Bypass Capacitors- Non-idealities

[dknguyen]

I need to bypass a circuit with some periphreals and a dsPIC chip (runs at 40MIPS with 80Mhz). I only have 10uF tantalum capacitors (and for some reason it's cheaper anyways if I buy those ones than the 0.1uF caps from the local store).

Anyways, is 10uF bypass caps too large for these frequencies? In theory, the larger the capacitor but better but in actuality the parasitic inductances of the capacitors can start to kick in, but I am not sure what at which frequencies and which capacitor values it will start to matter for. I just know that it has something to do with the parasitic inductances and resistances form a RLC network in the cap with a resonance frequency (with larger caps more likely to have a lower resonance frequency) and if you operate the capacitor above that frequency then the inductances start to kick in and the bypassing breaks down. Any insight would be nice.

ANd could I try and get the best of both worlds by using a large capacitor and a very small capacitor in parallel in order to maximize the capacitance and reduce the parasitic inductance? (I think the much smaller inductance of the small cap should dominate over the larger inductance of the large cap, if it is anything like resistors in parallel).

Thanks.

 

[Nigel Goodwin]

I would suggest that 10uF on it's own is perhaps a bit high?, I would probably add a smaller one in parallel as well - perhaps 1000pF or 10nF?.

 

[Paul Obrien]

tantalums have a different construction to the normal electros and do not suffer from inductance nearly as much. I'd love to use tantalums instead of electros but the cost is vastly different(or at least the last time I checked).

 

[raybo]

there are 4 leads caps that you can buy for hi switching power supply the low inductance is the sole purpose of making these caps and they are not cheap.

 

[dknguyen]

there are 4 leads caps that you can buy for hi switching power supply the low inductance is the sole purpose of making these caps and they are not cheap.

These are the bulk caps right?

 

[RadioRon]

Its common practice in narrow band and broad band circuits to bypass with a range of capacitors to match the range of frequencies that the circuit uses and the range of frequencies the circuit might misbehave with (ie be unstable at) if it wasn't bypassed properly. This latter range is typically much bigger than many people think and depends on the type of semiconductors used. You might be building an amplifier to work at 60 MHz but only use a 0.01 uF ceramic bypass cap, which would be a mistake because the amplifier might have a lot of gain at 2 Mhz and suffer parasitic oscillation because the power supply is poorly bypassed at that frequency.

All digital logic circuits are broadband circuits with bands of interest from DC to, let's say, ten times the highest clock frequency (rough rule of thumb). So it is best if you can bypass an 80 MHz processor up to 800 MHz or so. This requires a ceramic capacitor of small value, and preferably surface mounted to avoid lead inductance (you don't need a high Q type cap for this, just a regular one is fine). The effective range of a tantalum cap depends on the value and voltage rating, but let's just talk rules of thumb. Here's mine. I would not use an aluminum electrolytic as bypass above 50 Khz, nor a tantalum cap above 2 MHz. I would use ceramic above 2 Mhz. I would use a 0.1 or 0.01 uF ceramic for logic circuits capable of up to 30 MHz toggle rates, (in other words with significant energy up to 300 MHz) and I would add a separate small value cap, say 33 pF for example, to cover toggle rates up to 100MHz (ie. energy up to 1 GHz). The cap that is intended for the highest frequency bypassing must be placed the closest to the supply pin and ground pin of the IC you are bypassing. So the smallest value gets the place of honour, followed by the larger value ceramic cap. The tantalum can be anywhere nearby and the aluminum electrolytic just needs to be on the same pcb somewhere.

Have a look at some good radio schematics. Critical sections, like frequency synthesizer chip for example, are bypassed with two or three ceramic caps in parallel, usually a large value, and a small value, and sometimes a medium value too.

So, for your 80 Mhz processor, you are poorly bypassed using only a 10 uF tantalum and should add at least a 0.1 uF ceramic and preferably one of those plus one 100 pF cap (although this is perhaps too conservative for most hobbyists).

 

[dknguyen]

So tantalum/niobium oxide for large values and ceramics for small values.

Have a look at some good radio schematics. Critical sections, like frequency synthesizer chip for example, are bypassed with two or three ceramic caps in parallel, usually a large value, and a small value, and sometimes a medium value too.

Do you have any opinion about the following article:
**broken link removed**

I have seen some others also say (along with this article) that is is better to use symetrically distributed small caps of the same value rather than of different values because it creates an anti-resonance reducing the filtering capability across a certain range of frequencies(discussed on page 2 of the article).

I would think though, that if you can't symetrically distribute a large number of identical small caps, that a several caps of different sizes distributed asymetrically (by necesity) would work better since I think that the small caps at the end of the chain would be too far away from the IC to do much good.

Grr! so much stuff!

