Author: Robert Hyatt
Date: 19:05:15 12/04/99
Go up one level in this thread
On December 04, 1999 at 12:15:31, Tom Likens wrote: >On December 04, 1999 at 09:01:46, Tom Kerrigan wrote: > >>I don't see that superconductors would decrease the amount of metal necessary... >>And even if signal transmission in the wires takes more time than FET switching >>(it does??), the relative mu and epsilon of aluminum is almost ideal, so >>superconductors won't transmit THAT much faster... > >The wire delay is the biggest problem these days in most ASIC processes. >In fact the problem everyone is wrestling with now is the fact that the >wireload models used by the Verilog/VHDL RTL synthesis engines are so >inaccurate that we have to iterate too many times through the synthesis, >placement and routing cycle to meet our timing requirements. > >Copper wasn't used earlier because it is a *MUCH* more difficult metal to >work with then aluminum. Until recently only IBM was having any luck getting >yields up. Many fabs are playing a wait and see game, before making the >tremendously expensive move to copper interconnect. Look for most 0.13 micron >and smaller processes to use copper. > >Superconductors would help with power/heat problems, which is the other >major problem we are all dealing with as the geometries and frequencies >increase. Unfortunately, room temperature superconductors seem to be the >stuff of science fiction currently. Hopefully, one day (soon) that will >change. > >Interestingly, the other major piece of this that no one has mentioned is the >capacitance of the wires. This is just as important as the wire resistance. >Actually, because of the large fringe capacitances for deep-submicron designs >you could argue that it may be more important. Many fabs are experimenting >with low K dielectrics to lower the capacitance of the substrates. That's a good point. Hsu had some serious capacitance problems, and some even worse inductance problems. Long wires, parallel to each other, very closely spaced, inductively couple something fierce... It is discussed in his upcoming book... Those three problems, resistance, capacitance and inductance are all serious issues... > >It's an interesting time in the semiconductor business. > >--Tom > >> >>If higher conductivity is so great for modern-process ICs, then everybody would >>have switched to copper a long, long time ago. >> >>I'm not arguing that superconductors are bad. I'm just saying that as far as I >>know, they wouldn't help modern processes. (They may be great for future >>processes though...) >> >>-Tom >> >>On December 03, 1999 at 00:35:54, David Blackman wrote: >> >>>On December 02, 1999 at 04:54:52, Tom Kerrigan wrote: >>> >>>>I'm not an expert, but I think metals on chips are already almost ideal. The way >>>>to make a chip faster would be to use substrates that allow channels to form >>>>faster. Superconductors wouldn't help with this. >>>> >>>>Notice that IBM is using a copper interconnect process for G4s. Copper is almost >>>>twice as conductive as aluminum, but they're not getting very high clock speeds. >>>> >>>>Better conductors have lower resistance, which does mean less heat... >>>> >>>>-Tom >>> >>>These days on most CPUs the "wires" take up more chip space, more power, produce >>>more wast heat, and cause more of the delay time, than the transistors. >>>Superconductors would be a huge breakthrough if they could figure out how to >>>make them as cheap, reliable, and convenient as the metal layers used now. But >>>as far as i know no-one is close to doing that, and there are probably not even >>>any major development efforts trying any more.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.