Hi welcome to Fundamentals! Friday Today we're going to answer the question: why do you use multiple bypass capacitors? You've probably seen this in many circuits that you've got your chip. Here, you've got your power power rail, and you might have more than one bypass capacitor just on that one chip, or even just that one power rail on a chip that might have multiple power rails you. But for example, it's not that uncommon to find like a 1 micro farad cap, a hundred nano farad cap, a 10 nano fairy cab, or 1 nano fairy cap can have 2, 3, or 4 caps in parallel. Why what's going on here? Hmm, let's answer it now.

I Actually covered this very briefly back in episode 33 back when I was in the old lab, but it was only like a minute or two explanation. So I thought we'd go more in depth here and I've actually done a not video not that long ago on why you would use multiple electrolytic capacitors in parallel. and I came up with a huge list of nine different reasons why you would actually put more than one electrolytic capacitor in parallel. So click here if you haven't seen that video.

it goes in-depth and it does some art thermal testing as well to actually prove it. Now, we're not talking about electrolytic capacitors here. this is a different scenario. We're talking about different value capacitors for in particular chip bypassing.

Now, there are very good technical reasons why you would actually want to put multiple capacitors for bypassing applications in parallel, in particular different values and different types of capacitors But before we answer that, we have to actually look at what is bypassing. Now, in an ideal world, you wouldn't actually need bypassing to be completely pointless, because let's take a look at a chip like this. Ok, it does. Whatever this chip happens to do, We've got a battery or a power supply here.

It doesn't really matter, and we've got a load. So it's a consuming power inside the chip to do various switching and things like that. And that's what I'm showing here with these two our MOSFETs in there. Let's assume it's a CMOS chip.

doesn't matter, and so it's doing internal switching. It's doing all its business. And we got a like a totem pole output. It's driving loads.

It's driving lines. It's doing whatever chips normally do. Now, let's assume that we had a five volt supply here. Let's go old school now.

this 3.3 volt rubbish. And this five volts in an ideal circuit. You're going to get five volts directly on the pin of this chip in here because there's no internal resistance in the power supply. There's no internal resistance in the battery.

Whatever you happen to be using, there's no internal. There's no resistance on the PCB traces that you're using. There's no inductance. There's no nothing.

It's just an ideal world and our ideal chip. Everything's hunky-dory You don't need bypass, capacitors, and every other chip on your PCB as well. It's also going to get exactly 5 volts on that pin. It never moves.

It's rock solid so you don't need any bypassing in an ideal world. Unfortunately, we don't live in an ideal world in the real world. Unfortunately, everything has resistance and everything has inductance. Everything has capacitance.

All these parasitic elements. and take your power supply for example it's You can't get a perfect power supply. It's going to have some equivalent series resistance AC a resistor in series with it. your PCB traces going from your power supply like your power supply input connector on your board, for example, to you chip or to multiple chips.

The PCB traces. They're going to have resistance. They're going to have inductance. Every piece of wire has inductance.

No matter how small, it's going to have capacitance down to ground. but we won't look at that in this case. So what's going to happen if we have no bypass capacitor on our VCC power pen of our chip? Our 5 volts. If the chip is doing nothing and it's just static.

Ok, yes, we will get a straight. We will get just our 5 volt line on there. but the chip is switching. It's doing stuff.

There's lots of capacitance inside the chip. Capacitance takes switching currents and things like that. So you're getting all these pulses of current. so our waveform is not going to be straight like this.

On our VCC pin, it's gonna. It might jump up and down like this depending on the switching inside that thing. And then we've got our load as well. Our load is powered through the VCC pin through that top transistor up there to actually drive the load, whether or not a sinking current or sourcing current.

So that's going to contribute as well. And hey, depending on the value of these traces here, it can. You can actually get significant dips and it can drop below the operating voltage of the chip and start causing strange weird things. This is sort of like a gross generalization, but this is what sort of thing can happen if you've got no local bypassing on your chip.

But one of the big problems is not so much the resistance of the traces. It's more to do with the inductance of the traces, especially the higher frequency your chips get. Even in that high frequency can be a megahertz or so. Look at an old you know computer board from the 1970s or 80s.

you know, with hundreds of chips on them. They've all got a bypass cap next to each particular chip because of the inductance of all the power traces going there. And remember, we won't go into details, but remember, an inductor actually resists change in current. So if you're a chip or your load suddenly decides it needs to switch, your inductor goes oh no, I can't change that quickly I Can't do it.

