I have a small embedded system that read some sensors every 10 minutes and uploads the data to a server every hour using a 3G module. I'm using 4 Energizer Lithium L91 AA cells in series to power the entire system.

The 3G module is connected to a buck converter that reduces the voltage from the 4 AA cells to 4V. The module runs 60 seconds every 1 hour and draws around 500mA during that time.

The sensors and the MCU that controls everything only use the lower 2 cells and run every 10 minutes for about 3 seconds and they draw around 50mA. The rest of the time the system is put to sleep to preserve battery.

I'm facing a problem with the batteries where one of them gets really discharged when all the others are almost new. I will explain this a little more. This is my setup:

             +----------+          +------------+         +-------------+        +-------------+
   +---------+     1    +----------+      2     +----+----+      3      +--------+      4      +--------+
             +----------+          +------------+    |    +-------------+        +-------------+
3G Module                                            |                                                 GND
                                               MCU and Sensors

I put 4 brand new Energizer Lithium cells the first time and everything works as expected. After around 7 days the system stops working because cell 1 is almost dead (around 0.8V or lower) but the other cells have around 1.4-1.5V. I proceed to replace the dead cell with a new one. The next day, cell 2 is dead, the same history, I replace it and the system starts working again. After around 5 days cell number 4 dies, it has 0V and when loaded it goes to around -2V which is explained here.

I didn't try to buy more cells to try again (they are not very common where I leave so I'm still waiting for the shipping to get some more).

The only explanation I can find is that I just got bad cells? I expect they don't discharge exactly the same because of manufacturing tolerances but this difference is huge!. This question is similar. We are in winter now so the worst case temperature at night is 0 degrees Celcius, humidity is around 60-80% and I don't have any component that may be heating one battery more than the others, everything is pretty low power. Also, the L91 batteries should be able to supply the current I demand with no problems at all.

Energizer is a reputable brand so this makes me wonder about the quality of the cells. Each cell says 12-2036 which I guess is the expiry date so they are far from being depleted due to self-discharge too.

I also thought about putting the 4 batteries in parallel and use a boost converter to get the desired current but in this case, if one cell dies it's even worse because the other cells will start charging the bad cell!

Am I missing something? Is this something common? This is planned to be a commercial product at some point but if I can't even get reliable batteries that's a big problem!


I just got another device that is basically the same but without the 3G connection, it just saves the sensors data in memory to be retrieved later. This one uses 2 Energizer Lithium L91 cells in series directly connected to the MCU and the sensors and the same happened, one of the cells is completely dead while the other is still fine. In this case, the power consumption is around 50mA every 10 minutes for about 3 seconds so I'm really starting to think I got a bad batch of batteries because even Alkaline batteries should be able to work with this currents.

  • \$\begingroup\$ I’m not aware lithium cells would have any flatter discharge curve to support your findings since they seems to be extreme. If you load the remaining cells, one by one, do they still read high voltage? Side note: have you noticed how today’s remote controllers for consumer products only uses one battery whereas in the nineties two or four where commonplace? You are seeing now what they saw back then which pushed for single battery products. \$\endgroup\$
    – winny
    Commented Jun 25, 2018 at 21:05
  • \$\begingroup\$ @winny yes, on the remaining cells the voltage still reads high even when loading them, it only drops a tiny bit. \$\endgroup\$
    – Andres
    Commented Jun 25, 2018 at 21:10
  • \$\begingroup\$ @winny yeah, that could be a reason, it's really a shame, I will be forced to look for other types of batteries if that's the case. \$\endgroup\$
    – Andres
    Commented Jun 25, 2018 at 21:47

4 Answers 4


As a matter of interest, here is series/parallel switching arrangement to deal with the unbalancing of batteries where there is a long, low current drain at 3V, and an high current drain at 6V.

As the ration of idle/sensor energy to transmitter energy increases, at some point this becomes economic.


simulate this circuit – Schematic created using CircuitLab

D2 is a schottky and does carry Ismps, losing power, but more importantly worsening the dropout voltage. It can be replaced by a pfet active diode (note D,S)


simulate this circuit

A small dpdt relay is also an option, but you have watch the capacitors carefully. If there is a splat as the relay connects capacitors, the contacts will die quickly.

  • \$\begingroup\$ Thanks for the circuit suggestions, on my next version I will probably be using an ultra-low-power buck converter to get the 3.3V from the 4 series cells so they are all discharged evenly \$\endgroup\$
    – Andres
    Commented Jun 26, 2018 at 2:39
  • \$\begingroup\$ If your sensor task can run from the 2.4V minimum of 2 cells, then this is lower (zero) power, cheaper, simpler, and not single-source than a micropower step down. If your sensor can do sleep task on 2.4V, but needs (say) 3.3V for the sensor task, then it can be very energy efficient (and cheap) to use a step-up that you only turn on for the duration of the task that needs the extra volts. \$\endgroup\$
    – Henry Crun
    Commented Jun 26, 2018 at 6:34

Your excessive pulse current battery load is killing some batteries with additional pulses on sensors probably taking more than expected.

Yes the Pulse discharge is killing the battery due to undersized requirements.

If you notice the pulse response after 500mA, the datasheet voltage drops the 1.5V battery from 1.8V new to 1.56 but does loses 10% going back towards 1.8V each time. This is the double electric capacitor effect that takes time to restore the voltage but loses energy in the process. THe high density doubles the capacity gives it memory effect but if overloaded you don't achieve it. All batteries have memory inspite of what marketting has you believe. It is just the stress level and ratio of memory and long term effects that differs between chemistries.

The 3500mA rating is only for 100mA.

While 500mA pulses plus 50?mA kills the batteries unevenly. Series batteries must always be loaded evenly otherwise the weakest cell accelerates aging the fastest.

