Hi in my previous video on the IBM T two-to-one 4k monitor from the year 2000 which has been very popular. Highly recommend. You'll watch it I'll link it in down below and at the end if you haven't seen it cuz this will make a bit more sense I go into quite some detail on the board which we're gonna take a look at here. but I've got one comment on the video which got me thinking thank you very much Toad's gabber, it's not Gerber Let's go back anyway.

So how why do you have to make the tracers to carry the same amount of current as a properly executed jumper wire? Whole lot of road. Here's a fat jumper wire. You don't need roads. so this got me thinking this could make an interesting video and if you don't recall, you have to watch the video.

This is the main processor board out of the IBM 2 to 1 monitor that's got two huge Altaira Apex FPGAs which was state-of-the-art at the time. very expensive $40,000 FPGAs and this is the back of the same board and you can see here that there's these two big jumper wires on here which basically bypass the PCB traces on the board to get lower impedance. So I talked about this a bit in the video, but I thought this will make an interesting deep dive into PCB Plane impedances FPGAs margins Power supplies all sorts of stuff because I think this might be a real interesting example of how something seemingly as simple as a couple of wires on her board actually has really deep technical meaning behind it. And the engineers who designed this board probably spent me countless hours actually analyzing and solving this and why they actually needed to put the wires in place.

So I Think we're gonna go deep down the rabbit hole today. I Haven't planned this video at all, but I can think of stuff that is Yeah, Buckle up! Dorothy Kansas is going bye-bye right? So we'll do this with two high-res photos of the board here. this is the top side with the FPGA is here and then we've got the bottom side down here, which is like looking through through the board so you can see the connector here, for example, and that matches up with the connector on the top there. Okay, so everything's kind of like.

It's like when you're designing a piece of B you're looking through the board. essentially. Alright, let's start out what's going on here. Here's our power input connector.

This comes from our 130 watt Power brick here and you can see that they've got a bunch of Vias all in here like this and that of course, and the through holes of course go through the bottom side. So okay. nice low impedance connector there directly onto the board. They've got some fuse in here and just some filters there, and lots of via stitching in there to go down through to the bottom or inner layers.

so all the power comes through. This one connected the power brick 16 volts at 10 amps I believe it takes about 130. what? So we're talking. you know, just over 8 amps for the total power consumption here.

Now we have to look at where the power goes because the entire premise here is why did the designers choose to put these big huge thick gage jumper wires in here? You can see they've actually designed into the PCB because they've actually got this silkscreen. even though when Payne Whitney manufactured this, they just hot snot added here and ran the cables over here. I'm not sure why because they've actually the PCB designer went to the trouble to put silkscreen path right on there to show them exactly and even put little glue markers. Please put some glue there there there, it's very nice.

Ok, the PCB designer was thinking about this when they were doing it. So why if they run these wires, why couldn't they use the internal traces on here? Now this is I Haven't looked at all any layer markers on this board, but assume it's like an 8 layer board or something like that. Yeah, you might get away with 6 on something like this just, but it's more likely an 8 layer board. So usually you're gonna have a big one of those layers will be dedicated to one big ground plane and this is where we'll get into all sorts of tricky business and why You might want to run these wires directly over you.

You'll notice that they're running the negative wire as well, not just the positive. So obviously they're trying to bypass the internal ground plane in here. Why is it so? so? Why did they do this? well? There's usually two reasons behind doing something like this. one is that you simply ran out of room on your PCB to put in.

In this case, the high current traces required big thick traces in there. So you can get low impedance, low voltage drop. You simply ran out of routing room on your PCB. But on an eight layer board like this, you're really.

you're not going to run out of a room. So the next reason is DC impedance. Now there's two issues on a PCB One is AC impedance. One is DC impedance.

When you see a huge, big thick jumper wire like this means somebody's trying to fix a DC well resistance. Basically, you shouldn't say impedance. with DC A DC resistance problem. There's too much voltage drop on that.

They're trying to lower the resistance to decrease the voltage drop at a particular current. Remember Ohm's law voltages, current times, resistance. So the more current. Well, in this case, let's say we've got our eight amps flowing across our board like this.

