Daniel.Stodolsky@CS.CMU.EDU (07/17/90)
Several recent posters have commented on the fact that many systems are currently limited by the pinout of their chips, and shrinking transistor sizes won't make this much easier. One possible solution is to go to wafer scale integration, but this brings about problems of its own. Different packaging schemes can increase the pinout somewhat, but one still gets stuck after a while. One current approach to get around this problem is (time domain) mutliplexing of the pins. The I860, for instance, mulitplexes the data bus. But this doesn't increase the maximum number of signals one can pump in or out of a chip per cycle. IDEA: Why not voltage domain multiplex? On a given pin, one signal could come in at either -3 or +3 volts for 0 and 1, and a second signal could come in at -1 or +1 volts for 0 and 1. A little extra logic would be needed to decode the signal, but one could get a doubling of the number of signals for a given packaging scheme. Of course, if your voltage sources can be precisely controlled, there's no reason why you couldn't put 3 or 4 signals per pin... Comments? Dan Stodolsky Engr. Design Research Center Carnegie Mellon University danner@miracle.edrc.cmu.edu
henry@zoo.toronto.edu (Henry Spencer) (07/18/90)
In article <sacmZI_00hMNQmYF0k@cs.cmu.edu> Daniel.Stodolsky@CS.CMU.EDU writes: >IDEA: Why not voltage domain multiplex? On a given pin, one signal >could come in at either -3 or +3 volts for 0 and 1, and a second signal >could come in at -1 or +1 volts for 0 and 1. A little extra logic would >be needed to decode the signal, but one could get a doubling of the >number of signals for a given packaging scheme... I think the big problem with this is that the chippies :-) have enough trouble getting a binary signal from chip to chip in haste. Multi-level signals will narrow the noise margins, meaning more delay to let voltages settle if you want reliable transmission. Pin count is not quite the bugaboo it used to be, with pin-grid arrays and the like avoiding the limitations of DIPs. -- NFS: all the nice semantics of MSDOS, | Henry Spencer at U of Toronto Zoology and its performance and security too. | henry@zoo.toronto.edu utzoo!henry
mmm@cup.portal.com (Mark Robert Thorson) (07/20/90)
The limiting factor on pin driver bandwidth is what analog engineers call "slew rate", i.e. how fast you can change the voltage on a pin per unit time. Note that in your scheme, the transition time is different for 0 to 1 vs. 0 to 2 or 0 to 3. If you have an 8-bit integer bus, represented by four wires, the fastest you can sample the bus is the slowest transition between states, i.e. 0 to 3. Now if you were talking about multiplexing a binary voltage change with a binary current change, that might work.
mmm@cup.portal.com (Mark Robert Thorson) (07/21/90)
An even better method of sending two signals down one line is to use thermoelectric drivers and receivers. In a thermoelectric circuit, heating or cooling is produced when current flows across a junction of dissimilar metals or semiconductors. The heating and cooling effects around the circuit add up to zero, of course. Whether heating or cooling will occur at a junction is controlled by the direction of current flow and intrinsic properties of the metals or semiconductors (properties which are related to the work function of the materials). The multiplexing is achieved with four states: +5 hot +5 cold 0 hot 0 cold Note that a junction between material A and material B which heats up when current I passes through it will cool down by an equal amount when current -I passes through it, hence the heating and cooling effects cancel when the signal takes a jog through another material, for example if the aluminum bond wire on one chip passes through a copper pc trace to connect to the aluminum bond wire of another chip. This means that the materials used for the pc traces, lead frame, bond wires, etc., can be disregarded. The only important junctions are those in the driver and the receiver. Oh yes, musn't forget this --> :-)
merriman@ccavax.camb.com (07/21/90)
In article <31906@cup.portal.com>, mmm@cup.portal.com (Mark Robert Thorson) writes: [. . .] > > Now if you were talking about multiplexing a binary voltage change with > a binary current change, that might work. What ever happened to Ohm's Law?
mmm@cup.portal.com (Mark Robert Thorson) (07/22/90)
Yet another way to send two signals down one wire would be to have two power supply rails. One would be an ordinary 5V power supply; the other would be a source of muons (i.e. mu mesons). A muon is a negatively charged particle which can take the place of an electron in an atom. It's much heavier, though. The power supply would be somewhat expensive, because it would need to include a small cyclotron. Note that while our notational convention suggests current flows from positive to negative, the actual movement of electrons is from negative to positive. Therefore our two power rails would have to be below ground. Also note that in most logic families, the driver is a current sink at one logic level and a current source at the other logic level. We'd have to invent a new logic where the driver is a source at both levels, otherwise we would only have 3 logic states: 0 -5 (heavy) -5 (light)
merlyn@iwarp.intel.com (Randal Schwartz) (07/22/90)
In article <28851.26a72efa@ccavax.camb.com>, merriman@ccavax writes: | In article <31906@cup.portal.com>, mmm@cup.portal.com (Mark Robert Thorson) writes: | [. . .] | > | > Now if you were talking about multiplexing a binary voltage change with | > a binary current change, that might work. | | What ever happened to Ohm's Law? It got repealed by congress as an attachment the last year's "War on Drugs" mega-bill. Just wait 'til they do that with the Law of Gravity. :-) Just another example of "your tax dollars at work", -- /=Randal L. Schwartz, Stonehenge Consulting Services (503)777-0095 ==========\ | on contract to Intel's iWarp project, Beaverton, Oregon, USA, Sol III | | merlyn@iwarp.intel.com ...!any-MX-mailer-like-uunet!iwarp.intel.com!merlyn | \=Cute Quote: "Welcome to Portland, Oregon, home of the California Raisins!"=/