Why are these 7812 voltage regulators dying?

[ballen]

Over the past 7 years I have repaired a handful of Heidenhain VRZ 7xx Digital Read Outs (DROs) made in the 1970s.  These are used on machine tools, to indicate the position of the axes, typically with a precision of about 5 microns (0.0002").  They are CMOS digital logic (flip flops, counters, clock) with a bunch of 7-segment LEDs and drivers and mechanical keypad switches.  It's well-designed German electronics, classy cast case, high quality connectors, nice circuit boards and layout, etc.

In almost all of these, the only problem after 40 years of daily use is in the power supply.  Sometimes it's the electrolytic or tantalum caps, but the most common failure I have seen is a failed LM7812 voltage regulator in a TO-3 case.  This is fed 18 VDC and providing about 0.8amps of current, so dissipating about 5W of heat.  The TO-3 is mounted on a good sized steel heat sink, but with an insulating washer and a layer of paint underneath.  The 18VDC is coming from a classic linear supply: transformer secondary, four diodes in a full wave bridge rectifier, 2200uF of electrolytic capacitor in parallel with a smaller capacitor.

The ones that I have repaired have all failed the same way, with the failed LM7812 output at about 2V DC.  My question: why are these failing?

At first, I thought it was simple overheating, and when I've replaced these I have added some thermal grease to improve the cooling and also modified the primary connection for 240VAC rather than the default 220VAC to drop the input voltage a bit.  But the truth is that even in the initial configuration, the 7812 wasn't too hot to touch, and it's supposed to current limit if it gets too hot. 

So I am wondering if something else (voltage transients, for example) is killing the 7812.  Any ideas about that?  Should I add a reverse polarity protection diode from input to ground, and/or a 20V zener at the input or a MOV on the primary (or secondary??) to snub input transients? 

For now I've replaced the failed regulator with a L7812CT, which is specd for 1.5A up to 35VDC input. Yes, I could replace the 7812 with a little DC-DC switching module that would run cooler, but I'm not convinced that this would be more reliable over the next 40 or 50 years.

Cheers,
   Bruce

[mariush]

Maybe add a diode from output to input.

As alternative to 7812 maybe check out LM1085 or *1085 (as others make it as well), it's a 3A linear regulator with maximum 27-29v input voltage (but there are some versions that are limited to 18v so pay attention).

[T3sl4co1l]

The diode trick helps when there are multiple loads on the common unregulated supply.  If this is the only load, it shouldn't make a difference (you can still try it, it won't hurt anything).

Transients can include mains noise and surge, and stuff from the output side (ESD?).  Mains shouldn't be much, as the transformer and capacitor act as an effective filter.  Well, what kind of transformer is it, actually?  If it's toroidal say, it might still be able to source quite a lot of current, even for short durations.  I suppose that could be enough to push the capacitor voltage over the regulator's rating (the capacitor isn't ideal, but has some internal resistance which simply drops voltage in proportion to current flow).  The fix for this would be a TVS diode from GND to 18VDC, rated for maximum nominal voltage (so, 18-24V say), and with a maximum clamping voltage below the regulator's rating (so, SMCJ18A is probably good enough).

Tim

[bdunham7]

If replacing the regulator works and they aren't running hot, just repair them and maybe verify that the filter caps are good and the voltage and load are all correct.  It's possible that the batch of semiconductors used originally just have a 40 year life.  If you get a repeat failure, then perhaps dig deeper.

[TheMG]

It could be thermal cycling, 40 years of switching it on off on off (assuming they're not left on 24/7), even if kept relatively cool there is still variations in temperature and over time those thermal cycles can cause failures.

[bobbydazzler]

How large is the output capacitor for the 7812?  >6v output with > 10uf capacitor output can stress the regulator when input is shorted to ground.  Otherwise I'd say 40 years of thermal cycling is probably what killed it, I've noticed in a few atari 2600 repair videos they also need to change the 7805 regulator a lot.
To make it more reliable in the future I would definitely put a protection diode from output to input.  You could also put another in series with input(rated for 3-5amps here) as well as from ground to input for extra protection.

[Bud]

Yes, I could replace the 7812 with a little DC-DC switching module that would run cooler, but I'm not convinced that this would be more reliable over the next 40 or 50 years

So the equipment has been running for 40 years and the the plan is to run it for another 50  years ?     

