Decoupling Capacitors

[MrJammin]

I've read texts that recommend placing decoupling capacitors on the power supply lines of all ICs to electrically isolate them. How necessary is this action? Are there general guidelines as to when this would be most useful? Also, what values of capacitance would I have to use if I were to implement the decoupling capacitors.

 

[dknguyen]

It is absolutely necessary. The boards inside your PC would not be able to run without them and some PCBs have the entire backside just for PCBs. Newer electronics have a wall of capacitors around the chip.

-0.1uF for analog. 0.01uF for "general" digital. You want smaller capacitors for higher devices.

-You can also use parallel capacitors across the power supply since this parallels the parasitic inductance on the capacitor thereby reducing it (same goes for parasitic resistance but it's not as important).

-You can parallel different sized capacitors to bypass a broader range of frequencies

-THe capacitors should go as close to the pin as possible with the smallest cap being closest to the pin. The smaller the cap, the more useless it gets the farther it is away from the pin.

-Use ceramic capacitors as they have the least parasitic inductance of "affordable" capacitors. At uber frequencies other types must be used.

THEORY: Since capacitors have parasitic inductances that scale up with capacitance, giving capacitors a finite frequency range they can bypass. THis is why you use smaller caps for higher frequencies- because the increased parasitic inductance outweighs the additional capacitance at high frequencies.

Paralleling different sized caps to attempt to cover a broader range of frequencies works okay for most "slow" ICs but for really high frequency (like PC processor high) ICs, this doesn't work since the parasitic resistances and inductances form parallel tank circuits. Since they scale with capacitance if you use different capacitors it makes the tank circuits different, it causes different resonance frequencies between the tank circuits and this causes resonant impedance spikes at certain freqiencies and energy bounces between the two causing problems and not bypassing less effectively than you might have expected. It's not as simple as adding the bypass frequency range of the capacitors together. In this case (really high frequency circutis are where it is most noticable) you use many of the same sized capacitors (small capacitors since it's high frequency). The paralleling increases capacitance and reduces parasitic inductance and resistances at the same time.

You guessed it...it's a black art.

 

[Speakerguy]

Agree with the above, though I almost always use a 0.1uF on chips unless I run into problems with that. Usually you will have a larger electrolytic for several devices and ceramics at each device (for each rail if running multiple voltages, like a dual-supply opamp). MLCC type ceramics are the best, we use 0805 surface mount caps right at the power pins for everything we do at work. Since lots of IC's have power pins on diagonals I usually stick it next to the power pin and the other end to ground plane.

But yeah, it's necessary. You can get away without them a decent amount of the time, but it will screw you up more than often enough that it's always worth it to stick one on.

Also, don't do something like putting a .1uf and a .33 or .01uf cap in parallel thinking you will get better bypassing. You might, but it can cause problems when you parallel two similarly sized low ESR caps. Stick with a 100uF or larger board level electrolytic and .1uF chip ceramics at each IC.

 

[Souper man]

I rarely use decoupling caps across chips, but when you need a precise signal, it can help drastically. It smoothes the incoming voltage. when working with motors, awlays put a large value (3300uf) across the + and - terminals, because when you power up motors, they require a lot of mA compared to your chips' needs. It supplies the chips with additional power when the motors kick in. This helps maintain a steady speed or accuracy.

 

[Hero999]

3300:mu: capacitors are good for suppressing the surges generated by motors but you still shouldn't forget to add a 100nF capacitor across the IC and 100nF is also a good idea across the motor to suppress the high frequency RF noise it generates. For optimum RF suppression connect 100nF from each motor terminal to the motor's chassis (which if possible should be connected to earth ground) and put a ferrite bead over each of the motor's terminals.

 

[Souper man]

what is a 100nf cap? isnt it like 1uf? I never use nano in my cap measurements.
just pico,micro, and farad. some pretty big jumps, but works.

