PSA: do not use the TPS61099 boost reg in your designs

[jeremy]

Hi all,

Just reporting a rather poor experience I have had with a new boost regulator on the market. It works like a normal adjustable boost regulator in that there is a resistor divider to set the output voltage (you can also get fixed devices that have it internally). However, if the rise time on the power supply is too slow or there are glitches, it ignores the feedback loop and goes to maximum duty cycle mode and locks up. For example, using the TPS610995 (which is the only one in a DFN package), instead of boosting my 2.8V supply to 3.6V, it occasionally boosts it to a constant 5.7V on startup. The worst part is that I am using the exact supporting components from the datasheet, with the exact same layout as per the datasheet (well, as close as possible; they don't specify dimensions ).

Although this part is designed to be connected directly to a battery (they even have a reference design for it!), I have reproduced this problem using the contact bounce of just putting a battery in a holder. I assume that the same thing would happen with a particularly bouncy mechanical switch.

I wouldn't be so bothered if it would at least fail safe, rather than failing to max duty cycle and boosting to a high enough voltage to kill everything...

More info on the TI forum here: https://e2e.ti.com/support/power-management/f/196/t/756614 if you search the forum, it seems I am not the only one who has discovered this problem. But no satisfactory solution seems to exist.

I have tried to work around it with a voltage supervisor on the EN pin, but there seems to be a related issue I have found today wherein if the output voltage is higher than the input voltage on startup (for example, output capacitors still charged) the same thing can happen.

If anyone has any ideas about how I could work around something like this, I'm all ears 

Thanks,
Jeremy

[Siwastaja]

Thanks for warning.

Sounds like typical TI engineering. I use their power conversion/management ICs less and less - they are overcomplicated and have excess state space, and therefore tend to hit unwanted states - and a typical "wrong state" tends to mean: blow up everything.

In this case, looking at the TI E2E forum posts, it seems like an UVLO event is running some kind of special startup sequence, which blows everything up if happening more than once inside a small time window. The problem is, it's completely normal and unavoidable that UVLO events happen frequently, and it shouldn't cause erratic behavior.

These things are IC design equivalents of spaghetti code from a total beginner. If this was a software, a fix would be another overcomplicated "UVLO filter unit" which would reduce the number of issues, but increase complexity again. But ICs can't be updated.

[Gibson486]

That sucks...I have always had good experiences with TI's power chips (not a boost though). That said, I had a horrible experience with their high power op amps...and it was related to the enable pin as well.  I have always relied on Linear Tech for boost converter stuff. It can be pricey, but it usually always works. 

[SiliconWizard]

Damn, that's good to know. This IC looked interesting on paper.

Reading the datasheet, we can see it's got quite a few different modes of operation. Between the soft-start feature, the "down mode", "pass-through", bust mode and synchronous mode... eek, there's probably a lot of cases, depending on how Vin and Vout relate, to make it go bonkers. In your case, I'm guessing it's probably stuck cycling between two of those modes (or more) in an unexpected way. This IC looks like a bunch of good ideas gone bad.

Not sure what you can do to get something reliable, short of switching to another boost converter altogether. Maybe select one of the Linear boost converters that have a burst mode and very low quiescent current (although this one will be hard to beat). Never had any problem with them.

A zener diode at the output is probably a bit pointless. It may protect your circuit but with the common Zener voltages of 3.6V and 3.9V (off the top of my head), you are likely to be in the diode knee most of the time, wasting unnecessary current... and even if that protects your circuit, I'm not even sure this would get the IC out of this weird mode when it happens, so either way, your device would probably not work properly. Maybe adding a second voltage supervisor at the output with a threshold of say 4V (if 4V is still safe for your circuit), that would disable the boost converter when getting higher than this. But that may just get it into an infinite loop of starting and disabling... and it starts getting a bit much for just one converter. Sucks.

