David VanHorn says:
- You've probably seen a lot of versions of how to do this.
What I'm going to give you here, is the result of a lot of lab hours, and a lot of thought.
Feel free to make your own experiments and verify (or not) what I'm telling you.
- First, you have to understand what bypass (or decoupling) capacitors do, and how they do it.
The job of a bypass cap, is to supply short bursts of current when the chip needs it.
The job of a bypass cap is NOT to smooth out voltage spikes.
The bypass cap does it's job by storing energy when it's available, and being able to release enough energy, quickly enough, that the current required by the chip does not have to come from the main power supply in a short pulse.
- When you think about this, you need to think in the right scale. So, whenever you see a wire, even as short as a tenth of an inch, I want you to picture a big filter choke. Every wire has inductance, and that inductance can help you, or hurt you.
Let's take those "sockets with bypass caps in them", and get them out of the way.
They're better than nothing, but only by a little bit. Why is that?
Look at the capacitor. It's suspended right between the power and ground pins isn't it? Isn't that really good, you ask?
Well, as I said, it's better than nothing. Look at those leads. Picture a cap with inductors on either side of it. Got the picture? Your power arrives at the VCC pin, and can go to the chip, or the cap, but the cap has these long leads. The chip has the same long leads, but there's nothing we can do about that, and as you'll see in a minute, they are actually helping you, inside the chip.
The problem here is that when the chip draws a current pulse, it comes partially from the bypass cap, but also from the VCC tracks. Current traveling on a wire makes a magnetic field, and BANG you've got EMI. As I said though, it's better than nothing, where all the chip's current demands would have to be traveling over the VCC tracks. At low frequencies, the lead inductance on the cap isn't an issue, but also it's small value means that it can't supply enough current for long enough. As the frequency increases, the lead inductances increase, and the cap becomes less effective, even though it could theoretically be a lot better.
- So what's the right way to do this?
First, put the cap as close as possible to the ground lead. You want ideally NO lead length, so a surface mount cap is good. Also, use as wide a track as possible between the cap and the ground pin.
Second, bring the VCC track to the capacitor on as THIN a track as possible. Thin tracks have high impedance, so RF currents won't want to go down that track much.
Now, route a fat track up to the chip's power pin(s). This gives as low an impedance as possible, but the length isn't something we can change, and that will give us some impedance.
It is significant, that the traces should both go directly to the cap, not join at some point, and run from there to the cap. In the correct method, the track looks like it runs under the pad. In one side of the pad, out the other.
Now, look at this from the chip's point of view. It needs current fast, so the current flows over the power wire between the die and the pin. This is a series impedance. Then it flows through our wide power track to the bypass cap. This is another series impedance. The bypass cap though, is a paralell impedance to ground, and it is a good low impedance source of current, so the majority of current is supplied from the bypass cap. The thin track out to the system VCC is long also, and a difficult path for RF current. We've given it an easy path through the fat tracks and the cap, and that's where the majority of it will go.
You've seen this shape of circuit before, it's called a "Tee" filter.
The effect of this seemingly minor tweak is to drastically shrink the area enclosed by the loop of tracks that this RF current is flowing through. Reducing the loop area reduces the EMI radiation in proportion to the reduction in loop area.
This works both ways. A smaller loop is also less susceptible to currents induced by external EMI or ESD interference, making crashes due to induced current much less likely.
- "What sort of capacitors should I use?"
Electrolytic caps are much better than they were 20 years ago, but they still make rotten bypass capacitors for high frequencies.
The best caps for general purpose bypass use, are surface mount thin film caps. They have no leads to speak of, and low series inductance.
Howeve, there is one more wrinkle in all this.
Theoretically, the impedance of a capacitor, at a given frequency, goes down as the capacitor value is increased. This is true, but also, it's internal inductance and resistance rise. So, for any given value of capacitor, there is actually a band of frequencies where it will perform best.
If you have access to a sweep generator and scope, you can measure this yourself, on the caps you are using.
What I've measured, and what I use as a rule-of-thumb, is this:
0.1uF at 3 MHz
0.01uF at 30 MHz
0.001uF at 300 MHz
Take the crystal frequency, triple it, and start looking there, because square waves are composed of an infinite series of odd harmonics, decreasing in energy as the frequency rises.
Suprisingly, this tells you that for a 10 MHz processor, a 0.1uF bypass will not be as good as a 0.01uF bypass.
If you don't believe me, then try it yourself, but remember, any bypass scheme is only as good as the circuit layout. You will never get good bypassing out of a cap of any value or type, with an inch of 8 mil track on it's ground side.
This technique of steering the current works at any frequency.
A few years ago another engineer challenged me on this, having seen me route this way on a flyback switcher output running at 300kHz.
