Capacitors position close to IC. Why?

[Prime73]

I see sometime technical notes on schematics like:

* C11 close to
pin 13 (VD33),
R5 close to pin
14 (REXT).

Here is an example:


Are these caps have to be physically close to a IC chip and a trace length doesn't make any difference? or they should have a shorted possible trace on PCB? (or both) And the biggest question why it's matter in a first place? Are there any guidelines to read on this or when you design a board you go by your experience and see if it "works"?

[tszaboo]

inductance of the trace

[N2IXK]

Bypass caps are located close to the chips to minimize inductance and resistance in the traces, to shunt noise as directly to ground as possible.

For very critical applications, there are IC sockets available with the bypass cap built right in, for the shortest lead length possible:

www.farnell.com/datasheets/66800.pdf

[sacherjj]

Those caps are there to provide extra power when needed quickly by the chips.  This keeps power supply bus from dropping and becoming noisy.  Inductance reduces the increase in current flow.  So when a cap tries to provide a quick current hit to keep the supply voltage from dropping, the inductance of a long trace inhibits this. 

[mrkev]

By a looks of it, that cap is for decoupling. Every fast logic needs a close decoupling. Think of it this way, when the output change (f.e. from log.1 to log.0), there is very tiny time, when both transistors at output are not closed, plus you need some current to charge input of next device. This will make a current spike, (higher the speed of logic, higher spike for shorter time).
That is why you need cap at the voltage, so it can quickly draw charge from it.
Every route at PCB has few mohms and more importantly few nH. This means that if you get decoupling far from chip, it's effect drops significanly...

[Prime73]

thank you all for the explanations. So basically I would have to put these caps ( taking that schematics as an example) as close as possible to the chip and have the shortest possible traces.

[Rerouter]

The lower the ESR the better, considering how most modern ceramics have ESR in micro-ohms, you wont stand to benefit much, but for electrolytic's and other higher ESR caps then yes,

If you want to start learning a bit more about high speed signals, just plan out where the current will flow, both the signal and the return path, for low frequency stuff it will follow your ground trace wherever it may be (path of least resistance), for higher speed stuff it will want to go directly below where the current is flowing on the other side (or the path of least impedance)

Similar deal is wider traces have less inductance , and very close side by side traces will have capacitance (crosstalk)

[AlfBaz]

Here's a pic from the attached PDF showing the differences in inductance based on capacitor land and via geometries.


Curiously the attached appnote is marked for obsolescence and appears to no longer be available on Xilinx's website. I've seen some threads on this forum that seem to go against some of its recommendations
 

[PA4TIM]

A low ESR is often wanted, but not Always. And every cap has its pros and cons. Ceramics have low ESR but suffer from piezo effect. Some types of caps have huge voltage and/or tempco's others not. Mica's are not great at very low frequency (around 10 Hz) and MKT's are not great at very high frequency. Two caps parallel lowers ESR but increases selfinductance
Tantalum has low ESR and can last for ever, but are easy killed by over-voltage, some caps are low leak, others are can withstand pulses better.

Close to the IC has to do with the circuit layout. The trace between cap and IC, together with a possible input capacitance of the IC can make a nice LPF, sometimes wanted other times not. There are pcb's where the selfinductance of vias and traces can cause a lot of problems. The best place is making it part of the whole design.

Download the Agilent impedance handbook and there also is an appnote about ESL in caps, via's and traces.

A while back I had to filter some powerrails in an RF generator. I went a bit over the top with placing caps, ferrite, CLC filters and RFC's on strategic places. Measured impedances etc and it had to work great. It did work great in the  test with almost no load. But while testing with a dynamic load one rail started oscillating. After I found the cause a smal series resistor to increase ESR for a cap solved the problem. Not very strance, take some capacitance, add a sniff inductance, add some feedback and you have an oscillator.

[mrkev]

A while back I had to filter some powerrails in an RF generator. I went a bit over the top with placing caps, ferrite, CLC filters and RFC's on strategic places. Measured impedances etc and it had to work great. It did work great in the  test with almost no load. But while testing with a dynamic load one rail started oscillating. After I found the cause a smal series resistor to increase ESR for a cap solved the problem. Not very strance, take some capacitance, add a sniff inductance, add some feedback and you have an oscillator.

Well, everything can resonate... And i mean literary everything. I remembered "special devices" class, where we went to a devices that can handle microwawes (Our proffesor used to say:"1MHz? That's frikin' audio..."). Every cap has its nH (or pH), every inductor has its pF. What's total magic for me is anything that goes faster than 100GHz, at that point every flaw in package is critical as every fF makes the difference...