So you're going to get these huge dips and problems and all sorts of stuff so it all just becomes really nasty. And your five volt supply for your chip. Your power supply is not the solid power supply you're expecting, so that's why we add in a bypass capacitor in here like this, right at as close as possible to the pin of the chip, Because why does it have to be as close as possible? Because you're trying to avoid the inductance in the line here and every trace has inductance. So the further away you put your bypass capacitor from the chip, the greater the inductance and then causes all sorts of problems when your chip starts to switch at high frequencies.

So the goal of when you're bypassing is to try and produce a low impedance, low inductive supply element. Remember, capacitors store charge so they charge up and then when your chip switches and it requires a gulp of current, it comes from the capacitor instead of way way back on the other side of your PCB which has all these long inductances in series and all sorts of stuff. It comes directly from the local capacitor, so it minimizes the amount of duct inductance and resistance in series with it, so that bypass capacitor can supply that little gulp of current that your chip suddenly needs without being affected by the rest of your PCB layout and all the other parasitics. So let's take a quick look at what actually happens to an output pin here, for example.

which is really important because it's driving other chips as well as part of your system. So if you get issues on that output signal, it can cause corruption. the other chip may not read it properly. All sorts of issues like that and you may have actually seen this.

Let's take a look. So what with Drone is another waveform like this? In of course, the ideal world, your output will switch from 0 volts up to 5 volts. Here, it'll be absolutely perfect. There'll be no ring in that we know.

overshoot now. under shoot nothing. But of course I Use those key words: their overshoot and undershoot and ringing. What they're caused by is the inductance in the power supply here.

Even if you've got local bypassing bypass capacitors, there's going to be a little bit of inductance in the trace keys. Can't put it right on the pins, There's going to be a little bit inside the chip with the bonding wires for example, that actually you know because you died as like inside this. they've got to have the little bonding wire which goes over inside the chip that's got a little bit of inductance and all. That can actually lead to ring in on your signal like this and you've probably seen that before.

And then you can get some undershoot down here like this and causes issues like this. It's all to do with bypassing and the higher frequency content you've got, the more this becomes a problem. And I'm not just talking about the signal frequency itself. It could be you know 1 1 Kilohertz or 1 Kilohertz square wave for example, not high frequency as you would measure it on a frequency counter.

But remember that a change in digital signal like this, it's all. it's not to do the fundamental frequency, the time difference between here and here to do with how fast the rising and falling edges are, the faster the edge. You know, if it's a really slow edge like that, it's going to have low frequency content. If it's a SuperDuper fast edge that switches in a nanosecond or something like that, then it's going to have a really high frequency content.

That's your basic Fourier theory and all that sort of stuff. So even a 1 Kilohertz signal can actually have this real high frequency broadband content in there that causes all this ringing. And when you've got a complex system with many chips and everything else, well, it can cause a major problem even within if you've only got a single chip solution like this. If you don't bypass the caps and it's not going to clean power then internally - the chip is still going to get all this effective ringing and things like that due to the bonding wire inductance, your PCB trace inductance, and everything your ground inductance.

Here, it's not just your power line up here, you're going to have some inductance in here. You're going to have some inductance down here like this. So that's why it's important to have your bypass cap directly on the pins of the chip as close as physically possible. And of course, if you actually probe your power supply, you'll actually see this sort of stuff happening here.

Okay, you'll get your foot, you might have your five volts, but then you'll see that the ringing on the power slight yeah like that. So you'll get all these little. You'll see that if you actually probe correctly, there's high frequency probing techniques you need to use and everything else. But if your probe that you actually start seeing the switching on there.

and the if you have a no bypassing or not very effective Bypassing your ringing can be very big and calls all sorts of problems. So I know what you're saying Dave That's all great. but why not just whack one big bypass capacitor on there that can handle the most amount of current that this thing is going to pulse current that this switching chip and the system is going to take. Why do you need to have multiple different values and different types of capacitors on there? Aha trap for young players.

