Choosing an appropriate LDO for low power device

[rookie]

I am building a wireless device that is expected to work off of a lipo battery. to power the mcu/peripherals, I need a regulator and hence started looking at different devices available in the market. I am interested in a well performing LDO that can get the most out of my battery. I found a few devices and am looking to narrow down on one or two that I can test. My current requirement is Iout = 150mA and Vout = 3.3V
I am sharing some of my rationale for selecting from these devices, so may be someone can poke holes in my reasoning and help me make a better decision! I'm no EE guy, but this is my analysis based on spending some time reading basics of LDO and respective datasheets:


MCP1810 - marketed as lowest quiescent current ldo in the market by microchip:
(section 1.0 of datasheet)
Iq = 20nA typical
Ignd = 200uA at 150mA
Vdropout = 380mV at 150mA
PSRR = 40dB(100Hz)

MCP1811 - another low Iq LDO from microchip
(section 1.0 of datasheet)
Iq = 250nA typical
Ignd = 90uA at 150mA
Vdropout = 400mV at 150mA
PSRR = -50dB(1kHz)

TPS783 - These LDOs are from TI, one of their lowest Iq offerings.
(section 6.5 of datasheet)
Iq = 420nA typical
Ignd = 8uA at 150mA
Vdropout = 130mV at 150mA
PSRR = 20dB (100Hz), 15dB (1kHz)

NCP170 - These are low Iq offerings in IOT space from ON Semi
(page 9 of datasheet)
Iq = 500nA typical
Ignd = could be atleast as high as 100uA at 150 mA from figure 20 for Vout = 3.6
Vdropout = 180mV at 150mA
PSRR = 41dB (1kHz)

STLQ020 - Again, IOT oriented LDOs from STMicro. This one is actually 200mA Iout:
(section 5 of datasheet)
Iq = 400nA typical
Ignd = 100uA at 200mA
Vdropout = 160mV at 200mA
PSRR = 50dB(100Hz), ~ 34dB(1kHz)


Most these devices seem to work well with input/output caps and are available in small form-factors. Even though I could choose the device with the lowest Iq, my device is not going to consume 0 current even on standby, so other parameters will definitely matter. So here are my thoughts on the devices:

Even though the microchp LDO MP1810 has the lowest Iq, it seems to get beat in dropout voltage and Ignd.

The STMicro device still has higher Ignd even with low Iq and low dropout voltage.

The OnSemi device still has relatively high Ignd with comparable Iq and dropout voltage.

To me, the TI device has the decent specs in most departments and something I'm leaning towards - low Iq (not lowest), low dropout and really seems to win in the Ignd. Although I'm not sure how much that varies with Vout, I think their datasheet says 2.2V for the stated Ignd.

So what do y'all think?

[SiliconWizard]

From experience, depending on your input power source, I would avoid the NCP170. Those regulators tend to overshoot when the input voltage is rising (depending on rate and current draw at the output), so in some particular situations you could fry something. I really didn't like those LDOs.

[mariush]

Well, there's a problem here.

I am interested in a well performing LDO that can get the most out of my battery.

By design, a LDO takes an input voltage and gives an output voltage, throwing out the difference as heat. So no matter what LDO you may choose, it would be anything but efficient.

For example, if your input voltage is 4v, your regulator's efficiency will be 600mW in +quiescent current , 495mW out (3.3v x 150mA)... you get a 82.5% efficiency.

With such low efficiency, thinking about a small quiescent current is kinda silly.

The smarter thing would be to use a switching regulator to get higher efficiency. If voltage stability is really important, you may consider adding a LDO at the output... for example output 3.5v with your switching regulator, then output 3.3v with your LDO.

A Micrel / Microchip  MIC23050 is over 90% (89% at 1mA output) efficient and at around 200mA it gets close to 93% efficient: http://ww1.microchip.com/downloads/en/DeviceDoc/mic23050.pdf
... and it's cheap, at around 0.4$ in 25pcs.



If you want something easier to solder, PAM2301 from Diodes Inc is a decent choice. Not quite as efficient as the above, but can still go above 85% and gets better the closer your input voltage is.



The inductors can be very small, surface mount ceramics, because the above regulators work at 4 Mhz and 1.5Mhz (for the PAM2301)... so the capacitors on the output can also be ceramic and a tiny capacitance is enough.

There's also MCP1603 from Microchip, which claims above 92% efficiency ... 2 Mhz switcher : http://ww1.microchip.com/downloads/en/DeviceDoc/22042B.pdf  (see Figure 2-13, page 9)

[rookie]

@SiliconWizard , @mariush - Thanks for the suggestions!

