"NU180" - A U180 drop-in replacement for the 3458A.

[Kleinstein]

For the comparator I have a simple voltage divider towards 5 V or 3.3 V to shift the voltage a little up. So the voltage that reaches the comparator is no linger -0.6 to 1.2 V but more like 0.2 V to 1.2 V.
The NE5534 is indeed running a bit hot. With the resistor at the input there is actually no need for super low noise or the high speed. In the newer version with STM32 I have an OPA197 with more moderate power consumption.

The 74LV4053 switches so far work quite well. In my tests it was better than the old Cd4053 and HC4053 (not really a surprize) and max4053A.
Wanghar's tests are with ADG733 and mux1133 and these 2 types don't seem to work that well.

Low charge injection is somewhat desirable, but not the only parameter to look at.
The targets are about (lower is better, but one tends to need a compromise):
C_COM_off < ~ 25 pF    for the NU180 better < 15 pF as the frequency is higher and double the active ref. switches
Jitter < 1.5 ps     (random jitter), this is rarely specified - expected to come with high speed
Charge injection < ~ 5 pC ? a bit unclear, the DS values are only for the off part.
Leakage < ~1 nA   , maybe a bit more critical with 3458 because of more switches
R_on <~ 30 ohm, with the supply modulation as in the 3458 even 50 ohms may be OK, with better matching (15 V ref. case) even more

[wanghar]

modulation frequency of runup @330KHz, still 2.2nF integrator cap and 74LV4053. Not very good but not too bad.

Code: [Select]

0  246351.762  246354.581      -2.863       -0.0947     -8      0.00120
0  246351.766  246354.580      -2.861       -0.0945      9      0.00112
0  246351.766  246354.578      -2.841       -0.0943     12      0.00138
0  246351.774  246354.596      -2.806       -0.0948     -4      0.00128
0  246351.782  246354.609      -2.821       -0.0949      2      0.00080
0  246351.775  246354.599      -2.782       -0.0949     12      0.00145
0  246351.788  246354.605      -2.782       -0.0946      2      0.00132
0  246351.789  246354.611      -2.817       -0.0948    -10      0.00128
0  246351.796  246354.615      -2.810       -0.0947     -8      0.00101
0  246351.799  246354.624      -2.827       -0.0949     -8      0.00120
0  246351.815  246354.635      -2.825       -0.0947      4      0.00116


[wanghar]

@miro123

1. On the comparator's voltage threshold:
The signal from the slope amplifier is not only sent to the comparator it also goes to the amplifier circuit for the auxiliary ADC. To keep the ADC readings within a usable range, the allowable values for the voltage threshold are actually restricted. This is the reason for the variable resistor (trimpot) D27 there. On the actual board, this voltage sits at around 0.1-0.2V.

2. Looking for a better switch than the LV4053:
Especially for applications like the 3458A(NU180) with its 330KHz modulation frequency,  still searching for a better alternative. So far, aside from the MAX4619 (not available and tested yet), none of the other options I've tried have performed better than the LV4053, even though their specs look superior on paper. Of course, it's also possible that the board layout I'm using for testing has its own limitations.

3. Using the NE5534 as the slope amplifier:
The NE5534 does an okay job here. It does run a bit warm with its 4mA quiescent current. Replacing it with an OPA197 or OPA209 would definitely work, but the OPA209 feels a bit overkill for this. 

[Zondar]

Two things:

* I changed the earlier ADG734 test board to better reflect U180's approach, and submitted it for fab.

* Because of the recent discussion, I submitted a second test board based on the 74LV4053, as seen below.

I don't think either board will help much with measurements or decisions. They are more to have something on hand to play with in case that's desired.

I'm still a believer in low on-resistance for a number of reasons, and feel the 4053's is too high. In addition, I'm very much not in favor of putting switches in parallel, and I also dislike how 3 switches per package throws parts of the layout off. However, real data is valuable, and I'm watching that with great interest.

