AT90s1200-12 latched short circuit

I have a AT90s1200-12 with four switches (PD3-6 to earth) and a 
pizzo-electric transducer (PD0 also to earth.) The switches are 
diode-OR'd to INT0 (PD2.) 
The program spends most of it's time in power-down, until it gets a 
button press.

I powered this from 4 AAA cells. Since new 1.5V cells give over 1.5V I 
included a couple of diodes (1N4001) in series to pull the voltage down 
a bit. I measured it to be about 6.15V and decided that was close 
enough. (Maximum voltage for a AT90s1200-12 is 6V.) I don't have good 
current measuring capability, but I did what I could to check that the 
current was about 10-20mA active and < 1uA inactive.

The project worked for several days and then lay idle for a week or two. 
On trying it again it was not working (pressing the buttons - no 
sound.) Now, several days after noticing this, I just opened it and 
burned myself. The chip was too hot to touch, and the diodes were 
hotter. The voltage was 5V. The open circuit voltage was 6V. So it was 
sinking a lot of power. On putting the processor back in, it works OK 
and there's no indication of it failing again.

The only routes to either power line with no button pressed is via the 
pizzo transducer, or the reset pin (standard power-on reset circuit.) 
There is no limiting resistor on the pizzo, but even with the chip 
driving it hard I wouldn't expect it to develop much more than 20mA. 
That would not make the 1N4001s hot enough to burn my finger.

(The 6502 had a Halt-And-Catch-Fire instruction. It seems I may have 
found one for the AVR. ;-)

Has anyone seen this behaviour? Might 0.15V above maximum voltage cause 
this to happen? Or am I looking for something else?

TIA for any suggestions.

-- 
Richard Urwin