Does LM317 require a load to start regulating? (no load overvoltage)

[SL4P]

Just curious.
I built up an LM317 with two resistors (240R/720R) as per the datasheet - to drop 12V down to 5V

No load, just hacked it up, but the output was still up at 11+ volts!!!?

I couldn't see any mention of a minimum load in the datasheet.
Did I luck out with two dead LM317s, or did I do something else stoopid?

Obviously I tried not to cook the legs - but I did flex them a bit to an odd footprint for testing...
Did I really fracture *both* chips legs despite being careful?  Or just two non-LM317's

[MK14]

If you look at one of the datasheets, it says that the recommended minimum output current is 10 milliamps.
So it seems you do need a minimum load (which is what I thought, but wanted to clarify it with the datasheet).

http://www.ti.com/lit/ds/symlink/lm317.pdf

Recommended Operating Conditions
IO Output current 0.01 A

Also there are things in the datasheet like:

Minimum load current
to maintain regulation

Which gives 3.5 mA (typ) and 10 mA (max)

tl;dr
Yes, you do need a minimum load (current).

It also says:

7.4.3 Operation at Light Loads
The device passes its bias current to the OUTPUT pin. The load or feedback must consume this minimum
current for regulation or the output may be too high. See the Electrical Characteristics table for the minimum load
current needed to maintain regulation.

[SL4P]

So there!
I suspected as much.
Now i see it.

Next time, i'll read the full sheet, not the condensed version!
But i did look at the graphs, and they don't suggest non-regulation below that limit.

That could be tricky if i use it a project that shuts everything down to sleep, and the VR cranks the supply rail up!

[SeanB]

Simplest solution is to use a 120R resistor for the top of the divider, as that is the cheapest way to guarantee 10mA of load, plus you improve the regulation slightly as you now have less influence of reference current on the divider network.

[KerryW]

The current through the feedback resistors count towards the minimum load.  1.25V/240R = 5.2mA, which is above the typical minimum load, but below the max.  It's just bad luck that both of yours need more than 5.2mA minimum load.  You could change your resistors to 120R/360R and you would always meet the worst case minimum load.

[Audioguru]

If you look closely, the datasheet for the LM317 shows the more expensive LM117 that can use a 240 ohm resistor since its minimum load current is only 5mA (1.25V/240 ohms=) 5.2mA. The LM317 has a minimum load current of 10mA so it should use a 120 ohm resistor (1.25V/120 ohms=) 10.4mA.

[digsys]

I'd have to disagree with my brethren :-)
We use 1,000s of the fookers and at max, including high efficiency LED (an indicator is a must anyway), don't go above ~3-4mA max, and they regulate FINE.
(I actually use app 500R+ as the set resistor). You LIKELY have a unstable / switching problem. Did you do all the usual 0.1mono / 4u7 Tant (or low ESR),
both Inputs and Outputs?
Edit: I often find datasheets either over / under rated, or sometimes it's just the layout etc

[salbayeng]

I remember there was something odd about the LM317, it has a negative reference,  (like the LM10)  , so the substrate is the OUT terminal , which is also the middle pin of the package (as substrates always are).
If you look at the block diagram you can see all the "gnd" connections are to OUT. (Section 8.2 of http://www.ti.com/lit/ds/symlink/lm317hv.pdf ) .
n.b. the HV version still needs 3.5mA min load.
I'm not a big fan of LM317, useful for current sources and that's about it.

[joseph nicholas]

I bought a bunch of these thing from ebay and they all seem to work just fine.  One thing I don't like when they do stop regulating they seem to deliver a full unregulated voltage to the load when in self protected mode.  They never seem to self distruct though.

[SingedFingers]

Simplest solution is to use a 120R resistor for the top of the divider, as that is the cheapest way to guarantee 10mA of load, plus you improve the regulation slightly as you now have less influence of reference current on the divider network.

This is the correct solution. The default 240R resistor everyone picks is wrong!

[joseph nicholas]

Well, I tried the op'ers circuit.  It regulated just fine, with or without a 4 watt bulb as a load.  I tried the 120 ohm resistor and couldn't get it to work.  The data sheets probably don´t mention loads too much because they most likely don´t matter.

[StillTrying]

I'd have to disagree with my brethren :-)
We use 1,000s of the fookers and at max, including high efficiency LED (an indicator is a must anyway), don't go above ~3-4mA max, and they regulate FINE.

FWIW I agree - probably.
On the data sheet next to Minimum Load it has the conditions "VI - VO = 40 V", - so I've always concluded that the 10mA minimum load only applies when there's 37-40V-ish across the regulator, and that with a much smaller VI-VO you can get away with 3-4mA in practice.

[macboy]

Well, I tried the op'ers circuit.  It regulated just fine, with or without a 4 watt bulb as a load.  I tried the 120 ohm resistor and couldn't get it to work.  The data sheets probably don´t mention loads too much because they most likely don´t matter.

Minimum load is mentioned many times in the LM317 datasheet. 

[Audioguru]

n.b. the HV version still needs 3.5mA min load.

3.5mA is "typical" but where can I buy a typical one, without getting a maximum one? 12mA is maximum when the input is 60V and who knows how much of a load is needed at lower input voltages.

[digsys]

3.5mA is "typical" but where can I buy a typical one, without getting a maximum one? 12mA is maximum when the input is 60V and who knows how much of a load is needed at lower input voltages.