I spent some time on a new NU180 layout. The precision resistors are fairly large. I can just barely squeeze the most important resistors and their switches in the center of the PGA, where they typically have shorter traces, etc. The rest has to hang off to the side, with some awkward wires running across everything.

I'd strongly prefer to settle on which switches should be used before finishing a 5th revision.

Edit: Another problem with the 4053 is that it's a 0-5V type part that doesn't seem to allow for split supplies. I have tried to use U180's voltages as-is to the fullest degree possible. In particular, I use GND2 as in U180 for the switch's lower rail instead of GND (as the 4053 would seem to require). My presumption is that HP wanted that rail to be at least a little below ground (about -0.6V) so that signal ground isn't right against the rail. I approve of that choice and prefer to keep it.

[Zondar]

I did yet another layout of NU180 using ADG734's. It can be redone with others like the 4053's if desired. The latter comes in a smaller package and may fit in the same general space. I'm not completely happy with the layout, as there was some struggle to fit everything, but it's essentially ready to be fabriacated.

The 5 largest resistors (2010 English) inside the PGA are the 50k and the 80k's. These are the smallest that can be had with 0.2 ppm/C with these values (80k's being custom-made). The 7 medium sized ones (1206), also inside of the PGA are the 10k's and 20k's, which are also available in-stock with 0.2 ppm/C. The small ones (805) in the lower middle are the other main ones (24k, 4.14k, 6k, 37.5k, 16.7k), which are not available with 0.2 ppm unless custom-ordered, and are generally in the 10 to 25 ppm range. I did also add resistors in series with the switch controls as people advocate.

[Kleinstein]

The signal at the switches should not go much below ground. This should only be less than some 50 mV (one should get a reasonable guess from a simulation) from the integrator input settling. CMOS switches have no issue with this. It is only at some -300 mV when the "didoes" to the supply kick in and cause leakage.  The slightly negative voltage in the 3458 is likely there not because the signal excursions, but to get better blocking / off the the N channel fets.  If really needed, one could also use the slightly negative voltage to supply the switches (may than need to remove the same diode drop from the top). The slight shift would still give an acceptible logic signal at the input. This could still be a point testing upfront, as the sifted supply can cause extra jitter and complicate decoupling.
The supplies would be a thing that may want a few options for tests. Extra decoupling for the dual supply would also make sense, at least as an option to test.

The down side of the dual supply switch chips is that they need an extra level shifter stage and this adds a little to the jitter.

The 20 K reistors should not need super low TC. They work together with a 10 K 5 ppm/K resistor on the A3 board. A little better than 5 ppm/K could help with the zero drift for the super fast conversion. For the matching to the 80 K part there is anyway the R_on of the switches that limits things here. This also applies to the gain if only 1 switch is used for the input. With 2.5 ohm switch resistance and 20 K this would be some -0.5 ppm/K for the gain TC as a starting point. This would be still well OK for the fast mode. With much higher resistance switches doubled switched for the fast input would be a good idea. The 6K/4.14 K part should be reasonable stable as they effect the overall gain. With cheaper resistors I would consider the same series 6 K and a parallel resistor for the 4.14 K to get a bit better matching.  The 37.5 K / 16.7 K are not as critical and 25 ppm/K parts should be Ok there. Similar the 24 K and related 10 K for the 5 V smaller ref. are not that critical.

To allow using the same PCB also with lower grade smaller resistors (e.g. first test) it would make sense to have alternative footprints to the 2010 size.

[Zondar]

I understand that the voltage levels shouldn't drop much below ground, but I prefer to stick with HP's choice to provide about -0.6V headroom there. I also don't want to play games with CMOS thresholds, etc., so I want to use switches that properly allow a split supply. I'm also not going to get stressed out about jitter right now.

I am willing to make a version using 4053's (or others) if that turns out to be necessary or beneficial. But unless help with testing is available, I'll probably start with the ADG's or any better one with a split supply.