This is where we have to get into what a capacitor actually does and it's impedance versus frequency. Or let's go so in a real capacitor which I've shown in the previous video on electrolytic capacitors. If you're maybe want a bit more detail, it's not just a capacitor inside a capacitor. Here it is a real capacitor has an equivalent series resistance, which you might be familiar with.

the ESR which is a constant resistance value essentially in series with the actual capacitor itself, but crucially also inside a capacitor is a little tiny bit of inductance as well lead inductance, plus construction, inductance, and various things. and that's called the ESL the equivalent series inductance. So it's far from a just an ideal capacitor. it's an RLC circuit.

What happens with RLC circuits? Well, you can get resonances and you can get all sorts of funny things happening. And as you should know from your basic component theory for capacitors and inductors, they actually have an impedance or what's called a reactance or capacitive reactance and inductive reactance at a certain frequency. They effectively have like an AC resistance so to speak. and this is these are the standard formulas for your capacitive reactance and your inductive reactance and they change with frequency capacities is inverse with relation to frequency and the inductive reactance goes up with frequency and we're going to have a total impedance for the capacitor.

So a total a C resistance of the capacitor is actually going to be the ESR which is that constant fixed value in there, plus the impedance of the capacitor at whatever frequency you're talking about, plus the impedance of the inductor at whatever frequency you're talking about, So if we go over here and have a look at this graph here, we've got the impedance of the capacity Bypass capacitor. It's in Ohms of course. so the impedance in Ohms versus the frequency here. and you get this for a real bypass capacitor or a real capacitor we just happen to be using in a bypass situation, Real capacitor is going to have a response curve something like this, and this is sort of like an industry standard way to show it.

It is not a straight line like that because of course a capacitor will actually have infinite impedance at down at DC here, so it'll taper up like this. Now if we didn't, if this capacitor didn't have any inductance in it at all, of course this line would not be here and you just get a slope going down like that which changes with frequency and you can plot that yourself. Put the formula into Excel and you can do it yourself at standard basic component theory. But as I said, crucially, that little inductor in there.

it's tiny. It could be like Pico Henry's or something like that, but at a particular frequency, it's going to start to matter now. the capacitive reactance operates like this, but at some particular frequency here, which is the resonant frequency of this RLC circuit. Using a standard resonance formula, that's where the capacitive reactance and the inductive reactance are equal and that is going to be the resonance point.

At that point, then the impedance of the reactance of the inductor starts to dominate instead of the impedance of the capacitor. So hence White reverses and the resistance starts to go back up. And that's a very undesirable thing to happen. You don't want this thing to go back up at higher frequencies.

You want it to be down like this. Why? Because as we talked about before about the series, effectively this series resistance. the series impedance. You want the energy to come directly from the capacitor with no effect whatsoever with no impedance in the paths, no inductance in the paths.

When you start adding this real inductance either inside the capacitor itself or outside of the capacitor with your PCB traces, you're in inside the chip with a little bonding wires. Everything else then this can be a real problem. You impedance starts to rise and your bypass capacitor isn't acting like a good bypass capacitor anymore at these higher frequencies. And of course it's a these higher frequencies in modern devices say for an FPGA for example which take here have huge densities in a huge amount of switching, huge amount of logic and multiple rails, and they take huge amounts of current and everything else.

And they operate in extremely high frequencies like you know, hundreds of mega. They can switch at hundreds of megahertz, but the edges are even faster and you can get our frequency components into the gigahertz range fairly easily. And if your reactance of your bypass, the impedance in your bypass capacitor starts to rise at these really high frequencies up here. at hundreds of megahertz or a gig or whatever, and you're going to be seen serious trouble.

Your bypass capacitor may as well not even be there at these higher frequencies because yet, the capacitance is still there. It's still got. you know, one microfarad or whatever it is. Lot of capacity.

You can have a lot of energy stored in that one micro farad capacitor, but it's no good. It can't get into the chip if there's this massive series impedance in series with the capacitor. It just can't deliver the energy when that, when your IC actually requires it, Give me a big pulse of energy not. can't do it now.

