AVR Freaks

If there is just a few mm's of track between the separate VCC pins and separate GND pins, a single capacitor of 100nF is sufficient IF there is not a lot of RFI around. Like Doc explained. For the TQFP package and smaller that can be easily achieved.

For a professional (read: industrial) design, I would definitely go for a capacitor on each Vcc pin. Cost impact is minimal. Like Doc explained

Also note that the Arduino isn't designed for industrial use.

Sidenote:

A few years ago (time flies ...) I discussed a design with the folks from elecfreaks.com: we were considering to design an Anaduino: an Arduino variant with special attention paid to decoupling, specifically for the analog part. In the end the conclusion was that the market for such an optimized design was small. So we decided not to invest time and money in that design.

cheers

Nard

I scatter .1uF 6030 size caps everywhere: Every power pin or anything that looks remotely like a power pin. If there's any analog work to be done, I separate the analog supply from the noisy digital supply with a few microhenrys.

I've seen the addition of a single .1uF cap settle down whole systems full of glitches and unpronounceable problems. Consider them jellybeans, except that jellybeans cost more.

I've seen guys say the capacitor is a critical part of the chip's power supply. Given that our chips work at a wide range of voltages, I'd say the capacitor might be THE critical part of the chip's power supply.

I just haven't found any guidance on how to pick the bead, but next time I put together a digikey order, I'll get a few hundred tiny ones.

I use a hard to find Toko 0805 with an unusually large high-frequency resistance. There are three important parameters (beyond size and cost). They are: (1) DC resistance, (2) High frequency resistance, and (3) Lower corner frequency. To a first order approximation, they look like a resistor followed by a parallel combination of an inductor and resistor. At low frequencies, you see the series R (the DC resistance). At frequencies where the reactance of the inductor is larger than its parallel resistor, you see that parallel resistor (the high frequency resistance). The lower corner frequency is where the inductive reactance equals the DC resistance. 

I try to get the lower corner frequency well below the period of the likely signal of concern (usually the MCU clock). The lower, the better. In this case, better = more effective filter.

You also need to be a bit careful of the DC resistance. It generates a DC drop depending on the average current (if it is well bypassed on the load side). There can also be power dissipation issues with this resistance. There MAY be a maximum current rating so be careful of that; it is probably dissipation limited but I would not count on that.

Jim