EEVblog  

I see the problem in the schematic. The output stage uses a resistor from the output of the op-amp to the output pin. This R, plus a capacitive load adds a high frequency pole which causes excess phase shift and thus oscillation of the output stage. Move that feedback cap to the op-amp output instead of the board output and increase it a bit, I'd guess 100-1000pF. This was a latent design problem. Spice the circuit and you will see the marginal stability of that stage with a C load. A marginal design is like a blind man walking along the edge of a cliff. If he drifts one way, no problem. If he drifts the other way a tiny bit he falls off. The parameter that is changing slightly is the op-amp phase margin, changing slightly from batch to batch. Fix the design, don't select parts. Hope this helps. Dave Erickson

Send the chips to electronupdate for him to decapsulate and compare.

R15 is upside down, electrons have fell out

Not that rare! After 35 years of EE design, I can tell you that there are no equivalent parts. Even parts from the same manufacturer can change over time and lead to this type of problem. High volume production has to test for all of the 'corner cases' or you WILL always have gottcha's! Once its in the customers hands those gottcha's cost big bucks to fix. This is one of the significant problems with 3rd party manufacturing. I've got lot of stories about problems similar, some are significantly more weird than this.

Murphy should be paying rent with all the time he spends in the lab.

Are you going to add a capacitive load to your production test jig?

I think not having a capacitor between V- and GND or V+ and GND is actually a design error, as it puts the complete AC output current through the virtual ground supply opamp.

There’s no controlled impedance between the output of that op-amp and the power lines, and when you connect capacitive loads the impedance goes negative enough. The chopper op-amps are like impedance converters in this circuit. You can make it oscillate in ltspice too. Basically: you were lucky with the original parts. The circuit is under-designed. It’s not unlike designing stuff with discrete transistors and being lucky with beta. Or with differential in-amps and being lucky on linearity (been there, done that, still get a tear in my eye when I think of the brouhaha that resulted from that on mission of mine). I doubt it very much that the chip is fake, but even if it was: it’s certainly good enough to do its job most of the time, so the design is sensitive to whatever is different. And it can be cheaply made insensitive. Thus my assertion of under-design.

I have the same issue on both of mine, serial numbers in the 4100 range from November 2016! I bought them with the intention of looking at measurements on a scope but was very disappointed to see 100mV of noise on the signal which made it totally unusable. Connecting it to a DVM it worked perfectly so I put up with it thinking it was just a poor design and not intended for connecting to a scope. Now I know different and appreciate you doing a vid on this Dave with a great explanation of the cause. I should have queried it back in 2016!

My favorite eevblog in a while. Love a good mystery!

I think it would be really interesting to see these parts tested in isolation.

This really links in conceptually with your videos on what to charge for products, considering BoM, shipping, admin...and now considering potential product recalls and reworks.

This is great! You finally get a repair video that is a bit of a challenge and it's your own product! Kudos! Wonderful video!

This is a great video Dave. As an electronics enthusiast, it's oddly discouraging however as I've now had a glimpse at a level of electronics troubleshooting that I'm sure i'll never reach. Really cool to hear you walk through the progression of troubleshooting and techniques to isolate the problem. Thanks again.

It always makes me laugh out loud seeing your hands move in reflections. I love seeing how passionate you are about electronics.

Dave I really enjoy your videos. There is a slight amount of ego to your presentation but I rather enjoy it and under stand where it comes from. I was a professional automotive technician for 26 years and under stand the swagger and have come to appreciate quality workmanship and pride of doing it. I have had to switch professions into electronics and have learned more in 1 year from you than years in school. Thanks and keep the videos and rants coming...

First-world lab problem: too many scopes :D

This video is a masterclass example of general troubleshooting processes. It was very interesting, and frankly educational, to watch Dave go through the steps. I'm going to bookmark this one, this is great tutorial material

I used to work for ON Semiconductor. Do yourself a favor, Dave, and stick with the Texas Instruments part.

