raulmill@usc.edu (Raul Deluth Rockwell) (11/17/89)
In article <TOMS.89Nov15114003@omews44.intel.com> toms@omews44.intel.com (Tom Shott) writes:
;> Subject: Re: RISC vs CISC (rational discussion, not religious wars)
;> As we top 50 MHz for chip speed the biggest problem becomes getting data on
;> and off chip. You start needing delay cycles between reads and writes on a
;> IO pin to turn the bus around or a chip w/ lots of pins to get data in and
;> out faster. Putting a cache on the die has been beat into the ground.
;> One solution that has not been discussed is flip chip technology. In flip
;> chip technology many die are mounted directly on a ceramic carrier.
;> . . .
Another solution that I haven't seen much discussion of is using
optical fibers with an ultra-high-speed serial protocol as a method
for transporting data.
As the communication delays are non-trivial, this might be more
applicable to a vector machine, or to an architecure which can deal
with structured data (e.g. messages) than to a super-scalar
architecture, but there are a lot of good options in vector machines,
et al.
--lindsay@MATHOM.GANDALF.CS.CMU.EDU (Donald Lindsay) (11/17/89)
In article <RAULMILL.89Nov16083219@usc.edu> raulmill@usc.edu (Raul Deluth Rockwell) writes: >In article <TOMS.89Nov15114003@omews44.intel.com> toms@omews44.intel.com (Tom Shott) writes: >;> As we top 50 MHz for chip speed the biggest problem becomes getting data on >;> and off chip. >;> One solution that has not been discussed is flip chip technology. In flip >;> chip technology many die are mounted directly on a ceramic carrier. > >Another solution that I haven't seen much discussion of is using >optical fibers with an ultra-high-speed serial protocol as a method >for transporting data. Today's nifty interconnect technologies (like flip chip) are great because they reduce the interchip distance, the line capacitance, and even the inductance. IBM's Thermal Conduction Module technology did all of these. The new packaging technology in the VAX 9000 should be more-or-less competitive. (They use derated IC lithography to build up a copper/plastic circuit board, and TAB-bond naked chips directly to the board.) The major difference between optical signals and electrical signals is that optical signals don't interact: no crosstalk, the paths aren't antennas, the signals can even interpenetrate. The problem with using fibers is that they are made in a factory and can't be formed on the spot. Fibers would have to be placed and bonded in the manner of wires. Tedious! The answer seems to be that we should form optical channels directly onto a substrate, perhaps using a plastic. There was a recent announcement of a material which can be made nontransparent by exposure to (I think) UV. Advances like this suggest that we might be able to make "circuit boards" that take light directly from chip to chip. Next, we'd like to have integrated opto-electronic chips: that is, optical emitters and detectors on the same chip with logic. This has been a Holy Grail to the communications people, and they are about there. However, it might be a lot easier to use the new lift-off technologies. These allow one to build a chip, and then lift off the top ten microns or so. The ideal would be to build optical devices, lift them, and drop them onto logic chips. There should be room, if the chips only need bonding pads for power and ground. -- Don D.C.Lindsay Carnegie Mellon Computer Science