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

[Kleinstein]

The 10 K resistor for the AC / fast input path is on A3. However his is only for the fast part and this part is usually not that critical. The resistor on A3 seems to be a 5 ppm/K thin film type. The problem would be getting matching to the U180 internal 20 K resistors. So with the original U180 there is not that much improvement possible. As the external resistor is likely a NiCr part and U180 uses TaN the matching there may likely not be that good. So thin film NiCr may already be an improvement here.

The really critical resistors are the 50 K and 80 K resistors - 5 of the higher value resistors that should track each other.
Another pair of resistors are 6 K and 4.14 K. Here it's only matching not the absolute TC that matters.
The 2 resistors of 10 K for the inverter also only need to match each other. For 2 equal resistors there are many ready made pairs, like LT5400 and others. So these could use a cheaper solution than 2 foil resistors.
So I see 7 ciritial resistors where 0.2 or 0.5 ppm/K parts could make sense.
Others like the 24 K/10 K for the 5 V are already less critical and should get away with simpler (standard thin film, e.g. 10 ppm/K) parts.
It would still be more than $300 with the custom foil parts for 80 K / 50 K and 6 K / 4 K

First one would need to test if the ADG733 or ADG788 switches are suitable - too much jitter could still be a deal breaker, causing too much noise.
So a test with simpler resistors makes absolute sense. Not so good resistor matching causes higher gain TC. A higher TC for the 50 K tends to come with some U³ type INL. It could still be somewhat usable.

[iMo]

I doubt those 0.2 or 0.5ppm/K resistors are "specially manufactured" for that TC. They simply take a 5ppm/K (or similar) and select the ones with such low TC. You pay for the manual selection process (with some automation of course). Doable at home too, imho..
PS: the Q: what is the probability you will find X resistors with 0.2ppm/K (in the 40-50C range) in a set of XXXpcs. 5ppm/K resistors?..

[Kleinstein]

The super low TC foil resistors are a different alloy, not just selected units from the higher TC ones.
Vishay has different types (alloys C, Z,K ) with different typical curves and amount of 2nd order TC. The Z alloy is used for the 0.2 and 0.5 ppm/K types.
Much of the difference is with the 2nd order TC and the TC matching between resistors of the same type can still be quite good even with the 5 ppm/K resistors.

Vishay gives "typical" TC that is more like differential TC.
Many other datasheets give box TC as from the deviation at min temperature, 20 C and maximum temperature. With a large part from 2nd order TC this can give significant higher numbers than the differential TC near room temperature. So the numbers are not necessary comparable.

[Zondar]

Assuming that the critical resistors worthy of special orders include the 80k's, the 50k, and then the 6k and 4.12k, and adding in reasonable choices for the others, we get something like this:

Custom orders:

80k, 0.01%, 0.2 ppm, 2512, $41 each.
50k, 0.01%, 0.2 ppm, 2512, $41 (2010 version in stock for $40).
6k, 0.01%, 0.2 ppm, 1206, $34 (Best stock alternate: 5.97k, 0.05%, 5 ppm, $3).
4.12k, 0.01%, 0.2 ppm, 1206, $34 (Best stock alternate: 0.1%, 10 ppm, $0.75).
(All other 1206's are also $34.)

What I bought for basic functionality tests, which were the best available in stock except for any 0.2 ppm options. All are 1206 except 0805 for the 37 and 16k's.

50k: 0.01%, 2 ppm/C, $20 (5 ppm about $5 - I also bought one of these).
80k: 0.1%, 10 ppm/C, <$1 (best available in any size)
20k: 0.01%, 2 ppm/C, ~$2 (0.2 ppm in stock for ~28)
10k: 0.01%, 2 ppm/C, ~$5 (0.2 ppm in stock for ~$15)
24k: 0.05%, 10 ppm/C, <$1
37.4k: 0.1%, 10 ppm/C, <$1 *
16.7k: 0.1%, 25 ppm/C, <$1
16.5k: 0.1%, 10 ppm.C, <$1 (possible alternate to 16.7k)
4.12k: 0.1%, 10 ppm/C, <$1 *
5.97k: 0.05%, 5 ppm/C, ~$3. *

* Closest available values.

