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

[Rax]

One semi-realistic option is to order 25 of these for about $600:

https://www.digikey.com/en/products/detail/vpg-foil-resistors/Y174680K0000T9L/4232515

Specs are 0.01% value and 0.2% TCR. That matches what's in stock for a 50k.

If you only need 4, that's quite expensive (the best 50k's are about $40), but if you rationalize it as costing less than a custom order of 4, it can make sense. The second problem is the lead time: February of next year.

If that's all that kept us from success, I'd go for it.

Zondar: bdunham7 already mentioned going to https://webdirect.texascomponents.com/SearchResults.asp?Cat=1844. I've worked with them, they're very flexible to make custom resistors (literally any value to a couple of decimals) at .2ppm in very small runs (literally one if you must) and in very little time. They essentially print these to your spec, so I believe their cost is very low. It's, surprisingly, a little know source in the US of exceptional resistors and for not a high price. It may be ideal for what you're looking for.

If you're looking for the best TC (bar none) non-array, through hole or smd, resistors to be ordered in small runs and in short time, TCC is an unreal opportunity. I don't know the impact of tariffs and whatnot on their business - it's been a few many months since I last ordered from them - but they seem to cater to exactly this type of customer as we are.

[Zondar]

Hi Rax,

I'll be calling them on Tuesday. I'd have called already if it wasn't the holiday weekend. I'm very interested in hearing about their capabilities.

Thanks.

[aronake]

Here's a puzzle for anyone who is interested:

You have up to 8 resistor array packages, each containing 8x independent 10k Ohm resistors (total 64 resistors, but you don't have to use them all). Within a package, the TCR tracking is 5x better than between packages (e.g. 5ppm/C vs. 25ppm/C), which you want to take advantage of.

I made TC measurements on around 20 tdp16031002. The resistors in each IC are indeed tracking well. 6 closest out of the 8 resistors within 2 ppm/c. But from IC to IC there is huge variations. So probably a good idea to buy 3 times the number of arrays needed and do TC scans and selection.

[Zondar]

I made TC measurements on around 20 tdp16031002. The resistors in each IC are indeed tracking well. 6 closest out of the 8 resistors within 2 ppm/c. But from IC to IC there is huge variations. So probably a good idea to buy 3 times the number of arrays needed and do TC scans and selection.

Sorting can work. For fun I'm tracking a $0.01 10k resistor right now. It's doing very well, with nicely under 0.1ppm/C so far.

[Kleinstein]

I don't think one would really need to test the expensive 0.2 ppm/K resistors upfront. It may make sense to do some tests with simpler (e.g. 5 ppm/K grades) resistors if one can get enough matching from parts from the same batch and maybe even sort / select them.
A test may be possible with a good DMM and temperature modulation or in a kind of bridge circuit to see just relative changes. With 80 K and 100 K one may get away without 4 wire sensing at the bridge.
It makes relatively little sense to check the resistors together with the switches. The switches would add noise and uncertainty, especially if the temperature changes.

[Zondar]

OK, I guess you are right about premium resistors being a waste while looking at the switches.

A major focus needs to be on solving the 80k problem. I'll call that outfit in Texas tomorrow, or else I can spend the $600 and order 25 parts.

Before that, or meanwhile, a look at the easily available 80k's seems warranted. The best in-stock 80k's are Vishay thin-film, 0.1% and 10ppm/C, costing about $0.50 or so each in small quantities. I had wanted to compare them to their 0.2ppm counterparts, so I thought I'd try equivalent parts in 10 or 20k. But instead or in addition, I can array a bunch of 80k's (I sure wish I had a scanner card).

I don't know how I'd "sort" surface mount resistors without soldering them, though. I can't imagine that soldering, desoldering and resoldering them is good for their health.

I don't think that testing the switches alone will result in any big revelations (I'll certainly look, though). I had really wanted to observe them dynamically in conjunction with resistors. In other words, I wanted to test a partial circuit in a way that is similar to how U180 works, but without making a whole ADC. Any ideas?

