awneil 
This is a system with an ATmega328P.
Using 'Standby' mode, with a regular watchdog wakeup (for a timing 'tick'):
The sleep current consumption is ~ 90µA
Detail of the sleep level:
changing only the sleep mode, to 'Power Down':
detail of the sleep level of just one wake-sleep cycle:
I'm curious about the slow drop-off of the current when entering sleep - is that expected?
Full-size images attached.
Did you discuss this with your various capacitors & inductors? They make have a lot to say.
Also, how much leakage current is there when things are "off", say for 30 minutes
This is a system with an ATmega328P.
hmmm....
The power trace differences between Power-Save & Power-Down modes suggests you have an external crystal. is that so; you didn't make that clear ?
Interesting pictures... did you take care before of going Power Down to clear /as much as possible/ of PRR – Power Reduction Register?
awneil
Did you discuss this with your various capacitors & inductors? They make have a lot to say.
They are identical, so would be saying the same thing in both cases?
how much leakage current is there when things are "off", say for 30 minutes
Good question.
The 'Power Down' case seems to settle to ~ 5uA - which seems in line with datasheet values.
Here is 1 cycle of 'Standby':
The level remains pretty much steady; in the 'Power Down' case, the "drop-off" is ~ 30uA - which I'd have thought would also be visible in 'Standby' if it were due to "external" reasons?
The power trace differences between Power-Save & Power-Down modes suggests you have an external crystal.
Yes, that's true (3.5MHz); so the extended startup is expected - as the oscillator starts - is it also the reason for the slow "ramp-down" as it stops?
I also omitted to mention that this is at 3.3V. 
did you take care before of going Power Down to clear /as much as possible/ of PRR
Yes.
Yes, that's true (3.5MHz); so the extended startup is expected - as the oscillator starts - is it also the reason for the slow "ramp-down" as it stops?
A contrasting test using the internal RC oscillator might be interesting.
Oscilloscopes tend to have strange "recovery" effects after transient overloads. I would not highly trust the observed waveform if the transient is much more than 2x-4x full scale. Your mileage may vary with manufacturer, transient duration relative to the sampling rate, and the phase of the moon! And, probe compensation becomes even more tricky.
Jim
What tool is used to measure this waveform?
awneil What tool is used to measure this waveform?
This:
https://www.st.com/en/evaluation-tools/x-nucleo-lpm01a.html
and the pictures come from its associated PC software: https://www.st.com/content/st_co...
Oscilloscopes tend to have strange "recovery" effects after transient overloads.
So it's not an oscilloscope, but I'd have thought this effect, if present, would be the same in both cases?
But I will also try an oscilloscope ...
Can you believe it is possible...
awneil That's why I'm asking if it's expected behaviour!
EDIT
Sorry - which "it" were you referring to?
"It"- all, the board, the software, overall functions, everything. A new dimension of "know your power".
gchapman 1 KHz minimum bandwidth :
Real-Time Current Monitor with USB Interface Product Overview – ee203 (ee-quipment.com, bottom of page 2)
awneil
1 KHz minimum bandwidth
It seems a bit ambiguous about that:
But 1kHz really isn't much for this kind of thing.
EDIT
But still not bad for under $200:
https://www.ee-quipment.com/products/real-time-current-monitor-with-usb
gchapman I have a similar tool like that, maybe you can post and I can try it out and post graphs, but I don't have 3.5MHz external crystal.
Tool - Power Profiler Kit II
awneil Analog Hardware | Power Debugger
been there - the stupid thing can't save data: https://www.avrfreaks.net/commen...
Try setting up a transistor & resistor load ...switch it on/off via AVR & try your meas system on this load...does it accurately show a "square" on/off , going down to zero (a 2N2222A will shut down to handfuls of nanoamps at room temp).
Then you'll know it is/isnot the meas system settling out or having an offset. That will add (or lessen) confidence in your sleep measurements.
Try setting up a transistor & resistor load
I tired this with load resistor value of 100 ohm, VCC from my tool was 5V, switching the 2N2222A from AVR with base resistor 470 ohm. My room temperature is about 27 ºC.
I got square wave of current, with lowest being 280nA.
Looks okay?
May I ask OP to reveal, please:
wakeup source
sleeping time period (cannot read well from the pictures)
Fuse settings CKSEL
So your measuring system perhaps is working great...however your top current is WAY too high (50000 uA) to check if you drop to zero.
If you wan to know, did you drop down to, say, 0.1 uA ...your scope ADC steps ain't gonna show that!!
Start with something like 100 uA or 200 uA tops, most scopes read 256 levels (at 50000uA , that is about 195 uA steps @ 256 levels) and some maybe give 1023 levels.
...now the OP test system should also be checked.....just to be sure...then any results will be known to really represent reality
awneil
Wakeup source is WD interrupt; ~ 36ms period
@Heisen - what tool (HW and SW) did you use?
