Waveforms in a 74LS04 ring oscillator.

[joeqsmith]

Before testing the new parts, I wanted to see if there was any more speed gains in the F ring oscillator with temperatures below +12C.     Next few pictures show my free temperature chamber I  put together tonight.

Old computer heatsink.  Thermal epoxy the peltier to the heat spreader.   Epoxied aluminum block to the peltier.

[joeqsmith]

Built up a few layers of insulation.  Hot glued together.   Insulation block on the cover is used to press the perf board (IC lid) against the aluminum block.   

[joeqsmith]

After all of that, the results were less than impressive.   With the F part running at 5.5 volts, I can now get the temperature down to -7C, or about 19 degrees lower than before.     The speed was a wash at 110.97MHz. 

I then heated the part to 50C and kept the voltage at 5.5.   After it stabilized, it was around 104.28MHz.   

A lot more stable than I would have guessed. 

[dannyf]

Those are really good numbers.

One potential (but minor) issue I see is that on a typical vcxo, the control signal is a low-power signal. Here, it needs to be a power signal - being able to supply 10-20ma minimum.

That can be cured, however, with a emitter follower or even a regulator.

[JoeN]

Following this thread I built myself an oscillator, actually four so far, that work well out of 7404, 74LS04, 74LS14, and 74HC04 ICs (they all run 19-22Mhz except the Schmitt trigger one which runs more like 12Mhz).  I took the advice of the person who told me I should be the person to test out an ECL circuit and I am going to attempt to build that too, but I don't know anything about ECL so I think I will ask for some advice here.  I don't have an ECL inverter but I do have the part MC100EP105 (http://www.onsemi.com/pub_link/Collateral/MC10EP105-D.PDF) which is AND/NAND and can be made into an inverter by feeding the output of one stage into both the inputs (one differential pair to two differential pairs) of the next stage.  So this is the plan:

I0->I1 (first "inverter" to second)

Q0 ->!D1a 29->22
Q0 ->!D1b 29->20
!Q0->D1a 30->23
!Q0->D1b 30->21

so, pin 29 to pins 20 and 22, pin 30 to pins 21 and 23

I1->I3 (second "inverter" to third)

Q1 ->!D3a 2->14
Q1 ->!D3b 2->11
!Q1->D3a 3->15
!Q1->D3b 3->12

so, pin 2 to pins 11 and 14, pin 3 to pins 21 and 23

I3->I0 (third "inverter" back to first)

Q3 ->!D0a 6->26
Q3 ->!D0b 6->24
!Q3->D0a 7->27
!Q3->D0b 7->25

so, pin 6 to pins 24 and 26, pin 7 to pins 25 and 27

And a tap off of D0a as the oscillator output.  Vcc = +5V, Vee = 0V

I have to use an adapter board and traces are probably longer than they should be.  For traces above I will cut the wires to the same length to equalize propagation delays regardless of if they need that length or not.  Unfortunately, the adapter board itself has varying lengths on the traces.

Is it that simple or am I being naive about ECL circuits?  Thanks.

[T3sl4co1l]

If you have any kind of trace length (i.e., comparable to the rise time -- 100ps is only 2cm of trace or coax), you will need termination resistors.  And I forget, but suspect, that ECL needs/wants a termination resistor for correct bias anyway.  Follow the datasheet instruction with termination resistors and terminating voltage: bypass the termination resistor to ground as short as possible, using chip resistors/capacitors and ground plane.  The termination supply can be generated with an op-amp or something like that; it should be capable of bidirectional current flow (so, don't just use an LM317 or something).  Be sure it can handle plenty of current.

Likewise, your maximum frequency will be limited by propagation delay of the gates as well as the interconnects.  Should still be around a GHz though!

Tim

[hamster_nz]

I have to use an adapter board and traces are probably longer than they should be.  For traces above I will cut the wires to the same length to equalize propagation delays regardless of if they need that length or not.  Unfortunately, the adapter board itself has varying lengths on the traces.

Is it that simple or am I being naive about ECL circuits?  Thanks.

If you want to have a fun read about some interesting uses of ECL logic, have a look at the internals of a early Cray http://american.cs.ucdavis.edu/academic/readings/papers/CRAY-technology.pdf - full of interesting detailed stuff like:

With signal rise times (10 percent-90 percent) of 750 ps,
open line stub lengths must be less than 0.5 in to hold reflections
under 35 percent overshoot and under 12 percent
undershoot. Larger reflections can saturate gate inputs and
reduce noice immunity. Longer signal runs must use transmission
lines terminated in the line characteristic impedance.
The CRAY-1 boards have 7-mil wide lines and 7-mil spaces.
The 7-mil height above a ground (or power) plane results in
a 60-52 microstrip line with a delay of 0.15 ns/in.

[JoeN]

If you have any kind of trace length (i.e., comparable to the rise time -- 100ps is only 2cm of trace or coax), you will need termination resistors.  And I forget, but suspect, that ECL needs/wants a termination resistor for correct bias anyway.  Follow the datasheet instruction with termination resistors and terminating voltage: bypass the termination resistor to ground as short as possible, using chip resistors/capacitors and ground plane.  The termination supply can be generated with an op-amp or something like that; it should be capable of bidirectional current flow (so, don't just use an LM317 or something).  Be sure it can handle plenty of current.