Hmm, I think I'll add a charge-integrating amplifier to a switch-pair output, e.g. with 80k resistors. If I remember, the 3458A uses a 330 pF integration capacitor, right? Maybe an op-amp with around 50MHz bandwidth would do?

[Kleinstein]

A partial system to test the switches would be a kind of simple ADC. My ADC version uses only 3 SPDT switches and would be a reasonable test platform for switches. It is slightly different from the 3458, but still not too much. I have 50 K resistors to some +-14 V reference and the input.

I could even consider soldering up a SMD to DIP adapter (or even dead bug) and use my old board that has a DIP socket for the switches.
It would need a lot of leakage current for the switches to cause the ADC really fail. The switches would mainly be a performance thing, hardly a point to make the ADC not work at all.
I anyway will likely test a max4619 that is available also in DIP.

I did my initial ADC tests with separate resistors, some even only specified for 100 ppm/K (even better than one set of 15 ppm/K resistors). I don't remember big issues with drift.  For an initial test the 10 ppm/K resitors should be OK. To get some matching to the 50 K at the input I would consider having 2x80 K in parallel and 10 K in series as an alternative to just 1 good 50 K resistor.  8 x 80 K could get ~ 49.5 K with reasonable spread of power.

0.1% accuracy for the resistors should be good enough. The finest slope is only 1/256 and an 0.1% accurate ratio would be only 1/4 LSB DNL and better than needed. In many cases the last bit is anyway rather noisy. That last bit would be mainly a thing for short integration like <=100 µs.

The ladder part (37.5 K and 16.7 K ) does not need super low TC and 0.1% should be OK and even 0.5% may be acceptable.  For the right slope ratio it should be 16.666 K with some 60 ohm already from the HCT14 output. 16.7 K is a standard value, while 16.6 K would need a mix of 2 resistors ( e.g. 2.7 M in parallel to 16.7 K) . It might be worth to have that option for the first fine slope bit  - one could still worst case stack 2 resistors.

[wanghar]

Made a test on the on-resistance of the switch based on the 74LV4053. The test graph, schematic, and data are in the attachment. When Vi exceeds 1.0V, it basically stabilizes at around 14Ω, with the maximum Vgatesense voltage around only 2.8mV. After connecting the two channels in parallel, both Ron and Vgatesense were basically halved.

[wanghar]

Another test: Ran the 74LV4053, TMUX1133, and ADG733 respectively on the CPLD-based MS_ADC board, primarily for an initial noise performance comparison. After each chip replacement, both slope ratio and ADC ration calibrations were performed again. The original board was designed for SO-16 package, so a small adapter board was made to enable testing of these switches in TSSOP-16 packages. All tests were conducted with input shorted at 1PLC, with the last column showing noise_rms values in mV:

74LV4053:

Code: [Select]

0    1759.838    1760.244      -0.405       -0.0009     -8      0.00063
0    1759.843    1760.239      -0.383       -0.0005     10      0.00084
0    1759.840    1760.237      -0.367       -0.0005      4      0.00076
0    1759.843    1760.241      -0.403       -0.0005      2      0.00061
0    1759.843    1760.249      -0.416       -0.0008     33      0.00066
0    1759.847    1760.249      -0.400       -0.0007     16      0.00072
0    1759.845    1760.247      -0.412       -0.0007      2      0.00071
0    1759.853    1760.248      -0.393       -0.0004    -36      0.00071
0    1759.849    1760.253      -0.428       -0.0008    -18      0.00082
0    1759.852    1760.255      -0.416       -0.0007    -12      0.00060

ADG733:

Code: [Select]