@mariush - I have considered switching regulators and will certainly try out your suggested parts. There is also MAX38640 that seems really promising. The only reason I shy away from switching is that I don't have enough confidence working with them, and their effects on other components in the system like sensors, analog circuitry etc, let alone the wireless part itself! But the only way to learn is to actually try it out. I am limited by real estate, so I will have to think hard about incorporating a buck converter+ldo

[Jan Audio]

R-7833

[jbb]

For the LDO, dropout is likely king. What is the expected sleep current of your micro?

It may not be practical, but how about lowering your system voltage to 2.7V or similar? Then a micro power buck would be very effective.

[ThomasDK]

The TPS783 is horrible for RF applications. Been there, done that, lesson learned... The PSRR is a joke, but like me, you forget to take a look at the transient response... Over 100mV voltage sag at 10mA output step!

You don't say what kind of wireless you are planning to use, but TX current pulses are often higher than that (=even more undershoot), so expect lots of package errors!

[rookie]

@jbb - I'd say couple of uA tops from the micro only. My other sensors may add their contribution. I have some analog circuitry that needs slightly higher voltage (3),

@ThomasDK - It's 2.4GHz bluetooth low energy. 10mA peaks could be expected yes, may be a bit lower but certainly in that range. Interesting, what power architecture did you use eventually? higher capacity ldo or switching regulator?

[hamster_nz]

I'd say couple of uA tops from the micro only. My other sensors may add their contribution. I have some analog circuitry that needs slightly higher voltage (3),

Have you had a look at BD733L2FP3-CE2? 6uA quiescent, 200mA Output current.

Not cheap (about 4x that of the Microchip parts mentioned), but the low quiescent current might be worth it.

[magic]

The TPS783 is horrible for RF applications. Been there, done that, lesson learned... The PSRR is a joke, but like me, you forget to take a look at the transient response... Over 100mV voltage sag at 10mA output step!

Throw a capacitor at it, just make sure its leakage isn't more than Iq / Ignd of a higher bandwidth LDO

As for which chip to choose: you need to know how much current your circuit will draw when active, how much when sleeping, and the duty cycle. Say: 1% 100mA, 10% 10mA, 89% 100µA. Then you can calculate effective long-term efficiency of various parts, taking into account Iq, Ignd and conversion efficiency (if SMPS) and the answer will reveal itself. If you don't know, make up some guesstimates and calculate anyway, maybe some parts will show themselves clearly inferior under all resanoble conditions.
Worrying over specs if you aren't really sure what you need is the definition of audiophoolery
If it mostly sleeps, Iq and dropout voltage will be most important. If it's more active, Ignd may become significant. For very active and power hungry stuff, SMPS wins.

[NivagSwerdna]

I am building a wireless device that is expected to work off of a lipo battery

Why LiPo?  Capacity? Duty cycle? Iq and Iactive?

[ThomasDK]

The TPS783 is horrible for RF applications. Been there, done that, lesson learned... The PSRR is a joke, but like me, you forget to take a look at the transient response... Over 100mV voltage sag at 10mA output step!

Throw a capacitor at it, just make sure its leakage isn't more than Iq / Ignd of a higher bandwidth LDO

Too expensive ;-)
We went with another regulator with a couple of μA Iq. Can't remember which.

Anyway, since this the OP intends to use a LiPo, low Iq doesn't matter. The self discharge will probably be in the hundreds of μA, so who cares if the regulator uses 1, 2 or 5μA? It's smaller than the tolerance of the battery...

[Siwastaja]

One thing everybody's missed so far: your discharge low voltage cutoff. You'd want to be able to fully discharge your cell, otherwise you'll carry around dead energy storage. At least you need to know how much you actually get.

You see, you can discharge a li-ion cell way below 3.3V; especially if you have a considerable load current, or run at low temperatures, or run with an aged cell. Just 100mV of extra dropout from a regulator can mean extra 10% loss of usable capacity.

Let's look at an example curve of Samsung INR18650-29E from https://lygte-info.dk/review/batteries2012/Common18650comparator.php . Let's look at the 0.5A curve - your discharge current being only 0.15A, but OTOH I'm assuming your cell could be smaller (1Ah-ish), so a smaller current produces a proportionally larger drop, especially if you want to leave some margin for cold weather, or aged cell.

MCP1810 - marketed as lowest quiescent current ldo in the market by microchip:
Vdropout = 380mV at 150mA

Low-voltage cutoff at 3.3V + 0.38V = 3.68V -> from the curve: you only get 44% of the nominal capacity out of it! No go.

MCP1811 - another low Iq LDO from microchip
Vdropout = 400mV at 150mA

This will be even worse.