I hate the american measurement standards. I wish they taught everybody the metric system,which is so much easier. I self taught myself metric, because imperal sux

 

[ericgibbs]

what is a 100nf cap? isnt it like 1uf? I never use nano in my cap measurements.
just pico,micro, and farad. some pretty big jumps, but works.

I hate the american measurement standards. I wish they taught everybody the metric system,which is so much easier. I self taught myself metric, because imperal sux

Check this out:

**broken link removed**

 

[Speakerguy]

nano is the prefix between micro and pico. none of it is our stupid imperial system.

Personally, I am trying to develop a workable metric time system. 60s in a minute 60 minutes in an hour 24 hours in a day split into two sets of 12 with 365 days a year except every 4th year? who the hell thought of that....

 

[dknguyen]

Personally, I am trying to develop a workable metric time system. 60s in a minute 60 minutes in an hour 24 hours in a day split into two sets of 12 with 365 days a year except every 4th year? who the hell thought of that....

It was introduced with the rest of the metric system. However, time is so prevalent that it proved too difficult to change. What was it? 100s in a minute. 100 minutes in an hour, 100 hours in a day or something like that. But it doesn't matter anymore since the second is now defined by the cesium atom.

 

[Hero999]

what is a 100nf cap? isnt it like 1uf? I never use nano in my cap measurements.

Well it's very common.

I prefer to use nF for values between 1nF and 999nF and :mu:F for values between 1:mu:F and and 99,999:mu;F, any larger I use F and any smaller I use pF

just pico,micro, and farad. some pretty big jumps, but works.

Remember, it's every power of three it jumps to a new name.

For values <1:
[latex]p = 10^{-12}[/latex]
[latex]n = 10^{-9}[/latex]
[latex]\mu = 10^{-6}[/latex]
[latex]m = 10^{-3}[/latex]

For values >1:
[latex]k = 10^3[/latex]
[latex]M = 10^6[/latex]
[latex]G = 10^9[/latex]

Values outside the above range are vary rarely used in electronic engineering so I've omitted them.

I hate the american measurement standards. I wish they taught everybody the metric system,which is so much easier. I self taught myself metric, because imperal sux

I agree but nano is a metric quanitiy.

 

[jpanhalt]

what is a 100nf cap? isnt it like 1uf?

That question has not been answered per se: 100nF is 0.1 uF (microfarad).

As for metric vs. the imperial/American system. I use both and like both, but for different purposes. There is nothing wrong with dividing an inch into tenths, hundreths, thousandths, etc. NB: Something machined to 0.001 (typically the smallest graduation on a dial) is more precise than something machined to 0.05 mm (also a typical smallest division on a dial). Sure, one can divide the dial into 0.025 mm units, but it is also easier to count in "thousanths" than it is in "0.025 mm" intervals. Consider the problems people have in multiplying by 1000in their heads.

That leaves one to argue about whether the inch or the centimeter is a more "logical" unit. I like the anthropomorphic reality of an inch. A size 32 in. sleeve has real meaning and 81.5 cm is too big a number to visualize

As for taking fractions of various dimensions, including time, I prefer numbers that have more factors. That is, 36, 12, 16, etc. are easier to deal with than 100 and 10. Just how do the metric countries measure a third of a meter anyway? John

 

[dknguyen]

That question has not been answered per se: 100nF is 0.1 uF (microfarad).

As for metric vs. the imperial/American system. I use both and like both, but for different purposes. There is nothing wrong with dividing an inch into tenths, hundreths, thousandths, etc. NB: Something machined to 0.001 (typically the smallest graduation on a dial) is more precise than something machined to 0.05 mm (also a typical smallest division on a dial). Sure, one can divide the dial into 0.025 mm units, but it is also easier to count in "thousanths" than it is in "0.025 mm" intervals. Consider the problems people have in multiplying by 1000in their heads.

