Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: THE WATSONYOUTH who wrote (68370)	8/10/1999 11:53:00 PM
From: kash johal	Read Replies (1) \| Respond to of 1579770

Watson, Re:"It's not clear to me what the limiting factor here is. The Athlon in .25um technology MIGHT be limited by the active power. It seems from the AMD web site that at 1.7V, a 650MHz Athlon could generate 54W maximum. Maybe the package can't handle too much beyond that. If they simply raised the voltage to say 1.9V they would easily reach perhaps 720MHz. However,since power goes as (CV^2 X freq), the power would increase by perhaps 40% to a likely unmanageable 75W. So if they are power limited, (CV^2 X freq = constant) they can only increase freq. by lowering capacitance. Thus, they must run at smaller channel lengths. This put demands on both the device design and process capability. So perhaps 22 million devices does cost you something. If they are limited to 1.7V because of power constraints, then, if they were clever, they would choose a gate oxide thickness which just meets 10 year reliability at 1.7V. This is probably around 27A-28A. This is indeed much thinner than Intel is running for their .25um generation which I believe is about 35A. That is why Intel can raise the voltage to 2V without a problem. They have a large gate oxide reliability margin at 35A (probably 30A is the limit at 1.8V) Given far fewer devices, Pentium runs at lower power even at 2V. Alternately, perhaps Athlon is also around 35A in .25um technology.Then they are leaving performance on the table if they are power limited to 1.7V. Now, Intel's .18um process supposedly will run 20A gate oxide. This should limit the voltage to 1.5V (reliability limit) unless Intel comes up with an improved gate oxide process. Given so much lower operating voltage, Intel's device design at 1.5V must improve dramatically (much shorter channel lengths). From what I've seen, it has. So at 1.5V Coppermine power should decrease quite a bit even at much higher (825MHz) frequencies. If AMD also goes to 20A, then they should also be limited to approx 1.5V. But this is only .2V below where they are now. So, at higher frequencies (say above 900MHZ) power may not decrease at all. They may have to run even lower voltages at .18um to lower the power. This will put much pressure on the device design and process which will have to push to shorter channel lengths to recover the performance. Please be aware there are a lot of assumptions here. But, this should illustrate some of the tradeoffs. I do think there is a downside in using 22 million devices. Hopefully for AMD, they can get around it.." Yes, this voltage/power/speed matrix is indeed intriguing. I agree with you the devices is likely power limited and u also see a dramatic increase in max. Athlon power from 500Mhz to 650Mhz. I agree with on the Capacitance decrease when going to 0.18 micron will help of course. Also the redection in Resistance by using Cu will have a major impact as well. It seems to me that the Al version of K7 at 0.18 micron may not give a huge jump up in performance - maybe 20% max. But with Cu they may be able to get a much bigger increase in speed while keeping power reasonable. I must say (again) that the designers have done an awesome job and with a 100mm die they should have awesome margins next year. Short term I am of course worried. Regards, kash

To: THE WATSONYOUTH who wrote (68370)	8/11/1999 1:16:00 AM
From: Process Boy	Respond to of 1579770

THE WATSONYOUTH - <Alternately, perhaps Athlon is also around 35A in .25um technology.Then they are leaving performance on the table if they are power limited to 1.7V. Now, Intel's .18um process supposedly will run 20A gate oxide. This should limit the voltage to 1.5V (reliability limit) unless Intel comes up with an improved gate oxide process. Given so much lower operating voltage, Intel's device design at 1.5V must improve dramatically (much shorter channel lengths). From what I've seen, it has. So at 1.5V Coppermine power should decrease quite a bit even at much higher (825MHz) frequencies. If AMD also goes to 20A, then they should also be limited to approx 1.5V. But this is only .2V below where they are now. So, at higher frequencies (say above 900MHZ) power may not decrease at all. They may have to run even lower voltages at .18um to lower the power. This will put much pressure on the device design and process which will have to push to shorter channel lengths to recover the performance. Please be aware there are a lot of assumptions here. But, this should illustrate some of the tradeoffs. I do think there is a downside in using 22 million devices. Hopefully for AMD, they can get around it.. THE WATSONYOUTH> Nice post. PB