ECL

Browsing some manufacturer's websites (ON) I see that there are still some active ECL products out there. And, according to their datasheets, they have some incredible performance characteristics(onsemi Product Portfolio for example, 90ps propagation delays on this gate). What kind of products do these semiconductors get integrated into? These seem like the exotic of the exotic. What's the point here?

Very very fast logic.
We used to use ECL for very fast core processing, but it has always been expensive.

AWOL:
Very very fast logic.
We used to use ECL for very fast core processing, but it has always been expensive.

So you have a current production application for it? What do you use it for specifically if you can disclose this? From what I see, there are gates and counters available. What else? It seems like it would be dozens of devices before you have anything significant going. What does PCB design look like for an ECL application and what suppliers do you use? ON? Are there others? This is just idle interest for me. I wonder what this is used for.

No, I haven't used it in more than twenty years, though I imagine it is present in some of the test gear I use today.

I think that's a CMOS chip with SiGe ECL-style output stage (which is LVDS basically). The low input current
specs make it clear it's not ECL throughout!

The speed comes from using SiGe transistors in the output stage...

MarkT:
I think that's a CMOS chip with SiGe ECL-style output stage (which is LVDS basically). The low input current
specs make it clear it's not ECL throughout!

The speed comes from using SiGe transistors in the output stage...

Isn't that the best of both worlds then? If it achieves those speeds, who cares how it does it. What is interesting to me, not knowing jack about this, is how low the level of integration is. With only one gate or a 2:! mux on a chip, I would think you would have a board of these things before you have implemented something useful (a counter perhaps?) which then adds back a bunch of delays just because of trace lengths. What is this used for? Why did they never implement something like an 8-bit counter, shift register, or D-type flip-flops or something like that?

They use to be used a lot as input prescaler dividers ( a divide by 10 decade counter was popular) for frequency counters up to 1200 Mhz or so. They sucked up a lot of Vcc current and ran plenty hot, but they were the speed burners of their day. Motorola was a big supplier of ECL chips as I recall.

Lefty

retrolefty:
They use to be used a lot as input prescaler dividers ( a divide by 10 decade counter was popular) for frequency counters up to 1200 Mhz or so. They sucked up a lot of Vcc current and ran plenty hot, but they were the speed burners of their day. Motorola was a big supplier of ECL chips as I recall.

Lefty

Was that made out of discrete chips like this or did they have something with a better level of integration?

Some of the earliest ICs were ECL, I believe.

JoeN:

retrolefty:
They use to be used a lot as input prescaler dividers ( a divide by 10 decade counter was popular) for frequency counters up to 1200 Mhz or so. They sucked up a lot of Vcc current and ran plenty hot, but they were the speed burners of their day. Motorola was a big supplier of ECL chips as I recall.

Lefty

Was that made out of discrete chips like this or did they have something with a better level of integration?

The earliest used simple flip-flops wired as counter/dividers, but there were more complex counter/divider chips available in time. I recall using a 95H90 decade divider good for 250Mhz or so in the 70s.

Lefty

Motorola's MECL family dates back to early 1970's. (Circa 1972)

Parts like: MC10101 Quad Gate or MC1227F

Typical Speeds 85-120MHZ / 4ns switching time... not bad for 1972.

pwillard:
Motorola's MECL family dates back to early 1970's. (Circa 1972)

Parts like: MC10101 Quad Gate or MC1227F

Typical Speeds 85-120MHZ / 4ns switching time... not bad for 1972.

I have heard about ECL and how it was used on the first supercomputers. From Wikipedia, concerning the Cray I:

The new machine was the first Cray design to use integrated circuits (ICs). Although ICs had been available since the 1960s, it was only in the early 1970s that they reached the performance necessary for high-speed applications. The Cray-1 used only four different IC types, an ECL dual 5-4 NOR gate (one 5-input, and one 4-input, each with differential output), another slower MECL 10K 5-4 NOR gate used for address fanout, a 16×4-bit high speed (6 ns) static RAM (SRAM) used for registers, and a 1,024×1-bit 50 ns SRAM used for the main memory. These integrated circuits were supplied by Fairchild Semiconductor and Motorola. In all, the Cray-1 contained about 200,000 gates.

What I still don't get is the crappy level of integration. I guess this implies that there used to be a 16x4 ECL RAM unless there was a way to achieve 6ns with TTL back then. Why no ECL counters, shift registers, or an ALU for example? Can you imagine designing a processing unit out of dual 5-4 NOR gates?

Plus at those speeds everything was a transmission line and had to be terminated or you get lots of signal echoes.
When I was a co-op student at DEC (Digital Equipment Corp) in the very early '80s I was in a lab where we characterized 10K and 100K ECL parts. Fast stuff.

What I still don't get is the crappy level of integration.

Probably a heat dissipation issue.

What I still don't get is the crappy level of integration.

You would use these devices where propagation delay is an issue, so it needs to be pared to the bone - no extras!

[edit]
Also, back in the day, fabrication would have been more difficult, the less gates to fail, the better. Of course we still have massive fails, but they just become lower spec devices. Test downward until it works - "yerp that's a x speed device... Next"
This is after all how Sinclair got going, purchasing duff calculator chips that would only do the 4 main functions and put them in his digital calculator kits.
[/edit]

When I was a co-op student at DEC (Digital Equipment Corp) in the very early '80s I was in a lab where we characterized 10K and 100K ECL parts.

Side Note: I met my wife in one of those labs in 1978... it was called by DEC the "Central Incoming Test" group. Essentially, it's a bunch of wickedly expensive Tektronix/DEC Workstation collaborations where you figure out how bad your suppliers chips are... IE; how much they are lying to you about the quality of parts they sent you.

Digital created a few semiconductor labs in the northeast (Worcester and Hudson come to mind) as it became more and more expensive to have proprietary chips made... and at that time... ECL was common in certain sections of computer designs.