Note: Linear Regulator Woes – When is an AMS1117 not an AMS1117?

It seems like a few of my posts have been starting with something along the lines of “Well, I was shopping on a Chinese goods website …” and this one is no different! I guess I just don’t learn my lesson …

So, I was shopping on a Chinese goods website when I decided to take advantage of the fact that I would pay for shipping anyway so I decided to add some parts I might need to my cart. Of those was 50 pieces of the venerable AMS1117 SOT-223 3.3V linear regulator. This is a pretty unsophisticated, old but bulletproof regulator that traces its heritage back to the LM or “linear monolithic” series of regulators. For a list price of about AU$0.15 each, it seemed like a reasonable price to pay for what is essentially a “bread-and-butter” component.

Alas … it would seem there is a little more to the humble linear regulator than I first thought.

Would the real AMS1117, please stand up!

The number 1117 probably derives from the LM1117 integrated circuit, still being made under that name by Texas Instruments and On Semiconductor. This linear regulator is available in a range of fixed voltages and an adjustable version, but all universally have a “low” dropout, 800mA of current handling and internal protections against overheating. The AMS1117 is so named, as AMS is the initials of Advanced Monolithic Systems, a company that seems to have introduced a clone version of this chip.

Such changing of letters while keeping the numbers is a fairly common thing, but many clones often seek to match or improve over the original in some compatible way. The AMS-version of the 1117 became so popular amongst 3.3V microcontroller boards that in general, many people will search for an AMS1117 by name rather than the more generic designators.

Unfortunately, with such a popular part number, it seems other manufacturers of such regulators have jumped on the part number without distinguishing their offerings. As a result, now, there are actually many versions of the AMS1117 alone, not counting any LM/AZ/NCP/LD/LDL1117 versions from more traditional IC manufacturers.

Just looking around at the datasheets – even the real AMS version had a few datasheets with some variations in specifications – the below compares the absolute maximum input voltage, output voltage range, drop-out voltage at a given current, quiescent current and current limits.

In my case, I purchased my AMS1117 without any indicated specifications or linked datasheet. Which AMS1117 do I have? I have no idea whatsoever. Most of them are fairly “equivalent” in some way but the HT Wang/JinYu Semiconductor version can only handle 9V while also having a higher drop-out voltage, so it’s already clear that not all AMS1117s are made equal.

Putting it on the Board

The reason I wanted the AMS1117 was simply to throw it on one of the boards that I had recently made. This particular board was a quick rush design that I sent off to JLCPCB with my previous MOSFET Test Fixture just so I could make use of a few ESP-WROOM-02 modules I had lying around. I didn’t realise the castellated pads on those were so close together, so it is a pain to use it otherwise.

This quick design has a few issues including not dealing with Pin 15, a bug with the flash button and limited heatsinking area for the AMS1117 regulator. That being said, it does break out all pins so the board is still useful despite these issues, but perhaps rushing a design just to save on shipping is not a good idea. The manufacturing seems also to be done on the quick – the corners of the board have some “chips” probably because of the way they have been separated.

So I grabbed one of the regulators from the lot and soldered it down with plenty of solder just to make sure the heat conduction would be as good as possible. The regulator itself is laser-etched with AMS1117 3.38AGL39890T.

Quiescent Current

The first thing I wanted to find out was whether this regulator had a wide input voltage range and what its quiescent current would be. I did an I-V sweep using the Keithley 2450 SMU and KickStart 2 from 0 to 30V.

The regulator did not fail, although at about 30V, the quiescent current started to rise suggesting we had reached the limits of the chip. The quiescent current was about 1.63mA which is much less than the average 5mA value in the datasheet and even below the 2mA typical of the “best” AMS1117 in the datasheet roundup. Looks decent!

Protections? What Protections?

But what happens when I put on some load with the B&K Precision Model 8600 DC Electronic Load?

Performing the same sort of sweep from 0 to 30V under my own program control, the output held well with no load, but once 100mA was applied, bad things started to happen. By the time it had just eclipsed 16V, the voltage shot up to nearly match the input voltage. The regulator had “blown through” while dissipating about 1.3W. That would have got the regulator fairly hot … but it should be indestructible right?

A zoomed-in view shows that the 100mA graph shows the voltage rising somewhat before starting to drift quite rapidly before it blew through. I suspect the rapid drift is indicative of the internal band-gap reference drifting due to rapidly changing regulator temperature.

This chip was permanently damaged, so let me try again with a fresh sample, but only to 15V. Perhaps the first chip was bad, or perhaps I had exceeded its unwritten limits, so I decided to be a bit more conservative.

This time, it managed to keep itself regulated at 100mA but on stepping up to 200mA of load, the regulator blew through at 10V. This time, it survived dissipation up to about 1.2W but at 1.4W, it failed.

Looking closer, some signs of heat-stress are seen towards the end of the 100mA trace, and very prominently on the 200mA trace. The only reason the effects aren’t cumulative across current levels is because my script enforces a ten-minute cool-down period between each sweep.

I’m beginning to sense a pattern … but perhaps I’m still being too harsh. Out comes another fresh regulator and a more conservative set of parameters – testing only to 9V this time.