Then you're gonna get X amount of drop over that distance based on the resistance of the traces. It's nonzero. Even a nice big fat ground plane is not zero resistance. but I hear you're saying well Dave In practice, a big ground plane that's gonna be good enough for anything.

Well, that's not. so as we'll take a look when we go further down the rabbit hole. So now we're going to look at actually where this current is going and where these Y's are running. and why? Okay, so let's look at our board here as you can see it jumps from here over to here.

Now on the top side, that's our big-ass FPGA in there. Don't worry about the FPGA on the other side, We'll just look at this one for now. So it's jumping from here and it's going over to here. That's the red wire.

Basically, you can see all those little vias in there, so that's jumping over to there and then your ground is jumping over to this side. So these three caps over here. Just some bulk decoupling that's obviously powering. You can see it sneaking through.

so the positive wires coming in here and that's obviously going into this inductor here. and then that's we've got a switch mode controller of course, and then our outputs over to here. So this will be one of the rails of the like. The internal core voltage of the FPGA I'd Say you know that internal core voltage is probably going to take the majority of current in the FPGA The I/o is going to take less.

So this one here might be the five volt regulator because I believe all this logic is five volts. Okay, you can see that positive comes in here. There's an extra fuse here. This goes out to this connector which buggers off to the bottom board which is the I/o driver board.

that's neither here nor there in terms of the power consumption of this thing. Well, that'll actually then that'll be for the LCD too because that's where all the LCD is getting its power from as well. So and that'll be for the I/o board and the LCD as well. But I think most of our issue here is we're going to focus on the FPGA.

So how could you have gotten yourself in a situation where you're You have to run these wires across here? Well, one, it could be a placement of your board. You've got to look at that first. We'll go into voltage drops and you know, I've mentioned the voltage drops are the issue here, but why placement? Well, in this case, they physically needed this connector right here because it's on the back of the product. You know, usability reasons the user has to be able to plug it in.

This is all one big like interface on the back panel. so it's got to be here for example. So you might think, Well, you know you could say that the layout person probably should have if it was me I wouldn't have had my big switching rigs all the way distant from the power connector here. I would have like put them here.

Look at all this unused space around here. that's where I would have worked it. But the problem is no FPGA is here and all of this routing comes on the bottom side. Here's our big connector here like this.

so all this high-speed routings got to come out to go off to the LCD on the bottom here. and well, you can argue. Well, this could have been up the top and stuff like that and well, you know. Anyway, there's layout reasons why.

And so the PCB designer went well. Ok, I'm forced to put my switching regulators up here. That means I've got to get a reasonably low impedance path over here. and if you use the ground plane inside the board going from here over to here, that's a lot of distance there.

We're talking probably 20 centimeters there to get your 8 amps over to here and you've got your big FPGA under here like this. So any switch in of your switching power supply or your load on the other side of the power supply, even the FPGA itself is going to cause quite a few issues. So did they goof this from a layout point of view? Maybe. but I'm not gonna like.

You know, there's lots of reasons that goes into this. This is a month's worth of layout. So yeah, let's let's just not say that they goofed here, but it's possible now. Everyone thinks about when they think about PCB layout.

it's all about bypassing. You can see all the bypass caps in here under the FPGA right? They've done that well. You got your large bypass caps I've done a video on why you need like they've got three different sizes. one there like a like 1206 Oh a Doe 5 and then O6o 3s.

They've got all different sizes, all the different values for the different frequencies. Done a whole video on that. It's great. I'll link it in and I did.

but that there are two different things you have to consider when you're laying out PCBs and design in products like this one is your AC impedance and this has to do with all your decoupling and stuff like that. Now that's one thing, but a lot of designers will forget about DC resistance and this is why they've come. a guts are here and had to add these wires in. Because it's not uncommon.

at the end of a design, till you layout your board, you build it up your first prototype. you're testing it out. You know things aren't quite working, things are playing up and stuff like that and you might find that well, your ground paths weren't lower impedance enough. Getting it from one side to the other.