TPS783 - These LDOs are from TI, one of their lowest Iq offerings.
Vdropout = 130mV at 150mA

Low-voltage cutoff at 3.3V + 0.13V = 3.43V -> from the curve: you only get 81% of the nominal capacity. Acceptable I guess?

NCP170 - These are low Iq offerings in IOT space from ON Semi
Vdropout = 180mV at 150mA

3.48V -> around 75% available

STLQ020 - Again, IOT oriented LDOs from STMicro. This one is actually 200mA Iout:
Vdropout = 160mV at 200mA

3.46V ->around 77% available.

So if you need 3.3V out, you can clearly rule out two regulators.

In some cases, you use the LDO just to drop the voltage beyond maximum ratings (such as 3.6V max), but are OK if it drops below 3.3V.

Most switching regulators are likely out of question if you mostly sleep. It doesn't matter if your linear regulator efficiency is "only" some 60-70%. The switching regulator is likely much worse at low loads, and the zero-load consumption is easily orders of magnitude more, even with pulse-skipping modes.

200uA Ignd is actually quite a lot of no-load current, most micros would happily sleep below <20uA. This poses a problem if you have a small cell, you run it almost flat, your micro does what it should and puts the device in sleep to protect the cell, and then the user forgets the product for a while. This can damage the cell by overdischarging it. Assuming 1Ah cell, discharge stopped at 20% (200mAh left), and shelved by the end-user, it would take 0.2Ah/200e-6A = 1000h = 41 days to self-destroy beyond repair. IMHO, such a product is unacceptable, I always design for a bare minimum of half a year of margin between "run-flat" and "put into charger" events, preferably more.

So, put together, I wouldn't accept anything else than Vdrop < 0.2V and Ignd < 50uA, only satisfied by TPS783.

[Siwastaja]

Anyway, since this the OP intends to use a LiPo, low Iq doesn't matter. The self discharge will probably be in the hundreds of μA,

A nice example of the classical false assumption which leads to all those nasty crap devices which self-destruct.

Okay, take a small li-ion pouch cell, say around 1Ah. The self-discharge of an empty cell is not "hundreds of uA", not even "tens" - if it was, the cell would self-destruct by self-discharging itself beyond low-voltage cutoff, finally causing low-voltage damage (copper dissolution). But this does not happen. I have tried to measure the self discharge of empty(ish) cells, and it just is impossible to measure given standard external means. During my 1.5 years of test time, none of the dozen samples lost any measurable charge below about 50% SoC, not even at elevated temperatures.

Now, if you indeed assume that the cell self-discharges at "hundreds of uA" anyway, and so it is OK to add external load in a similar range, you end up with this classical broken-by-design gadget:
1) The user buys your battery-powered product (can be a $1 toy, a $10000 special instrument, or a $100000 EV car, the principles are the same)
2) They charge it, turn it on, play with it
3) The battery goes empty, the user turns it off,
4) The user has something else to do
5) The user comes back to the product after a month or two - with worst offenders, just after a week - it's a brick.
6) If(expensive && userbase_has_clue) warranty_repair_nightmare_ensues(); else user_throws_it_into_bin();

This is surprisingly commonplace in both cheap Chinese throw-away toys, and expensive "professional" products. This is actually a very common problem even in high-end battery management systems, I have seen many which have destroyed good packs instead of protecting them, which was their only job!

Yet it's trivially easy to calculate how to avoid this problem.

Look how much charge your cell has left at the cutoff of your choice, and divide it by the worst case "off current". You need to decide one parameter, which will be "time from turn-off to self-destruction", which, unfortunately, cannot be infinite. If you don't calculate it but do it based on hand-waving or just ignore the issue altogether, it may accidentally become just a week or two. Yet you'd like to have something around a year or so.

Regulators / MCU sleep modes with Ioff/Isleep < 10uA, even <5uA are very much needed in the li-ion battery powered low power device market.

[tszaboo]

You have to define the:
- Battery chemistry
- Expected battery life
- Battery mAh rating
- ratio of sleep time / on time
- sleep current of your device
- ambient temperature range
- typical application
- budget

Than we might be able to help. Otherwise we could suggest something, but selecting the optimal converter is not an easy task.

[schmitt trigger]

I've personally witnessed what Siwastaja posted, although at the time I did not understand what was going on.

[magic]

Look how much charge your cell has left at the cutoff of your choice, and divide it by the worst case "off current". You need to decide one parameter, which will be "time from turn-off to self-destruction", which, unfortunately, cannot be infinite. If you don't calculate it but do it based on hand-waving or just ignore the issue altogether, it may accidentally become just a week or two. Yet you'd like to have something around a year or so.