These resistors are available in-stock with 0.2ppm TCR: 50k, 20k, 10k.
This is available in stock with 2ppm: 24k.
This is available in stock with 5ppm: 6k.
The remainder are available with 10ppm: 80k, 37.5, 16.7, 4.14.
Of course, we are discussing buying custom-ordered 80k's with 0.2ppm.

Except for the 0.2 ppm resistors, these are mostly inexpensive enough that you might as well use the best ones available. The real question is whether any of the others also need custom levels of ppm's. So is 5ppm for the 6k and 10 for the 4.14 good enough? Would you care to review them resistor by resistor and comment explicitly on any that need more attention?

Yes, a few of the footprints need to be changed to use cheap versions of the 80's and 50k, which I will definitely do to start with, but that is less than 5 minutes of work to change.

Unfortunately, I will probably have to redo the layout yet again, because I may need to go to the next size up still for the 80's and 50's. (It seems that higher resistances need larger areas to meet all specs, or else with the foil ones there just isn't enough range in a smaller package.)

If so, I probably won't be able to fit any switches inside the PGA area. Instead, I'll try to fit all the resistors inside the PGA area, and all switches outside. The problem is that the only room is to the right, and just a little to the left, with a cut-out needed to avoid an annoyingly-placed test point. (Then I started fantasizing that maybe all the resistors should be on their own tiny "hybrid" daughter-card. Hmm, maybe not.)

It's funny: When I look at the red and blue top and bottom metal in KiCad, what my brain incorrectly "sees" is polysilicon and metal1 in a CMOS circuit. And I keep using the (wrong) hot-keys from the CAD software I used 20 years ago. It drives me nuts!

[iMo]

Perhaps a fully OT idea at this stage (as there will be a lot of iterations of that board, imho) - when all the resistors will be placed inside the NU180 pinout, you may, for example, put a heater from the bottom side, including a temperature regulation (like keeping them all at say 50C). Then some cutouts on the pcb at the right hand side may help - isolating the switches and FPGA thermally a little bit.. 

[Kleinstein]

For the ADC only the slow mode (via 50 K input resistor) is really critical and needs low TC. The fast mode (e.g. for AC) is though the external 10 K resistor anyway and thus a higher TC. So I would not care that much about this part.

The 50 K and 80 K resistors are the most critical ones. The 50 K in additional also effects linearity, not just the gain TC.

The 20 K resistors for the reference should get away with 5 ppm/K resistors as they are relative to another 5 ppm/K.
They should still match to some 0.2 to 0.5% or so to the 80 K

The 6K + 4.14 K effect the gain TC, with roughly 40% sensitivity. So a bit less critical than the 50K/80 K.
With 6 K + 13.3 K for the 4.14 K one could get away with 5 ppm/K resistors. Accuracy for the ratio is not that important (e.g. 2 % off could be OK)
The absolute value should not be that critical. One could also use slightly larger values here, like 7.5 K and 5.1 K  or maybe even 5 K and 7.5 K from a 4 x 5 K array (e.g. MORN) with a slight change that the SW would complain with too low a ref. voltage.

For the 10K+10K for the 12 to -12 V step the matching should be good, like 0.1%. I would consider an array like Susumu RM3216 or MPMT.... with good matching tolerance. The absolute value does not matter at all (e.g. +-50%)
The 24 K and relate 10 K for the -12 to 5 V step should not be too critical, like 0.2% maybe 0.5% and a TC of 10 ppm/K maybe 25 ppm/K.
Similar are the 37.5 K and 16.7 K. Here the right value may be more like 16.6 K instead of 16.7 V though with some 0.2 to 0.5% tolerance this may not matter. If needed on could add some 2.7 M in parallel to the 16.7 K at pin 19. The 37K and 16.7 K connected to pins 21 and 23 are even less critical (e.g. 1% would be OK) and no need to worry about the details there.

For the reference resistors (80 K and 20 K) the nodes to the switches should be not too much capacitance. So ideally have these connections no too long. The other parts should be less critical.

Instead of a heater, one may consider trimming the gain TC with a suitable resistor in the 6 K / 4.14 K pair. So build up first with a mix of resistors and than exchange one resistor with a suitable TC.