Over here 20 centimeters was just too far and you getting too much DC voltage drop and that can interfere with your FPGA And here. which is why we need to go into the videotape. We need to go to the datasheet of the FPGA and have a look why this might be a problem. Why DC can be even more important than AC Decoupling All right.

This is a data sheet for the Apex 20k logic family and this dates from Nightmare 1998 Something like that. So quite an old device, but we've got the EP 20k 400 here, so I'll spare the details. Let's go all the way down all the way with LBJ and what we're looking for is voltage. And here we go: Operating conditions: This is what we have to look at and this is just A.

This is not even a deep dive into Fpga data sheets. Never done a video on that anyway. Supply voltage Here we go. so 5 volt tolerant.

Recommended operating conditions: Our internal voltage FPGAs will have an internal core voltage as well as an I/o voltage like this. and sometimes they've got even more than that. But this is a fairly old-school FPGA The more modern ones they can have, you know, four or five, or even more different voltage calls. For various things, you have separate ones for PLL's and all sorts of stuff we won't go into anyway.

Look here we go: Minimum at two point three, seven, five volts and two point six to five volts. so you can so one of those switch mode converters that we looked at on the board that's going to be a 2.5 volt switching converter to provide the VCC integer. and as I said and the other one will be VCC IO which is your nominal 3.3 volts which of course powers every all the other 3.3 volt stuff on the board as well. And this is a 5 volt tolerant R Fpga so it will tolerate 5 volt.

II And even though it's only got a VCC Io 3 point 6 volts. Now look your three points volt rail here. Okay, it's got a fairly wide margin. It's got nominal three point three, so it's got a 300 millivolt minimum is 3 volts of 300 millivolt margin on the low side, but have a look at VCC In here, that's only two point three, seven, five volts.

That's only a hundred and twenty-five millivolts less than the nominal 2.5 volt coal. Voorhies So I've only got 125 millivolts to play with. So that's going to include the drop not only on our power rail, but also on our ground rail Because remember, it's a big complex system. so not only do you need a minimum of two point three, seven, five volts at the FPGA chip itself, if you don't apply proper star grounding techniques, we just have the input connector and then you just run the ground off separately.

Then the interaction between your different chips can cause logic, threshold issues, and all sorts of stuff. So if the DC voltage drop on your power and your ground connections go into that FPGA A greater than 125 millivolts. Wah-wah-wah-wah You've come a gutter and your FPGA here and now operating outside of the recommended conditions and all bets are off. So 125 millivolts might sound like a lot to play with.

But let's run the numbers on this and remember, this is an old FPGA This is not one of the newfangled ones with very low core voltages. If we quickly check out a sheet or one of those, we'll see. we probably don't have much margin to play with. So although it's not directly relevant to this particular almost 20-year old design, let's look at em modern.

say out arrow: Stratos FPGA For example: they've got data sheets just for the AC and DC electrical switching characteristics of these chips, and the data sheets are phenomenal. It's not just one data sheet, it's multiple ones. They've got one Darla sheet just dedicated to getting the power right on. FPGA is this is how important it is.

So if we scroll down here, look at all the different types that got VCC which is the core voltage which is also called VCC inter. Yes, they've changed over the years. VCC PT Programmable power technology for the configure just for the configuration pins, they've got auxilary ones, they've got battery backup power supply, they've got the IO pre driver power supply and they've got the IO power supply and not done yet. the PLL digital power supply, the PLL analog power supply.

So how many we up to now? Two, four, six, seven, eight, nine nine, and then regular V I Set nine ten different power rails Ten ten different power rails modern FPGAs are insane. Like I said before, it's not uncommon to get four or five or more in this particular case with these high-end FPGAs And but let's go down and have a look here at the requirements: VCC Here's the requirements for the just the core voltage of the FPGA pain in the arse through this naught point nine volts, the core voltage ridiculous, and then for the programmable stuff 1.5 then you need another 2.5 Then you need a year 3.3 s and 1.8 for all your different power buffers and some low and then your programming voltages, your PLL's or another 1.5 and 2.5 Often, you can't share them because you need clean power for your PLL's For example, you don't want to switch that in with you call voltages or your I/o voltages and stuff like that. Anyway, the key point I want to make here is no point. no volts.

