Dan's blowing out his rear again. 5nm gate lengths is not a realistic number. Some people have reported AMD reaching 70nm gate lengths on their .18u process (and that's right now, not in 1999), which is very impressive, given that this is small, even for .13u manufacturing. Intel's .18u transistors have advertised gate lengths of 130nm, although they may have improved since then. The notched gates on the 1GHz Pentium III were 100nm, though, so I doubt Intel's current transistors are much smaller than this, since on the Pentium 4, they are no longer notched. (Intel's .13u process is supposed to have transistor gate lengths down to 65nm, so I would expect much faster performance once that process ramps up.)
I think there's some confusion here. The notched poly was used beginning with coppermine on 0.18 micron which resulted in 100-110nm gates. As far as I know they're still using it even on the P4 (since it's the same process). If you have references showing Intel dropping the notch on the P4 I'd like to see them (although I'd be very surprised to see them). Intel is indeed advertising 65nm gates for 0.13 micron, which is about the functional limit of 248nm lithography + tricks. Anything smaller requires 193nm which still has numerous issues.
As far as the world's smallest transistor gate lengths, Intel publicly holds that record. In fact, no one has yet claimed to beat Intel's previous record of 30nm transistor gates (and that includes IBM). Dan3 obviously made up that 5nm gate length number, since he couldn't remember exactly what the slide show had indicated. Congratulations in catching him in another lie.
BTW, here's the link to Intel's 20nm gate length press release.
Intel set the first record in 2000 with 30nm gates, then topped it the next year with 20nm gates. Quite an accomplishment as both are straight wall (non-notched) gates. |