[Zondar]

I did another layout, this time with the main resistors within the PGA. This was more difficult than the prior version, and yet the earlier version is slightly tighter. However, this one accomodates 2512 resistors, whereas the prior one topped out at 2010.

I put the 50k Ohm and the 80k's in a centroid fashion. The 20's somewhat also.

[Kleinstein]

With the really good resistors the symmetry of the resistors in the layout should be less of an issue. The tricky parts are more decoupling and ground for the switches (to get low jitter and similar). I would definitely test the PCB first with simpler resitors to check if the layout and the switches are acceptable (no too much extra noise and INL from delay modulation).

[Zondar]

I wasn't very happy with the prior version (squeezing everything in got too chaotic and messy), so I started over again. This time I began by planning out one of the worst problems: getting so many tracks from one end of the board to the other while having to snake in-between the PGA pins. The result is much cleaner. I also committed to 2512-sized resistors for the 50k and 80k's, as there are more high-valued foil resistor options in this size.

This version finally feels like it's worthy of a real prototype, and I'll put it into fabrication soon. 

[Zondar]

My next task is to design a small "motherboard" to plug the NU180 into. At a minimum, I want stable and convenient access to its pins for power, references, clock, control signals, resistors, etc.

So aside from breaking the pins out, are there any suggestions for inclusion? Also consider that the genuine U180 could be plugged in for measurements too.

Thank you.

[coppercone2]

yeah maybe current shunts, so you can probe it

and probobly local high frequency decoupling

[Kleinstein]

For the U180 testing board, one could consider some kind of relatively simple ADC. So a bit like the board from Wanghar and the U180 replacing the 4053 type switch and precision resistors. 
So supply regulators for some +-15 V and 5 V ,  a voltage reference (likely LM399/1N829 or similar),  some CPLD or µC for the control, an input buffer (could be a bit simpler with a single OP-amp - an OPA182 / LTC2057 should be pretty linear) and MUX for the input.
Compared to the original A3 board one should get away with a 2 OP-amp integrator and a LM311 comparator.  No need for the Zjump part.

p.s.:
It would make sense to test points also accessible with the NU180 installed.
Chances are good that one can get away with a 2 layer PCB.

[Zondar]

I submitted a version of NU180 with size 1206 resistors for the 50 and 80k's instead of 2506 for better availability, and ordered parts for it.

Here is the breakdown of the resistor values purchased. Note that these parts are for a functionality-test prototype using less expensive resistors. The proper version should have 0.2 ppm/C resistors for the most critical components, and better parts elsewhere if available too.

50k: 0.01%, 2 ppm/C
80k: 0.1%, 10 ppm/C (best available in any size)
20k: 0.01%, 2 ppm/C
10k: 0.01%, 2 ppm/C
24k: 0.05%, 10 ppm/C
37.4: 0.1%, 10 ppm/C *
16.7: 0.1%, 25 ppm/C
4.12: 0.1%, 10 ppm/C *
5.97: 0.05%, 5 ppm/C *

* closest available resistance values.

I'm still working on the NU180 test "motherboard" since I'm still figuring out what I want to do with it. It will probably be bare-bones, though.

I think I'm going to custom order the 50k along with the 80k's. I don't think the price of the 50 will be too different from what's available in stock (>$40 each), and the hope is that matching might be better given that all 5 parts should be from the same batch.

(Oh, and I found and corrected an error in the schematic and board: one of the switches had its power and ground reversed - yikes!)

[coppercone2]

you can build a drift matcher with a light bulb in a styrofoam shipping box and a thermocouple.


I wonder if anyone built a characterizer that is more compact & fast yet, for small resistors

[Zondar]

I received the NU180 boards back. I had half expected a comedy of errors, and unfortunately that turned out to be the case: I used the wrong footprint for the switches. Darn! I have to respin the boards. The proper footprint is larger, but my latest design had put the switches outside the PGA, where they can take up a little more room without too much trouble. (Also, DigiKey messed up my order, and still hasn't fixed it.)

