atmega8 to atmega168 porting problem

mpo · Joined: 18 Dec 2009 Posts: 39

Hi,
Not sure if my question is more bascom or hardware related, but I start here.

I have device based on atmega8 which has feature: "write few bytes to eeprom when power off".
It's done by monitoring power line. When the voltage drop, interrupt occurs and chip still has enough power from condensator to write to eeprom and stop.
This works perfect with atmega8.
Recently I switched to atmega168 (because of bigger flash) and it doesn't work anymore.
In both cases BOD is disabled, and internal 8MHz clock is used.
What I forgot to change, when I put atmega168 to my device??? Rolling Eyes

Relevant part of the code:

Arera · Joined: 23 Sep 2007 Posts: 386 Location: Wuppertal, Germany

To tell weather your problem is HW or SW, I would change R23 from 0R to 1K. Then pull down the regarding input, while power is still up and stable.
If the code works fine, you need more capacitiy on your backup-C!

Marc

mpo · Joined: 18 Dec 2009 Posts: 39

I tried this.
When INT1 is just pulled via resistor and I ground it manually, interrupt occurs and eeprom is written.
But I see no reason why backup capacitor is not enough. I tried to add 1000uF more and still nothing. According to datasheet, required current in active state is about 50% less than in atmega8.
Weird...

MWS · Joined: 22 Aug 2009 Posts: 2262

Evert :-) · Joined: 18 Feb 2005 Posts: 2156

and what power consumption did your measured with the M8 and the M168?

mpo · Joined: 18 Dec 2009 Posts: 39

Neill · Joined: 08 Dec 2008 Posts: 50

Is it too late to make some changes in hardware? I would use a comparator to generate the interrupt signal, with a crude reference like a Zener on one pin and a Voltage divider from Vcc on the other.

I think some AVRs even have an analogue comparator built in, perhaps it can generate an interrupt, something to look into?

Arera · Joined: 23 Sep 2007 Posts: 386 Location: Wuppertal, Germany

Signal forming is a good point, but I think doing that on the SW side using any kind of debounce should do.
What I woory about is the potential on the input.
If Vcc drops, there is no more Vcc. But does that mean there is a proper GND?
A 1k from Vcc to GND could help, but not for signal forming but to get a proper potential at the input.

six1 · Joined: 27 Feb 2009 Posts: 553

the +5V line is connected to other parts, so there might be a big capacitance inside which holds the 5V up!

Take another Diode to INT Input with little C and 10K R for defined drop down!

MWS · Joined: 22 Aug 2009 Posts: 2262

laborratte · Joined: 27 Jul 2005 Posts: 299 Location: Berlin

I agree to Neill. Monitoring a power line is a task for analog circuts.

If the falling time in power line is too slow (i.e. a big cap in power supply) the avr power behind the diode will also be fallen when the int occur. INT as digital input has a LOW voltage of about 0,2 to 0,3 Vcc, so it is depended on your voltage behind the diode. In worst case there will be not enough voltage for writing EEPROM when the interrupt occurs. AVRs are very sensitve on voltage for EEPROM write access, that is the reason why it is recommend to use BOD when working with EEPROM.

It seems that the mega8 could handle the EEPROM write with the cap's energy but the mega168 (PA?) fails because of slightly different electrical characterisics on LOW voltage and needed voltage for EEPROM write.

If you can change your circut use the analog comparator of the AVR. You can make a voltage divider in your power line (i.e: 3:1) which gives you a nominal 1,25V. Feed this in negative input of analog comparator and put the positive to the internal bandgap reference (1,1V). The comparator then can fire an interrupt when the voltage drops below 1,1V, which equals a 4,4V on power line. This gives you enough headroom for writing the EEPROM. Or choose a voltage divider which fits your needs, but make shure that you have at least 2,4V for 5ms after interrupt occurs. This is in fact an BOD which points to an isr rather than a reset.

BTW your isr should also handle the case where main power returns before the AVR stops working, i.e. a short power drop. Could also be done with watchdog...

mpo · Joined: 18 Dec 2009 Posts: 39

Well, the PCB is fixed, so there is no room for additional comparator.
On the output of 7805 I have only one 100nF cap. Total current is about 100-200mA, so voltage should drop immediately.

Some progress...
I tried 1k resistor in parallel to C2 - no changes.
I removed ceramic C2 - no changes
I replaced ceramic C2 with tantalum cap 1uF - random trash in eeprom
I added again 1k to 1uF - much better, but sometimes random eeprom address is written with random value

Unfortunatelly I have no possibility to check how fast voltage drops before diode and how long it's high enough on backup capacitor.

laboratte,
As I wrote, I don't have big cap on power supply line.
This used to be worked with smaller chip and I'm very surprised with this situation.
According to datasheet mega168 (not the picopower version in my case) works properly from 1,8V so it should be even more time before energy from cap will exhaust.
What do you mean by "fails because of slightly different electrical characterisics on LOW voltage and needed voltage for EEPROM write."?
Regarding 'power return situation' - I have watchdog running. When avr stop after eeprom write, it will be restarted after timeout.

I wonder if it will make difference if I disable pullup on D3 (INT1) pin?

MWS · Joined: 22 Aug 2009 Posts: 2262

You can try (if your circuit allows for it) to disable the output pins in the voltage-low ISR, maybe one of the pins sucks out the µC's power, so disabling them can buy you some time. The same, disabling the pullups can help therefore.
As you run in a stop-loop you could also use NOSAVE, which spares a few cycles. However, shouldn't make a difference, compared to the cycles used up by eeprom-erase/write.
But try it out...

mpo · Joined: 18 Dec 2009 Posts: 39

I added some sugested changes to isr

Arera · Joined: 23 Sep 2007 Posts: 386 Location: Wuppertal, Germany

Here we go again:
DON'T WASTE TIME IN ISR!
just set flag (bit) in the ISR.
Check for that flag in the main routine and do your stuff if it's set. Don't forget to reset the flag after your are done with the things to do aftre that int occured.

Marc