I think I mentioned before that not only do you have different values here, but you have different packages as well because the package actually makes a difference. As a general rule of thumb, the smaller your package gets, the lower inductance it's going to have the lower your internal inductance here. So it let's assume that this one is a Oh six, oh three. For example, you know a SMD package.

Then if you've got an OE a five, it's going to look something like that. It's going to have a higher value. so that could be a 805 and then you could have. And you guess that Oh 402 package Looking something like that.

They're actually going to have different values for the different packages, so it's actually better for higher frequency stuff to use the smaller packages. But of course, the big question is, why do they use different values? Well, different values have remarkably different frequency characteristics as you'd expect the bigger value capacitors in this case, say 1 micro Farad, for example, is going to have a reson appoint at a much lower frequency, so it's going to cover the lower frequency range is going to have a lower impedance at a lower frequency. Once again, it's not this V-shape it's You know, it's going to be something like this, right? So it's going to actually cover a much broader range at a lower frequency right down here. But yeah, work with me.

Okay, and then you're going to have different values for assuming like this, all the same package. For example, hundred n is going to be high in frequency and then a ten in again and then a one nano farad, then 100 puffs if you want to is going to be much lower. So what you get and the answer to the question, Why do you use multiple bypass capacitors? It's so you get the lowest impedance across the largest frequency range possible. So if you've got all three of these values in here, your final curve is going to look like this: Tada.

So you've got a much broader lower impedance. so you've got a more effective bypass capacitance over a bigger frequency range and that's why you do it. So there you have it. That was a bit longer explanation than what I intended.

What was it? 20 minutes or something to explain how bypassing works and why you use multiple bypass capacitors I could have just jumped straight to this and said this is why, which is what I did back in episode 33 or whatever, but there's good background information there to explain exactly what's happening here. So I thought I hope you found that interesting. but hey, I think we might be able to reproduce this on the bench and actually show. You Could be a little bit tricky, but now let's give it a go.

now. ideally to measure this, we would use a network analyzer. big expensive bit of kit which I don't actually have here in the lab I need to get myself one but hey, we can use our red Pattaya which you've got seen in a previous video and now even I'm now powering it from an external plug pack to amp plug pack by the way over via the USB which seems to have solved the rebooting issues I was getting before even though before in the previous video I was actually powering it from a USB 3.0 port which is supposed to be capable of supplying two amps, but now I don't know Anyway, so it's working a bit more reliably now. But I'm still having a few issues with the impedance analyzer app, which we're going to use today.

So I'm going to use three channels here, and here's a diagram of how it's actually hooked up. We've basically got a 10 Ohm shunt resistor in there and then the device under test. Now, the reason I'm getting this convoluted arrangement here with the bit of Erebor and the wires and everything else is that you're probing in this sort of thing and your wiring test cabling is actually quite critical. If I actually ran, coax is off here and stuff, we'd find that we'd be getting all sorts of issues in our impedance plot the higher up in frequency we go.

So yeah, I'm often just dangling wires like that can be better. So I'm converting my SMA to B and C, then converting B and C to a banana and a binding post here. so I can just hook that up and should be right. It's a little bit, you know it's a bit crude, but hey, we should be able to show the concept at least now.

The good thing about the very board here is that it has two convenient strips like this that allow us to put multiple capacitors in parallel. So I've got a cap in here? I've just been testing the thing to make sure it all works and we've got our 10 Ohm shunt resistor there. Something just put as many caps in here as we want. but with something like this we're dealing with, you know, high frequency.

We're going to go up to 60 megahertz today. I sweep it all the way up to that frequency. So what we want to do first is actually replace the capacitor with a shunt resistor in there because we've got a 10 Ohm sorry. I've got a 10 Ohm shunt resistor, replace the capacitor with a resistor so that we can actually check to see our frequency response responses flat and there.

We're not getting any weird effects caused by a cable in all the test setup. So just like we discussed, what we want to get is an impedance versus frequency graph. So basically anything that goes up to and megahertz should, we should be able to see something like this. This thing's 125 makes samples per second.

you know, analog bandwidth 50 60 megahertz? Something like that that'll be good enough to see various Arc capacitors in parallel. Hopefully now you'll have to forgive me for not doing this live so to speak. But not only does it save time, but trust me. I Spent a lot of time around with this thing actually trying to get a result because the test setup is actually quite crude.

