Megaprocessor - built from individual transistors

[Vincent Slyngstad]

From: David Griffith: Thursday, June 25, 2015 4:16 AM
> On June 24, 2015 1:27:56 AM PDT, Pontus Pihlgren <pontus at Update.UU.SE> wrote:
>> I like to think that you could shrink the computer by, at least, a
>> factor of four. Probably smaller.
>
> I don't think thatt making the flip-chips double-sided will be desirable.  I 
> want to shrink the machine and minimize any need to reengineer the backplane. 
> By the way, can the backplane of a straight-eight be realized as a PCB? 
> Replicating that component without that will suck.

I've laid out a few of DEC's backplanes as PCBs before, mostly for the
later designs that use both sides of the edge connectors.   They tend
to require about 4 signal layers.  It can be helpful to bus power separately
(using wire much as was originally done), and to route on the diagonal,
as the funky sockets have larger chaseways in the diagonal directions.

Also, the device is relatively high current.  Some of my earlier attempts
used trace widths that, in retrospect, weren't wide enough (especially
power traces), trying to keep the number of layers down.

I can't particularly recommend the original backplane form factor.
They are all based on a 1/8 inch grid instead of the industry standard
0.1 inch, and not at all easy to find or inexpensive these days.


From: Pontus Pihlgren: Thursday, June 25, 2015 6:39 AM
> Now, I've looked at the backplane of a PDP-12 and it looks to be
> 4 or 5 levels deep at the most and fairly "roomy". So, with my
> limited experience it doesn't seem impossible to recreate in a 4
> layer PCB.

Agreed.  6 with power and ground planes might work well, too.

> I'm not sure what would be a suitable edge connector
> though, that would probably dictate the size of the flip chips
> more than anything.

To my mind, the obvious choice for a new edge connector would
be 0.1" headers and sockets.  They are easy to find and produced
by the millions, which means they are also relatively affordable.

> Please correct me if I'm talking complete bulls**t. This is all
> new to me.

The only other concern I've had during my thought experiments
along this line related again to the current/power involved.  The
device is likely to become difficult to cool if you achieve a 4X
volume reduction.

Oh, and I guess the nightmare of soldering many thousands of
surface mount discretes and transistors to manufacture such a
thing.

    Vince