[PartialDischarge]

Not saying the IC is perfect but I see some fault on your part for not assuring a stable Vin before enabling the converter. You mention connecting directly the battery, or that the same would happen with a switch, but that is not how you are suppose to turn on the converter, that is what the enable pin is for. This chip has true disconnect so the battery is supposed to be connected at all times and use the enable pin to discoonect.
First connect the Vin source, of course if the impendance is high an input capacitor helps, then enable the converter. I personally would use a resistive divider in the E pin.

I have designed circuits with chained switched converters, and needed to control the enable of each one, assuring that the output of the previous converter was stable.

[jeremy]

Fair enough mastertech, but the datasheet and reference design doesn’t have anything connected to the EN pin. See http://www.ti.com/lit/ug/tiduee2/tiduee2.pdf

You make a good point about the resistor divider, I’ll need to think about that more.

In the “test setup” section, it even says to specifically connect the EN pin to Vin *before* applying power (I’ve also tested what happens when the EN pin is floating; it ain’t pretty )

I don’t think it would be too much of a stretch to see this happening upon connecting a battery using a connector, or when using a depleted cell with a bursty load like a cellular radio. You can only increase input capacitance so much, and given the primary package of this device is a CSBGA I can’t imagine they expect you to use a bunch of bulky electrolytic caps.

[PartialDischarge]

Caps are only needed if you are using a weak battery with lowish mAh. That helps at the startup.
The EN pin can be connected directly to Vin only for a source that has enough low impendance to turn on the converter, and that doesnt seem to be your case.
In summary, your problem lies in the low energy of your battery, so you will have to use input caps + control the EN pin.

[jeremy]

Although I agree on the overall cause, we will have to agree to disagree on the scenario it occurs in. In particular, the chemistry of the cells used in the reference design are intended to be extremely long life and as such are installed in difficult to reach places. My experience is that due to this, they are used until they are completely and utterly dead, so at the end of this cycle you will almost certainly have a cell with very poor impedance which no reasonable amount of capacitance will save you from. In this case, a low mAh cell is the same as a depleted one.

[PartialDischarge]

But the reference design, unless I read wrongly, only starts up once. It is a very low Iq design designed to be on forever. The problem lies at the startup in these converters, always, not in steady state.

If you can’t change the requirement that your system be turned on by connecting a battery repeatedly, then you have 2 choices, come up with a way of controlling the EN pin in an effective manner, or choose a boost converter with a programmable soft-start, that allows very looooong startup time.

[iMo]

The DS Fig23 shows a nice sharp Vin leading edge

Btw - I was using the similar https://datasheets.maximintegrated.com/en/ds/MAX1722-MAX1724.pdf - except lower Iout and almost 20y old.
Also it was not so easy to get what it claimed - the efficiency achieved with some special L and C types/materials only.  But it worked from 0.5V up, as I can remember, and the package was easy to solder.. 

[jeremy]

But the reference design, unless I read wrongly, only starts up once. It is a very low Iq design designed to be on forever. The problem lies at the startup in these converters, always, not in steady state.

If you can’t change the requirement that your system be turned on by connecting a battery repeatedly, then you have 2 choices, come up with a way of controlling the EN pin in an effective manner, or choose a boost converter with a programmable soft-start, that allows very looooong startup time.

Yes, the reference design is sort of only intended to startup until the battery is depleted, but the contact bounce of connecting the battery initially could cause the issue. Also, “forever” is relative, a typical cellular modem in this configuration will draw 100mA for many seconds (I’ve measured it). Forever could be years, yes, but not literally forever.

[Bud]

I am suggesting we start from where you sourced the parts. If out of eBay or Aliexpress, the conversation should end right there.

[tszaboo]

These converters sometimes go through a silicon revision. If there is a TPS61099A arriving, that is basically TI improving on it's silicon to avoid this. It wouldnt be the first time, and wont be the last.

[T3sl4co1l]

they are overcomplicated and have excess state space, and therefore tend to hit unwanted states - and a typical "wrong state" tends to mean: blow up everything.

This, this and ALL OF THIS!!!