The output diode, capacitor lead, and output pad, were arrainged as corners of a square, (one missing of course) with about 200 mils between them. Naturally, I used a 100 mil track, and ran from the diode to the cap, then from the cap to the output pad. This other engineer didn't believe that this would make a measurable difference, as opposed to making one big 'blob' and connecting them all together.
Well, when the boards came back, we built one up, and used a spectrum analyzer to measure the noise on the output.
Then, we jumped across from the diode to the output with solder braid, and agreed that this was a reasonable facsimilie of a solid copper triangle.
The output noise, at the switcher frequency was up by just over 2dB.
Granted, it still wasn't at a significant level, but 2dB is a lot of effect, just for running a track one way rather than the other.
Question: I've noticed over the years that the 'standard' has shifted. 25 years ago, you were supposed to put a 0.01 uF ceramic across power on TTL packages that generated incredible noise and used incredible power. Then the recommendation shifted to 0.1 uF ceramic or tantalum. Then in 2001 the recommendation shifted to 1 uF. Now we should have a combination of a high value and a small value. People are including large filter capacitors on the outputs of regulators.
If board space and cost are key issues, why include multiple capacitors? If package noise and power usage have dropped dramatically, why require more bypassing than 25 years ago? A 7805 regulator was only good to +/- 5% 25 years ago, but now its +/- 1%. Why do we then need more, not less, bypassing?
Answer: Its the logic type familys. CMOS switches from rail to rail very quickly, requiring a short pulse of current. CMOS inputs are also referenced to the supply voltage so if the supply drops because of something switching, so does the logic0/1 threshold. TTL on the other hand, has thresholds set by transistor junctions so it is less susceptable to this.
As the switching speeds increase, the need to get that current pulse out of the cap becomes more important. Big caps just can't do it so a compromise is required. The little caps can supply the needs of the chip but also need to be charged up again and fairly quickly. The next capacitor up the chain can be larger and a bit slower to provide that function. Those caps in turn, need to be topped up and if the regulator isn't fast enough, yet another cap can bridge the gap.
If space is at a premium, you can replace the large, low frequency electolytic caps and the small high frequency ceramic caps with a single (or fewer) Tantalium cap(s). They are able to react faster than the big caps and have more capacity than the small caps. Be aware though, Tantaliums are not nice components...see Capacitors
You want the caps to absorb well at the third harmonic of the clock. 0.1uF does well at 3 MHz, 0.01 at 30. 0.001 at 300. It's a broad response, so dont think that there's one specific value. However, if you use 0.1uF on a 20 MHz part, you won't get the supression that you could if you used 0.047uF.
Russell McMahon says
Practical capacitors are unfortunately not purely capacitive nature. They also have inductive and resistive components. Larger value capacitors generally have higher internal inductance and lead inductance than smaller valued capacitors of the same type. Consequently the impedance of a given capacitor will generally have a a minimum at a certain frequency and the impedance will be higher and both lower and higher frequencies.
The rule of thumb that was mentioned above is in the order of correct for ceramic capacitors. ie The "correct" value with short leads for decoupling in the MHz plus region is around 0.1uF. As the frequency rises a smaller capacitor will be optimum. The excessively enthused can even specify capacitors based on the series resonant combination of lead lengths and capacitance.
[the] ARRL handbook gives these figures for series resonance (optimum bypassing) for disk ceramics with total lead lengths of 0.5 inch.Cap uF Freq MHz 0.01 15 0.0047 22 0.002 38 0.001 55 0.0005 80 0.0001 165
Sounds like we should be using 100 pF decoupling caps with 100 MHz Scenix's ! :-)
RF practice (and serious microprocessor practice in some cases) is to group several capacitors of different values together to combine the characteristics of each = effectively a rather broad bandpass filter. Use of small ceramics and larger valued distributed electrolytics (tantalum for the brave, solid aluminium for the wise, wet electrolytic for the adventurous) can be useful.
Place the cap at the ground pin, and route power to the cap first, and from there to the chip. Otherwise, you compromise the bypass. Murata and panasonic make through-hole 3 leaded caps which help eliminate this by having ground, in and out leads (I and O are reversible)
Peter Cousens says
A trick I use on RF boards is to have non plated through holes included for small decoupling caps of around 100pf. I put surface mount caps (805) through the hole and solder it on the top and bottom, one of which will be the ground plane.
see Capacitors, Gotchas@, Noise@
|file: /Techref/power/decouple.htm, 14KB, , updated: 2010/6/9 11:08, local time: 2021/4/11 00:15,
|©2021 These pages are served without commercial sponsorship. (No popup ads, etc...).Bandwidth abuse increases hosting cost forcing sponsorship or shutdown. This server aggressively defends against automated copying for any reason including offline viewing, duplication, etc... Please respect this requirement and DO NOT RIP THIS SITE. Questions?|
<A HREF="http://techref.massmind.org/techref/power/decouple.htm"> Decoupleing</A>
|Did you find what you needed?|
Welcome to massmind.org!
Welcome to techref.massmind.org!