>Or, put another way. A .22 caliber bullet puts a nice round hole in the
>side of a barn. But the barn doesn't fall over.
>
>Reduce the barn wall to a .22 inch circle, the bullet eliminates it.
look at it from another angle..
if a quarter-inch barn only costs a penny, what's to stop you from blowing
$20 and spreading a couple thousand of them around the farm? keep
redundant copies of the harvest in all of them, and you don't have to care
if one gets blown away.
smaller basic technology means a lower cost of redundancy, and higher
redundancy means higher reliability. today's chips devote a certain
number of transistors to on-board testing hardware. three processing
units do the same calculation in paralell and compare the answers they get.
if two are identical and one differs, the consensus value gets passed on.
if all three differ, the calculation has to be run again.
if individual transistors get small enough, we won't necessarily see
smaller chips, but we're almost guaranteed to see additional redundancy in
the same designs. if you reduce a transistor to half its present size,
you can produce a chip which is the same size, and does the same thing, but
has 100% redundancy on every component.
as a flight of fancy, imagine a CPU which is 100 x redundant. you could
overclock the hell out of the thing, because each transistor only sees a 1%
duty cycle and the die is less vulnerable to overheating. gang the
transistors in groups of five under the voting protocols described above,
and you have a massively reliable system in which each component only sees
a 5% duty cycle. blow 40% of the components, and there's still a decent
chance that every operational unit will still be working and redundant.
>> Chips are not
>> shrinking in any significant way, and are not at all likely to
>> become microscopic. Where would you attach the pins?
another flight of fancy.. change the design of the chip and socket.
eliminate the pins entirely and replace them with highly redundant contact
points along the entire inside surface of a hollow package:
+-------------------------+ +-------------------------+
| /---- contacts ----\ | | |
| +=====================+ | +---------------------+ |
| | | | | |
| +=====================+ | +---------------------+ |
| \---- contacts ----/ | | |
+-------------------------+ +-------------------------+
side cut-away end
and put similar contact surfaces on the mounting post which would replace
today's ZIF socket. alignment of the contact surfaces can be expected to
be reasonable, but not perfect, so toss some electronics into the mounting
post which probe for registration points at startup. once you know where
the registration points are, you can adjust the patching between the
connection surfaces and run further diagnostics.
even at a conservative estimate, such a package and mounting system should
be capable of several thousand discrete connections, and frankly, it would
be easier to produce than a contemporary CPU package. the contacts could
be microengineered into the wafer itself, instead of welded in mechanically
after the fact. chip designers would no longer be bound by the
restriction of pin scarcity, or the need to make the pins line up in some
symmetrical pattern.
put a contact for chip ID in a known location, and the chip can program its
pinout into the mounting post immediately following the registration phase.
different chip designs could have comepletely different internal pinout
maps, but still connect to the same motherboard through the same mounting
post.
sure there are problems with the idea, but they're engineering problems.
you could build a working version of such a system today, if you don't mind
getting a 6-pound monster the size of a ham sandwich. make the components
smaller, though, and suddenly it doesn't look so bad.
the packages themselves don't have to get smaller, but we can get
significantly more power out of the same sized package.
mike stone <[EMAIL PROTECTED]> 'net geek..
been there, done that, have network, will travel.
____________________________________________________________________
--------------------------------------------------------------------
Join The Web Consultants Association : Register on our web site Now
Web Consultants Web Site : http://just4u.com/webconsultants
If you lose the instructions All subscription/unsubscribing can be done
directly from our website for all our lists.
---------------------------------------------------------------------
