Hi in a previous video which I'll link in at the end of this end down below number 859 for those playing along at home, a very popular one on bypass capacitors and why you need them and how they work and all that sort of stuff. but I've always wanted to do a video doing some Munson what is Munson Wow let's actually have a look I Made a quick look Madmen Months was they used it like a car salesman back in the day. A nice old TV sets and things like that. no link all these in down below aren't there's an old Bob P's article.

what's all this Munson stuff Anyhow, and he talks about how Madmen Months was famous for carrying around a pair of side cutters like this. and whenever the engineers you know designing the TV sets would show him the TV what they proud of their creation, he'd go in there, starts nipping out components probably capacitors, resistors, whatever start snipping them out, and if the TV set still worked then that component went. once the TV sets stopped working up, put that last component back in, ship it. Well, that's the story anyway.

how true it actually is I don't know. but anyway, Munson is the term for snipping out or removing components until your product works and then shipping it because obviously those components weren't needed in particular, bike paths, capacitors And got me thinking, can a product work with absolutely no bypass capacitors? Well, the obvious engineering answer to that is maybe or no depending on the circumstances. So I thought it'd just be an interesting, fun experiment to actually take a product, remove all the bypass, or snip them out one by one, and see where it stops working. If it does stop working at all, let's go.

So I Thought we'd have a try with this Gigatron TT or microcomputer because this is a classic double sided PCB layout. You can see all the tracers running vertically like this, on the top mostly anyway, and you can see them all running horizontally on the bottom with you know, just some a sprinkling of ground fills. There's you know, a couple of stitching points, the ground layer, and stuff like that. So I'm not sure I don't have the PCB fold a hand to have a look at that, but anyway, it runs all our 5 volts.

so higher voltage. Our excursion rails than your typical were 3.3 volt one once. it does 6.2 5 megahertz so it's not particularly quick. But if we can get something like this to fail, then that'll be really interesting.

And you note that it has lots of bypass caps on here practically I Think there is one for every chip actually. So in, you know, traditional time-honored engineering practice, they've put a there's just specified in the design a bypass capacitor for each and every chip, whether or not they're actually required. In terms of the layout and the impedance of the traces and and the loop paths and all sorts of things, it doesn't matter that you know a typical double sided 5 volt TTL digital board. And of course, the other good thing about this is that it has a ready output for us to see if it's failing or not, so we can just get it to do stuff.

like if the computer if the VGA output starts to wobble or or do something weird or you know anything like that, we should be able to actually see fires pretty readily at the system level. So first of all, we'll start by actually I just probe in some key signals around here. Obviously, we can't probe everything. it's all too hard.

There's just too many signals to probe, but we won't know specifically what point actually failed. We might be able to analyze it later if we can actually get it to fail, but let's measure some points just so that we have a baseline and of course, using proper signal integrity techniques to measure them and then start removing some caps. See what happens now? At this point, it's actually very important to get your ground in correct. I've actually shot more than 10 minutes of a waffle talking about the correct ground probing technique in different probes and things to use and how you set up your scope.

So I decided to rather than do that now I left it to the end of the video, so jump to about that. Just over the 19 minute mark on nine and a half minutes and you'll be able to see that right at the end, right? So now it's Madmen months in time. I'm gonna get my psyche out. it's not going to go in there and hopefully not short anything out.

Be careful. and I'm gonna trim off bypass caps and see if anything changes. And I've got my Mandelbrot I running in the background. So I'll be able to see if anything changes.

So here we go: I don't know. I'll go over to the condition decoder. You see that? Yeah, and I'll just trim off some caps. So here we go.

Yeah, we go. It's one two three. Everything's still working now. Mandelbrot Still going.

This is our instruction register kind of important. Nothing's changing yet looking pretty jazzy. but I wouldn't expect it to be a big deal. Well, yep, shorted it.

Oops, let me run that Mandelbrot again another one Gonski sorry, that's all around the instruction register, data register, bus buffer, bus, access dakota, and the condition decoder. They're all without decoupling in that section. Address Mode decoder: Gonski It's this made man months. Our technique is it's working Instruction decoder.

