30 Aug
2008
30 Aug
'08
5:48 a.m.
A couple points made in the past day that I can't resist following up
on with my contradictory experience:
I think, rather, that that's a good portion of the
explanation of why they
were so slow.
Ballpark example, take a 12AU7: the sum of the grid-to-plate and
grid-to-cathode capacitance is around 3 pF. Suppose the network resistance
feeding the grid circuit is 250 KOhm, that's an RC time constant of 0.75uS,
a little better than just 1 MHz. (R can be reduced of course but power
consumption is then on the climb.)
(Not to say there weren't other reasons they were slow..)
In fact, if you build similar flip-flop circuits out of transistors
(essentially RTL) you'll end up with very similar counting speeds.
The same tricks that speed up transistor flip-flops will speed up
tube flip-flops: First you add some "speedup capacitors" to help
overcome grid/base capacitance storage effects. Then you start
eating away at your component variation/supply voltage variation
margins to get extra speed at the cost of hand-picked parts.
Then you start thinking about active pull-ups instead of
just resistive pullups.
Then you go to totem-pole structures. The concept works the same
in both tubes and transistors. But until IC's you don't see it
used an awful lot because with discretes, totem poles mean more
parts. Even into the 80's NMOS and PMOS were still cutting
edge and then the chip industry didn't go whole-heartedly
into CMOS until the mid-80's.
Brent wrote:
And how feasible would it have been to replace those
ring counters with 4-bit
binary decade counters to save a lot of tubes? Perhaps it wouldn't have helped
much because of the way the values from each counter were transmitted around
the machine. Binary counters were around by then of course, I'm not sure when
binary counters were first wrapped into decade counters ( by late 40's at
latest).
Both BCD counters and ring counters made out of tubes were around
before WWII, and of course both were used a lot in WWII radar equipment.
The earliest reference I know is Williams, "High Speed Thyratron
Automatic Counting", Proc Roy Soc May 1932, mentioned in Electronics,
July 1932, p 232.
The flip-flop tube count difference is really very very minor: 5 tubes for a
Johnson ring counter, 4 tubes for a BCD counter, and the BCD counter takes some
cleverness to avoid the use of diodes (in the 50's could've been either
vacuum or solid state) for gating. And the ring counter is
much more easily decoded into ten digits for display.
From a purity point of view, the ring counter has some
huge
advantages in glitch-free decoding. But just like 70's TTL
implementations,
most designers were blissfully unaware of the glitches that messed
up many of their circuits :-).
Note that a lot of tube decade counters even into the 60's didn't
use strict BCD or string ring arrangements, instead they always
chose an arrangement that was most easily decoded for the application.
Note that even in solid state IC form both schools are thoroughly
established: The CMOS 4017 is a ring counter, the TTL 7490 is a BCD
counter. Usually the ring counter was used where there was no need
for any BCD coding. And the TTL nerds who didn't understand ring
counters always used 7490's followed by a 7442 decoder when they
wanted 10 decoded outputs, when they should've been using a ring
counter to begin with!
Tim.