This project has very little power use from the AVR - nearly everything is high impedance except maybe one LED which will be less than 5mA. Does it really need a bypass cap on all four VCC/GND pairs? That seems a bit overkill. What about one for the VCC/GND and one for the AVCC/GND. How are they connected internally? Is this a good idea or a bad idea?
mega1284 tqfp capacitor on all 4 vcc/gnd pins?
Author
Message
Level: Moderator
Joined: Wed. Mar 28, 2001
Posts: 29086 View posts
Location: Sydney, Australia (Gum trees, Koalas and Kangaroos, No Edelweiss)
I would put 4 of them in...but you can always try with just the 2....but are you ready to modify the board if things play up unexpectedly? 
Put pads for 4; use what you need.
Actually, the problem is not load currents. It is switching (changing logic state) of thousands and thousands of internal gate output nodes. Or, more correctly, it is the capacitance at each node. This takes a current surge to charge up (and discharge) those thousands and thousands of little capacitors. Back in the 8051 days, I measured power supply current spikes around 100mA that had a pulse width of maybe 10ns. Geometries are smaller now, making for smaller node capacitance but the gate count has gone up substantially. The spikes may be a bit higher and they are also narrower.
The external caps provide a reservoir for delivering the charge needed to change all those internal gates. That is what the whole bypass game is about. If you had to pass FCC or CE emissions testing, you would learn this VERY fast.
Jim
Last Edited: Thu. Aug 22, 2019 - 04:31 AM
And that begs the question - how close is a capacitor on one of those vcc/gnd pairs to the wire bonding inside? Is its capacitance sufficient for what is going on? You guys are probably right and I should just throw 4 on there.
I generally find that "close" is not much of an issue. Other than external crystal, bypass caps are the ONLY components I have close. I use a ground plane on the back side of the board and drop vias right next to the MCU ground pin pads to ground. Then, bypass caps are close to the power pin pads with a via to the ground plane where-ever practical close to the part. I usually have the Vcc end of the bypass cap within 1/4" or so of the MCU pin using 0805 ceramic caps; the bypass ground via is usually within 1/8" or so. I make them tight, even when it is not perceived to be "necessary" (due to board constraints).
That is what I do.
Cheers
Jim
Level: Moderator
Joined: Wed. Mar 28, 2001
Posts: 29086 View posts
Location: Sydney, Australia (Gum trees, Koalas and Kangaroos, No Edelweiss)
And that begs the question
Ours is not to question why, ours is to cap the die.
alank2 wrote:
And that begs the question - how close is a capacitor on one of those vcc/gnd pairs to the wire bonding inside? Is its capacitance sufficient for what is going on? You guys are probably right and I should just throw 4 on there.
Presumably the internal wiring is "good enough", or the devices would be unreliable and thus unsellable. Definitely a good choice to allow for all 4 caps. I think it is often the case that different power and GND pins actually supply different sections of the chip (one case being to allow more I/O current through the GPIO pins than a single bond wire or chip trace can handle), so it is not as though using 2 caps vs. 4 will give the entire chip half the bypass capacitance, but rather that some sections of the chip will be bypassed while others will not.
js wrote:
Ours is not to question why, ours is to cap the die.
<groan> Nice one.
alank2 -- Do what you want with your own designs. There are many treatises about decoupling cap, for AVR and otherwise. If you have a design with a low clock rate and negligible power requirements, it will probably run on a breadboard with no caps. Now, do you really really want to use that lashup to control your sprinkler system and pool and security system, under all rated conditions?
Thanks everyone; I'll just put the 4 on there and have no fear!