Tuesday, August 28, 2007
Process variability below 65nm is more than 20%. And rising.
Now, with transistors and other chip components down to around 65 nanometres across - about the width of 130 silicon atoms - variations can be as much as 20 per cent of the components' total size. Let extrapolate: 32 nm technology, at its 64 atoms transistor's channel, might miss up to 40% - up to 25 atoms. In that case transistor should switch with only 40 atoms. And 22nm transistor by the year 2011/2012 with only 22 atoms in channel. That can cause chips to fail or perform poorly. For example, differences in the sizes of components on two different chips might make those chips generate wrong signals.
That is why AMD K10 is in delays. Everything is fine, but when you engrave 4 cores in one chip manufacturing variations will make their collaborative functiong pretty unpredictable and full of the glitches.
Only Intel has no problems even in 45nm. They invented perfect process, or they are perfect liars?
That is why AMD K10 is in delays. Everything is fine, but when you engrave 4 cores in one chip manufacturing variations will make their collaborative functiong pretty unpredictable and full of the glitches.
Only Intel has no problems even in 45nm. They invented perfect process, or they are perfect liars?