Let's do the instruction decoder down here. We don't need to do any, it's still working of course. you know, like, but don't necessarily expect these to change you would. You would definitely see the signal integrity change if we were actually probing directly on these chips here.

So what I'll do now is I'll actually do the one on this all go like right in the nearest point down here to the signal. but of course it's not coming out of that chip, so it's probably not going to do much. if anything not. Yeah, we got because the signal integrity has to do with the loop area of the ground and the signal.

So the clock signals come in from either up here. it's coming from wherever and it's It's all forms what the signal integrity is all part of a big effectively a loop in here. I'm what's called a loop area and that's the most important thing for generating and reducing electromagnetic interference. For example, if you've got a if your clocks over here and your loads over here on your board and the current has to run right across the board and all the way back and you really got very high-frequency edges in there, then that loop area is what's going to generate huge electromagnetic interference.

It's going to act as a huge antenna, so you know this thing would not. If you had a design like that, it wouldn't pass like a EMI EMC compliance and things like that to just be spewing out radiating crap out there. You can alleviate that with ground planes and other things. This thing doesn't have a solid ground plane, it's just got like split ground planes with you know, lots of hickety pickety cars in their to do with wherever they could drop the vias down on the board Anyway, still working.

Let's go. No. I'm gone. Oh, that was me.

I'm just actually got this load by here because they see something going across there. I've snipped a few more now I Forgot to press record. they're gone. Ski.

What made me and Muntz is right. Saved a few cents on our design already. Ok. I've now removed all the capacitors bypass caps from this side of the board where the instruction register the RAM the ROM everything is and it's no problem whatsoever.

I Should have actually measured a few more signals around here. so the clocks actually weren't really a good idea. And really, you know you should see the LEDs up there. Stop by chasing - the program stops I Have snipped the leads on every night night.

one right near the SRAM right near the Airstream There we go. I Snipped every single bypass cap on that board except for the main filter cap on the input this board. His computer runs fine with absolutely no bypass caps. I'm not that surprised if we turn out our reference waveforms there.

You can see that you know a little bit difference in there. geez, where we're really talking very minor differences. but hey, you can see him right? Nothing on the blue waveform that looks absolutely identical, but the clock one which is down in the middle of the board, down near the or gates down there. Yes, a little bit of change, but hotly maybe a little smidge up there, but basically nothing.

It's identical. All the bypass caps are gone so firmly. zoom in on that waveform where it's two nanoseconds per division. You can see that's probably like two and a half, maybe three nanoseconds rise time or something like that, so you know it's It's a pretty sharp edge for a TTL signal.

Okay, what I've done now is actually taken out the the big bulk electrolytic cap there on the USB input. So there are literally unless I've missed one, not a single bypass capacitor on this entire board, even the bulk input capacitance or the ceramic hundred in bypasses on there. And that's what we get there. It basically is deviated very little.

Oh, our blue one is now our blue ones now changed a little bit from our reference point. Look at that. But apart from that, like like it's practically no difference whatsoever on those clock signals. and quite frankly, I wouldn't expect difference much else and the computer still works fine.

Not one bypass cap like I'm not kidding. they've all been lifted every single one of them. Like I said, that one is not a bypass cap. that's like a 47 puffs there, your crystal ones there.

but everything else. They've all lifted every single one of these caps on every single one of these chips and things still works just fine. Now of course some people might jump in and say, oh, look, it's the ground plane. This thing's got a ground plane, so in combination with the power traces on there, it forms a capacitor.

And when you've got big ground planes like internal layers like a four layer board, If we had a giant ground plane in the middle and then the capacitive plane next to it, then yes, that is a thing. You can actually get away. Bulk distributed capacitance. In fact, you could write a PhD thesis on this actually analyzing bulk capacity of ground planes versus individual caps.