Had a lot of issues with the red pitaya. there's software and things like that and the test feature with even with you know the short wires that I'm using, the BNC s make a difference. The adapters, all that sort of stuff all comes into play. so you know I Didn't really engineer a proper setup for this, so I was actually lucky to actually get a usable result out of this.

But I should be able to show you something here. So what we've got here is an impedance response graph just like we saw on the white board their impedance in Ohms versus frequency there. In this case, we're sweeping from 100 kilohertz up to 60 megahertz on a logarithmic axes there. I Tried to set it to start at a higher frequency, but I just wouldn't let me.

I'm not sure what's wrong with the app anyway. Um, you can see that started off at 100 kilohertz there and right down a DC. Of course it started off as a nice perfect 10 Ohms. Exactly what you'd expect, so that just verifies that the system is working.

But of course now the parasitics of our test setup come into this. You can start see it around about two megahertz there. or so you know it starts to roll off and you know it's usable. Up to say you know, 20 megahertz might be usable.

It's down to measuring. You know, 7 1/2 Ohms or something like they're You know, good enough a ballpark, but at the higher frequencies of course. then it becomes you know they're all the parasitics of the Vra board and everything. test fix should come into play.

You can see a bit of noise right at the high frequency. That's because there's not much out signal-to-noise ratio there. but in this case that in accuracy at the you know greater than 20 megahertz range isn't that bad because some of the impedances as you'll see actually go up to you know, hundreds of Ohms and things like that. So you know it, it's kind of usable.

So I'll sweep it to 60 megahertz, but just keep that in mind that yeah, it's a little bit off up there. And I'll start out by showing you some large value capacitors. This is a 10 Micro Farad 1206 packaged ceramic capacitor. very typical large value bypass cap and as you can see it does have that characteristic V-shape response.

As I said before, quite, you know, much broader than what we saw on the white board. What's there? You can see there's a resonant point about one megahertz there and then it tapers back up. Now here's a 10 micro farad are tantalum capacitor and you can see it's actually higher in value, goes up to like 1.75 ohms at. you know, 60 megahertz or something like that because see, it's got a similar shape, similar sort of resonant frequency around 1.5 megahertz and now this is a just as a curveball.

47 microfarad electrolytic capacitor. you can see it resonates about, you know, 8 9 megahertz or something tapers back up and obviously that big tail down at the end is Judas and parasitics on the Vero board. Now here's a very typical hundred Nano Farad Oh 805 bypass capacitor. You'll find in practically every product and you'll see notice that the impedance scale has now gone up.

It's auto scaling and this Red Potato software was a little bit annoying that I couldn't actually manually a scale the thing to see the data, but it's changed significantly. We're talking about hundreds of millions before, but now down at 100 Kilohertz, we're talking like up over 15 ohms. quite large valued, no good for low frequencies. and you'll notice that the resonant point is now up to you know, around about five or six megahertz, so higher than it was with that larger value capacitance.

And now we'll take a look at a a 10 Nano Farad capacitor same Oh a 805 up package. but you'll notice that the Y scale has changed even again by an order of magnitude down at 100 Kilohertz. But let's up up over 150 ohms or they're about 10 times more than what it was before and you'll notice that the now the resonant frequency is right up near 40 50 megahertz or something like that. In fact, this test setup isn't good enough because we're talking about.

you know much higher frequencies here, so but as you can, they are actually quite broadband. You know tens of megahertz for these values like you know, quite low impedance. Now if we actually combine a 10 microfarad ceramic with a hundred n ceramic and a 10 n ceramic, you can see that we have look that rise around about 8 megahertz there, so it very similar to the combined peak response we got on the whiteboard. Now here's an interesting little trap for young players, which we didn't discuss before, but what happens in reality Now you can see on the left-hand side the same graph we had before of the combined 10 microfarads plus the hundred in there and you'll notice the big lump in there in the middle that around about 8 megahertz or so.

Now this is actually undesirable because look at the one on the right as we saw way before, this is just the 10 microfarad cap on its own and you'll notice the y axes are very similar. It's actually a better result just to have the 10 micro farad capacitor there. In this particular case, With this particular, these particular values on this particular very board with all its particular parasitics and everything and the values and the hole and the test setup and the whole works. it can actually be detrimental in some cases.

