Intel does know how to make integrated circuits and microprocessors but they sure do leave a wake of confusion about their new design of transistor to be rolled out at 22nm.
The idea seems to be not only to reduce feature-size from 32nm to 22nm, a normal step of Moore’s Law, but also to make the transistors more than silicon sandwiches. The long-standing model has been a layer of silicon, a layer of insulator and another layer of silicon or metal to make a FET (Field Effect Transistor). The new model is 3D in that the features are not just layers but 3D features with horizontal and vertical components. This should permit better “off” condition and better performance at low voltages. Basically, the electric field which pinches off the current flowing through the transistor is stronger because of the geometry, the electrodes are closer to the drain and surrounding it. By Gauss’ Law you get a stronger electric field around a blade or a sharp point for the same or lower potential difference.
The normal step of Moore’s Law should give a transistor of half the area (22nmX22nm=484 nm2 while 32nmX32nm=1024nm2) allowing about 50% power consumption all else being equal. The new geometry allows the “all else” to be more optimal, mainly lower voltage. However, Intel is only hinting at “less than half” the power consumption.
It’s all very confusing. ARM is ahead of Intel at any resolution so far. This trick will improve Intel’s status in mobile and may help increase density in servers and regular processors, but ARM should still have an advantage, IMHO. Intel is out a factor of 3 or 4 at the moment. Both technologies will converge near zero power consumption eventually. At some point the power advantage will disappear. Nevertheless, long before that happens, ARM will have intruded into Wintel’s traditional spots. Neither Intel nor M$ can stem the tide.
To see the confusion in posts about this technology, read
- ZDnet:“Intel noted that it couldn’t merely continue to shrink processors and keep Moore’s Law going”
- The Register:“It will be used in PCs, notebooks, and servers, and Perlmutter indicated that chips will arrive in volume early next year. The company did not give a timeframe for Atom’s switch to 22nm.”, and
- Intel and
- Intel’s details
I am all in favour of improving technology but this seems to be hyped somewhat. For example, the 3Dness will likely hit the wall of Moore’s Law sooner than than 2D design. My point is that the 22nm is a “feature” size. The horizontal and vertical parts are both “features” so this transistor is not that small. Is this a one-time thing showing Intel’s desperation to catch ARM in power/performance? Will it scale to 14nm? I doubt it because atoms of silicon are a just about 0.1 nm in size… so the gates will be just a few atoms across. Intel states the technique will work at 14nm and 10nm but that remains to be proven. The technique will certainly permit a given chip to go faster but speed is not our problem currently. Mobile needs lower power and ARM can do that without resorting to a new design of transistor. The technique can certainly yield higher-performing chips and that is what Intel has announced. They will not immediately use this in Atom.