0    1775.494    1775.705      -0.233        0.0073      2      0.00078
0    1775.492    1775.713      -0.212        0.0069    -17      0.00092
0    1775.506    1775.716      -0.193        0.0073    -33      0.00084
0    1775.508    1775.723      -0.247        0.0072    -24      0.00096
0    1775.507    1775.728      -0.198        0.0069      0      0.00092
0    1775.509    1775.728      -0.270        0.0070    -22      0.00091
0    1775.505    1775.718      -0.212        0.0072     14      0.00084
0    1775.517    1775.729      -0.224        0.0072     26      0.00078
0    1775.508    1775.728      -0.235        0.0069     38      0.00088
0    1775.511    1775.728      -0.193        0.0071     16      0.00080

TMUX1133:

Code: [Select]

0    1827.979    1828.525      -0.585       -0.0066     32      0.00376
0    1828.026    1828.566      -0.433       -0.0064      0      0.00434
0    1828.168    1828.666      -0.578       -0.0047    -18      0.00471
0    1828.024    1828.571      -0.579       -0.0068      5      0.00414
0    1828.010    1828.564      -0.750       -0.0070     34      0.00319
0    1828.056    1828.588      -0.368       -0.0063     34      0.00358
0    1828.132    1828.659      -0.556       -0.0058     50      0.00403
0    1828.225    1828.768      -0.566       -0.0067     -1      0.00339
0    1828.125    1828.661      -0.568       -0.0062    -12      0.00416
0    1828.144    1828.671      -0.559       -0.0059     -4      0.00315

Conclusion: On this test board, the 74LV4053 still demonstrated the best noise performance. The ADG733 was decent but slightly worse than the 74LV4053 (the 220pF capacitor before the integrator had to be removed to test the ADG733). The TMUX1133 exhibited significant noise.

[Kleinstein]

How is the RMS noise measured ? So how many readings and is it directly single values or the AZ mode difference of 2 conversions ?

A point for the effect of the switch depends on the frequency of the modulation during run-up. The higher the frequency the more important the switches get.

[iMo]

@wanghar: perhaps a little bit off topic - while looking at your board I see no resistors around your FPGA - people usually put there in the i/o lines small resistors, like 22-33ohm (when interfacing analog-like circuitry). What is your experience with that?

[wanghar]

The rms calculated as this function:

Code: [Select]

def Checkrms():          # sum up and check if ready for output
    global outcnt, sumdu, sumq, sumu1, sumu2, avdu, rms, u1m, u2m
    outcnt = outcnt + 1;
    sumdu = sumdu+du;
    sumq = sumq + (du-avdu) ** 2;
    sumu1 = sumu1 + u1;
    sumu2 = sumu2 + u2;
    if outcnt == m:
        outcnt = 0
        checkrms = True
        avduold = avdu
        avdu = sumdu/m
        rms = sqrt((sumq - (avdu-avduold) ** 2) / (m-1))
        u1m = sumu1/m
        u2m = sumu2/m
        sumu1 = 0
        sumu2 = 0
        sumdu = 0
        sumq = 0
    else:
        Checkrms = False
    return Checkrms

The data in above post , m=50. ADC runs at 1PLC, AZ mode.

if m=2, the data jumping alot looks like as:

Code: [Select]

0    1566.935    1567.547      -0.613       -0.0026     12      0.00125
0    1566.951    1567.540      -0.589       -0.0017     20      0.00090
0    1566.958    1567.544      -0.585       -0.0012      6      0.00055
0    1566.955    1567.550      -0.595       -0.0019     11      0.00068
0    1566.926    1567.534      -0.609       -0.0027      6      0.00084
0    1566.939    1567.518      -0.579       -0.0012     12      0.00142
0    1566.949    1567.518      -0.569       -0.0007      6      0.00060
0    1566.947    1567.518      -0.571       -0.0009     12      0.00028
0    1566.923    1567.525      -0.602       -0.0019    -26      0.00102
0    1566.914    1567.530      -0.616       -0.0028      4      0.00096
0    1566.918    1567.518      -0.601       -0.0022     -9      0.00065
0    1566.918    1567.514      -0.595       -0.0018     10      0.00041
0    1566.911    1567.520      -0.609       -0.0023      5      0.00055
0    1566.940    1567.524      -0.584       -0.0014      4      0.00098
0    1566.946    1567.537      -0.591       -0.0014      9      0.00037
0    1566.945    1567.532      -0.587       -0.0019     12      0.00069
0    1566.953    1567.528      -0.575       -0.0007     15      0.00123
0    1566.947    1567.537      -0.589       -0.0017     -1      0.00098
0    1566.938    1567.533      -0.595       -0.0018     -2      0.00011
0    1566.945    1567.532      -0.587       -0.0015      1      0.00026