You should defenetly call in to Mouser / digikey or even call in On- Semiconductor / fairchild to let them know that this problem exists. This is an fatal production error that they realy should correct

Brilliant detective-work to track-down this fault. Thankyou for sharing!

Thanks, Dave. I really enjoy these videos. I'm not a trained EE, just a lowly systems integrator (which makes me little more than a glorified technician.) I hate to say that (at least where I work in the defense sector) this level of debug is starting to become a bit of a lost art. I work with a lot of EEs who I'm not certain have ever built anything outside of MultiSim/OrCAD/whatever. All well and good when everything works as expected, but they tend to flop and twitch when things go awry. Videos like this, and your test equipment and basic electronics principles videos have really helped me professionally. I appreciate all that you're doing and always look forward to the next one.

A really interesting issue. I just checked my own stock, and I have some LMV321M5 (Nat Semi, marked A13 like your TI ones - these were purchased 2010 prior to the acquisition), and some LMV321WG (Diodes Inc, marked BX0jD). I might just swap in the Diodes Inc part on my uCurrent Gold (#00624, with LMV321 marked RC1F) and try this same test with my 121GW.

This is the reason why big companies in the automotive market like Continental or Bosch require a customer nofication from the supplier whenever something is changing in the device; and I'm not talking about the design or the technology process but even a change in a transistor dimensioning counts. And they usually want to have proof that the parts that they will receive will still be in specification. As an IC test engineer in the automotive industry this is quite a spill you have Dave :)

Good problem and troubleshooting Dave! In my younger days as a summer intern at an electronics company, we had a mysterious problem in a discrete uvolt analog to frequency converter that was affected by too much lead in the silkscreen paint that caused leakage paths between components! It taught me that you have to take everything into consideration and not discard what you might believe could never be the problem or, at least contributing to the problem.

Interesting video! Looking forward to your tests of the "similar" parts.

Thank you for absolutely brilliant peace of information.

Dave's voice hit Top 'C' at least three times during that first segment. You can tell this one was a real doozy to solve!

*Interesting video with added value!*

Learned alot from this video thanks for that break down it was legendary!

This was very interesting. Great video!

Very informative even though I have suffered seemingly harmless production substitutions myself. It’s terrible with DRAM and flash where the manufacturers do buyouts and rebadges monthly

A capacitor from the negative rail to the +ve input of the voltage divider opamp is normal, taken from one end of the decoupling cap, to avoid unity gain stability problems. So is using wide short tracks for decoupling capacitor connections for the same issue (unity gain stability) 15:56 shows the teeniest of tracks which could have been a lot shorter and wider. Not saying that would cure it, just that it's normal practice to help keep Murphy away!

I know some are (they are often pilloried relentlessly for their forthright honesty) , but ... Iimagine if car makers, politicians, PRESStitutes, public servants, fakebook, computer companies, drug companies and medical administrators were *ALL* this honest. What a wonderful place the world would be!

Once I was repairing LCD monitor where backlight was not working. But CCFLs were ok, high voltage PSU also but it's driver was cutting it off after one second, probably detected that CCFLs are worn out a little bit. At some point I hooked up my multimeter to HV output, and bingo, backlight was just working with my leads attached to the output. And these leads have some capacitance. Soldered 1pF 1kV ceramic cap to the output and monitor fixed...

GREAT JOB

Dave, have you considered just adding bypass capacitors between your virtual ground & the power pins on the 321? Seems to me that that would stabilise it. (Lowering your voltage divider resistors from 100K might work too, but of course that'd increase power consumption.)

Can you remove the oscillation with decoupling capacitance between the virtual ground and the + & - rail ? maybe pop a couple of 1uF in ?

Hi Dave, something I had show up on multiple occasions. Always attributed to stray capacitance that the fab house doesn't gurantee equal amounts of board capacitance in the order of 4pF. I got boards that exhibit 50 to 80 pF that I have to compensate for in design. Need to add as well, chip manufacturer designs components to specs valid in house which are at times harder to mimic in test. The closer the specs to acceptable margins means probably out of spec performance.