Kleinstein: I know you have questioned whether a few of the values should be slightly different. Please advise especially for the 4.12k, which could be adjusted slightly if custom ordered.

Also, I know switch-related jitter keeps being brought up and that it's a risk, but I don't have a good way to measure this at the moment. I don't feel it's an actionable topic until it is. Also, do we know anything about the jitter in U180's switches, or compared to any other alternative?

Edit: It was my understanding as well that sorting is insufficient, and also that obtaining suitable performance at higher resistances requires larger SMD packages. I'm a little nervous about soldering several hundred dollars worth of resistors, especially if moving them between boards becomes necessary (minimum 3 heating cycles). Leaded resistors such as in that video can be used too, at least if one allows bending the leads (there isn't very much room in height). It's possible that this would even make the layout a little easier, and perhaps slightly better performance can be had too.

[Kleinstein]

The slightly different values would apply to the 37.5 K and 16.7 K. Ideally these should be  3/4 and 1/3 of the 50 K ladder resistance. So the ratio would be 9/4 to 1.  The 34.4 and 16.5 K combination should be about as good as the original 37.5 and 16.7, just on the other side of ideal. For the highest bit one would still want 37.5 K and could than get away with 37.4 and 16.5 K for the rest.

For the 6 K and 4.12 K it is about the ratio only. The nominal ratio should be right. One would only consider changing the absolute value a little. This would ideally want an adjustment (R168) on A3 to compensate the ground current. A slight difference should not be that bad, as it would be a constant current.
For the cheap version one may get away with 4.7 K and 6.8 K as parts that are available in 5 ppm/K and 2 ppm/K (0805 size at least).

For the 2 x 10 K resistors there are quite some options for matches pairs, small arrays.

Data for Jitter are rare. To reach the observed noise level the switching in U180 should be better than some 3 to 5 ps of RMS jitter. That is about as good as 74HC logic is. 74 AC logic is supposedly about 1 ps_RMS jitter, similar to the usual crystal oscillators.
Measuring the jitter could be done in a test setup, maybe the nu180. Use the 4 +- ref. channels to have 2 x positive and 2 negative and swap the 2 sets regularly. With some low pass filter the noise would depend on the jitter and the ref. inverter noise. With tests at a low and high frequency (e.g. 10 kHz and 1 MHz) one could separate the amplifier part (fixed) and jitter + switch capacitance part ( ~ sqrt(frequency).
One may also just test in the 3458 and look at the noise. The capacitive part can be roughly calculated.

[julian1]

Did you test 74AC logic versus LV/LVC logic for jitter?
A comment from TI suggests that AUC ( ie. SN74AUC2G79) is the fastest logic, which may correlate with improved jitter performance.
https://e2e.ti.com/support/logic-group/logic/f/logic-forum/594284/74ac11074-about-low-jitter-trigger

Perhaps more than one flip-flop could also be paralleled (with ouptut resistors)?

Something similar to an external re-synchronizer might be possible with a fpga/cpld by hiding/breaking the clk relationship from the synthesis tool.
If a separate dedicated clk input pin was defined, then it is at least open to the tool/synthesis to emit a single DFF output synchronization primitive (no sram lookup), independent of the normal clock domain logic.
To complete the circuit, the extra package pin would connect to the xtal oscillator with a separate pcb-trace.


Edit,

Ok, the DS actually show vastly better performance, of AUC versus LV. Eg. for positive clk-edge flip-flops

sn74lv175.  4x FF
tpd    7.9ns typ. 16.6ns max.   fmax 105MHz.     2.5V.   Cl=15pF.    25C. 

sn74auc2g80     2x FF .           
tpd.   0.3ns typ 1.3ns max      fmax 275MHz.     2.5V.   Cl=15pF

[Zondar]

If you are interested in the topic of jitter in this circuit, I would suggest starting with the switches and working backwards. As a proxy for jitter, which is hard to find specs for and hard to measure, you can first look for fast control switching, say <=2-3 ns on/off switching speed.