Since I have little else to report so far, here's a review of OSHPark, which I tried for the first time:

They are super smooth to work with if you use KiCad. You just upload your project, and they do the rest. They present clear images of all layers for you to check out before confirming your order, which adds confidence. Ordering stencils is similarly easy. The boards and stencils are made in the USA.

For small boards of a few square inches, the cost is quite reasonable, but for larger boards the cost will be higher than some of the other usual outfits. The 6 layer NU180 board cost $15 per square inch, for a total of $48 for three parts including free shipping by U.S. mail (faster options available).

I'd say they are a little slow relative to some alternatives, taking 2-3 weeks including time in the mail. But out of the four boards I've ordered so far, they put two on their higher-cost "swift" service without charge, which saves a week or so.

It's a matter of taste, but I don't like the purple finish that's your only option for most boards. The boards also come back with a few annoying sharp points along each edge, but they are trivial to sand off.

I'll probably get a corrected version out to fab by Monday.

[IanJ]

Since I have little else to report so far, here's a review of OSHPark

Try PcbWay next time, the repeatability of quality is higher IMHO.

Ian.

[Zondar]

Try PcbWay next time, the repeatability of quality is higher IMHO.

Ian.

I wanted to try a US manufacturer this time. Who else is left here? Who else will be left? Anyway, I think OSH is fine for the hobbyist-scale projects that are their niche, and I didn't see any quality problems aside from the pointy panelization remnants. And like I said, the process was very smooth.

[MiDi]

Which Dev-Software and Programmer do you use to program CPLD ATF1502AS?

[Zondar]

Which Dev-Software and Programmer do you use to program CPLD ATF1502AS?

I went with the ATF1502 because it's about the only part in current production that is 5V compatible.* But I haven't used them before and so I'm just now trying to figure that out.

From Atmel, the standard free software would be the ancient "WinCUPL." I note that they have just now come out with "WinCUPL 2", but it's in Beta with no documentation. Since the task is very simple, I can probably get by with WinCUPL.

My experience with FPGA's has been exclusively in the Xilinx ecosystem, and I can probably find one of their programmers around somewhere if that will work. Otherwise, I'm uncertain about which will do and what software will operate it. There are a lot of ~$20 programmers out there, but I don't have experience with them.

If anyone has recommendations, please let me know. Thank you!

Edit: Atmel provides support for programming the ATF1500 series via part ATDH1150USB and the ATMISP program. The device is a bit expensive at ~$100 given that it (officially) supports only those devices. I suspect there isn't anything special about it vs. other JTAG programmers, but at least there's some confidence that it will work.

The good news is that the faulty NU180 board is still useful to test programming and running the CPLD, and so I've soldered that up. If I can find my old Xilinx programmer, I'll try that with the ATMISP program, and if that doesn't work, I'll order the ATDH part.

* Having 5V outputs turned out to not be necessary after all (no digital signals are sent to the A3 board), but I'll use this part for now anyway.

[MiDi]

A cheap way might be:
- dev with WinCUPL (or manually with editor & compile with CUPL)
- convert JED to SVF with ATMISP or fuseconv.py
- FT232H as JTAG programmer
- ISP over JTAG with OpenOCD or Afterburner

Sources:
https://www.hackup.net/2020/01/erasing-and-programming-the-atf1504-cpld/
https://github.com/peterzieba/5Vpld

[voltsandjolts]

If high propogation delays of 'a few us' could be tolerated, an mcu might function as the control logic at much less power than these PLDs consume/emit.

[Kleinstein]

For the U180 replacement there is not much spare time for delays. It's a 20 MHz clock and for the sync flip flops and thus less then 50 ns, likely more in the 20-30 ns range of setup time. So already the slow version of the ATF150 may not be enough, but the faster one should be OK.

A µC may be an option for the control of an ADC circuit to test the U180 replacement, though it could still be a bit tricky in parts.