0'C also kills the capacity 10% (ideally) but in your case with pulse load maybe 20%. To explain this requires modelling of the double layer effect in supercaps and batteries with unbalanced currents at low temp.

For a cutout voltage of 0.8V the rated service hours for 1.5V/cell*500mA = 750mW is 5 hours or 4 hours at 0'C and with your duty cycle of 1/60 it should have lasted 15 days. Except your Buck regulator of 4V probably becomes less efficient when the the battery approaches 1V per cell then drops from ~80% to 70% or so.

So you definitely need to rethink your battery load requirements for 4V pulsed and 3V lower pulsed with a different dual regulator solution that can withstand cold temps for same or less cost than these rare pricey batteries.

Plan on a capacity of 2 watts with a capacity of 2 months at 1/60 duty cycle or 24 Wh and expect 50% efficiency with your present regulators, temperature and pulse loading. There are better ways , but we don't know the scope of this project.

Or figure out how to compress your power budget.

  • \$\begingroup\$ The 3500mA rating is only for 100mA. yes, but at 1000mA load current the capacity drops to around 3200mAh only. Also on page 13 from the L91 application manual (data.energizer.com/pdfs/lithiuml91l92_appman.pdf) says that with a pulsed current of 1A the capacity is around 3200mAh too. So, how is 500mA killing the batteries? Am I misinterpreting the datasheets? \$\endgroup\$
    – Andres
    Commented Jun 25, 2018 at 22:21
  • \$\begingroup\$ The datasheet I read on Digi-Key showed 10% worse loss at 0’C This manual says it is hard to compute service life for all pulses and duty cycles at all temperatures but your frequency and temperature makes the higher ESR calculations that they say to ignore become true effective ESR from the two chemistries circulating current thru higher ESR every 3 minutes which is far worse due to long time constant of memory rise and internal losses. So they are misleading. \$\endgroup\$ Commented Jun 25, 2018 at 22:52
  • \$\begingroup\$ Pulsed currents can be high but need to be short duration and lower than 1s intervals to avoid the voltage swing and circulating current losses in each cell. Therefore it is better to put all in parallel and have 2 boost converters with sleep / disable mode on the 4V to avoid the weak link stress in series with -V target \$\endgroup\$ Commented Jun 25, 2018 at 22:56
  • \$\begingroup\$ OP states that the first cell to die is #1 - which is not running the sensors, and would be expected to die after #3,4 if the extra sensor load was the case. (nb: I suspect that currents and battry#s the OP has given are not accurate) \$\endgroup\$
    – Henry Crun
    Commented Jun 25, 2018 at 23:50
  • \$\begingroup\$ In the datasheet data.energizer.com/pdfs/l91.pdf#page=2 , the graph "Industry standard tests" + "Personal Grooming" gives a usage pattern/currents very similar to OPs, suggesting a 12 day (>300min) life. \$\endgroup\$
    – Henry Crun
    Commented Jun 25, 2018 at 23:52

This is really interesting. I have never seen any datasheet info on the distribution of capacity between cells from the manufacturers, which is interesting in its own right. However taking a wild guess, I would expect 5-10% and never 50%.

If all is as you say, it is very unexpected. Your current profile seems to be well within the use case for these cells. In the datasheet data.energizer.com/pdfs/l91.pdf#page=2 , the graph "Industry standard tests" + "Personal Grooming" gives a usage pattern/currents very similar (but 750mA not ~400mA), suggesting a 16 day (>400min) life @400mA.

I suspect that something is not really as you say it is. The actual life is so far from a (conservative) calculation (7 vs 16 days) that something is wrong. A bad batch is probably unlikely, but only testing a new batch will prove or disprove that.

The currents drawn by a GSM module, and the on-times are probably very unpredictable, being dependent on loading of the cell site (how fast it responds), available operating bands, propagation conditions etc.

The sensor current may also be higher than you think. It is also very easy for a micro that you think is sleeping, to either come on more often, or for longer, or not go to sleep sometimes, or have higher standing currents (e.g. floating input pins or mid-voltages applied to input pins).

Some voltage regulators (both linear and smps) also draw much higher currents when they approach the dropout voltage. For this sort of application, you need to check the idle current of regulators over the full possible voltage range.

As Tony said it is no really good practice to have some cells have extra load, but you state that cell #1 failed first which is not consitent with that being an issue.

None of that really explains cell-cell differences between #1/2 and #3/4

You should do some controlled experiments on new cells, with a timer and a fixed load mimicking your system load, to eliminate the variables, and prove both the capacity and matching of the cells, independent of your system. Then report back - we will be very interested.

BTW, you can get cheap DC power meters off aliexpress for $10, that use true energy meter chips, and will totalise the energy actually used by your system over long periods.

  • \$\begingroup\$ Thank you Henry for your answer! Yes, 3G connection current can vary a LOT depending on signal conditions. My sensors current is pretty constant and I'm pretty sure my MCU is in sleep most of the time as expected (I've tested it) so I would expect all batteries to die sooner due to high 3G currents but only one of them died first, that's what makes me wonder what's happening. I will have a new batch of batteries next week, will run some tests on them and will report back with the results! \$\endgroup\$
    – Andres
    Commented Jun 26, 2018 at 2:11

I finally got new Energizer Lithium L91 AA cells to test once more and they now work as expected! They discharge mostly at the same time as expected! The system has been running for 20 days as of now.

So yeah, apparently all this problem was due to bad cells. I got the new cells from a different provider just in case. These new cells have the marking 12-2037 instead of 12-2036 so they should be 1 year newer than my previous ones.

Thank you, everybody, for your suggestions! I hope this experience will be useful to someone out there.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.