ECL

Sure, I just did a rudimentary ALU, it required only a 100 amp supply...

It's been a while, and then usually a bit off-label as analog parts, like here:

Well, mostly, ECL is no longer available, the fab processes are gone, so you are limited to ancient stock in a dusty warehouse in Malaysia.

The nuclear research community used differential ECL to communicate logic signals between modules. A fair bit of this old gear is still in use. So, I build new stuff that INTERFACES to ECL. But, due to the difficulty getting the chips and the vast power consumption, it is possible to find some differential receivers, at least, that can take the ECL logic levels and convert to CMOS. Not so sure about a differential ECL driver, though. There are the ancient 10124 and 10124 chips that can still be had.

There is NO REASON to actually build any hardware with 10K or 100K ECL chips, most FPGAs will run rings around SSI/MSI ECL constructions due to the very low integration level and the board trace capacitance.

Other than the power consumption, ECL WAS a REALLY nice logic family! Especially if you bit the bullet and used differential signals between all boards. The signal rise time was comparable to the gate propagation delay, which greatly eased crosstalk issues while giving better speeds than 74FAST and other high speed TTL families. You got wire-oring, which made bus schemes easy to do. Great for CPU design. But, the power was a BEAR! Forget convection-cooled gear.

Jon

On PLL chips it's still better to use ECL/PECL for frequencies above ~500MHz-1GHz... lower power.

That's the way my Atmel WiFi chip design was... 5.5GHz down to

Show us.

Hey, I did that for a huge wire chamber detector for CERN, used 10KH logic as amplifiers. That worked fine after I got the layout guy under control. Nowadays I'd use a fast opamp or a MMIC.

Newer parts, like EL and EP, have much higher gains and speeds and would be hard to keep stable. Old ECL gates had gains not hugely higher than 1.

I need a /4 or maybe /8 Johnson counter running at around 500 MHz, and the most convenient drive is from an ADCMP comparator, ECL or PECL. I wonder if ECL will continue to be available.

Onsemi and Microchip and Micrel still make ECL, mostly the faster 10EL and EP parts. And there is the extreme stuff, Gigacomm or whatever, multi-buck multi-GHz stuff.

I need a Johnson counter at 500 or 600 MHz, and I don't want to waste nanoseconds going into and out of an FPGA. FPGAs are crazy fast internally, but the I/O is slow.

We use 10EP logic, and some of the faster stuff, in some of our products, with no unusual cooling. But not many parts at once.

NB7V52 is a nice little single 10 GHz flipflop for a mere $10.

There are some exotic 50 GHz sorts of gates and flops around, for hundreds of dollars per function.

How did you get such huge numbers ?

The 4 bit 10181 ALU is only about 150 mA, not much more than 74 S 181. The 10800 series bit slice processors were not that bad either.

While the current consumption might been a bit high for contemporary designs, but the big 24 pin DIL packages helped in getting away with the heat and of course, you couldn't fit too many big packages on a PCB, so no heatsinks required on the chips, just a powerful fan (5 m/s).

BTW, what kind of logic chips are used for 10GbE and 40 GbE Ethernet systems ? I would guess running 80 channels (colours) DWDM with 10 GbE each through a single rack would require serious cooling :-).

ON-Semi still seems to be selling ECLinPS parts, so the answer has to be ye s.

The good thing about ECL is that it is current-steering logic, so ECL rails are pretty clean, and you can mix it with low-level analog without having to go to a ridiculous amount of trouble to keep noise on the logic power ra ils out of the analog ground tracks.

It's fast, and it drives terminated 50R transmission lines. There have got to be situations where this is handy.

I/O at those rates is usually CML.

One doesn't use "logic chips" per se, just great big ASICs or FPGAs that interface directly to the optics socket. I find I have fewer big chips on my boards now that I did 10-20 years ago. Basically there's one big chip (plus a few memories) that does all the processing, and half a board full of DC/DC converters to supply the one big chip.

If I had a picture of my latest 100GbE design handy, I'd post it.

Regards, Allan

Scaled sensibly, the DC/DC converter section requires about 20% of the physical surface area. I expect that most of the real estate actually serves to dissipate heat from the one big chip body.

Mind you, I'm sure the DC/DC section is flattered. Normally it's expected to reside miraculously in whatever bits of real estate and hardware are left over, after all the development budget of time and $ has been used up.

RL

I've always appreciated their linear interface applications in low impedance circuits. Sometimes current drive just makes more sense.

RL

Wasn't the Cray-whatever supercomputer done primarily in ECL, immersed in a bath of inert flourocarbon coolant?

John,

What are you planning with ECL logic ? AFAIK logic ECL threshold levels are something weird nowadays --> hard to interface with other logic family. Compare with CMOS, BJTs are not easy to implement in microelectronics. the reason of abandoned techno, IMHO.

H.

He did it with old discretes BJTs ... from Russia. All soldered on a huge piece of wood. :-D

We know now from where Tchernobyl catastrophe comes from.

That's certainly a recipe for disaster. It's not an uncommon design process, though.

Sure, most mainframes of the era were ECL. IBM only did one non-ECL (TTL) large mainframe before the '90s. Cooling was always a big-budget item.

You have it backwards. The reason BJTs are not easy to implement is because they're not used. ECL isn't used because it's too power hungry. Complexity (that CMOS allows) finally beat speed (of ECL).