Look at the tolerance. 30 millivolts, 13 millivolts. That's all you have a point. Eight seven is the minimum.

If you go below that, all bets are off. your FPGA may not work. or maybe it start playing up doing weird stuff and you don't want weird stuff and you have PG A They're already weird enough as it is and incredibly difficult to debug. So at 30 millivolts drop, that's nothing you remember.

We had a hundred and twenty five millivolts with the 20 year old Apex FPGAs these ones 30 millivolts and that includes your positive and your negative rails. You can now see how absolutely vital is not only to get an accurate power supply, but ensure, but there's a minimum drop on your power and your ground. Traces go into the FPGA so often you will tweak. It's very common to actually tweak your supply voltage up to instead of nominal 0.9 You don't want to set it to that, you might want to set it to 0.93 knowing that you're gonna get some drop on your tracers and then instead of 30 millivolts, you've now got the difference between that and that.

So you've now got 60 millivolts drop to play with when you're laying out your PCB traces and we'll go to a calculator soon and calculate all this stuff. But as you can see, it's not much to play with there. But thankfully the Apex one is a bit more tolerant 125 millivolts anyway. But let's go back to our power supply here because we're not done yet Those 125 million volt and 30 million volt margins as I said.

and it's got to get over to your FPGA over to here like this. not only power but the ground as well. It's got to get through there. It's got to get through all the vias.

It's got to get through the local bypassing. Remember these are DC values has got nothing you can have all the bypassing in the world. It's not going to help you. That's the difference between AC and DC Characteristics of a complex device like this: FPGA These are the most complex devices available today.

These FPGA is you think your until you're bleeding edge. Intel Processors know they're like 5 times smaller in die size, 5 times less transistor counts and some of these high-end FPGA is today. These are the most complex beasts on the planet. But it's not only DC you've also got like low frequency transient characteristics as well.

So let's say your voltage is like that and then it suddenly starts to draw a bit of current. More might drop down like that a little bit at higher current and go back up like that. So that is your dynamic characteristics of your converter. Because these switch mode converters in here I don't know which chip they're using, we can probably look it up, but anyway, these are going to have dynamic characteristics as well.

The output impedance of these switching converters is something at a DC and something slightly different. Again, only has to be a smidgen difference because remember, we're leading-edge FPGA Oh yeah, 30 millivolts to work with. That includes not just your PCB traces, but all your switching characteristics of the output MOSFETs and switching elements inside your switch and converter. So this is a deep rabbit hole.

If you actually want to do this, you could spend a designer can spend weeks and weeks or a month just getting this power system right. Bloody scrolling. Anyway, so it looks like on the bottom there there's our MOSFETs. You can tell they're MOSFETs because all the the pins are tied together like that.

There's just the one gate pin here in here. Okay so these these MOSFET switch in here for our two converters and then the controller was that on the top side these have internal voltage references which set those DC characteristics so all of the dynamic switching characteristics you could have if you don't have proper layout in your switch mode converter like this if you don't keep your tight loops and I've done videos on this and if you don't keep that tight you can get extra DC error or switching or dynamic switching errors when you when your output loads. So if your output switches like this it might have you know two different current levels. you might be drawing say five amps up here then a might drop down to one amp down here different modes and it might do this at low sort of lowest frequency.

might do it at higher frequency and stuff like that. then if you don't get your if this is you know is that this is very crude. but if you don't get your star grounding points like this like this one is going off to your load for example and this one might be going off to you know, some other load or whatever and this one goes to your chip. it goes to your converter for example.

if you don't, if you have your large switching currents going up like this and then you connect here your chip off to this point here, then you're going to get their voltage drop all the way along there. That's the very crude diagram, but then you've got an internal reference voltage inside this thing that you've just like your dynamically switching when your load switches Karagounis which you might have only five or ten millivolts, but that could change your output reference voltage from your note point nine volts. It could easily change it down to naught point eight and you're screwed. And that's just with the layout of your like little switch mode controller chip that doesn't include actually going off to your big arse FPGA up here and the voltage drop on those so that that's just ground wear didn't like I'm just considering ground not considering power here.