To put capacitors in parallel, you can form these resonant peaks there and sometimes it might interact with your hardware in ways that you didn't intend. so you know it's not just magic, you can't. Just you know. put Y 10 different values and whack them all in.

You know you could actually get an issue with resonances between caps, so it's a potential pitfall. Just watch out for in this case is not particularly bad. but look, just the 10 microfarads on its own would technically be better in this particular case. Now here's a better response if we actually combined four caps a 10 mic, a 1 mic 100 N and a 10 M.

Once again, all SMD ceramics in various size cases and you can see that that 8 mega Hertz peak has gone away. It's you know, still at where we can argue that this is a bit better than the original, our 10 microfarads just on its own. But yeah, it's hard to see this because the higher frequency ones really need a higher frequency response test system which we don't have here. And here's my for ceramics in parallel here: 10 microfarads, 1 microfarad 100 N, and 10 in various different package sizes and the package sizes are going to make a big difference in terms of the SR and the impedance response of the individual capacitor.

It's not just the capacitance value package plays a big part, so I couldn't really get lots of visually good results with just the SMD ceramic capacitors. they're just too good. So I got like a really poor axial sorry radial lead at 47 Microfarad electrolytic capacitor and put that in parallel with a 10 nano farad ceramic on there and you can see that you know peak around. you know, 15 16 megahertz or something like that.

but the extra 10 different ceramic brings the impedance of that way back down again at the higher frequencies, which is desirable of course. And that little Matteo back in up after 40 megahertz is just due to the test system. as we saw right back at the start. the 10 Nano Farad's would allow much better high frequency performance into the hundreds of megahertz and things like that that the 47 Microfarad electrolytic on its own.

It just keep going up and up and up and there'd be hundreds of ohms at there free. It would just be way off the scale at that frequency and you may as well not have it at all. So that's a reasonable example. Visual example of how combining those two I caps actually can, you know, get a reasonably smooth response over a very broad range from 100 kilohertz right up to you know, maybe a few hundred megahertz or something like that, but we can't see it.

But yeah, it would be quite decent performance over that big entire range. So you use the 47 Micro for decoupling big heavy current bursts and the 10 Nano Farad for all the high frequency switching. Now here's a little interesting aside, you may have seen weird looking service mount caps like this in a wide package like this. Well, why you're not you and you may not have thought anything of it.

Well, these are actually special low E inductive capacitors designed specifically for this application. Now if we have a look at this little snippet from an AV X app. Note on these low inductance or the evolution of ceramic capacitors Here you can see that say a 1206. you standard 1206 one has about 1200 pica Henry's or there abouts of inductance right? But if you take that exact same sized chip the 1206 and you put the caps on the sides, the conductive caps on the sides instead of the ends.

same size cap but a hundred and seventy puffs. And if we have a look at this TDK datasheet for their RC series, their specific low ESL equivalent series inductance that we've been talking about. They're called reverse geometry and they just put the conductive end end caps actually on the side of the capacitor instead of on the ends and it makes the world of difference. And if you're designing in our high frequency switch mode, power supply or something you might see you know real performance-critical are stuff where the you know the bypassing is really going to matter then you might typically find these low ESL caps in there.

So there you go I Hope you enjoyed that rather lengthy look at how bypass capacitors work and why you put multiple values and types in parallel. There's some real good reasons for it and sorry I couldn't really you know comprehensively. show this. This test setup is pretty crude.

It's not the best thing you really need you know, a really high frequency high performance system and carefully lay it out, test setup, and everything else. but hey, you know, just with this we were able to see so it did actually take quite a lot of mucking around and trial and error just trying different caps and different sizes and packages and values and things like that just to try and get a response. And I probably you know ultimately could get a more realistic sample of what I showed on the whiteboard there. but I hope you get I Hope that was good enough and you really get an idea of how it can really make a difference, especially at really high frequencies you can imagine.

just you know. Extrapolate those graphs right out and assume we've got a perfect test system and can make one heck of a difference. Anyway, if you liked the video, please give it a big thumbs up and all that sort of jazz you know. We discuss it and links down below for data sheets and other uploads and things.

Catch you next time you.