The author is a complete idiot.
that's not very nice or scientific in fact it just your opinion
I have been doing overclocking of pc's for years and with a regulated environment you can exceed the specification that the manufactures provide as they are just a guide anyways
my final year dissertation was on "practical overclocking to increase processing performance" intel said chip runs @ 2.66ghz I proved it is still running @ 3.8ghz almost 2 years on, so am I an idiot too?
results from exceeding specification vary with silicon batch as every batch of silicone has some differences but all work within a set tolerance just like passive devices +- x%
Like anything else in this world, crystals inherently possess imperfections, or what we often refer to as 'crystalline defects'. The presence of most of these crystalline defects is undesirable in silicon wafers, although certain types of 'defects' are essential in semiconductor manufacturing. Engineers in the semiconductor industry must be aware of, if not knowledgeable on, the various types of silicon crystal defects, since these defects can affect various aspects of semiconductor manufacturing - from production yields to product reliability.
atmel do sell a 3.3v part and why would they tell anyone to buy a part and under/over clock it when they can sell you one designed for xyz purpose, its cheaper for large production to make lots of one thing and just change the label thus more profit.
if this was an MOD product all part would have better high/low temp tolerances and better voltage range and cost that extra ££ more
If I was atmel I would do the same make 1 datasheet with very conservative figures to allow for manufacturing defects and as AMD have shown with the TLB bug one error can lead to end of a product, memory manufactures use a testing process called "speed binning" to determine which chip gets the faster or slower product code within a product family.
the only way to see if your batch of atmel chips will run at xy frequency is to test it and look for errors in the output of your function, no datasheet will be able to tell how that bit of silicone will react in your circuit as even the board traces can add resistance/capacitance/inductance to a running circuit.
I have a working pic rated as 8mhz running at 12mhz (same product family as a 20mhz chip), I also have a atmega8-8pu with a 16mhz crystal and a arduino 16mhz bootloader and that runs fine, maybe i should take my hot air gun and heat it until I get errors :D
I have also had the atmega32-16pu running error free @ 20mhz 5v
my temp range is about (12c-35c) for my projects if you needed very high or low temps or very accurate data that has no error handling then it would be best to stay within datasheet guidelines.
if like me you love to experiment then try it and report your findings to us :)
I have not done any testing of these chips at 3.3v lowest for me is 4.7v so I am intrested
What he is saying that the few he has tried work so far. I am so not impressed with this sort of shabby approach.
to push the boundaries of knowledge with experiments and testing is what science is about 8-)