45nm in 2006???

[SatinSpiral]

http://www.theinquirer.net/?article=25512

Okay I know it's the Inq, but it's nice to speculate... Here's hoping we get some good news confirming this at IDF.

 

[pxc]

There will not be any 45nm chips for sale in 2006.

One quote is interesting, if true: "Merom will get to 9W in a ULV dual core part, but Yonah is down at 5.5W, single core though"

A 9W dual core chip? Amazing. Intel wouldn't call a 45nm die shrink of Yonah, "Yonah". It would have a new name. I have a hard time believing that Yonah will survive past 65nm since it lacks 64-bit capabilites.

 

[SatinSpiral]

I think the most encouraging thing I got out of this little rumour is the flat out statement that leakage has been solved at 45nm. I was getting very concerned that technology had finally hit the physical limits of what silicon could do, and that smaller transistors would simply continue the trend of worse and worse leakage, rendering them nearly unusable. Regardless of when we finally DO get 45nm processors in retail channels, its comforting to know that we will actually get there without a titanic struggle against thermal dissipation or severe frequency limitations.

 

[FreiDOg]

I think the most encouraging thing I got out of this little rumour is the flat out statement that leakage has been solved at 45nm.

Indeed.
Probably means we're going to see high k dielectrics (on the transistors) and metal gates.

 

[mwarps]

Leakage has not been "solved" - Leakage has been reduced to levels where the chips will yield above Intel's lowest requirements. Making blanket statements like this .. well, it's hit the target audience quite well, as we can see.

 

[Jonsey]

65 nm isn't even supposed to sample until Q1 2006, right?

Here's speculation on the prospect of thermal leakage being solved:

http://www.overclockers.com/tips00825/

 

[SatinSpiral]

Leakage has not been "solved" - Leakage has been reduced to levels where the chips will yield above Intel's lowest requirements. Making blanket statements like this .. well, it's hit the target audience quite well, as we can see.

do you know something that we don't about what's going on inside intel, that you can say it's not really solved?

 

[Jonsey]

do you know something that we don't about what's going on inside intel, that you can say it's not really solved?

As I understand it, leakage occurs when electricity continues to flow through closed transistor gates. The electrons in effect tunnel through the gate. The word "solving" implies stopping this phenomenon. Since this is a quantum mechanical effect that gets worse as distances get smaller, I don't think it's something you can entirely stop. That implies that the problem has been reduced somewhat. The amount it has been reduced is open for arguement.

 

[BigTaf]

65mm is the next step as far as I know

they say mid 2006, but could have similar problems as the 90nm shrink, in fact I can almost gaurantee they will because everything ive read seems to say current leakage will continue to be a problem and going to a smaller die size usually increases this. 45nm is a pipe dream right now I think.

Intel would have one hell of an Ace in the hole if they could nail a smooth 65nm transition now.

 

[starbuck8968]

Well considering that they've already shown Yonah running, I guess Intel has the 65nm figured out as well.

 

[FreiDOg]

As I understand it, leakage occurs when electricity continues to flow through closed transistor gates. The electrons in effect tunnel through the gate. The word "solving" implies stopping this phenomenon. Since this is a quantum mechanical effect that gets worse as distances get smaller, I don't think it's something you can entirely stop. That implies that the problem has been reduced somewhat. The amount it has been reduced is open for arguement.

well, there are several types of leakage current.
Subthreshold leakage occurs because transistors are not perfect on/off switches. There is a point (the voltage threshold) where the transistor allows a substantial amount of current to flow, this is sort of the 'on' point for a MOSFET. Below that point, current still flows, but in exponentially lower amounts. (depending on the process bewtween about every 60 and 80mV higher voltage up to the voltage threshold leads to a 10 fold increase in current flow) So long as your chip runs above absolute zero there will be a thermal voltage, and consequently subthreshold leakage.
If you're able to use slower switching transistors, you can use high voltage thresholds, and substantially reduce subthreshold leakage. On the 90nm process a high VT transistor had a subthreshold leakage about 1/10 that of the low VT transistors.
By moving to highly efficient, highly parallel, relatively low clock speed chips, allows intel to take several power saving, but speed hindering, steps, including using low VT transistors much more sparingly.

The other thing they were trying to tackle was largely what you were talking about, leakage through the insulating material. Up through 90 (and apparently 65) nm processes, Silicon Dioxide is the basic insulator around gates. It's a fairly weak insulator (a 'low k' material) that allows for very low gate capacitance (a smal charge can facilitate a voltage swing, making them switch on/off very fast), but also allowed for signfigant leakage through the insulator. There have been some success in creating a physically thicker insulating layer without raising the gate capciatnce, but it was not going to stop leakage, especially at very small processes where even then the thickened insulator would only be 3 or 4 atoms thick.

A high k insulator could (effectivley) provide a stop to leakage through the insulator. Few years ago high K metal gates were slated for introduction in the 45nm process, reports of 'fixing' leakage may indicate they hit that mark. Intel was projecting something like a 50% increase in gate capacitance, and a 99% reduction in leakage through the insulator.

At 90 to limited extend, 65 and eventually 45nm to a greater extent, wires are becoming a major component of delay. As that happens the speed at all costs approach to transistors looses some of it's appeal.