[David Hess]

If you have a microscope, cut the top off of the failed TO-3 packaged 7812s and inspect them.

What is the condition of the input capacitors where the 7812s failed?  If the bulk input capacitor went open after decades, then a large change in load current can allow the leakage inductance of the power transformer to generate a voltage pulse high enough to destroy the regulator.

[amyk]

If you have a microscope, cut the top off of the failed TO-3 packaged 7812s and inspect them.

There is another German member here who might even help you with that.

[Zero999]

You could try adding a series resistor and capacitor, to filter out spikes, but care needs to be taken to ensure there's still enough voltage headroom.

[ballen]

Thanks for all the helpful replies.  I'm impressed by the amount of expertise and knowledge exhibited here.

Mariush: a protection diode from output to input makes sense.  In looking at the datasheet again, I see a few cautions to protect again reverse biasing output wrt input.

T3sl4co1l: the transformer has three secondaries. One is for an unregulated 9V regulated 5V circuit, the second (this circuit) is for unregulated 18V regulated 12V, and the third is for unregulated 4.5-4.9VDC which drives the LED display and where one can toggle two different levels of brightness.  There is no other circuit sharing this secondary.  The tranformer is a conventional one, not torriodal.  I didn't know about the existence of TVS diodes, the SMCJ18A is a great way to stop transients.  Response time is picoseconds, quite amazing.

bdunham7: Yes, I've already replaced the regulator, and I think it probably will last another 40 years, see below.

TheMG: Regarding thermal cycling, see the failure mode below.

bobbydazzler: output cap is 5.6uF in parallel with lots of little bypass ones.

Bud: these DROs work extremely well, and I expect that lots of them (and the machine tools that they are attached to) will still be around and working when we are gone.

David Hess: I did what you suggested.  I made a mounting post, put the transistor on the lathe, and carefully turned off the top.  Then took some photos with a USB microscope which I bought for one of my kids a few years ago. See photos below.  You can see that a corner of the chip (attached to the input pin) has cracked.  I am fairly confident that this did not happen when I removed the top of the TO-3.

amyk: I saw your comment too late to take advantage of it.  But it only took an hour to make the mounting post, turn off the top, and take photos.  Will be quicker next time (:-).

Comments welcome, about why the chip failed like this.

PS: I am following the instructions to embed photos, but it doesn't seem to work. What am I doing wrong?  Photos can be seen below, but I wanted to embed them larger and with text around them.

Cheers,
    Bruce

Here I am making the mounting fixture

" alt="" class="bbc_img" />



Here it is in the lathe, getting ready to remove the top



Pictures of the 7812 chip.  Broken corner is by the input wire

[T3sl4co1l]

Wow, cracked right off!

The inline attachments thing is broke and we're not sure why (more precisely: it works but corrupts images, so it's been disabled).  There is an alternative: right-click the thumbnail, copy link (or expand and copy image link) and paste it between [ img ] tags.  It works like any other random hosted image.

The funny thing about the TVS response time is, at the junction itself, it responds that fast.  But between the low impedance of the junction, the high impedance of the leads (they're short, but at ~ps timescales, we're talking a wavefront of sub-mm length -- everything matters!), and whatever the circuit is doing, means you'll see at best some nanoseconds.  But surges are in the microseconds, still more than plenty; even the fastest high voltage electrical transients (ESD, EFT) are in the single nanoseconds range.  (A direct hit to a properly wired signal line sized TVS, might have a peak voltage of 20-30V, out of an initial 8kV or so -- not a bad ratio for a tiny little chip!)

Tim

[Runco990]

I had a piece of gear a few years ago.  EVERY SINGLE 12V regulator died in it, one after the other.  Replaced without any additional problems.

The one thing the old regulators had in common.... same brand and date of manufacture.  Perhaps a run that had some flaw.

[bobbydazzler]

Possible it was mounted too firmly to the heatsink, eventually cracking it like that.  Could also have been from repeated thermal cycling but it doesn't sound like it gets super hot in operation. 
Preventing future failures could mean not screwing it in too tight to the heatsink(did they seem to be screwed in super tight?), and soldering the pins in after you've attached it to the heatsink or at least reflow the solder joints after the heatsink's attached.