BTTF reference in your multimeter part numbers? YESSSS.

I read on analog.com that when using a virtual ground you need to place a capacitor in parallel with the output of the resistor divider to preserve power supply rejection. So maybe whacking a cap across R7 would fix the problem.

Decap the chips!

Had a very similar problem with the MAX4239 oscillating at low / no differential signal in a similar current sense circuit which previously worked fine (albeit single supply rail). After many approaches we ended up changing the opamp. Would be interested to know if replacing the MAX4239s with older also addressed the problem

I hope you have fixed this. 3:30 - The protection resistor R8 should be at the output pin, not before the feedback loop, right? Here the output load can feed in unimpeded noise through the C4.

Got really interesting here, because it is a real life problem. Sometimes in my job it is just hard to find the time to get in this deep with a problem.

“Rings a bell”. If I recall, one Fluke meter has 4 nF input capacitance on the low ranges. Caused some surprises until I figured it out. Motorola MJ802 used to be one of our favorite robust transistors and then it suddenly changed to be anything but. After various queries, we found that Motorola had moved the manufacturing to another bigger plant. Bigger wafers and other changes in the process. BAH! We went around and were promised great robustness by a company called Solitron for their version of 2N3771 or something of that number series. We tried and they failed on a short notice. When we pursued that issue, Solitron engineer eventually said that that we must stop using their transistors, because our application with 1 to 20 Hz pulsing caused thermal cycling stresses on the chip to base bond (soft solder connection). Would be fine with DC or over 100 Hz pulsing. Thank you very much for that clarification! We never found a good replacement for the old MJ802, although MJ15015 survived a little better than any of the other substitutions.

Time for “Approve Supplier” specification paperwork and new test procedure with capacitive load

Interesting vid. Part swapping is convincing evidence of the oscillation correlation but does not identify the underlying oscillation mode. For example, is the LM321 actually oscillating or causing the Maxim amps to oscillate through the Vgnd/load interaction? Your comments suggest that the Vgnd circuit action is trivial. Maybe the DC action is trivial, but the AC action is certainly not. Would be very interesting to learn the underlying oscillation mode. Thx.

Dave, could you explain why your DMM and the Keithley have such a "high" reactive load? Does it have a purpose? A side effect from something desirable?

For best results, one should try to buy from the producer. Also, one should try to select a producer that continuously supports it's ICs. TI has an online store and they do seem to care about quality preservation over shipment.

On the output op-amp, the capacitor should go from the op-amp's output pin to the inverting input. This provides a negative feedback that is not delayed by capacitive loads.

I have always trusted TI, Maxim and Analog Devices over other companies, i feel like they have met or exceeded their specs more often than even LT

Screwups happen. Digi-Key were kind enough to send me a new batch of ICs at no cost when they screwed up my old order and I got some async binary counters instead of 8-bit transceivers, and with no need to return the incorrect parts, since I was working in small quantity and the shipping was worth more than the chips themselves.

The LMV321M5 is the former National Semiconductor IC while the LMV321IDBVR is the TI one. They may be even still made at different plants because of effects exactly like you found. Imagine TI buying ON Semi and then only making oscillating LMV321 chips at the ON Semi plant, that would piss some engineers right off. So at one point TI may officially declare the LMV321M5 obsolete and introduce the LMV32IDBVR as a successor.

Could you share where to get that the variable capacitor and variable precision resistor devices that you've been using?

You should remove battery before using hot air. Otherwise it may pop right into your face :)

Noisy boy... Dave's been watching AvE, I'm calling it right now.

Hello, I will give any comments to anyone rarely...but this video just demonstrated the real engineers cabability to give feedback to customer in any situations.. I have noticed that most of the time customers at (carribean area)industrial level are really happy when i tell bad news to them... Some times i have just told that this is fuck up, but i cannot fix it now because it is not my scope and they are still like heaven... This is because lot of service engineers dont have guts to point up other issues they have encounted..anyway customer is pleased that they know issue an can later fix it...