So here's the actual chip used on this board. It's the TPS 56 O2. This is the controller and you can see there's our MOSFETs that we saw before uses external MOSFETs and they are the two biggest inductors that we saw there and you can go down and we can find lots of various information about this that is pertinent to this particular case. Once again, you're gonna have like your voltage reference is, that's gonna be a thing.

and it's kind of vary over temperature and stuff like that so you'd have to look into potentially things like that. But really, what we're interested in is like dynamic eye characteristics for example. So let's go down. Circuit protection.

all sorts of hysteresis, voltages and things like that. We won't worry about any of that typical application circuits. switching frequency. Of course, switching frequency is going to change.

It's a complex equation which changes with all sorts of things that they can affect various things. And I you can come and guts and just there alone. that'd be a whole video Anyway, Under Voltage: Here we go: our output efficiency system Efficient Load regulation. There we go.

Three points Free volt Output Load regulation. That's pretty tight, so no worries there. Over like five amps. So that's load regulations.

Not a problem, but usually something that you've got to consider. That's load regulation is how tight as you can see how tight the voltage remains over the entire output current capability of this thing. 1.8 Bolt Output Load Regulation: That's pretty good. So only yeah.

5 millivolts. You know, like no, not even just a couple of millivolts so you know it's nice and tight. But look, output voltage ripple is you can come a guts just on the output voltage ripple. Look at this.

I mean we're talking like 50 millivolts. We've only got 125 millivolt margin that's not including any of our DC drops. So when you include the dynamic characteristics like oh well, the outputs not our dynamic characteristic, but it's a fixed in DC rail characteristic output voltage ripple and you add that onto your drop juicer, you add the ripple onto the drop due to your PCB traces in your PCB. You might calculate your PCB traces just fine and it could be hunky-dory It could be with inside that hundred and twenty-five millivolt limit and but you add on your ripple you forget about that oops or your product into some other different operating mode or something like that and and it draws an extra current and then the ripple gets bigger and bigger and and that can change with temperature and all sorts of like what like it's all over the place right? This is complex pair of sequency.

We don't care about that. Transient response, transient response. This is what we want. Look at how much it can change.

This is a hundred millivolts per division. remember or any were talking about a hundred and twenty milli volt margin here. that's not including the DC margin right? That's that's including everything. It doesn't care if you transient outside that for a split second.

then as your the FPGA might change modes, your product changes modes or whatever It chops from. you know five amps down to one amp or something like that you're gonna have a transient response you get coming outs are right there on these dynamic characteristics under shooting, over shooting and that's all that. ons on top of your DC characteristics like there's a lot involved here I Know you're saying Dave just switch to A They just use huge ground and power planes and it's gonna solve all your problems. Don't worry about this star grounding and layout and routing and all that sort of stuff.

Well a couple of things. One is that you can often look at all these vias in here. There's V as V as V as V Z is everywhere that's chopping up your ground planes every time you do it year you can have blind and very deep buried vias where if your boards like this then your via only goes between a couple of layers. It doesn't go, it's not drilled all the way through like this and so if you've got your ground and power here they can be a nice solar ground plane.

but then you got a more expensive board and it's a trade-off and stuff like that. But even if you had a big gigantic ground plane in here which is very common for these high-end FPGAs for that Stratus when we looked at with your 10 power rails oh you know we no common implementation. You might have seven or eight rails. Something like that.

You might dedicate not a whole ground plane to each one. You would have one or two ground planes, but then you would dedicate like a couple of your layers to power. and you'd be routing. You'd be like routing your different paths for your power as well.

Typically, you'd have them all coming back on the one ground plane. But the problem is big ground planes aren't magic because it goes back to Tada. Bring back you. Remember this if you've been watching for a long time.

The resistor grid. It's what was this video. 5 or so 10 or something? I Don't know. Anyway, it was quite old and you can think of your ground plane as a grid of resistors like this because that's effectively what it is you're current it.

the higher fare. I've done a video on this. The higher your switching frequency, the more the current is going to remain. So you're switching for your, your source and your loads here.