[T3sl4co1l]

Maybe noteworthy that really old stuff (50s-60s) can have poor encapsulation and such.  Which may include welded metal types (like TO-18, 39 and the TO-3 of this thread) if they used compounds for die attach, encapsulation/passivation, or just fill the can with goo to take up space, or absorb possible moisture, etc.  Materials and processes continued to improve through the 70s, so you don't see a lot of failed components from there.

Like, people restoring minicomputers and mainframes from the 60s, the things either have a lot of metal-can semis that are basically okay, but the carbon composition resistors say are drifting; or early ICs that are decaying.  A bit better from the 70s (e.g. Xerox Altos, HP workstations, etc.), things work for the most part but you'll often have to track down faulty chips.  And, I don't think much late 70s/80s+ stuff tends to fail?  Aside from obvious, gross / environmental problems like mechanical and corrosion damage.

So, it could be problems along that sort of thing.  Maybe it's just thermal cycling, maybe just bad luck, who knows.  I'm not real clear if that one is soldered or glued (if it's glue, it's silver-loaded epoxy or something like that), but probably soldered.  So it should be reliable.

Tim

[David Hess]

Possible it was mounted too firmly to the heatsink, eventually cracking it like that.  Could also have been from repeated thermal cycling but it doesn't sound like it gets super hot in operation. 
Preventing future failures could mean not screwing it in too tight to the heatsink(did they seem to be screwed in super tight?), and soldering the pins in after you've attached it to the heatsink or at least reflow the solder joints after the heatsink's attached.

Improper mounting can definitely cause that sort of failure.

[magic]

You can see that a corner of the chip (attached to the input pin) has cracked.  I am fairly confident that this did not happen when I removed the top of the TO-3.

Can you check if it still behaves the same? I'm somewhat surprised that there was any output whatsoever.

I wonder if it's possible to chip off some other corner to make it pass full input voltage

[ballen]

Magic: you are right, it is now open circuit.  With Vin =18V I find Vout = 0.  But keep in mind that when I tested it before, it was in-circuit.  There are two other power supplies in the same box.  The 5V supply is not connected to the same circuitry as the 12V supply, but the LED 4.5-4.9V driving circuitry (transistor/resistor and inverters) is connected to the same chips. So in the absence of optical couplers it's quite likely that the 12V line was pulled up to 1.8V by that stuff.  I measured it with a high-impedance voltmeter, I didn't try to draw any current from it.

I am quite sure that my handling on the lathe didn't crack the chip, and there is no sign of any impact.  My support is machined flat, so screwing it down could only have pulled the base flat it if were already deformed.   Removing the cover might have taken welding or other stress off of the base, but again it was being retained on a flat surface.   So I'm going with the simplest assumption: the chip was cracked before I opened the case.

David Hess: the document from Freescale that you linked was very interesting.   The mounting surface  is an aluminium casting, with a layer of textured hammertone paint on it.  So it might be at the 0.004" per inch spec, but probably well over it.  Then there is a silicone insulator washer underneath the device.  Also I've taken the plastic cover off the 7812 and left the box running for a few hours now, and the part is just warm to the touch, perhaps 28-30C max.  So I now think that the mounting is indeed putting the part under stress, and when that TO-3 case was made in the 1970s, it was probably not as robust against mishandling as the current generation.  Bottom line, I will put the box onto the mill and skim cut the mounting face down to bare metal to ensure that there is a flat and heat-conductive surface.

bobbydazzler: The mounting screws pass through plastic bushings but have no belleville or spring washers.  If I can get some little belleville washers I'll use them.  My solder joints were done after mounting, but I bet in the original manufacturing they did it the other way around, since they were doing assembly not repair.

Runco990: Yup, that could also be it.  Maybe Heidenhain bought a box of 10k 7812 chips from a single run and they lasted many years.  If the mounting surfaces generated stress on the chip bond, but it only appeared after some decades, that run of chips would have been long gone before anyone in the repair/service division understood that there was a problem.

T3sl4co1l: yes, at 30 cm per nanosecond, light makes it about 0.3mm in a picosecond.  For propagation that timescale, the normal circuit analysis picture has be be replaced with impedance matching for stripline and planar conductors. Nevertheless for traditional non-relativistic circuitry, that switching is pretty damn fast.