And there's you know, a lot of science that actually goes behind that. and there's a lot of theory that says, and in practical demonstrations, they can show bulk distributed capacitance can be better than individual bypass caps like this. So in theory, if you're doing a real high frequencies design, you want to put your ground planes. That's why.

yes, stack up in your PCB When you actually stack them up, you don't want the ground plane down here and the positive playing up the top with these signal traces in the middle, because not only is that a pain in the arse, you can't access your signal traces on the outside, but then that just screws up the capacitance between the layers. You're better off sandwich in them right together as close together as possible on the internal prepreg what's called when they make up the multi-layer PCB And then you get the distributed capacitor effect. To remember: a capacitor is just two plates separated by an insulator, And that's exactly what a PCB is. Two big planes, ground and power, for example, in this case, five volts separated by the Fr4 dielectric material in there.

So it's one big bulk capacitance. But anyway, that doesn't really apply in this case because we don't have any full ground planes or full power planes or anything like that. so the power with traces is just going willy-nilly so there might be a little tiny bit I Don't know, you know, half a bee's decapitates of capacitance in there. But yeah, it's nothing.

So it's not that this thing simply works fine with our bypass caps. but I've worked on systems before with bypassing problems in them, and just the act of probing. Just the act of the capacitance of your probe is enough to cause your circuit to suddenly start working or failing. Depending on where Murphy is.

on that particular day, the capacitance your probe affects the signal under test. and it could you know once again. As I said, if this ringing goes low enough to affect the threshold, then your gate could switch and then your system can be completely screwed and your capacitance can have a positive or negative effect on that. Okay, now let's actually just randomly probe some other signals on here.

I've just got in one of the pins of the donor whether it's a data address of the ROM chip and you can see, of course there's data switching in there, we're referencing it to the clock here. we could actually reference it to Channel 2, but you can basically see what in the trade is known as an eye diagram in here. It's actually still pretty good. Like there's nothing really dipping down hugely low.

This is with absolutely no bypassing at all. It's it's really quite nice. That's why the computer is still working. But although the computer is working herein lies some of the problem.

now. I've just probed pin 5 of u29 here, which is part of the X register. I Don't know what it's doing, but it it we are have just single-shot captured this and it's interesting. Look at this look.

We have liked this little runt pulse going up here and that's obviously caused by you know, grounding type issues somewhere else in the system. And then we've got this little glitch over here and things like that so you know the computers working, but you could really come a guts are on these, especially if you're right. You know your power supply is varied and something like that because once that passes the signal threshold what they like, the TTL thresholds either positive of Vol or voh then it can change the level of your chip or I can. If we're talking about a flip-flop or something, you can get into a metastable state I think I've done a video emitter stability and things like that and it can really ruin your day.

So that's you know, a pretty horrible looking waveform there. So what I've done is actually re-soldered the bypass caps around that particular chip and the source of it over here. and I've put those ones back. We'll see if it makes a difference.

We're still going to see ugliness, but and maybe some of the little glitchyness might go away. Okay, so here we go. his single shot capturing some stuff so that looks that looks better than before. Of course you know we're still got.

you know, stuff happening out here. Wiggle wiggle wiggle wiggle. Yeah, over here. No.

caused by our you know, a ground bouncing elsewhere like in other parts of the circuit of switching over here. Actually, that one's going right down like that. That's a legitimate Sal edge it iment thing. like a legitimate end in a second pulse happening there.

So yeah, but you know you still got some other issues, but obviously it's a maybe it's a bit better. Hmm. now check this out. This is a rather interesting experiment in terms of like grounding layout and loop area and are switching how the you know the grounds relate to the various switching elements and going across our ground domains.

and I won't go into all the details but there's lots of This is where ground plains really come into play which this one doesn't have and I'll show you some more variation on this now. I'll go to this output register up here, which happens to be very fairly clean and you'll see what I mean by that in a second. Okay, look at this. Check out the ground signal.

The ground there? Okay, there's very little in the way of like you know, there's your regular noise down there. but there's no periodic switching noise as you'll see in a second. So note that as we say, go over to this chip over here. I Know, you know I'm just like probing random pins, right? But look at that.

look at the amount of switching noise on there. and no, this is not due to the bypass caps. I've actually sold in. Most of the caps are back on this port.

