How much of a difference will SOI make?

[Dputiger]

I recently read an article at The Register (www.theregister.co.uk) that made me stop and wonder about the future of the AMD vs. Intel war.

What the article brought up was how Intel is going to decisively beat AMD to deploying a .13 micron process--which is going to allow the P4 to scale much higher in the next six months than it has done so far (much like the Thunderbird effortlessly scaled past the P3 last year).

Now the Hammer processors have been pushed back because AMD is moving them to a .13 micron process (and Barton, the T-bird upgrade after Thoroughbred, will be seen at roughly the same time).

The Register concludes that Intel may well roll out .10 or .07 micron processes before AMD can, thereby giving AMD another hurdle to overcome.

So, after all of that, here's my question: Isn't the whole point of SOI to drastically improve yields and reduce heat? Under that logic, won't AMD be able to use .13 micron processes much longer than Intel will, to produce faster chips?

[Target]

I read that article as well Dputiger....

I don't personally believe its going to matter much if Intel goes to .13 micron before AMD does. In a way, Intel NEEDS to do so more quickly than AMD does just so their P4 can compete. We've already seen several head-to-head comparisons between higher clocked P4's and AMD Athlons (sometimes as much as a 300-400Mhz difference), and while the P4 wins some of the benchmark comparisons, the Athlon handily beats it in a majority of them.

So, while moving to the .13 micron process would also allow AMD to scale their processors higher than they currently are, I don't believe they are as pressed to do so as Intel is. I see the quicker move by Intel in two lights...... 1st, they have the fabs and the money to do so more quickly than AMD does. 2nd, they need something to defend against the current and future designs from AMD.

The folks at Intel are not completely dumb. They know that the real techies can see their P4 CPU is not as superior to the AMD offerings as their marketing folks think it is. And certainly not as superior as they lead us to believe it would be. However, being able to say you have the "fastest" CPU available (when using MHZ as the only quantifier) is a huge bonus when it comes to marketing your product to a consumer base that doesn't know the difference.....and the .13 micron process should allow them to scale to these higher speeds.

The smaller process also allows Intel to manufacture more chips per silicon wafer....which "should" reduce the over-all cost of producing the chips. Since AMD has been priced much lower than similar offerings from Intel, again, Intel needs this to attempt to remain price competitive as well.

I don't see the delay by AMD on the Hammer and Barton because of the move to .13 micron as a big negative. Sure it would have been nice to see the CPU's released on-time (or early)....but I think AMD has made a very calculated decision here, and one that should pay off huge dividends in the end.

They too will see the same advantages that Intel does in the move to .13 microns. Lower manufacturing cost, higher scalability, lower operating voltage and temperature, etc.. I believe AMD's decision was based more out of confidence in their current product line being able to remain competitive with Intel's until they are ready to make the move themselves. I also think they are confident that the Hammer is a more superior design than the P4, and that rushing it in any way, shape, or form is a mistake at this point.

AMD has made enormous strides in changing the tides that were once flowing only in the Intel direction. But there are still an awful lot of people out there that think the AMD Athlon success might have been a fluke....that Intel simply stumbled, and will wake up to once again produce the best CPU's in the world.

AMD can't afford to allow those people to be right. They have to be able to show that "lightning can stike twice". And if as I am assuming, they are confident they can do so, what better way to prove your point than to wait until the Intel giant has laid out the best it has to offer.... only to have AMD come along and one-up them again in the high stakes game?

As for Intel moving to the .10 or .07 micron process before AMD does.......I think its probably a certainty. But again, I think it will be for the reasons I listed above. We'll see if its actually viewed as a hurdle/barrier for AMD or not at that time.


The point of SOI is to remove the theoretical limitations of CMOS technology at around .10 micron or below. Whether or not it allows for higher yields is as likely to depend on the manufacturing process as the design technology itself, but it has shown dramatic improvements thus far. However, it should do all of that and more.

It should allow for use of lower voltages, and higher clock speeds because of more efficient transistor operation. IBM lab tests showed a 20-25% improvement in transistor cycle time for SOI based chips when compared to similar CMOS chips. SOI also boasts an average power requirement of 40-50% less than thier CMOS based counterparts. Sure sounds like more speed and less heat to me!.... It also reduces the occurance of "soft errors". Soft errors are those anomolies produced by cosmic rays, radio active background signals, etc..

Will all of this allow AMD to use the .13 micron process longer than Intel and to produce faster chips?? I think the answer is a definite yes.....especially if Intel doesn't make the move to SOI as well.

Rob Rich wrote an incredibly impressive article on SOI for Sysopt, and is the only reason I have even a small understanding of the technology. If you have not seen it, I'd suggest it as mandatory reading. You can find it here:
http://sysopt.earthweb.com/articles/soi/index.html

Would be interesting to hear Rob's take on your questions as well, so I will drop him a note about this thread.

~Target.

[This message has been edited by Target (edited 06-14-2001).]

[RobRich]

I would offer an continued opinion, but it appears Taget has covered most of the importat marketing reasons for deploying SOI in great detail. Intel will need to implement SOI before AMD, simply to offer a more price and performance comparative P4 as compared to the Athlon. The Athlon at .18 micron offers severely limited scalability, especially as Intel believe its approaching .13 micron P4's should be viable upto 3+ GHz. Also, Intel will be releasing the .13u P3 Tualatin shortly, with yeild rates upto 1.5+ GHz quite probably. Current 1.13 Ghz T-256 samples are reaching 1.4+ with just standard heat/sink fan cooling.

Robert Richmond

[Dputiger]

Rob,

You said Intel will switch to SOI before Intel--did you say that backwards? As I recall, Intel is not working on an SOI solution.

I recently wrote a 20 page paper on processor yields and overclocking for a computing class--in researching it, I noticed something interesting:

When chips are 'in their prime' both Intel and AMD CPU's tend to OC about 15-20%. Chips hold well to this, save as the P3 approached 1 Ghz, the yield dropped off.

Looking at current Athlons, they still appear to be OCcing well--so at what point do you believe AMD MUST make the switch to .13 micron in order to keep the chip current?

Palomino's decreased heat requirements would seem to help the chips lifespan some.

[RobRich]

Intel is also working in SOI based solutions, it just isn't publicized quite as much as AMDs. Other processor manufacturers alos working with SOI include Compaq, HP, IBM, Sun, and many more.

When Intel moves to .13u, the poor scalability of the Athlon core will be fully known. The .18u Athlon, even the latest Palamino, will loose yeild rate effectiveness as the speeds ramp towards 2 GHz. However, a .13 P4 has potential for upwards of 3GHz, and more like 3.5 GHz with SOI. The lack of scalability is simply a byproduct of the Athlon's older core implementation. When the Athlon was introduced AMD expected that 800-1000 MHz would be a peak cap rate for the proceeding 2-3 years, thus we are now seeing the introduction of a 1400 MHz Thunderbird with only moderate yeild rates as compared to similar 1 - 1.2 GHz cores. As the speed increases, AMD will be forced to move to .13u, or they will loose the MHz race.

While raw MHz doesn't equal performance, it does equal higher sales rates since most of the public doesn't have a clue about core latencies, IPC values, pipeline lengths, etc. If AMD wants to market a 1800 MHz Athlon while Intel markets a 3000 MHz P4, then AMD will surely see a significant loss of revenue within the consumer desktop realm.

Robert Richmond