[wanghar]

@iMo  Yes the small value res should be there and close to CPLD chip. I've placed them on some I/Os, but not all — for example, the MUX0-2 pins. I'll probably add them in the next revision.

[Kleinstein]

The OP-amps are connected as voltage followers. To be closer to the integrator it should be more as a slowed down inverting amplifier.
As a second point the switches are all working between 2 voltages and not with one input at ground. So the conditions for the switches would be quite different than in the NU180.
I would at least connect one, maybe 2 connected more as current steering and the amplifier as inverting amplifier.

[Zondar]

Actually, the op-amps are intended to act as charge-integrating amplifiers. The resistor is to be 1M Ohms or larger, just to provide weak DC feedback for low-frequency stabilization.

The switches are operated between VDD and VSS, where VSS is intended to be -0.6V as in A3 (i.e., via Pin 52, "GND2"). However, VSS can be connected to normal ground if desired (or even +/- 2.5V).

I'm not sure it will work, but the idea is to operate the switch via a waveform generator and alternately integrate charge towards VDD and VSS. The duty cycle can then be adjusted to achieve balance.

Edit: I did have the op-amps hooked up wrong! Thanks for catching that. Here it is again:

[Zondar]

I heard back from Texas Components. They do seem very responsive.

Three different Vishay part numbers, some through-hole and some surface mount (Z201, VSM2010, FRSM2512), were offered for 80k resistors (0.01%, 0.2ppm). Prices varied a little, but let's say that 10-12 parts costs the same as 25 of the "maybe in 6 months" parts.

I also inquired about a 9 resistor network (1x 50, 4x 80, 4x 20), but it will take a little longer to work out if that's practical.

[Kleinstein]

To get some charge integration the input signal must go to the inverting input. So the OP-amp part needs a change.

As shown the switches would see quite different conditions as in the 3458. The voltage at the switch pins would change quite a bit. In the 3458 all 3 pins from the switches would still stay close to ground.
If one wants some +-references (like +-5 V or +-7 V) one should consider including the inverter for the reference.
One may even get away with zero voltage as input.


For testing the resistors, a set of 4 resitors in a ring would be a minimalistic bridge. Just the resistors connected to one point would help relatively little.

The 20 K resistors for the fast part would be far less critical. They would only effect the really fast part and this already depends on the external 5 ppm/K TC resistor R181 and 2 rel. simple OP-amps.

[Zondar]

I did correct the schematic and layout, or is there some other error in it still? Also, I know it's not the same. I just wanted something a little dynamic to look at.

Since 0.2 ppm/C 80k's are available after all (at a price), I don't think there's much purpose left in trying to sort 10ppm resistors. I'll certainly use cheaper resistors for the first trials, though.

My focus will be back on the NU180 board now, since I have to redo it from scratch yet again (resistor footprints have to be made larger, which messes everything up).

I noted the noise tests by wanghar, which are a little startling. I wonder which parameters in the spec sheets, if any, reflect noise? I suppose bandwidth differences are a possibility. The TMUX version he tried quotes 200 MHz under certain conditions. The ADG version quotes 160 MHz, -3dB. The 74LV4053 doesn't seem to list that, but its on-resistance is far higher, so one presumes that the bandwidth is lower. It's not clear to me what the bandwidth needs to be, but when using multi-kilo-Ohm resistors, it can't be too high. What other parameters might indicate noisy vs. quiet switches?