Likewise, your maximum frequency will be limited by propagation delay of the gates as well as the interconnects.  Should still be around a GHz though!

Tim

Ha, this is what I was thinking the issue was.  I am going to have to have a small PCB made to try this out.  Good thing boards are only $5 a square inch these days for three.  Not only is there really no way to breadboard this circuit, it sounds like it would be next to impossible even with solder on a perfboard using a QFP-32 adapter, which is what I was thinking of doing.

[T3sl4co1l]

A QFP adapter, maybe not; but building it up on copper clad wouldn't be impossible.  Just tedious.

If you have shaky hands, hand-carving a board is probably beyond your scope, but if not, 0.05" pitch traces are easy to create with a utility knife.  Vias are probably harder (you need to poke a hole and solder or rivet in something conductive!).

Tim

[joeqsmith]

Following this thread I built myself an oscillator, actually four so far, that work well out of 7404, 74LS04, 74LS14, and 74HC04 ICs (they all run 19-22Mhz except the Schmitt trigger one which runs more like 12Mhz).  I took the advice of the person who told me I should be the person to test out an ECL circuit and I am going to attempt to build that too, but I don't know anything about ECL so I think I will ask for some advice here. 

Sounds good!  Looking forward to seeing what you come up with.   


Added a little more insulation to my chip cooler. 

[joeqsmith]

Ran the voltage sweeps for the four new parts.  See attached.     The Motorola MC74AC14N is the second fastest part I tried at about half the speed of the F.     Wanted to see what the edge rates on this thing look like.   Pretty good. 

Changed cables to allow me to use the DSOs with the cooler.   

[dannyf]

it doesn't have to be a ring (3x gates). One gate works just fine - connect its input to its output.

[joeqsmith]

it doesn't have to be a ring (3x gates). One gate works just fine - connect its input to its output.

I think people understand that.  If you looked at the LVDS oscillator I showed, that was one gate. 

Cooling the Motorola part to -10C slightly improved the fall times and frequency.   

[joeqsmith]

Going back and looking at the low voltage operation, the Motorola MC74HC04N won out.   I ran the part at 22, -5 and 50C.  Part runs at a slightly lower voltage at higher temperature.   0.846 volts is about it with sub KHz output. 

[JoeN]

If you have any kind of trace length (i.e., comparable to the rise time -- 100ps is only 2cm of trace or coax), you will need termination resistors.  And I forget, but suspect, that ECL needs/wants a termination resistor for correct bias anyway.  Follow the datasheet instruction with termination resistors and terminating voltage: bypass the termination resistor to ground as short as possible, using chip resistors/capacitors and ground plane.  The termination supply can be generated with an op-amp or something like that; it should be capable of bidirectional current flow (so, don't just use an LM317 or something).  Be sure it can handle plenty of current.

Likewise, your maximum frequency will be limited by propagation delay of the gates as well as the interconnects.  Should still be around a GHz though!

Tim

OK, here is my schematic and on the PCB I try to keep the traces about equal length and the resistors very close to the pins they are terminating.  Am I even close to being right on this?  I read the ON guide to termination as well as one from Micrel and as far as I can see this seems to be the way it is done.  One guide seems to say that you don't need bypass caps for ECL because it is inherently balanced but I included two anyway.  Device is MC100EP105.  Resistors and caps are 0805.  Resistor is metal film 1%, caps are ceramic.  The output is an SMA edge connector.  Any idea about what I should do for unused inputs and outputs on the unused gate?

[joeqsmith]

Looks like you are off to a good start.   Check the pg 8 of the data sheet you provided the link to for the termination.  VTT=VCC-2.0.  I would use that last gates to drive the DSO.      I checked and only have the real old 80s Motorola book that covers the 10K and earlier.   Does not seem like you would want the series terminator but could be wrong.

[JoeN]

Looks like you are off to a good start.   Check the pg 8 of the data sheet you provided the link to for the termination.  VTT=VCC-2.0.  I would use that last gates to drive the DSO.      I checked and only have the real old 80s Motorola book that covers the 10K and earlier.   Does not seem like you would want the series terminator but could be wrong.

So, on closer inspection, you recommend terminating it this way:



So it recommends terminating to Vtt instead of Vee, right?  Vtt would be 3.0V in this case.  Should I throw on a regulator to create that voltage?

And why no series termination?  I see 50 ohm resistors in series on this schematic so that is why I put them on my schematic.  I could leave the 0805 footprint in there and drop in a 0 ohm resistor and see if it matters.  Does that sounds OK or would a 0 ohm resistor affect the circuit?  I'm going to get 3 boards in any case and I have five of these ICs total.

Micrel seems to recommend terminating to Vee which is why I did that at first:  http://www.micrel.com/_PDF/HBW/App-Notes/termination.pdf

By using the last gate to "drive the DSO" are you meaning the RF output and assuming an oscilloscope on the output?  What advantage does this give, will having a buffer square up the output a bit since it is no longer also driving a gate?