Unfortunately, those switches usually have other downsides. For example, it's easier to get high on/off switching speeds if the switched transistors have smaller gates, which then leads to higher on-resistance.

The ideal switches would have:

* SPDT switches, preferably multiple per package.
* Low on-resistance.
* Low charge injection.
* Low analog I/O capacitance.
* Good analog flatness, especially near ground.
* Includes a negative supply rail for higher performance near ground.
* Fast on/off times.
* Operation with or tolerance of 5V control signals, and more...

Unfortunately, many of these parameters conflict with each other. For example, lower on-resistance often means higher capacitance and charge injection.

The ADG734/788's have all the desired features and were selected for their lower on-resistance plus being good-all-around in other specs, but there may still be better choices. You can even consider making your own switches starting with individual FET's, but I'd consider that only after exhausting other options.

Of course, even a 1ns on/off time is huge compared to Kleinstein's target of <=5ps jitter, so merely looking for high on/off speed could be a weak or ultimately useless metric. It's best to develop a reliable way of measuring jitter first.

[Zondar]

Two updates:

I received the new boards and built one using the functionality-grade resistors. The photo below shows it sitting in it's little test "motherboard." I'll spend a few days programming it and checking it out. Assuming it looks good, I'll then have to work up the courage to hack my A3 board and socket the U180 hybrid. Yikes!

I also bought a set of resistors for the "real" version. It's perhaps a little premature, but they honored their (erroneous?) on-line prices "this one time" so I figured I should jump on it.

It's getting closer to a real test!

[Zondar]

I'm having a problem with the new board. First, sharper eyes than mine (apparently) may have noticed that the CPLD was installed incorrectly in the above photo (rotated -90 degrees), which obviously wouldn't work. I rotated it into the correct position, and it still didn't work. It was using virtually zero power (should be ~20 mA) and would not accept programming. I checked every power and ground pin, and they are all correct.

I swapped in the CPLD from the previous board (already programmed and known-good), and that behaved the same - zero power use, non-operational. I swapped the suspect one onto the old board, and it's operating correctly there. So the chip was not damaged due to being rotated.

It would seem that there's some pertinent difference between the boards? But although the layout changed between the two board iterations, the CPLD's connectivity was identical. And since the power and ground pins all tested correct, I just don't see how the power consumption should be approximately zero. The "approximate" could just be due to the switches, which use a tiny amount of power. These are not the "L" version CPLDs, which have a low-power stand-by function, by the way.

Further checks: All CPLD pins look correct. Applying a 10 MHz clock does change the power consumption, but only by 10 uA (should be several mA). I'm sure this will wind up being something stupidly obvious once I figure it out, but so far ... clueless.

Maybe I'll load the new, unprogrammed CPLD (currently on the old board) into a new board, alone, and see how it does there. I think I'll order some more "functionality grade" parts, too.

Any other ideas? Thanks.

[Kleinstein]

Not beeing able to program makes me suspect a missing supply (e.g. broken trace/via). Another point could be entering a JATG debug mode.

[wraper]

Further checks: All CPLD pins look correct. Applying a 10 MHz clock does change the power consumption, but only by 10 uA (should be several mA). I'm sure this will wind up being something stupidly obvious once I figure it out, but so far ... clueless.

Did you actually measure that there is VDD present on the IC and GND pins are actually connected to GND on the connector?

[wanghar]

Need to check if pins 9, 17, 29, and 41 of ATF1502 have 5V to GND, and whether pins 4, 16, 24, and 36 are all reliably connected to the GND net.