That leaves one to argue about whether the inch or the centimeter is a more "logical" unit. I like the anthropomorphic reality of an inch. A size 32 in. sleeve has real meaning and 81.5 cm is too big a number to visualize

As for taking fractions of various dimensions, including time, I prefer numbers that have more factors. That is, 36, 12, 16, etc. are easier to deal with than 100 and 10. Just how do the metric countries measure a third of a meter anyway? John

I don't have a problem with dividing inches into multiples of 10. What I do have a problem with is dividing feet, yards, miles into weird multiples like 12, 16, etc. Although I understand the convenience of using multiples of 12, 16, and 36 when counting on your fingers, most technical things don't seem to have such luxuries in which case it seems to make everything harder. It's also silly when you consider that there is no imperial equivelant to measure things like capacitance, resistance, magnetic fields, etc.

 

[jpanhalt]

Although I understand the convenience of using multiples of 12, 16, and 36 when counting on your fingers, most technical things don't seem to have such luxuries in which case it seems to make everything harder.

How so? BTW, most people have only 10 fingers.

Binary also seems to work better than base 10 for computers. A lot of coding systems work on hex. Some people confuse anything that is decimal with metric. Metric refers to a specific system. Its usage is not as standardized throughout the world, as some might think.

It's also silly when you consider that there is no imperial equivelant to measure things like capacitance, resistance, magnetic fields, etc.

Do you really mean "silly." The American system simply incorporates those units in it. Just like our language can assimilate words from other languages. There's nothing silly about that. In fact, it makes perfectly good sense.

 

[dknguyen]

How so? BTW, most people have only 10 fingers.

Binary also seems to work better than base 10 for computers. A lot of coding systems work on hex. Some people confuse anything that is decimal with metric. Metric refers to a specific system. Its usage is not as standardized throughout the world, as some might think.

Do you really mean "silly." The American system simply incorporates those units in it. Just like our language can assimilate words from other languages. There's nothing silly about that. In fact, it makes perfectly good sense.

Math is not language. Mixing unlike things in math causes problems. Counting fingers refers to everyday counting like how many apples to buy at the grocery store.

A lot of my resentment comes from having to write down lbs/32.2ft/s2 over and over and over and over and over again during my mechanics courses in place of m.

 

[Hero999]

Binary also seems to work better than base 10 for computers. A lot of coding systems work on hex.

Binary is good but inconvenient for everyday usage. Also it can only accurately represent fractions which are powers of two.

Metric refers to a specific system. Its usage is not as standardized throughout the world, as some might think.

Most of the world uses the decimal system for wieghts measures and even currency.

Do you really mean "silly." The American system simply incorporates those units in it. Just like our language can assimilate words from other languages. There's nothing silly about that. In fact, it makes perfectly good sense.

Form whatever opinion you will but I feel that the decimal system is just easier to use mainly because it's consistant; It always uses base 10. Imperial is inconsistant, for example there are 12 inches in a foot and three feet in a yard and this only makes calculations harder as you need to mess around with remainders and fractions.

The imperial and decimal systems both make perfect sense, however as decimal is vastly easier to use, using it in preference to the imperial system makes more sense.

In my opinion the only reason why some people want to keep the imperial system is because they're used to it.

 

[Souper man]

I knew it was a borderline (close to 0 uf, between 0.1uf and 1 uf) value, but I wasnt sure.

 

[RadioRon]

I want to comment on the original poster's question about why use decoupling caps so I apologize for interrupting the discussion about nF.

When a logic gate switches states, it does so very quickly. While switching, many (all?) logic families require more current than when they are not switching. You can call this a surge current if you like. So, while the gate's output is swinging from high to low or from low to high, its internal circuits take extra current and its output stage takes extra current to force the output voltage to change quickly despite the capacitance that the load inevitably puts on the output. This current surge is very short in time, and in some types of logic can be very large in amplitude and all of this surge must be supplied by the DC power supply system.