I do not have any new idea, you may question the measuring method... the first step you know, #21: reduce a peak current, drop the maximum current to be a fraction of.
Ask the manufacturer which is /measuring/ response to a step function /you have/.
@Heisen - what tool (HW and SW) did you use?
I am using Power-Profiler-Kit-2 from nordic semiconductor. I use their provided software named nRF Connect - Power Profiler. It looks something like this and has 100 ksps sampling rate.
Start with something like 100 uA or 200 uA tops, most scopes read 256 levels (at 50000uA , that is about 195 uA steps @ 256 levels) and some maybe give 1023 levels.
Okay, done. Now it looks like this.
Still showing the same current 290 nA. Looks like it's accuracy is good enough. Now that we concluded that my tool is working okay (phew !!), let's go back to the actual thread.
I tried the awneil setup (as much as I could understand), a ATmega328p on breadboard with watchdog interrupt, waking the MCU from power down sleep mode every 32ms (and immediately going back to sleep). Only difference is mine is running at 5V and at 16MHz. I tried running at 3.3V at 16MHz the waveforms were 99% same only difference being slightly reduced current consumption. Anyways here is what it's looks like.
The current waveforms matches with what awneil showed. Let's zoom in on those wakeup pulses.
I see some switching happening at the end. Answering the main question.
I'm curious about the slow drop-off of the current when entering sleep - is that expected?
No, I don't see the slow drop off of current. Not sure if this is how it's suppose to be. Let's zoom in a bit more on those pulse's where they end.
The current consumption literally drops to zero in between those switching and remains zero a little bit after, but then it slowly (but still quick) rises up like that and remains at 6.2uA (average) in deep sleep like shown till the next cycle.
Very nice!! Also, it seems like your "meter" has more than 8 bit, 256 level, resolution (nice).
What happens if you let it sleep a LONG time, say 20 seconds? Does that 6ua ramp up you show level out & stay at 6uA, or does it fall back down during long sleep?
If it stays at 6 uA for a long time, you can poke around & maybe find where it is going.
awneil
Analog Hardware | Power Debugger
been there - the stupid thing can't save data: https://www.avrfreaks.net/commen...
I dug it out anyhow:
So that is showing something of a drop-off - but not as pronounced as the ST tool.
But then it does look a lot smoother - so maybe it's just filtered stuff out?
Apart from not being able to save the data (as already noted), the Atmel tool is pretty useless for getting any actual measurements - no cursors, no graticule lines, and no summary data for the current view.
gchapman Apart from not being able to save the data (as already noted), the Atmel tool is pretty useless for getting any actual measurements ...
Microchip Gallery | MPLAB-Data-Visualizer-Standalone(Windows)
...
MPLAB® Data Visualizer v1.2.956
May 6th, 2021...
[mid-page]
WHAT'S NEW
v1.2.956:
- Visualize power consumption on supported platforms
...
awneil
Microchip's Data Visualizer, instead of Atmel's, and the most recent version?
...
MPLAB® Data Visualizer v1.2.956
May 6th, 2021...
Hmmm ... that's within days of when I was looking at it before - I wonder if that's a coincidence ... ?
I'll have to check again...
EDIT
Indeed, the version I have is 1.2.835 - but it reckons it's up-to-date;
That's the Standalone Visualiser - I guess they haven't updated that?
I really don't want to have to install the whole MPLAB IDE just for this!
EDIT 2
No, it definitely says it is the Standalone one:
But why would you have to "open it in the IDE" - if it's "Standalone" ?!
Is this some strange, new Microchip meaning for the word, "Standalone" of which I was not previously aware?
theusch 
Is this some strange, new Microchip meaning for the word, "Standalone" of which I was not previously aware?
Must be "Cheese"...
The farmer in the dell
The farmer in the dell
Hi-ho, the derry-o
The farmer in the dell...
The mouse takes the cheese
The mouse takes the cheese
Hi-ho, the derry-o
The mouse takes the cheese
The cheese stands alone
The cheese stands alone
Hi-ho, the derry-o
The cheese stands alone
awneil But why would you have to "open it in the IDE" - if it's "Standalone" ?!
You don't.
It is, in fact, a standalone download and it installs a standalone executable - there is no need for the IDE.
Apart from not being able to save the data (as already noted), the Atmel tool is pretty useless for getting any actual measurements - no cursors, no graticule lines, and no summary data for the current view.
Does now seem to do all those things.
gchapman But why would you have to "open it in the IDE" - if it's "Standalone" ?!
Was unaware, until your post, of the dependence of DV EXE on MPLAB X.
edit : Oh (#30)
But why would you have to "open it in the IDE" - if it's "Standalone" ?!
You don't, but the IDE has a way to manage packages, i.e. download, unzip etc.. in this case its a standalone package, you can download and unzip manually if you so please.
Jim
awneil in this case its a standalone package, you can download and unzip manually if you so please.
So why does the web page not say that?