[Zondar]

Yes, I checked that all 4 power pins and all 4 ground pins are connected properly. I measured resistance to the supply inputs (normal) as well as checking voltage levels (normal). I probed right on the chip's pins themselves, not just the tracks they are connected too, etc. Unless I'm absolutely blind, I'm also certain that it's in the correct orientation now, and that it's in the same orientation as on the older working board, too.

It's not that the chip couldn't be programmed, it's that it wasn't recognized by the programmer as being there at all, which is consistent with the lack of power consumption. Even if it's not programmed, it should be using about 20 mA.

Tomorrow I'll move one to another new board, this time by itself, and see if it behaves any differently. I only have two CPLD's right now, but I'll order some more parts tomorrow as well.

Thanks.

[iMo]

When starting with the new board - do solder the cpld chip only onto it, no any other parts.. Then do check the programming and basic functionality of the cpld..

[Zondar]

I hand-soldered the unprogrammed CPLD into a fresh (updated version) board, with no other components, and it draws the appropriate amount of current. I haven't programmed it yet, but I'm pretty sure it is functional.

I remain baffled. I've checked over and over, and there is no difference in the CPLD's connectivity between the two boards. The additional components attached to the CPLD consists of decoupling capacitors, 33 Ohm resistors on its outputs, programming pins, and the switch's hi-Z control inputs. None of that should cause any problems in principle.

I've wondered if my using a stencil and hot-plate (peaking at ~230 C for standard tin/lead solder paste) damaged the board or any parts, but that shouldn't be likely. The connections all seem good anyway, and the extracted CPLD seems fine too.

I don't have enough parts to populate the rest of the board unless I transfer some of them over from the built board. I guess I should do that. Perhaps I'll start by removing the switches one by one and see if anything changes.

I did order more "functionality grade" parts this morning, which should hopefully come in 2-3 days. I have one clean board left, which I'd want to save for fresh components, to be built using the stencil.

The custom precision resistors should arrive in about 2 weeks. I'd definitely want to be fully confident that a board will be good before using them. But that needs to wait until I test a "functionality-grade" board in my 3458A, and also make updated boards with 2512 footprints for the 50 and 80k's.

Edit: It occurred to me that I can remove the 33 Ohm series resistors from the CPLD's output path, and that would disconnect it from the switches. I might do that first. I'm still puzzled, though, and mostly don't think any of that will help.

[MiDi]

About the "pin keeper" function that MiDi mentioned: In the ATF1502, this is programmable. There is no information on how to turn this on or off in the Wincupl manual. However, I found this in an example, to be added immediately following the header block:

PROPERTY ATMEL {PIN_KEEP = ON};

That will only set the keeper function for declared inputs, undeclared are unaffected.

So either declare all unused pins with e.g. = 0 or electrically connect to GND/VCC (CUPL does not allow to set pin keeper individually AFAIK).

I wanted to check if the CPLD's V_OH level might affect the on-resistance of the switch:

The answer is no. The on-resistance of the switch remains essentially constant with a control voltage of 4V vs. 5V (about 2.4 Ohms). It does start to increase slightly, by ~0.1 Ohm, as the control voltage approaches the logic threshold, but that won't happen.

The power consumption is also nearly unchanged. If anything, it goes down very slightly with the control voltage at 4V vs. 5V while clocked at 100 kHz. The chip uses very little power anyway, consuming perhaps 40 nA with a 100 kHz clock at 5V. (Spec sheet says 1 nA with a 3.3V DC control input and 3.3V supply.)

Unless people have something to suggest, I think I'm done with preliminary tests until the new board arrives in a week or two.

That is the expected behaviour as the switches have a driver built-in that decouples the input, but good to have that checked.
Unexpected is the change in Rdson when approaching the logic threshold.
I assume that the supply voltage is kept constant and only the logic input voltage is changed, changing the supply voltage will definitely change Rdson and other parameters.

[MiDi]

My PoC version of the U180 replacement: https://github.com/EleDes/3458A_U180_replacement_PoC

[Zondar]

Looks good!