I'd like to state another thing: According to your schematic your feedback path in the OpAmp-amplifier has 9kOhm||10pF resulting in a 1.8MHz cutoff frequency. Notably, your oscillations are in the range of that frequency!In your µCurrent specs a -3dB-bandwidth of 300kHz. So you may additionally think about increasing the capacity from 10pF to 33 or 47pF. I am quite sure that would add additional stability and it may make the batches better comparable: Your 10pF are effectively increased to 12 to 15pF by the pads and line capacities - a quite big range that is significantly reduced by a larger cap.

Stuff like this makes me wonder how any consumer electronics ever actually work out in the real world. No way you could possibly test everything, every little crazy combination, etc... wow

20:04 The Mouser datasheet seems to be / link to the Fairchild datasheet of the LMV321. The OnSemi datasheet (from their website) does indeed have the list of markings on page 12. www.onsemi.com/pub/Collateral/LMV321-D.PDF

O'Toole's law says Murphy is an optimist. When dealing with electronics even the best laid plans of mice and men get fouled up by Murphy. About 15 years ago we had automated controls on our pumps installed. All so that we could operate them from a remote computer. Every so often we had a pump shut down for no known reason. All of our equipment was designed to be stable even in the event of an EMP event including hardened CPUs designed for use in space. And everything was opto isolated so that nothing should get by. Wrong! It took two weeks for it to show up on the chart recorder but finally we caught the little spike that Murphy was throwing at us. It was getting by the batteries, the filter capacitors in the power supplies, the opto isolators, and triggering a fault inside the controller.

Well, it is not the only problem with oscillation, as I discovered and fixed: www.eevblog.com/forum/testgear/problem-with-ucurrent-gold/ And in my uCurrent is a RC1F opamp.

Isn't the avoidance of ringing/oscillation part of op amp 101?

So the lesson of the day is keep good records.

Are you going to change the test jig/procedure?

There is no surprise that Fairchild part oscillates. It has a big hunch on it's Frequency Response vs C chart. From my experience it is a clear sign that it will oscillate on that frequency under some conditions. And TI datasheet has no hunches on it's Gain vs Capacitive Load chart.

Looking at the schematic, I would be wary of the high impedance present at pin 1 (the non-inverting input) of the LMV321. I would have liked to see the scope examine the 321's output at pin 4 when the MAX4239 opamps were oscillating to see whether the oscillations were present there as well. If not, it's difficult to see how the different 321s could cause the oscillations. If oscillations were present on the 321, then my first instinct would be to decouple pin 1 to each of the supply rails to reduce the ac impedance at that point.

Hmm, so whether or not virtual ground generator is the "culprit", something is clearly able to feed through the capacitive coupling, since the error depends on a capacitive load on the µCurrent outputs. So this could reasonably be narrowed down to whatever is connected to the outputs. What sticks out to me is the connection of R8, the 270 ohm resistor. Firstly, this resistor makes the output gain 10.27 instead of 10. But that's at the amp output. The 270 ohm resistor forms a voltage divider and the meter output is at the 10:1 tap. The precision loss from a 1% 270 ohm resistor would be below the 0.05% when taking into account the full scale of the resistor network. So far so good. And none of that would create any oscillation issues anyway, just a small measurement error. The second thing that sticks out in relation to R8 is the connection of C4. The MAX4239 datasheet calls for a capacitor in series with the feedback element. As it stands, the feedback element is R13+R14+R8 whereas the capacitor is only across R13+R14. This could have two effects: 1) Not limiting the bandwidth of the feedback properly. 2) Somehow feeding interference back onto the input with whatever effects that might have. Regardless, the capacitor technically isn't connected as advised in the datasheet. Things I'd try: 1) Move the C4 connection directly to the output of the MAX4239. 2) Move the R13+R14 feedback connection directly to the output of the MAX4239 as well. R8 would now be in series with the output but no longer part of the feedback loop. This should only be a problem with a very low impedance load which might sag the output appreciably.. 3) Come to think to think of it, 2x 100n for the whole circuit seems a bit anemic given the symptoms. Might want to increase that a bit and see if that stabilizes the circuit. 4) Might also want to add decoupling to the virtual ground as well as some people have suggested.