As I've mentioned before - We use 1,000s of the fookers, including HV. I've spent a LOT of time on destructive / stability testing, and have found a few mA is plenty.
And, as I've said, datasheets are NOT always accurate or concise. Often, datasheets are over-spec'd, just to be safe, and allow for user margins.
If you purchase from reputable suppliers, legit dies, you should be fine. PLUS, it's very easy to test for yourself. We've been making these things for 8-10 yrs.
YMMV
PS: I have actually run them ALL quite successfully at 1mA and LESS !! But I decided to add plenty of headroom (ie LED)

[KL27x]

+1 Minimum load is only necessary if you need to meet full spec at that tiny amount of current. That's where the manufacturer will stand behind their specs. And as others pointed out, the feedback voltage divider counts. So unless you're running something in microamp range or very voltage sensitive device like laser diode, it wouldn't even cross my mind to bother with meeting minimum current.

I would guess you wired your circuit, incorrectly. If you can measure 11V on the output, even with no load other than divider, you did something wrong.

Practically speaking, 12V to 5V is a lot of drop. You may want to investigate how to calculate proper power resistor to put inline with Vin. 12V input limits your output to like somewhere south of 100mA without a heatsink. And average LM317/78xx dies a very slow death, causing grief down the road. Slow dying 78XX regulators have been bane of my early tinkering, causing lot of hours of repair. 5 years later, even. Series resistor gives much more output/safety margin when Vin is so high.

Ironically, I was given same advice back then, specifically regarding LM317/78xx and unpredictable death... and I ignored it. I thought if it works for days, it will probably work forever. For me, this is not the case with 78xx. Even supplying tiny constant load, like small panel meter or even a 10mA logic circuit/LED.  Calculating proper resistor for max current draw required takes a little extra time, now, but it may save hours down the road. I think over 50% of my linear regulators eventually died when run with high Vin, before I started using series resistor. Even for tiny current draw, I am taking time to place resistor (and retrofitting said resistor - and other fried parts - into older projects when they eventually die, out of the blue, under same conditions they have been working for years.)

[salbayeng]

12V input limits your output to like somewhere south of 100mA without a heatsink.

Agree, I have no idea why they are labelled as "1 amp" ,  at the lowest dropout voltage (say 2v) and 1 A , you have 2W, 
But with a Rtheta of 65K/W  that a temp rise of 130K ,  in reality you are limited to 1W, or 500mA at dropout of 2v.

The only way to get 5v  at 1 A from 12v supply, is to add a helper PNP in a TO-3 package on a heatsink .

[SeanB]

65K is junction to ambient, no heatsink. there is a better case with a heatsink, but still it will only deliver the full 1A with under 5V across it, and a decent size heatsink.

The failure with time is insidious, the thermal cycling as it operates tends to separate the die from the heat spreader in the package, as they have slightly different expansion rates, and this slowly cracks the bond so the die eventually sits with a small gap on the bottom, meaning it can no longer transfer heat as well.  Running cool means a longer life before this happens or it never doing this, but running it at full cook means that it eventually will run at thermal shutdown, and eventually migrate the pass transistor diffusions through the die, shorting the device.

Adding an external resistor to dissipate most of the heat in a fixed application is best, as it provides added protection to the load when it fails, allowing you to perhaps have a thermal fuse to disconnect power.

[tszaboo]

12V input limits your output to like somewhere south of 100mA without a heatsink.

Agree, I have no idea why they are labelled as "1 amp" ,  at the lowest dropout voltage (say 2v) and 1 A , you have 2W, 
But with a Rtheta of 65K/W  that a temp rise of 130K ,  in reality you are limited to 1W, or 500mA at dropout of 2v.

The only way to get 5v  at 1 A from 12v supply, is to add a helper PNP in a TO-3 package on a heatsink .

It has 4K/W junction to case. With a TO220, you will have approximately 0.2-0.5 case to sink using thermal paste, and it is easy to pick up a heatsink, that has 10K/W ish thermal resistance. Any of the extruded ones will have something like that (the sheet metals are worse). That way you can dissipate 8-10W. Add some airflow, better heatsink, you get to 15 ish W. In best conditions, a TO220 can only do 60-70W, but that requires water cooling and top quality FETs with ultra low thermal resistance.

[StillTrying]

I would guess you wired your circuit, incorrectly. If you can measure 11V on the output, even with no load other than divider, you did something wrong.

Yep, for there to be 11V+ on the output (and VI-VO=1V) then the current through the 2 resistors must be over 13mA. There must be something wrong somewhere.

[KL27x]

The failure with time is insidious, the thermal cycling as it operates tends to separate the die from the heat spreader in the package, as they have slightly different expansion rates, and this slowly cracks the bond so the die eventually sits with a small gap on the bottom, meaning it can no longer transfer heat as well.  Running cool means a longer life before this happens or it never doing this, but running it at full cook means that it eventually will run at thermal shutdown, and eventually migrate the pass transistor diffusions through the die, shorting the device.

Anecdote: I once debugged an "unpredictable" electrical problem on a motorcycle, symptom of which was battery dying. Bump start, and it will work fine, for a time. I wired a panel meter to the power rail and I discovered the problem was in fact very predictable. The regulator works fine if it starts up cold. But once up to temp, if you cycle the power, the regulator will not start up. So if I ride to destination, park for a spell, and ride home, there is no problem. But if I make a quick stop for gas along the way, the battery will drain from that point on. I learned these parts commonly need replacement after about 4 years.

I regooped it and installed a fan over it, and this eliminated the problem. It would stay cool enough to restart when hot. It worked fine for several years after that, up thru the day I sold it.