For example, then your current at high frequencies judo is spreading. Inductance on your board will fault. The loop will follow the path directly under the power traces on your board, so there won't be as much current flowing out here and here. But DC In theory it kind of will start flowing out, but once again, you've got spreading resistance in here.

It's not magic. Think of it as an array of resistor. It is an array of resistances like this. and this is how that you actually analyze.

But unfortunately, to analyze something like a ground plane, how much drop do I get on a ground plane? How much voltage drop for a given current? Not easy to calculate. In fact, there's no simple calculators out there to do. It Might be a couple of rules of thumb, but they're not hugely accurate. so you have to use what's called a finite element analysis.

you have to. You can get real expensive tools to do it, but they do very complex mathematical modeling. In this particular case, modeling the drop like if we had a big ground play. Modeling the drop right across this board, over from this power connector over to here like this and then this over to you and there will the output.

sorry to if you FPGAs and stuff like that, it's really. you know, it's almost. Don't bother calculating something like that so you can do it in theory, but in practice. Yeah, no.

Okay, so that's why they probably went for the wires because it's actually a more predictable resistance. We can actually get in here like this, so just putting in a thick gauge wire you can calculate the resistance of that. It's a really easy and if you're running PCB traces For example, if you're running power the positive one on the internal layers. If you had a big you know I would make the trace like hugely wide like this.

You know a big thick trace going over there like that. you know, like 20 millimeters wide or something going right over or even. It depends on how much room you got maximize your amount of room. But anyway, you can start.

Once you get traces like that, you can then start to calculate the DC resistance and the resultant voltage drops. And another thing to overcome the voltage drops at high currents coming from the output of our switch converter over to our FPGA Over here some switching converters, you can get a remote reference voltage so it might take a reference trace like it's a 4 terminal measurement so to speak. It'll take around and power references from under here so you might have a big power square under here for example. It's quite common.

you'll lay out like a big power square like that for example and all. Yeah, so that's how you get high frequency bypassing, but it's also just one big brick. But of course all the vias that you've got to drop through from your pain in the arse BGA package on he dropped through. they all split up that plane.

So anyway, it gets complicated I haven't done a video on that. but anyway you have a big thing like this and then you'll have figure me it's vanishing. Then you'll have one little trace, a sense trace coming off which goes back to your power supply. You have you no sense traces and then it actually adjusts for the voltage actually on square and it eliminates any of your voltage drop going across here.

Like this a compensates for it. But as I said, there's dynamic switching characteristics as well to take into play. so it's not necessarily that simple and if you only got 30 millivolts to play with. Okay, so let's do some calculations now.

I Have actually recommended this tool before and I recommend again. It's the best thing out there and it's free. It's the Satin PCB design tool. Search for it.

It's absolutely fantastic. It does. It does everything. differential pairs of ear resistances ie.

diamond does Ohm's law, does heat sink, thermal stuff. It's got pad stack calculators, crosstalk, conductor via impedance, and parts-per-million calculators, embedded resistors on your board, substrate resistors, and IDEs just. it's insane. It's the best tool ever.

All right. So we'll go to Conductor Properties here. And as I said, there is a DC and AC element to this. We're not interested in our AC characteristics.

we're just looking at our DC voltage drop here. So let's not complicate things. so let's just set it to DC mode here. Well, let's have a look here.

We'll have to go over to Imperial and on this micrometers. Rubbish. No, I'll be swapping back and forth. it's just a habit.

Anyway, let's look at our base copper weight, which we got in our PCB Now everyone's used to one ounce copper. Now you don't automatically get one ounce copper on a big eight layer board like this. You might get one ounce copper on the outside layers, but more often than not, you'll get no point five ounce copper. Or you might even get quartz copper.

But half ounce copper is the most common on the inside. And then you've got the plating thickness here. This is whether or not your board is plated. In most cases, your traces are not going to be plated or your internal layers.

they're just bare copper. It's just they just etched it as bare copper. Then they sandwich the extra layers on top and on top and on top. And even your top and bottom layers usually aren't plated.

That's what's called solder mask over bare copper or Sm OBC. So if you scrape away, that's why if you scrape away the solder mask gonna trace, you'll just get the raw copper. So there is no plating. But if you had a tin plated or something that you can add that and you might see that the resistance down here might have yep, about half if you played it something like that because you're pleading might be another half ounce played in for example.