I would rather have 10-100 years to be honest. Are there no simple protection ICs you just put in series with the cell and connect a 3rd sense terminal to the opposite pole that would solve it while drawing nanoamps or less on their own? Basically, I would want a MOSFET with well defined threshold at 3V and very high transconductance, latching and hysteresis could be a bonus

[Siwastaja]

I would rather have 10-100 years to be honest. Are there no simple protection ICs you just put in series with the cell and connect a 3rd sense terminal to the opposite pole that would solve it while drawing nanoamps or less on their own?

Of course, but the market is full of protection devices (ICs or professional modules) that utterly fail to do that job. So the designer needs to know that such traps exist, and needs to understand the importance of the current draw after a low-voltage cutoff event.

In some cases, it is very appealing to do it without an extra switch and extra controller, for example if all loads in the system (often just a single microcontroller, really) can be turned to low-power or sleep modes reliably, with low enough off current. Extra switches and extra controllers bring extra problems to be solved (cost and PCB real estate being the most obvious ones, but things like inrush current in high-power devices can be very demanding traps for young players "just using a protection IC").

[ThomasDK]

Anyway, since this the OP intends to use a LiPo, low Iq doesn't matter. The self discharge will probably be in the hundreds of μA,

A nice example of the classical false assumption which leads to all those nasty crap devices which self-destruct.

Okay, take a small li-ion pouch cell, say around 1Ah. The self-discharge of an empty cell is not "hundreds of uA", not even "tens" - if it was, the cell would self-destruct by self-discharging itself beyond low-voltage cutoff, finally causing low-voltage damage (copper dissolution).

Self discharge is not constant, it tapers off over time. The first day or so is the worst.

To take your 1Ah example (first hit on Google):

https://www.sparkfun.com/products/13813

Excellent long-term self-discharge rates (<8% per month)

That's 80 mAh in a month = 80/(31×24)=108μA.

Now, if you indeed assume that the cell self-discharges at "hundreds of uA" anyway, and so it is OK to add external load in a similar range, you end up with this classical broken-by-design gadget:

The self discharge will probably be in the hundreds of μA, so who cares if the regulator uses 1, 2 or 5μA? It's smaller than the tolerance of the battery...

5μA is not a similar range...

Regulators / MCU sleep modes with Ioff/Isleep < 10uA, even <5uA are very much needed in the li-ion battery powered low power device market.

...but you seem to agree

[Siwastaja]

Excellent long-term self-discharge rates (<8% per month)

Well yeah, a typical li-ion cell (I measured about ten different cells) self-discharges at approximately 0.1 to 0.5% per month. Which, of course, is less than 8%!

Excellent ones are those near to that 0.1% rate or below, the Sony VTC5 cell was the best in this regard in my tests and some others were close.

However, if any cell self-discharges at rates near 8% of month (at room temperature at least), they are completely dead crap. No one would accept that, it's more than an order of magnitude worse than industry average.

I'm quite positive even the cheapest no-brand cell sold by Sparkfun doesn't do that. This specification is just total bullshit. Or is it? A destroyed cell (caused by manufacturing failure, or abuse such as a serious overdischarge event) may leak more, of course. So it is possible that Sparkfun sells failed cells from a QC reject bin of a Chinese li-ion cell factory, and specify their leakage this high to avoid complaints. Such cells should be disposed of, however. They may be dangerous as well. In this context, "excellent" is an insider joke. Like: "excellent sports car, top speed 15 mph".

The exact self-discharge figure measured from the full cell doesn't matter, however, because the leakage rate goes to basically zero when the cell is near empty, which is the relevant SoC for this question. All cells I tested were equal with completely non-measurable leakage below about 50%, even the "worst" ones, even at elevated temperatures. And, the samples deliberately overdischarged to approx. -1% SoC did not leak below the original voltage measured at the start of the storage period, not a single one of them, during 1.5 years.

Only seriously damaged cells would continue leaking while near-empty.

So just don't assume that you can add external leakages because the "cells would leak anyway". They won't - your external load is what kills the cells. They would survive otherwise, unless they are bad to begin with.

Yes, I agree with you that most often it doesn't matter whether Ioff is 1, 2 or 5 µA (although 5µA is starting to get into the range of mattering in some cases). I wasn't commenting on that part.

In any case, let's assume a 1Ah cell, with 5% SoC (50mAh) margin at low-voltage cutoff event:

1 µA: time to self-destruction: 5.7 years
2 µA: 2.8 years
5 µA: 1.1 years
10 µA: 7 months
100 µA: 20 days

A smaller gadget, with a 200mAh cell, with the same 5% margin, would already die in less than 3 months given 5µA external leakage. So yes, sometimes it matters.