When a surge is so short, and has the shape of a pulse with short rise and fall times, then it contains a lot of high frequency energy so we have to think about high frequencies when considering how effective our DC supply system is. This seems contradictory, but that is the key to why we use bypass/ decoupling capacitors. In a circuit, the faster the logic gates switch, the wider the bandwidth of the DC power supply must be to support it.


The DC power supply is very complex from the point of view of high frequency energy. It usually includes some sort of regulator IC somewhere on the circuit board, followed by a distribution network of copper traces and planes feeding positive supply and ground to the ICs. This network of copper combined with the voltage regulator has an impedance that is too high to allow the logic ICs to get the full amplitude of current that they need during switching.

Don't forget that ohm's law always applies. If the IC demands a large surge of current while switching, and we want the voltage to the IC to stay constant, we need a low impedance in line with that current. If we don't have a low impedance, then every time the IC switches, its power supply voltage (as measured at the logic gate IC pins) will drop and then pop back up. This happens very fast and since there is too much inductance in a pcb trace, the voltage overshoots and "rings". This puts a huge amount of undesirable AC noise on the DC power supply which is also a bad thing.

We can claim that in many cases, it is impractical to lay out the + supply and ground copper to minimize the power supply impedance and it is impractical to expect the average regulator to behave ideally at frequencies beyond 10 or 20 MHz. We have found that the most convenient and economical way to lower the impedance of the power supply system as seen by the logic IC is to put some local charge storage right there at the IC pins. The capacitor is our local charge storage device, acting like a reservoir of local coulombs attached right where it is needed. You can't use any old capacitor since you want to be able to move current in and out of the capacitor very very quickly, so we need a capacitor that behaves correctly at high frequencies, hence we choose the ceramic capacitor.


The value of the capacitor should be related to the amount of surge current demanded by the IC and by the rise and fall times of that surge. But we don't have the time or ability in many cases to measure or calculate every surge current requirement by every IC and ceramic capacitors are very cheap, so designers generally use rules of thumb in determining the best value of each bypass capacitor. These rules of thumb are stated in other posts here. In every case, the basic rule of thumb includes 1) use a ceramic capacitor and 2) put one capacitor on every logic IC directly from the +V to ground pins of that IC.

 

[Souper man]

Very nice reply. That taught me more than I knew

 

[Bob Scott]

I hate the american measurement standards. I wish they taught everybody the metric system,which is so much easier. I self taught myself metric, because imperal sux

Souper, don't get US mixed up with Imperial. The American SAE system is different from Imperial.

A US gallon (128 US Oz) is 3.78 Litres
An Imerial gallon (160 Imperial Oz) is 4.5 Litres.
A US pint is 16 Oz.
An Imperial pint is 20 Oz.
And the volume ounces are different too. US Oz. are bigger.

Force/mass conversions can be confusing. In metric, KG can be used for weight and mass. The unit of force is the Newton. In Imperial, Pounds can be used for weight and force. The unit of mass is the Slug.

Do the British still use wrenches that are uniquely British size?

I'm also curious..Do you still need to buy licenses to own televisions and radios in England?

Bob

 

[HarveyH42]

Souper, don't get US mixed up with Imperial. The American SAE system is different from Imperial.

A US gallon (128 US Oz) is 3.78 Litres
An Imerial gallon (160 Imperial Oz) is 4.5 Litres.
A US pint is 16 Oz.
An Imperial pint is 20 Oz.
And the volume ounces are different too. US Oz. are bigger.

Force/mass conversions can be confusing. In metric, KG can be used for weight and mass. The unit of force is the Newton. In Imperial, Pounds can be used for weight and force. The unit of mass is the Slug.

Do the British still use wrenches that are uniquely British size?

I'm also curious..Do you still need to buy licenses to own televisions and radios in England?

Bob

If you go down to the pub, and order a 'pint', do you get 16 or 20 ounces?