The 74LV4053 isn't as appropriate for this project since the control scheme is different, the on-resistance is higher than any of U180's switches, and 4 channels would be better if possible. I'm still open to suggestions, including a little higher on-resistance ones if there are other benefits. I had a quick look at the TS3A5018 10-Ω chip, for example. It has higher on-resistance but other parameters are a little better.

[Zondar]

More questions for the experts:

* Any soldering has to be brutal on finely-tuned resistors. So for best ability to move resistors from one board to another without damage, which is preferred, through-hole or SMD?

* Vishay's VSM and FRSM line of resistors have virtually identical specs, yet are distinctly different. Any experience or recommendations?

Thanks.

[Kleinstein]

The TS3A5018 does not work at all: the 4 switches have a common control signal. So not 4 x SPDT but  1 x 4PDT.

For the test board the amplifier part looks OK, but the swiches are still a different configuration and thus testing different parameters.


I see 4 ways the switches contribute to the noise:
1) Jitter or timing noise
2) the capacitance at the COM pin. This capacitane is connected and disconnected from the integrator many times during run up. This gives so called reset noise, which is kind of the equvalent to the Johnson resistor noise but for a swit ched capacitor circuit. So here the capacitance is the relevent parameter. This part can be calculated - I still miss the frequency of Wanghar's test, but I don't think this part would explain the whole difference.
This noise contribution is proportional to the square root of the capacitance.
3) the same capacitance also works as a charge pump. It causes extra charge from residual voltage at the integrator vs ground for the switches. Here the added noise depends on the noise in the residual voltage. In addition to noise, this mechanism may also contribute to the INL.
This noise contribution is proportional to the capacitance.
4) The net charge injection of a switching cycle can depend on the supply voltage and CM voltage at the switch. This gives some coupling from supply voltage noise and again possible INL contributions from variations in the supply the correlate with the signal. I have tested the supply sensitivity with my ADC and it is not a big factor. The charge injection quoted in the datasheets in the turn off part and this is different from the sum over the on/off cycle.

Most of the parameters are not given in the datasheets. It is mainly the capacitance that is usually noted at least for some cases. What matters is the off state capacitance for the COM terminal, that may not be conventional measurable for the SPDT switches without an enable. In addition one could argue to add some of the gate capacitance. The BW reflects the R_on to capacitance ratio but would not be directly relevant. It can be a first hint on the capacitance.

The R_on of the LV4053 is not that far off from the U180 original. The meaured resistance was 14 ohms for the input path and the ref. paths should be 8/5 of this or 22.4 ohm. Thus just at the typical 23 ohm for the LV4053. For slightly better (still not ideal, but acceptable) matching one would have 2 switches in parallel for the input. Because of the higher resistance the LV4053 would however want the supply voltage modulation as in the original U180. One might still get way with a selected resistor instead of a dummy switch. Without a change in the resistors on the A3 bord one would need 2 of the LV4053 switches for the dummy.
The LV4053 has 3 switches per chip, but the configuartion is suiteable (just the unused enable pin).
One would likely need 3 x LV4053 and 2 lower resitance switches (e.g. ADG733/734) for the fast part.

[wanghar]

@Kleinstein  One my board（2.2nF  integ. Capacitor）the Vrefs input should be switched at the frequency of 45.9KHz. This frequency is much slower compared to the 3458A's 330KHz.

[Kleinstein]

One could do a noise test also with a 2nd higher modulation frequency. There is no need to change the integrator capacitor for this (larger capacitor is OK, just a missed opportunity to get even less final charge noise, which is anyway small).
The comparison of 2 (maybe 3) frequencies could tell how much of the noise is frequency dependent and maybe even decide between a more ~f or ~ sqrt(f) part.

If the switch noise is already visible with 50 kHz modulation it would be quite a bit worse in the 3458 with 330 kHz . Double the active switches for the references makes things worse for the capacitive parts but it helps with the jitter with a faktor sqrt(1/2).