It's a bloody switcher, what does it need all that logic for?  Do they start with an MCU and strip out unused bits?  Just look:
https://zeptobars.com/en/read/Ti-TPS62321-3-MHz-Step-Down-Chip-Scale-CSP

They're great when they work, but it's all too easy to trip over this bullshit. It's not a new thing, I've got a question on E2E (that shall remain unanswered, it seems) regarding the apparently old TPS40210, which seems to have an analog discrepancy.  Also relating to startup conditions, oddly enough.

I wish they'd put the transistor count on the datasheet, just so I can avoid these stinkers.  But who knows.  Transistor count doesn't really mean anything, it's the shitty design that makes it shitty.

Tim

[jeremy]

I am suggesting we start from where you sourced the parts. If out of eBay or Aliexpress, the conversation should end right there.

Digi-Key 

But good point nonetheless.

[amyk]

It's a bloody switcher, what does it need all that logic for?  Do they start with an MCU and strip out unused bits?  Just look:
https://zeptobars.com/en/read/Ti-TPS62321-3-MHz-Step-Down-Chip-Scale-CSP

It's more like "start with an MCU and add power transistors"... I wouldn't be surprised if there was some sort of primitive microcode and CPU-ish logic in them.

I'd be interested in a die photo of a simpler switching regulator, for comparison.

[T3sl4co1l]

I mean, this might be a little too simple, but it illustrates the point:
https://zeptobars.com/en/read/MC34063

I've synthesized similar functionality to the UC3842 with a mere seven transistors, lacking some functions like low-current startup and UVLO, which would hardly take a dozen more.  It's not hard to do, they just don't care about analog design anymore.

Tim

[DaJMasta]

Even if there are some ways to reduce the likelihood of a faulty start to near zero, they should probably be described in the reference circuit in the datasheet and much more importantly: this kind of over-boosting failure mode shoudln't ever be acceptable in such a chip.  If you have a faulty start and need extra stuff to clean up power, fine, it may not be a treat to use in your design but it will work well enough.... if a potential failure mode for a boost converter is to overdrive the target voltage by like 50%.... that's a design failure.

Maybe they can update their datasheet to include clamping circuitry, but they really should be looking into respinning it and considering recalling the parts - it's not going to be a converter of choice if the BoM inflates with larger clamping diodes.

[radioactive]

What values did you use for R1/R2 feedback?  Did you see this in the datasheet? 

For the best accuracy, the current following through R2 should be 100 times larger than FB pin leakage current.
Changing R2 towards a lower value increases the robustness against noise injection. Changing R2 towards higher values reduces the FB divider current for achieving the highest efficiency at low load currents. 1-MΩ and 249-kΩ resistors are selected for R1 and R2 in this example. High accuracy resistors are recommended for better output voltage accuracy.

Also, another thing that would be worth trying is switching to the fixed 3.6V  (TPS610995)?

Another last ditch type effort would be to have an external voltage ref / comparator monitor for over voltage and the drive enable pin low.  Probably not worth the cost unless you really need to use this part.

[jeremy]

I am using the

What values did you use for R1/R2 feedback?  Did you see this in the datasheet? 

For the best accuracy, the current following through R2 should be 100 times larger than FB pin leakage current.
Changing R2 towards a lower value increases the robustness against noise injection. Changing R2 towards higher values reduces the FB divider current for achieving the highest efficiency at low load currents. 1-MΩ and 249-kΩ resistors are selected for R1 and R2 in this example. High accuracy resistors are recommended for better output voltage accuracy.

Also, another thing that would be worth trying is switching to the fixed 3.6V  (TPS610995)?

Another last ditch type effort would be to have an external voltage ref / comparator monitor for over voltage and the drive enable pin low.  Probably not worth the cost unless you really need to use this part.

I am using the fixed 3.6V option partially because I didn't want to be bothered with problems like this  . I have been playing with various voltage supervisors, etc and I just can't seem to get the circuit right. It's fairly easy to delay the startup of the converter with one, but the overvoltage condition is what is stumping me. Really, you need a circuit which de-asserts the enable pin if it goes too high *and* discharges the capacitors on the output stage. A comparator with wide hysteresis and a voltage ref is probably what is needed, but cost and complexity starts to be a problem there. Also it has to be crazy low Iq. Right now my circuit uses 4 times the sleep current of the entire rest of the system.