Why is there a 100nF capacitor between V+ and V- in two different places in the schematic (C1 and C2)? Is the second one just a duplicate and thus is it redundant? Thank you for clarifying!

So this shows as out of stock on your store page. Is that because the problem you identified is being fixed? Any idea when this will become available?

I am so electronically illiterate but for whatever reason, I watched the whole video... It's like the adult version of Dora the Explorer. Every time you say AHA! I feel like I'm about to be enlightened and still, I feel like im in the dark..

This op-amp uses Fairchild top marking, according to the binary date code scheme was manufactured in 2006 work-week 16-23. The confusion is because Fairchild was adquired by ON-Semiconductor, the making scheme between both companies were different and current ON-Semicondutor documentation in the website is a mess.

Wow @12:10 "Stable as Uluru" :)

What is the device, a microammeter to be connected in series? Is there a microvoltmeter available instead? To trace small currents in circuits, detecting voltage drop in pcb traces .

Just to add to your serial number database: in Nov 2016 I reported an issue where "Short" reported a -5.5 mV offset (on one of your multimeters!). My serial number is #004118. With zero response, I "filed" this dud somewhere. I'll find it and e-mail you direct to discuss what to do about it

Murphy got you always :D

At the return path of U2: LMV321, pin 4 to pin 3. Would adding a compensation i.e. R&C pole & zero do the job?

Hi, i like this little capacitance-"box". Thinking about building one or by one if it is not too expensive. Is this type available to buy somwhere?

why do you have such small paths? Would it not be sensible to have thicker lines between the components when you have the space on the board?

I hate SMD :) Exactly same story here few weeks ago with DAC chip.... Tiny package, only one letter difference, and bang! The original driver FPGA (with no documentation and copy of the firmware) is became a super-fancy, super-hot egg cooker...

I solved this mystery 3 years ago. The output stage is oscillating because of the 270 ohm resistor plus a capacitive load causes too much phase shift. It is marginally stable, so some work, some oscillate. See my reply and the design fix below.

What does a capacitive load say to an opamp? hasta la visat, phase margin

What changes in circuitry would be needed to compensate for all chip variance? if possible

If you are going to make a video showing the problem by directly adding a capacitive load to the output, can you also test when the TI part starts to oscillate?

Daiyve, your next video will be a detailed account of your losses and possible damages to your very fine reputation. Or, you could just load up the Toyota and head out to PIne Gap, Australia. Lots of crap oscillating out there!

It's a well known fact that "identical parts" can vary a lot between manufacturers. A while back, I did a comparison of quad op amps. ee.david.promo/OpampShootout/ There was a surprisingly large difference between an STMicro and a TI LM2902 that were supposed to be identical...

Seems like the battery has too high of an internal resistance!

I had a huge oscillation issue with ON LM324 in a power supply.

Hold on, datasheet doesn't tell you what what the variant of the chip does? How else are you supposed to know? Or is it normal to have fingers crossed getting a reel that it'll work, can't believe that.

Check the ESR ratings have a similar situation. On your new part.

Where can I buy the "µCurrent"? (preferred in Europe)

Will you contact Fairchild to see what they say about the part in your circuit? After all, they can't fix it if they don't know about the issue.

silly question, but you do not sit while working at the bench?

Dave - i recently bought one no: 5692 How do I know if this has the potential issue ? (can you confirm what batch numbers could be effected)

Stable as Uluru - cool, didn't know Aussies actually said that.

I have come across opamps that, in a dual rail config, were not happy with a single decoupling cap across the rails, needing one cap per rail. Could it be something similar here?

Why not just decouple your voltage divider to stop it feeding back into the virtual ground amp? After all, the tiniest noise on that power rail will get into the input... it's a DC rail generator, so no frequency response is needed. I guess a 10nF would do it.

Sometimes a squeak, is much more than a peek.

15:00 The resistors have been put backwards!