Not enough to care.
Using multiple values can make things worse, because each pair of capacitors has an antiresonance (impedance peak) between valleys. In the video, Dave does not draw this correctly. The impedance of a parallel group is not simply max(Z1, Z2, ..., Zn).
Correct design process is not to ask "which is better", but to synthesize the network based on the requirements. Whatever the maximum impedance required is, that determines first of all how much stray inductance you can have from pin to the first bypass cap. Then the value of that cap, and the stray to the next, and so on. Conversely, an existing network (layout) can be approximated as trace lengths, and simulated to measure its impedance.
Most often, you will have a sequence of series trace lengths, and parallel bypass capacitors: a ladder network, which implements a low-pass filter, or lumped-element transmission line. Which therefore has a definite characteristic frequency and impedance. Impedance matching applies, so that real loss resistance must be provided (usually at one or both ends of the chain) to dampen the network.
Normally, you use large capacitors to provide the damping, but contrary to popular interpretation, the values are not large because they store energy. Indeed, the energy stored in those 'bulk' capacitors has very little relevance at all in the impedance of the network! The real purpose is to provide damping, through internal ESR. Tantalum capacitors are most suitable for this, because their ESR is convenient and stable. Electrolytics are good too, but because their ESR varies with temperature and age, they aren't as good. Ceramic and polymer* capacitors have very low ESR, so you often need to add explicit ESR (a resistor) in series with them for best results.
*Polymer types (in both aluminum and tantalum varieties) are available with modest ESR, but by and large, the most common types have very low ESR (under 100mohm), which means they are only suitable for very low impedance networks, like low voltage, high current, processor-core supplies, with careful layout accordingly.
Any added inductance in the network needs the same consideration, for the same reasons. Simply adding a ferrite bead and ceramic bypass cap can make things considerably worse at a modest frequency (a few MHz, usually), worse than simply leaving the supply unfiltered. Again, the best option is usually a one-two punch: low ESR ceramic bypass, with a larger value, lossy tantalum (or ceramic + R) in parallel with it.
Tim
Home
Help
Search
Login
Register