Even if there are some ways to reduce the likelihood of a faulty start to near zero, they should probably be described in the reference circuit in the datasheet and much more importantly: this kind of over-boosting failure mode shoudln't ever be acceptable in such a chip.  If you have a faulty start and need extra stuff to clean up power, fine, it may not be a treat to use in your design but it will work well enough.... if a potential failure mode for a boost converter is to overdrive the target voltage by like 50%.... that's a design failure.

Maybe they can update their datasheet to include clamping circuitry, but they really should be looking into respinning it and considering recalling the parts - it's not going to be a converter of choice if the BoM inflates with larger clamping diodes.

I suspect this is a marketing problem perhaps (though I suppose Hanlon's razor says that its a case of nobody who can do anything about it knows about it). This is clearly designed for a tiny board area (otherwise who would bother to use the CSBGA), and clamping would make that a lot worse. You can also kiss the ~1uA Iq goodbye as well. At that point, this converter basically has no market; who wants a low Iq boost converter that isn't really that low Iq, takes up a bunch of board space but still requires you to solder and inspect a CSBGA, and still might get into a glitch loop where it draws too much current.

[IDEngineer]

they are overcomplicated

This, this and ALL OF THIS!!! They're great when they work, but it's all too easy to trip over this bullshit.

Oddly, this reminded me of the C vs. C++ thread raging in the CPU/MCU forum.    Here it is expressed in hardware... excessive complexity that often ends up being more of a hindrance.

[T3sl4co1l]

Oddly, this reminded me of the C vs. C++ thread raging in the CPU/MCU forum.    Here it is expressed in hardware... excessive complexity that often ends up being more of a hindrance.

Sort of.  Notwithstanding style issues ("blargh blargh don't use C++ because I don't like its syntax and when I use it it's always so bloated!", etc.), most of that is semantics and syntax -- complexity used to generate code, but not complexity in the code itself.  The result can be quite lean and orderly.  You just have to make sure you aren't telling it to compile with more code features than you need.

So too, HDL can be quite complex at the design level, but the synthesis doesn't need to be complex.

It seems we have the worst of both worlds here.  One should not blame the tools (necessarily), but the commitment to quality is what's lacking.

Tim

[IDEngineer]

Sort of.  Notwithstanding style issues ("blargh blargh don't use C++ because I don't like its syntax and when I use it it's always so bloated!", etc.), most of that is semantics and syntax -- complexity used to generate code, but not complexity in the code itself.  The result can be quite lean and orderly.  You just have to make sure you aren't telling it to compile with more code features than you need.

Agreed, but I was thinking more in line with your comment:

It's a bloody switcher, what does it need all that logic for?

...and how that syncs up with the original intent of C, to be a portable Assembly language - not a "high level" language. Just as a "simple" switcher doesn't need "all that logic", so too does a "mid-level" language not need all of the baggage of a true high level language. There are plenty of complex IC's, and complex high level languages, to choose from. Why corrupt the intentionally simpler IC's/languages?

Sorry for the hijack, back to our regularly scheduled programming....

[BrianHG]

With TI, it's not just their power management ICs.  Their new norm on documentation SUCKS shit.  On their clocking ICs, I tried to get information as their formulae in the data sheet made no sense and had non-standard characters.  I was told from TI support, just use their PC configuration app to get the IC2 configuration settings.  NO... The IC said I can program any frequency I wanted, both setting the reference and feedback dividers and their config software didn't always make the most stable oscillator configuration unless I manually tweaked it.  Their own engineers couldn't decipher their own documentation...  It took 5 days and a few friends who specialized in math and physics to reverse engineer the notation in their .pdf data sheet and to get the IC to dynamically be programmed to do exactly what I want.

I also found bugs in a few of their other ICs.  If you stray anywhere from the documented examples with any TI part above opamps, be warned...  They do not proof their complex devices or test for extraneous bugs at all.

[schmitt trigger]

Subscribing to this thread.

Hope that I can find some documents to share, related to some Intersil multi-phase buck converters, that were an unmitigated disaster.