So anyway, so there you go. So we'll set a bare PCB Like this: Assume we're going like an internal layer. half ounce copper. I'm going to go over and in external or internal layers.

Makes a difference. Well, it doesn't make a difference to the conductor resistance. here. the conductor resistance is going to be the same, but it'll make a difference to power dissipation and whether or not a plane is present here.

you'll see that it only affects the power dissipation figures which we're not really. You know where, that's not a concern. All we're concerned about is the DC resistance and the current and the voltage drop at a particular current. Because in this case the plane.

you'll see how when you enable that it puts on a distance to plane like that 10 mils 10 fell between the planes because then you'll get conduction or you know radiation from the power trace through to the power plane and then that power plane can kind of act as a porous heatsink and that's why your power dissipation will be more if and further away. If you drop, you will notice our power dissipation. Well can we solve that? There you go, it goes down. So you know hundred mils like there.

Anyway, that's got nothing to do with what we're doing today. It's just a side thing, but there's lots of complex stuff that goes into PCB Layout People PCB Layout Designer: They just lay out some traces and Bob's your uncle know if there's a ton of stuff if you've got bleeding edge parts pulling, lots of power like we do on this one and all these different voltage rails with all these different DC AC dynamic and Static power conditions to meet very tight tolerances, you know it's it's not stun of stuff involved. I Can do a video on every single one of these properties in every single one of these tabs. Anyway, calm down.

Dave Let's go over to here. Let's say you've got a 20 millimeter trace. You know at the big thick trace on your board, right? the conductor lengths in this case what? I say it was 20 centimeters. So let's go 200 millimeters like this and the PCB thickness doesn't matter.

It makes no difference whatsoever. And our let's call it 10 milli ohms, they're out. Can we saw? Yeah, they were solved it. They're ten milli ohms there and the current down here actually has to do with the temperature rise.

and I've done this in a separate video. So if you've got a that 20 millimeter wide trace over 200 millimeters at not normally 10 milli ohms resistance there it will. That trace will rise by 10 degrees Celsius it'll actually go out. It'll heat up when you pass 7.2 AB straw.

And you can see that we were like talking about 8 amps here for example. And at 10 degrees, there's a bit of rule of thumb for temperature rise. You really don't want any more than that. A lot of people set it to 5.

all often said, like harv it to 5 I don't want to 10 degrees rise and stuff like that anyway. 10 milli ohms Drop on that trace. The 8 amps we're talking about 80 millivolts drop. That's only one of them.

that's just the power trace. What about the ground trace? Bingo, You've got a hundred and sixty millivolts already. We're already over our our text datasheet limit. Where was it? Hey, you remember it was only 125 millivolts.

So if we take our power trace like this over here and it was 20 millimeters wide for example. nice big huge beef and power trace, we're gonna get a voltage drop across there of 80 millivolts at 8 amps. There you go and then then you've got the extra drop for the the same shop. again for the ground or whatever.

but in this case that's going to the switching converter. I Probably should have drawn that the other way. Let's just say that we had our nice big square in here like this and then we had our 20 millimeter trace coming over. If it was like, let's just say that's a hundred millimeters there, then we'll get half the voltage drop on that.

So we'll get a 40 millivolt drop coming from the output of here. Over to our assume: let's assume that once it's gotten under the FPGA Like that, you get your one big solid thing that there's no more kind of little drops in here. For example, there will be there might be, you know, two millivolts or something like that going from one side to the other. Whatever.

It depends how many vias are in there. like actually our breaking up that that big, nice big solid block you've got in there. but you might have. You could easily have 40 millivolts drop going from there to there at 8 amps.

easy. And that's just the power, let alone the ground connection as well. But you can argue that if you've got a nice big solid ground plane like this, for example, then you can kind of round it down to zero. You might add on 5 millivolts or something like that, but you know you might.

But there you go, right? You remember we only had a hundred and 20 millivolt margin, and remember that doesn't include any dynamic characteristics that doesn't include any layout issues and how much you know drop. If the LCD Oh, you remember, we've got our LCD over here as well. it's going to be drawing. It's going to be drawing its own currents and you're going to get extra drop across there.