I think I'm only missing one or two. Okay, so let's go over this chip over here. just random probing. Look at that.

Look at all that switching crap in there. you can see it isn't it horrible and that is basically just due to the layout and signals, crossing ground pars and doing all sorts of stuff. So this is like the Hi free. that'll be the high frequency clock in there doing that.

Okay bonus. let's go down to a chip over here for example and you get the same sort of thing. Check this out. This one here.

fairly clean. That's part of that accumulator. Look at that. Interesting, huh? It's all to do with the layout.

There's another one right next to it there. I Don't know. Let's go down to this instruction decoder down here. What's that going to do? It's nothing on there.

There we go. It's clean at that point. it starts to switch at that point. So that's interesting in that maybe one of the signals that's crossing one of the grounds going to this particular chip or part of the ground loop in there is not switching at this time.

but then it suddenly started to turn on at that point, that point, that point, and so on. So no amount of bypassing is going to fix that problem. That's a fundamental problem with like a double-sided board like this where you can't do a nice solid, low impedance, low inductance ground plane over the whole thing. If you stitched all these pins together, did this on a four layer board, you wouldn't be getting stuff or you know it's it's much more minimized.

This sort of switching effect. I Mean you know, have a look at the size of some of those excursions. and yes, I've put the bypass caps back. Okay, the first thing we're going to do when we're actually probing something like this we care about signal integrity and signal integrity has to do with the types of probe you use and the probing technique you use.

So the first thing you want to do above everything else is get rid of this antenna ground lead because that is an inductor. Any piece of wire of any length is an inductor. You want to get rid of it. so that's gonna cause ringing on your waveform.

You're not going to be able to measure the true wave form in your circuit by using one of these. You're just not unless it's a sine wave. Then you're okay. But when we're not, we're measuring TTL digital signals here.

Not very fast. six megahertz. But it doesn't matter if it's a 1 Hertz signal, it doesn't matter. It's the transition where the waveform goes through right up like that that contains high frequency content, where your ground lead is going to come a gutter on you.

So what you want? This little puppy that came with your probe should have. Hopefully you didn't toss it out, You put it around there and now you've got a nice low inductance ground path like that. And then we can go around and probe our circuit. Yes, it's not very convenient because you can only reach that sort of distance, but if you want to probe things properly, that's what you have to do and it just so happens and that we have a couple of test points on here for the main clock.

So we'll just do the main clock. so we have a ground point here and one of the clocks here. In this case, it's a system clock so we can just work that In there like that, our ground plane and hook that on this. So we now have a beautifully low impedance test point there to measure our clock signal with.

So that's how we measure signal fidelity properly. that will do proper high frequency probing techniques as basically what we're doing. Now the next thing we want is the choice of probe that we're using. In this case, we're using a passive probe, but I'm using the absolute best passive Pros: I've got here in the lab.

These are one gig Bandwidth: Tektronix TPP 1000 for those playing along at home. Fantastic! One gig bandwidth analog Their passive probes I don't know how they do it. It's absolutely black magic technology. It really is.

And you want low capacitive Pro's 3.9 puffs, 10 Meg fixed times 10 probes and these particular probes are matched for the system, right? So these have got like the probe interface. There's like a a one wire chip in there that identifies the probe and you'll notice that there is no compensation adjustments on either on the end that plugs in the scope or on here. There's no traditional trimmer caps on here because these probes. what you do is you hook them up to the calibration port on here, and then you run the auto calibration routine.

It identifies these probes, it knows the serial number, and it actually stores it inside the scope here. So if we actually go over here. Oh, by the way, I'm using the Tektronix MD Hope 3000 here because this is the highest bandwidth scope I've got in the lab 1 gigahertz bandwidth and it has the match in best probes I have here in the lab. so we can actually go into our probe set up down here and you can see that it's all right.

I've actually compensated these probes already and it stores them. So even if I swap this probe over to this channel, the compensation follows it because it's smart enough. It knows that's the serial number and then it follows. It's absolutely brilliant.