I also did a few tests with different switches (HC4053 from Ti and ST) and max4053. These have less / comparable capacitance and still showed more noise. The extra noise there is very likely from jitter.

[wanghar]

Quick test result @ 153.8KHz modulation frequency, observed a noticeable jump in noise. The last column is noise_rms values in mV:

74LV4053：

Code: [Select]

0  184933.967  184934.778      -0.831       -0.0178     58      0.00096
0  184933.975  184934.789      -0.755       -0.0179    -63      0.00102
0  184933.963  184934.774      -0.841       -0.0178     85      0.00105
0  184933.951  184934.763      -0.790       -0.0179    -62      0.00112
0  184933.956  184934.765      -0.838       -0.0177     62      0.00094
0  184933.970  184934.778      -0.760       -0.0177    -46      0.00104
0  184933.959  184934.768      -0.792       -0.0177     26      0.00104
0  184933.959  184934.774      -0.816       -0.0180    -36      0.00104
0  184933.960  184934.776      -0.807       -0.0180     20      0.00085
0  184933.956  184934.772      -0.777       -0.0180    -10      0.00103

ADG733:

Code: [Select]

0  314516.297  314516.542      -0.277        0.0059    -20      0.00130
0  314516.290  314516.549      -0.213        0.0052     12      0.00134
0  314516.294  314516.541      -0.269        0.0058     15      0.00130
0  314516.290  314516.549      -0.242        0.0053     -9      0.00114
0  314516.299  314516.555      -0.260        0.0054     10      0.00113
0  314516.298  314516.552      -0.254        0.0055    -21      0.00145
0  314516.308  314516.567      -0.276        0.0053     24      0.00146
0  314516.301  314516.568      -0.271        0.0049    -60      0.00131
0  314516.313  314516.572      -0.242        0.0053     76      0.00104
0  314516.311  314516.571      -0.329        0.0052     60      0.00139
With a 2.2nF integrating cap @ 153.8KHz, the Vpp on the cap is only 1.3V.  The integrator's efficiency is way too low . Try a 330pF or 470pF C0G tomorrow.

[Kleinstein]

The low voltage at the integrator output should not be an issue. 1.3 V are still not that little. The rundown part is uneffected by how fast the modulation during run-up is. I have done my test with higher frequencies (AFAIR up to some 350 kHz) also with a 2.2 nF integration capacitor. A smaller capacitor should not make a difference.

The quite a bit higher noise with 150 kHz moduation suggests that there is still some kind of jitter or switching related issue.
This not necessary only the switches but can also be the clock of FF with it's supply / decoupling.
Normal jitter and the anvoidable capacitive part should only go up like the square root of the frequency.
Some 400 nV of noise are expected for the non switch related sources. With the LV4053 at 50 kHz this would leave some 570 nV for the switch part. Multiplyed with square root 3 this would be around 1000 nV and together with 400 nV from the rest some 1070 nV, which is a bit more than observed, but still in the ballpark. Normally there should be less switching related noise with the LV4053.

[miro123]

The current hardware uses a 12-bit MCP3201 for residual voltage detection and a TI TLV3501 as the comparator. These two components were chosen mainly to balance 3.3V level matching and performance. Compared to the μC version, programming flexibility has been greatly improved.

The design look like lift up to more modern design.
Few thinks still bother me
1. The slope amplifier output generates voltages at TP4 within the range of (-0.6....+1.2V) - the comparator is powered at 0...3V3. Kleinstein is doing the same. Do I miss something?
2. Is there a better switches choice than 4053? - Does somebody tried some modern low leakage low charge injection SDPTs from TI or AD?
3. Does NE5534 look like thermonuclear reactor on FLIR images - there are plenty of other options - e.g I have some OPA209 at my shelf. They cost $1.4855 for single pcs. and $1.355 for 10+ at lcsc.com