[T3sl4co1l]

No, those aren't series terminations, those are indicating to use 50 ohm transmission lines.

Tim

[joeqsmith]

I am only suggesting to check the data sheet.   Micrel app note you link to, see figures 3&4.    Take the time to read it.   

Zo is not a resistor. 

Adding any discontinuity, like a zero ohm jump will cause some reflection.   Higher speeds, need to pay more attention.

You could make a few different boards on the one panel to try different ideas out.   

By using the last gate to "drive the DSO" are you meaning the RF output and assuming an oscilloscope on the output?  What advantage does this give, will having a buffer square up the output a bit since it is no longer also driving a gate?

Yes.  Isolation.   

I have not used ECL in many years.   You should have no problem getting something to work on a circuit board.   This is a very short video I put together showing a single PECL FF on an a VERY crude X-acto knife board.  The videos not real exciting  but does show you can clock some of them well into the GHz range as mentioned.   

https://www.youtube.com/watch?v=k1pUlJwJzAI&feature=youtu.be

Looking forward to seeing what you get working.

[JoeN]

I am only suggesting to check the data sheet.   Micrel app note you link to, see figures 3&4.    Take the time to read it.   

Zo is not a resistor. 

Adding any discontinuity, like a zero ohm jump will cause some reflection.   Higher speeds, need to pay more attention.

You could make a few different boards on the one panel to try different ideas out.   

By using the last gate to "drive the DSO" are you meaning the RF output and assuming an oscilloscope on the output?  What advantage does this give, will having a buffer square up the output a bit since it is no longer also driving a gate?

Yes.  Isolation.   

I have not used ECL in many years.   You should have no problem getting something to work on a circuit board.   This is a very short video I put together showing a single PECL FF on an a VERY crude X-acto knife board.  The videos not real exciting  but does show you can clock some of them well into the GHz range as mentioned.   

https://www.youtube.com/watch?v=k1pUlJwJzAI&feature=youtu.be

Looking forward to seeing what you get working.

Thanks for the info.  If you say those series resistors are definitely not necessary then they come off, no problem.  Probably save a few cents on the board and keep the traces a bit shorter.  My idea of a regulator to supply 3.0V for the 50 ohm parallel terminator resistors, is that a good idea?  If so, maybe I can get this design out tomorrow to OSH and then we have to wait a couple of weeks.  Thanks for your help.

[joeqsmith]

From your Micrel note:

"As a result of ECL/PECL’s differential, high input impedance, very low output impedance (Open Emitter), and small signal swing (andresulting low EMI), ECL/PECL is ideal for driving 50? and 100? controlled impedance transmission lines. A signal trace is considered a transmission line, thus requiring termination, when the signal’s rise/fall time is faster than a trace’s round-trip propagation delay. In some applications, if the distance between two devices is short enough, then termination may not be necessary. Another way to express this is: If, T RISE(signal) < 2 × TPD(trace), then the trace is a transmission line and proper termination is required"

I am not sure how the Micrel note applies to the parts you are playing with.

My idea of a regulator to supply 3.0V for the 50 ohm parallel terminator resistors, is that a good idea?

Personally, I would just use what is shown in Fig 3 or 4 and forget using another supply.   Doubt you really care about the added power.       Really if you want to play with it, I would recommend you try all three.  Then one without termination and one with series termination just to see what the effects are.     Seems like all 5 circuits should fit on a small board.   

[GK]

Potato semi's so called GHz logic might be interesting to play with/try. Parts can be had from their on-line store in small qty.

http://www.potatosemi.com/

[dannyf]

74HC132 -> 120Mhz for single gate and 40Mhz+ for 3 gates, all at 5v. 3 gates work down to 2Mhz @ 1v and 60Mhz+ at 10v (yes, 10v).

[JoeN]

Potato semi's so called GHz logic might be interesting to play with/try. Parts can be had from their on-line store in small qty.

http://www.potatosemi.com/

I am sort of surprised I haven't heard of them and never got a hit on them on eBay during other searches.  Sort of interesting.  I do think they need a new logo and name though, the current choices are pretty lame.  I know it doesn't matter as far as their product goes, but really it does actually matter at some point to generate enough confidence that the company is a serious entity that will be there in a few years.  I wonder if they have any serious customers.

[tggzzz]

Potato semi's so called GHz logic might be interesting to play with/try. Parts can be had from their on-line store in small qty.

http://www.potatosemi.com/

I am sort of surprised I haven't heard of them and never got a hit on them on eBay during other searches.  Sort of interesting.  I do think they need a new logo and name though, the current choices are pretty lame.  I know it doesn't matter as far as their product goes, but really it does actually matter at some point to generate enough confidence that the company is a serious entity that will be there in a few years.  I wonder if they have any serious customers.

Good question. Personally I am put off by them still having the Vcc/GND pins as far away from each other as they can. I disliked that in the late 70s, and haven't seen any reason tochange my mind.