Both. If the IC is expecting a very low supply network impedance, then more caps nearby may be desirable.
If better filtering is required (e.g., quiet analog circuit, EMC filtering?), the ladder design (input one side, series of branches, output from the far end) is better.
But neither circuit is really correct. It's more correct if we draw in some parasitics:
This illustrates the trace inductance between capacitors. It might be 3nH for 0603-1206 sized chip caps spaced close together. This suggests a low impedance lumped transmission line sort of circuit, with an impedance of sqrt(3n / 2.2u) ~= 0.04 ohms and a cutoff of 1 / (2*pi*sqrt(3n * 2.2u)) ~= 2MHz. Expect good attenuation above 2MHz, modest damping around cutoff (due to inevitable ESR), and expected behavior below (i.e., the 10uH into the total 4.5uF will cutoff at some point, and so on).
But that's still not very correct. Let's go deeper:
The ESL of an MLCC in the 0603-1206 size range is about 3nH. Assuming direct connection to (well-stitched) ground plane on the ground side -- I'm sort-of showing this by showing individual grounds on each capacitor.
Likewise, ESR is roughly typical, and I'm using some artistic license, assuming that 10uF is maybe, say, electrolytic instead. Thus giving very different ESR/L parameters. Some of these you can find in datasheets or appnotes; some you just have to infer from experience and experiment. Much depends on layout, when you're looking at this level of detail.
The power inductor inevitably has some DCR, capacitance and equivalent loss. The impedance curve (vs. frequency) isn't a straight upward line like we want from an inductor: it always peaks somewhere (where Xc = XL, only the loss R remains), then falls back down again. And sometimes has peaks and valleys after that (transmission line effects).
So now it no longer looks like a pure lowpass LC ladder filter, but the cap ESL's pass a lot more at high frequencies than the nameplate LC values should, urged on by the low impedance components of the power inductor.
Note also, here I've indicated the input/output connections as coaxial connectors; transmission lines. In general, you're always connecting a source to a load, with energy flowing potentially at all frequencies, from DC to RF. Back and forth -- and exclusively only back, or forth, never just sitting around. Energy always flows at the speed of light, and our task is to allow the right frequencies to pass, and reflect (or absorb) those that shouldn't.
You can also tell, from the inductances and capacitances, that there may be other problems afoot. If the output end is a modest load (many ohms, or constant current, as most electronic loads are), then there will be little damping on the overall CLC filter (10uF, 10uH, 4.5uF), and it will resonate easily at the cutoff frequency. This is a problem for loads of varying current draw (a periodic or step change in current draw excites the resonance, causing supply ripple), and for input ripple or transients (same outcome). It should perhaps be damped by adding another 10uF cap, with about the right ESR value as shown (or using a low-ESR ceramic with an external resistor added in series).
You can also tell, based on equivalent circuits like this, what the high frequency response might be like. Got a noisy switcher? Run a simulation on this and see what it does. The performance above 10MHz will not be impressive! What to do? Add more inductors (smaller and better ones -- most power inductors have a fair bit of capacitance and loss, if they're rated for impedance at all!), and use smaller caps, to mop up the HF range.
Throughout the process, make absolutely sure the ground is as contiguous as possible! The power distribution network becomes even more complicated if we draw those in: what happens to your circuit when "ground" isn't? Use shielding to your advantage, and surround traces and components with solid grounding (well stitched) where possible.
By the way, last year I designed the power conversion stages and EMC filtering for a medical LED lighting device. Apparently it's the only time, in many years of product design, that my client's managers have seen something breeze through EMC first try. I guess that means I'm too good; I probably could've saved a few cents skipping some of those 0.1uF capacitors!
Tim
Home
Help
Search
Login
Register
