the "Apollo 7" that I heard about over a year ago.
Moto is surely gimpin' hard if all they can muster is 20%.
I bet it's just the tip o'the iceberg of their problems.

 

[pellucidity]

Originally posted by RoofusPennymore
I'm really quite surpised that Moto has found it profitable to stay in the semi buisness. They don't seem like they want to stay in the semi business.

Moto are doing just fine, if you factor in that Apple and the desktop are not their priority. They do well in the embedded market. Cheap chips that only run as fast as needed, not faster.

Apple's problem is that the chips they needed from Moto aren't the same ones Moto could shift in bulk. Hence the G4 doesn't scale because unlike Intel/AMD, Moto isn't willing to invest the money and R&D to push the process. Same reason we have fake DDR- Why should Moto put a faster memory controller on-die when only Apple needs it?

This is the real reason Apple are better off with IBM. It's not just that IBM are the last game in town for Apple, it's that their goals are the same. Both companies need the chips to scream, to do SMP, and even SMT. Both need fast memory interfaces (not that that is done on-die yet!), and both need fast flexible buses.

Apple/IBM makes a lot more sense. If a low-end chip (MPC74xx) can't be sped up enough, why not try to cram in a high-end chip (Power4)?