Yes, I expected that switch behavior as well, but I did want to check to be sure. The "slight increase" when the control voltage is near the threshold could be explained by the switch's gate voltage dropping slightly due to a sag in the control circuit's output. It's nothing to worry about, though, it was just an observation.

I'm continuing to be driven insane by the erratic working/non-working behavior of the CPLD's. For my last attempt, the power consumption was normal, but it refused to be programmed. The error was "Hardware and software settings miss-match." I don't think I was doing anything differently, though.

If you have any ideas about why the CPLD, programmed or not, would change from using ~20 mA to ~0, please let me know!

I have more parts including more CPLD's coming tomorrow.

[iMo]

Except an hw error the erratic behavior could be caused by the faulty bitstream and/or faulty programmer.. I do not know how you test the cpld, but the easiest way is to program an activity (ie. an N-bit counter) clocked by the clock and watch the outputs. The unused inputs should not float while testing - all inputs have to be defined as 1 or 0 (otherwise may cause the erratic behavior too)..
PS: to be "driven insane" is a rather standard situation when working with cplds/fpgas

[Zondar]

My first and biggest problem is that the CPLD's have sometimes appeared dead:

* My first try was a hand-soldered CPLD-only test, which was successfully programmed and tested without drama.
* Next, a fully-populated board (fancy resistors and all), made using a stencil and hotplate, used no meaningful power and could not be programmed. It just appeared dead.
* Next, I made a partially-populated board (all electronics except U180 resistors) via hot-air and hand touch-ups. This showed normal power consumption, but an error while trying to program it.
* I moved a programmed, working CPLD to a different board, and it stopped working.

I ordered another 4 CPLD's, but maybe 10 would have been better! 

Edit: I also ordered 3 more boards. I wanted to reuse the same stencil, so only a few changes were made: I added a few more power/ground pins, and I grounded all unused CPLD I/O pins.

[Zondar]

After days of stress trying to get the CPLD to run, it turns out everything was (mostly?) OK. I still don't fully understand why some CPLD's started out life at 1 mA vs. 20 mA, or why I had trouble programming them earlier. But ~1 mA prior to programming is OK after all, and during my last attempt, one like that on a fully populated board was successfully programmed without any issue at all.

Onward! The next step is to shake down the fully populated board.

[Zondar]

I've tested most of the functionality of the fully-populated board (not all 128 possible control signal combinations). So far everything seems good.

One curiosity was that the 16.7/37.5k ladder was not measurable with any certainty, since no matter which taps were probed, the value would drift upward. The other resistors held steady. The 50k (0.01%), for example, measured 49,999.7 +/-0.2 Ohms over a few hours via a 34465A.

Do people have any questions or suggested tests that I should perform? Otherwise, I'm near the point where I have to hack my A3 board.

[Zondar]

Removing U180 was not very easy. It took over an hour and a lot of patience. However, here it is with the replacement installed.

IT WORKS!!!

I have a 10V ("10.00000", i.e. supposedly good to ~10 uV) voltage source on loan. A new full calibration would be required at this point, but after an ACAL, it has been reading about 10V +/-40 uV, slowly wandering (covers not on). My original U180, by contrast, was reading vastly worse at an awful 9.946V(!), visibly drifted as you watched, and would usually fail completely after a few minutes.

It's not free from problems: A DAC convergence error popped up after ACAL ALL, but not after ACAL DCV. After several hours of operation, I got one error: Multislope rundown convergence, but it's back to running fine now. I wonder how much the precision resistors and a proper recalibration would help?

So, there's some work left to do, but the "functionality-grade" NU180 is already performing significantly better than my original U180!

[bdunham7]

I wonder how much the precision resistors and a proper recalibration would help?

I would assume lower tempco resistors will help a LOT!  The acid test will be assessing the unit's linearity.  You could do a very rough check by measuring the VREF-10 on the 100V range. 

[aronake]

IT WORKS!!!

Absolutely magic work! May not be 8.5 digit grade yet, but just getting here is absolutely awesome!

A good solder gun helps for the U180.