What is the point behind the two internal oscillators? More specifically, the 32MHz one, considering that the 2MHz one can be PLL-ed to 32MHz?
Maybe the 2MHz osc is more accurate and "safer" for start-up code, which has to jump through the unprotect function to set the 32MHz osc? The 32MHz is called a ring osc, so maybe it needs more time to stabilize?
I Like to Build Bikes : http://www.AtomicZombie.com
I Like to Hack Stuff : http://www.LucidScience.com
You can make the 32MHz oscillate at 48MHz to use it with the USB module, while the 2MHz with the PLL can be used for the CPU.
the USB module
At this point I am using only non-U versions. Therefore, I did not even think about USB. I guess that brings another question. Are the internal oscillators on the U versions really that good? I.e. good enough for USB, without external crystal?
There's a special setting for the internal 32Mhz clock that uses the timing of USB (packets/bytes/bits?) received to calibrate it to 48Mhz. This is then suitable for running the USB module. It's rather neat, and means that there's no need for an external crystal to do USB.
The 32MHz oscillator uses more power than the 2MHz oscillator. If you want very low speeds (say 500KHz) then it is better to divide down the 2MHz oscillator.
Also note that the PLL needs to generate at least 20MHz, so if you want to generate 8MHz you would have to set the PLL to 32MHz and then divide it down by 4. Not as efficient as simply dividing down the 32MHz oscillator, and the start-up time of the PLL is longer too (when coming out of sleep mode etc.)
I just want to bring up this old post again. Some comments was a lot of guessing, but also some good observations.
Let's say that we want the CPU at maximum speed, 32 MHz. Then, as far as I can see in the datasheet, the current consumption is MUCH lower by using the 2 MHz and PLL than using the 32 MHz directly. Only disadvantage I see, is the time to get locked and stable (if sleep mode is used). Is my observation true? The frequency stability and quality seems to be exactly the same, when calibration is enabled. I haven't fired up the spectrum analyzer yet (and the datasheet doesn't discuss this matter), but I assume a bit more phase noise using PLL. But this could be a good thing in digital systems (EMC), if not using ADC or DAC.
Does anyone disagree on my assumption on power consumption?
Then there are the special situations where e.g. USB or Hi-res is used....
The A3U datasheet says 270uA for the 32MHz oscillator, 85+220uA for the 2MHz oscillator + PLL.
So there's a little current advantage to using the 32MHz oscillator over PLLing the 2MHz oscillator up to 32.
277,232,917 -1 The largest known Mersenne Prime
Measure twice, cry, go back to the hardware store
Small current advantage, better temperature stability.
I'd guess that there is probably slightly less electrical noise too, if you are making precision ADC measurements etc, but I have not confirmed that.
I see, but I am going to use the PLL anyway, because I want to synchronize to another clock.
©2015 Atmel Corporation