26 Nov
2014
26 Nov
'14
7:47 p.m.
Will wrote:
"On the other hand, ECL does not suffer from the spikes, burps and
farts TTL tends to make.
Late model Cybers tend to have very little in the way of filtering,
simply because the ECL load is "nice"."
ECL, if done properly, uses a terminator on every signal line. On MECL it is usually
it's 560 ohms to -5.2v or 220 ohms to -2.2 volts. Because most circuits have two
ouputs, signal and not signal, termination gives you almost a steady load. This
termination and balanced lines gave the Cybers very fast signal propagation with little
noise like Will mentions. And it requres a lot less regulation and filtering requiremets
for the power supplies.
It was a joy to design with ECL, if you ignored the horrendous power required. We could
weld with the Cyber power supplies! And if the freon cooling failed, you got some
interesting smoke within seconds.
Billy
26 Nov
26 Nov
8:03 p.m.
On Wed, Nov 26, 2014 at 8:47 PM, Billy Pettit <bpettitx at comcast.net> wrote:
ECL, if done properly, uses a terminator on every
signal line. On MECL it is usually it's 560 ohms to -5.2v or 220 ohms to -2.2 volts.
Because most circuits have two ouputs, signal and not signal, termination gives you almost
a steady load.
That most ECL circuits have differential outputs is a common
misconception. Look through a 10K or 100K databook. The vast majority
of the parts have single-ended inputs and outputs. When a single-ended
signal crosses a power supply domain, if the supplies aren't very
close, bad things can happen.
One approach to dealing with that would be to use only differential
parts, but that rules out any of the MSI parts (counters, adders, most
registers, etc.), so you end up using a lot more ICs. Another
approach is to use single-ended on a module, but only differential
between modules. To do it properly, the differential signals between
modules need to be twisted pairs.
The KL10 mostly uses single-ended signals between modules, with
differential over twisted pairs used mostly for clocks.
27 Nov
27 Nov
9:23 a.m.
On 11/27/2014 01:00 AM, Mark Linimon wrote:
On Wed, Nov 26, 2014 at 09:03:47PM -0700, Eric Smith
wrote:
Not much, any more. The old 10K and 100K ECL has not been
produced since
the late 80's, I think. There is still a fair amount of new
old stock somewhere,
maybe people are salvaging chips off boards. It is still a
standard method
of sending trigger signals around in nuclear data
acquisition systems, and
MC10124 and 10125 TTL-ECL translators are used a lot, but
otherwise
ECL logic functions are RARELY used. Now, everybody in the
nuclear business
is moving to LVDS, gives all the advantages of fast
differential signalling without
the huge power dissipation.
(My day job is in the area of nuclear signal processing, and
I designed a
bunch of ECL gear about 20 years ago. We still use that
stuff, and I maintain
it, but haven't built anything new except occasionally using
a 10124 or 10125
to connect to the old stuff. For logic, we can do stuff
with FPGAs that will
run rings around the old ECL.)
Jon
Look through a 10K or 100K databook.
I
used to have one of the former around somewhere :-)
Related question: is ECL still used in anything these days?
mcl
6:28 a.m.
When a single-ended
signal crosses a power supply domain, if the supplies aren't very
close, bad things can happen.
This is a problem that the power supply folks solved a long time ago.
If they still can not get a number of supplies tracking each other
properly, then fire them and get new guys.
--
Will
9:29 a.m.
On 11/27/2014 08:28 AM, William Donzelli wrote:
Well, when you have hundreds of Amps flowing across the
backplane, it gets
hard to have all boards having the same voltage. Even if
you have, maybe 4
boards in a block, there will be significant voltage drop
When a
single-ended
signal crosses a power supply domain, if the supplies aren't very
close, bad things can happen.
This is a problem that the power supply folks solved
a long time ago.
If they still can not get a number of supplies tracking each other
properly, then fire them and get new guys.
from the feed end
to the far end. If you make
the power planes too thick, it
becomes impossible
to solder the backplane connectors, so you have to go to
press-fit contacts.
That is all well-known technology now, but might have been
on the cutting
edge when the KL10 was designed.
Jon
10:04 a.m.
Well, when you have hundreds of Amps flowing across
the backplane, it gets
hard to have all boards having the same voltage. Even if you have, maybe 4
boards in a block, there will be significant voltage drop from the feed end
to the far end. If you make the power planes too thick, it becomes
impossible
to solder the backplane connectors, so you have to go to press-fit contacts.
Yes, I realize the problems, and I am not even a power supply guy. I
do know that it is enough of a concern that *qualified* power supply
guys need to be hired to come up with a working system.
That is all well-known technology now, but might have
been on the cutting
edge when the KL10 was designed.
No, balancing big parallel power supplies goes back quite some ways,
into the 1950s, at least.
--
Will
12:14 p.m.
On Thu, Nov 27, 2014 at 7:28 AM, William Donzelli <wdonzelli at gmail.com> wrote:
That was my point. The KL10 used separate regulators with remote sense
to achieve that. If they had used "one big regulator" as claimed
elsewhere in this thread, it would NOT have worked.
When a
single-ended
signal crosses a power supply domain, if the supplies aren't very
close, bad things can happen.
This is a problem that the power supply folks solved a long time ago.
If they still can not get a number of supplies tracking each other
properly, then fire them and get new guys.
9:10 a.m.
On 11/26/2014 09:47 PM, Billy Pettit wrote:
Will wrote:
"On the other hand, ECL does not suffer from the spikes, burps and
farts TTL tends to make.
Late model Cybers tend to have very little in the way of filtering,
simply because the ECL load is "nice"."
ECL, if done properly, uses a terminator on every signal line. On MECL it is usually
it's 560 ohms to -5.2v or 220 ohms to -2.2 volts.
There are two ways to do it.
You can have a terminator at
the far end of the line and
a pull-down at the driver end. The pull-down will be as you
mention above, but the
terminator will be a much lower resistance, somewhere
between 51 and 91 Ohms,
typically. The other way is to have the terminator resistor
combine both functions,
so you only need one resistor at the far end of the line.
Very short lines can be
done with only the pull-down resistor. Generally, you can
only get away with that
for a couple inches. The terminator power is so large in
ECL systems it can
often exceed the power dissipation in the chips themselves.
The terminator needs to match the characteristic impedance
of the transmission
line. It is probably around 80-90 Ohms for wire wrap wire
over a backplane.
Differential ribbon cable is 110 Ohms, and coax can be
anywhere from 51 - 91
Ohms. So, if you have a 91 Ohm terminating resistor to -2.2
V, and the logic
high is -0.8 V, then the terminating resistor draws 15 mA!
Now, assume 1000
single-ended terminated lines and you've got 15 Amps. Since
damn near EVERY
line needs to be terminated, it is real easy to get to 1000
terminators. So, that's
why short lines only use pull-down resistors, to save
power. A logic low on the
same scheme draws about 6.5 mA, but the 220 Ohm pull-down
only draws 6 mA
on the worst case (logic high).
Jon
3454
days inactive
3454
days old
8 comments
5 participants
participants (5)
-
bpettitx＠comcast.net -
elson＠pico-systems.com -
linimon＠lonesome.com -
spacewar＠gmail.com -
wdonzelli＠gmail.com