 

[iso9001]

My rule of thumb has been a ceramic .1uF at every IC. An alum 22uF for small circuits, 47uF for medium, and a 100uF for my larger boards.

I should figure out some day how to accurately gauge my alum cap needs.

 

[Hero999]

Wow, I wonder what Intel does with a 3.2GHz processor, do they use onchip bypass capacitors or something? At 32GHz even the PCB traces would have a large enough inductance to cause the smallest SMT capcitor to resonate.

 

[dknguyen]

I think interplane-capacitance and 40-50 bypass caps for the Intel chip.

It seems that it's completely impractical to try and bypass past the noise from DC up to 10x the clock frequency. How do you decide which noise frequencies are most important and will affect the IC most? Really high? Really low? 1/10th the clock frequency to 10x the clock frequency? Or is it that depending on the circuit, noise is mostly exists at certain frequencies (and you have to test for yourself to find out...bleh!)

 

[Hero999]

The bypass caps aren't really for noise suppression, they are to reduce the effect of the inductance and resistance in PCB traces which increase the impedance of the power supply at high frequencies. Bypassing at ten times the frequency of the clock is a good idea as the clock signal contains odd harmonics up to a very high frequency, in fact 10 times is wrong, it should be nine times or elleven times as the harmonics are odd not even.

 

[eng1]

The bypass capacitor removes noise that couples into the device and is usually in the range of 10 uF - 100 uF, with a small parallel capacitor that has a higher self-resonant frequency.
In addition decoupling capacitors must be mounted as close as possible to the IC. Those caps create a low impedance loop and high frequency components (that are generated by the device and not coupled into) stay in that loop without coupling into other areas/boards and as a consequence EMI emissions are lower.
For high frequency devices I'd use a 10 nF in parallel with 100 pF (keep the ratio of 100). The dielectric must be properly chosen: for high frequency strontium titanate dielectric might be better.

 

[RadioRon]

Now we begin to see how the complexities of a circuit expand enormously at higher frequencies. As for bypass caps, one does what one can, and experience tends to narrow down your preferences. When we are designing a radio, for example, we don't use rules of thumb to guide us in placing bypass caps, we look at each and every IC pin, study the data sheet and application info, and then judge what values and types of caps that pin uses. Usually the key factor is to guess how much current is going into (or out of ) that pin and at what frequencies, and then tailor the capacitor choices to that. The point of the bypass capacitor is really to provide some charge storage local to the pin so that when the internal circuits begin slewing from on level to another and demand some coulombs, there is a charge source with very little series inductance (and therefore very little series impedance) in its connection, that can be tapped for quickly satisfying their thirst.

It also helps that the typical RF engineer often has a vector network analyzer standing by on which one can actually measure the impedance of a capacitor across a very broad frequency range. So it is possible to get some first hand knowledge of a capacitors behavior rather than relying on anecdotes, but, as usual, some instrumentation is needed.

Choosing bypassing capacitance is something of an art form so don't be discouraged if you don't get it right the first time.

 

[dknguyen]

I don't quite have a feel for the difference between 0.1uF and 0.01uF cap in terms of the difference in filtering of frequencies...I just bought a bunch of 0.1uF because they were cheaper (oh, and because there didn't seem to be any 0.01uF caps).

Here is something I am wondering, my chip runs at 40MIPS with a 80MHz clock. Now if I used an external clock, I would try to bypass for the 80MHz and its harmonics right? But what if I decided to use the internal oscillator and PLL of the IC? I would still need to bypass for all 80MHz right? I couldn't bypass any lower just because the clock signal is no longer working outside the IC? Because the internal gates of the IC still switch and draw power at that frequency? (I'm not quite sure how to word this last line)

 

[Nigel Goodwin]

I don't quite have a feel for the difference between 0.1uF and 0.01uF cap in terms of the difference in filtering of frequencies...I just bought a bunch of 0.1uF because they were cheaper (oh, and because there didn't seem to be any 0.01uF caps).

Here is something I am wondering, my chip runs at 40MIPS with a 80MHz clock. Now if I used an external clock, I would try to bypass for the 80MHz and its harmonics right? But what if I decided to use the internal oscillator and PLL of the IC? I would still need to bypass for all 80MHz right? I couldn't bypass any lower just because the clock signal is no longer working outside the IC? Because the internal gates of the IC still switch and draw power at that frequency? (I'm not quite sure how to word this last line)

Don't get too carried away, stick a capacitor across the supply and forget about it! - you 0.1uF should be fine.

 

[raybo]

Hero you realy don't think that 3.xxxx computer frequency is running all over your motherboard or do you think so?.And basicaly why capacitor become inefective at hi frq. because they are made with coil laminates and that becomes coils to hi frq. now a small mica will do you more good then 10ufd, and location is inportant not just hang a little C.