So this is the best probing solution we could possibly have for measuring true signal fidelity of the clocks in this system as we remove those caps. Now technically speaking, I do actually have better probes here in the lab. I have active FET probes and if you really want to get serious, you'd be using an active print for a FET Pro: This is a cow chest CT 41:21 It's a 1.2 gig band with 10 to 1 it's 1 Meg in parallel with 3 puffs 3 Pico Farad's so that ones are better than the 3.9 of the Tektronix there. And of course it comes with all these different little probing attachments that allow you to probe things with you know, real low impedance and everything else.

Sometimes you've got to even roll your own. Well, I do have one slightly better. I've got this Agilent in 27 to 796 AAA active probe and it's only one picofarad input, one path input capacitance. do you believe it? Absolutely crazy similar sort of thing.

and it's even look it looks like looks like a prawn or something. and it's got little lights on there and everything. Anyway, these are, you know, real. If you really want to probe signal fidelity and high-frequency stuff properly, then active FET probes are the way to go.

But in this case, for our six point two five Meg fundamental frequency, the harmoniously enemy, much more. we're just going to use our these one gig passive probes are more than enough we can do it with. You know we can do it on our Rygel You know, two hundred mega a golf scope as well. but you know I'm just using the best scope and the best in terms of bandwidth and the best passive probing system.

I've got in the lab here and we've got another clock signal up here which we can measure as well. This one's a little bit tricky act, sort of like hold that in place. It's a little bit how are you doing but we can do it. No workers.

Alright, so we've got our live signals here now. Setting up your scope is for a proper signal fidelity measurement. It's just as important here and you can see that well. we can actually turn on our frequencies.

You can see that both Channel 1 and Channel 2 both two volts per division. So the yellow waveform here is o'clock one down towards the bottom of the board and the blue waveform is clocked to the top of the board there. and they're both six point two, five megahertz. You can see that there's some askew actually between them that's inherent in the board.

Because yes, I have D skewed are these probes? So if we go in here and look at our probe setup here, we can actually Rd skew them. You can actually set up a propagation delay 5.3 nanoseconds in this case and I've this has an automatic desk you and all that Serge's so if one probe was longer than the other, you might be out. Like, you might have to change the UH skew on these things. or you can actually compensate for the skew on your board as well, not necessarily in your probe.

but we don't want to do that. Okay, so we're happy, but we're not actually interested in the timing of the waveforms, it's just interested in what the waveform looks like. There is our waveform. Now, here's the thing right.

we can actually go into a choir here: I'm actually in average mode at the moment. so I've actually got I'm displayed I've got a record length for 100k. fast acquisition mode is off and average, and I'm doing sixteen averages here. Of course, if we go in the sample mode, that's the crap that you get because the waveforms are always.

you know, changing. There's always noise and whatnot crap on there. so if we turn on our fast acquisition for example, that will. that will give us that.

In really, that's updating. You know, hundred thousand or a million waveforms per second or whatever this scope is capable of doing, but it's only got like two hundred and fifty samples or something. So let's go down to our record length. You can see here how the record length can change some things.

Just notice how it's slowing down. All that crap is still there. Look at that. The channel r1 down the bottom of the board is the worst here, and all that crap is coming from other switching elements inside your computer here.

So actually, let's see if I can get it to change. Let's turn on the Mandelbrot Dosha program does things you might see differences here. So let's have a look at this. So I'll press Start and reset it.

Boom. We're back at the menu. Maybe I don't know. Not as much variation, but in in theory, right? that is possible.

If we do something in our circuit, we could see variations in our signal fidelity because of all the you know, the non-optimized ground layout, and now stuff that we've got on this board. Yeah, you can see the record length of how it's slowing down, but really, you know I'm happy to get an average. If you're looking at signal integrity, you don't want the other crap in the to affect it. Really, it depends on what you're after those.

Sometimes you may want to see how other parts your circuit. In this particular case, if we're chopping off some of the capacitors on the other side of the board. for example, that might upset the clock at this particular part of the circuit so that we're actually probing. You know, so you're may or may not want to see that.