This time, it passed at 100mA and 200mA of load but at 300mA, it blew through dissipating almost 1.6W. The amount of power it dissipates at failure has increased a bit because the sweep is now getting shorter so the regulator doesn’t have as much time to “build up” in temperature.

Zooming in, it seems that 200mA shows a very minor sign of temperature stress but 300mA clearly shows the output voltage deviating quite a bit.

It’s becoming clear that this particular AMS1117 doesn’t seem to have a working over-temperature shutdown! They just keep blowing … but granted I haven’t provided it the optimal heat-sinking condition.

Can it work at a more gentle 7V? Here comes regulator number four …

This time, it passed 400mA of load but blew through at 500mA. This sort of failure mode is potentially devastating as it means the connected load would just receive the full input voltage. That is often likely to fry the connected device right away.

The dissipation reached 1.7W at the point of failure. Perhaps being even more gentle with just 5.5V input would allow for more current …

… and indeed it does – up to 800mA just fine but at 900mA it blew through. Peak dissipation was close to 1.9W this time.

On the whole, it’s clear that this regulator doesn’t have over-temperature protection and likes to fail with input and output nearly-shorted. Given internal protection is a feature of LM/AMS1117 specifications from all vendors, this seems to be an especially poor clone of the device. The “drop-out” voltage is also a little high – at 800mA, it measured 1.32V. Buying this version of the AMS1117 seems to have been a mistake.

Improved: Now With A Little More Heatsink

Seeing the results above suggested that over-temperature due to a lack of protection was the cause of failure, I broke out a sixth regulator and decided to solder it to a different scrap of PCB. This one featured copper planes on both sides with a pair of large plated-through holes which I filled with solder. Soldering leads directly to the regulator and clipping connections on, I endeavoured to provide a level of heat dissipation commensurate with a highly-considerate design willing to dedicate excessive amounts of copper to cool the regulator.

In this way, I tested it with a sweep up to 7V and managed to have proper operation up to about 700mA. At 800mA, the output actually tried to fold back a few times before the sweep finished. By 900mA, the fold-back occurred once or twice but then the regulator blew through.

The effects of heating are clear, but the increased heatsinking allowed dissipation of up to about 2.6W to be handled. Failures occurred when passing 3W. It seems that perhaps there may be an attempt at some thermal-protection in the design, but it just doesn’t operate properly in most cases.

Cracking the Case?

I heard that some others have experienced variations in AMS1117 devices before and some have suggested that small dies are present in clones. Since I had a pile of busted regulators, I decided to try and take a look inside.

Putting the package into a vice, I managed to snap the plastic encapsulation off the metallic contacts. Scraping with a set of tweezers, I managed to expose the back-side of the silicon die. Each of the contacts appears to be connected by two bond wires.

A macro photo of the die placed next to a ruler with 0.5mm graduations leads me to believe the die to be 1mm x 0.5mm. I’m not sure what the regular size for such a regulator is, but it does seem a little small. Unfortunately, whatever it is encapsulated in is too hard to scrape away, so the investigation ends here.

Conclusion

I thought that the humble, old, simple, nearly bullet-proof low-dropout linear regulator would be a safe part to buy from any source. Unfortunately, this experience seems to prove me wrong. While the AMS1117 part number is quite popular and originally stands for “Advanced Monolithic Systems”, there are many other vendors producing devices with the same part number and subtly different specifications. Unfortunately, it seems that this particular anonymous device doesn’t even correspond to any of the data I found.

Instead, this particular AMS1117 has one particularly disturbing trait – the lack of a functional over-temperature protection fold-back. Linear regulators had a reputation for being indestructible because they would shut down when they were overheated, but this particular regulator instead blows through and becomes nearly a dead short between input and output, exposing your load to an overvoltage condition that will probably destroy it.

If the demands are low and the cooling is sufficient, users might not notice this nasty trait until the load grows (e.g. a project gets more peripherals added on) or the cooling fails (e.g. due to higher ambient, failed cooling fan). Once the conditions are sufficient, the regulator will blow-through and expose the load to an overvoltage condition.

The drop-out voltage of the regulator under load is not particularly stellar at 1.32V at 800mA, but the quiescent current was a tame 1.63mA. But even for the price of AU$0.15 each, had I known about this flaw, I would not have chosen to purchase this.

Instead, there are plenty of potential regulators from reputable manufacturers which would be equivalently capable to the AMS1117 or better, but are more expensive when supplied through global distribution chain brands. Such parts might include the:

• TS2940CW-3.3RP which is about $0.37ea but has a higher quiescent current of 10mA,
• AZ1117CH-3.3 TRG1 which is about AU$0.40ea and has a more ordinary 4mA quiescent current,
• NCP1117LPST33T3G which is about AU$0.49ea with a nice 0.55mA quiescent current,
• LDL1117S33R which is about AU$0.81ea which has even lower 0.25mA quiescent and a much lower drop-out of 0.35V and a higher 1.2A rating.

I’ve opted for the latter, but such parts being less ubiquitous also means they are less likely to be counterfeited, thus making for peace-of-mind in purchasing. But perhaps I should learn the lesson – not to trust Chinese websites to supply anything semiconductor-based.