But of course, if you want the signal integrity at that point, the best way to do it is to simply well. you can turn on high-res mode which does a boxcar averaging of like you know, like sixteen or something like that in a row at the sample thing. But the best way to do it is just with a nice average like that. There you go.

We've got our two waveforms. now. we can actually store those as reference waveforms. Okay, so I'm just going to choose a hundred K there as and sixteen averages there as our reference.

so just stop that. And now we've got our nice to reference waveforms with all our capacitors in place. So now we can actually go in and store these. We can go into menu here: Source One, Destination Reference Waveform One By the way, the Scopes A bit weird.

If you want to actually store your references, you don't press your reference button. You actually have to go into savor it all down here. Maybe it's intuitive, maybe not. Anyway, there we go.

So we save that. and then we can save our source. save channel to and save like that. And now we have out what.

You probably saw the white reference waveforms here. So now we can turn our reference waveforms off and on. So now we can actually see if our waveform changes. There you go.

Okay, it's doing that because of the reference, but the reference is still there. And so we've always got that reference wave waveform to compare to. But there we go. So if we see any changes, we can just use the reference to highlight them.

Oh, by the way, yes, these are actually like real little things happening in the circuit Because we're using proper high fidelity high frequency probe into makes here this stuff's actually happening. There's some sort of transition thing that's causing a grant, like a little bounce you know, in the ground system Whenever this edge here is pretty good on the top of the clock. But look at that. You know, look at the ringing on that.

It's pretty terrible writing an undershoot on there, right? But this is actually what's happening in our circuit. This is real stuff. Because we're probing it properly. We know there's very little effect.

you know, negligible, half a bee stick to do with our probes. So we're doing the absolute best we can. So that's actually what's happening in the circuit anyway. So there you go I Hope you enjoyed that video and a look at Munson Was madmen months right that you can move capacitors? Your product still works well.

Yeah, obviously yes, in this particular case, we removed them all and it still works I Still, you would have to try this over like the voltage limit range like so the 5r plus minus 5% over temperature and all sorts of other qualification type things if you were going to ship a product with no bypass capacitors which I do not recommend by the way. so do not take this video as a recommendation that you don't need bypass capacitors because you do, In fact, they can be absolutely critical and a but don't go overboard with bypass capacitors because it's a whole science and you don't want the things to resonate anyway. I've got that whole video looking at bypass capacitor and that really only scratches the surface. like I'll link in some like papers down here.

Here's like a IBM paper on decoupling caps and ground planes and and capacitor impedance and you can go to town. look at that, look at the mappings there. God it's just absolutely insane. You can do like PhD theses on just on bypassing I mean it's absolutely crazy you know and and lower your inductance which from your pad to you via to drop down to the ground plane it can make it hell of a difference in there.

And there's tons of stuff and you know TI and almost every manufacturer has discussions about bypassing because it is actually critical. You know if you go into our FPGA is for example they've got like how many pages is this document I don't know. this is just a document on power supply, designing power supplies and bypassing your FPGA and power delivery systems just for the Xilinx FPGA s and and stuff like that like it goes into the same stuff. I've done it in the video and then shows you and then tells you how long your traces can be to your buyers and you know all sorts of stuff.

it's just placement around the under the chip and all sorts of stuff. I've done probably previous videos on that and they go to town and they go to town for a reason because bypassing is important so this was just a bit of fun to see if Munson worked and it kind of did in this case. be nice to get and I might do it in future videos. get more advanced products and actually remove probably can't snip out little.

I Don't recommend snipping out little surface-mount ceramic caps a shadow, then T your pads off your board ruin your day anyway. it would be nice to you know, get a more high-tech product perhaps and do some Munson Anyway, um I Also may there'll be an excellent I'm hoping spin-off video to this one as well in the future, so maybe look out for that anyway. if you liked it, please give it a big thumbs up because it always helps a lot subscribe and all that sort of stuff. You've got to get that notification build where is it I don't know down here somewhere up there down there, whatever you know or youtubers say the same crap.

Catch you next time.