And if and if you did have one big ground plane. yeah, sorry, you can't see the can. my drawing cursor. whoops.

Anyway, if you've got one nice big ground playing like this, getting extra power that has nothing to do with the FPGA across from one side to the other, that's got to share that space. You can get the voltage drop inside this power plane going from here over to here. and You know that could be that could be an extra I don't know, 50 millivolts or something like that. Who knows.

It depends on the load that you get in and that could interfere with your voltage drops inside your FPGA Like this people can't my guts up. Let's get rid of all that. That's why you would have a star ground. For example, you would separate if I was saying.

let's say this was all the power in. well it is. This is all the power going off to our LCD panel over here. I would literally split my ground like that and I would have ground going over separately and then I'd literally have a split in there like that.

I'd have two separate grounds. They'd eventually join back here. so we've actually split or it's effectively like a star ground. This is our this is our point and then of course then we might have say another ground playing going over here to the rest of this.

FPGA and maybe if we had some circuitry up here that then needed its own ground plane we didn't want to interfere with the others then we would actually split the ground plane like this. for example, under all the memory and the FPGA that would have one ground plane then this would have another one and it all come back to the star grounding point like this. So this space in here you would leave blank that doesn't have any ground plane in it. So you've isolated your oh yeah, you've isolated your ground planes like this so that the current flowing from here around here ground and power to to the LCD connector over there doesn't interfere with your FPGA or switching in your FPGA doesn't interfere with the PAL going over there and vice versa DC and AC characteristics as well.

But remember when you split power planes like this, you have to be incredibly careful because it can be really bad news for EMC. So in this particular case, look we've got the the connector goes down here. This is the high-speed connector that goes off that's driving the LCD. So in this particular case, the Altaira the signals coming out of here.

they're going through these buffers and then they go on across here. And if you split your ground plane across here and that ground plane is used for those switching signals as the return path for those switching signals, you are screwed. It is one of the cardinal sins of PCB layout to split your ground planes and then have the signals running across them like that because then the currents have to flow all the way and then your loop is much bigger and larger loop creates greater EMC I've done a whole video on that and yeah, don't do that. So in this particular for this particular layout here when I suggest doing the split and ground playing like this, I don't mean split up the signal ground play which takes that one up I'd probably have separate ground planes, one that just handled the current for the LCD connector over here and one that and then a separate one that handled all the FPGAs and stuff like that.

So yeah, I just would have devoted one section on another layer for that. but in this particular case, the FPGA ground would have extended down here like this on a different layer and you can have like grounds overlap in other grounds. That's fine. So then we can have the separate on another layer than the ground for the power and LCD connector coming over.

So ultimately I think the reason for the wires is is pretty obvious given the proximity and also this power connector going off to the dryer driver and LCD board is because the driver an LCD takes a their own fairly large amount of power and they didn't want that flowing right across the board like this which could screw up everything they want to flow in across the ground planes and they probably didn't have enough internal layers to add some extra ground planes or whatever to do that. and sometimes you can fix this if you rip up your whole board and redo it like you might realize this at the last minute. and like all you know, you're two weeks into your layout and you might realize oh damn, I forgot about you know, I forgot about. the extra power coming over here because often the designers of the schematic and like they may, if they're good, they'll put notes on the schematic explaining how they want stuff done.

But it's ultimately up to like which things are important. saying, you know, look it like they'll show it as a star, ground and things like that. So the PCB layout person has to know that and stuff like that beacon R it you know your brains not engaged. You can spend a week stealing your layout, then you're coming out.

So I couldn't be bothered redoing the whole thing. You got a tight deadline or whatever. Okay, well, let's just add some wires on there. It's not how I own product.

she'll be right. But of course those wires aren't magic. They're gonna have voltage drop to in this I think it's about maybe 15 AWG wire. You know, almost one and a half millimeters.

Something like that. Ohm's per kilometer. It doesn't allow you to. We'll have to convert that.

Let me get the confuse. Er could have done that in my head. Divide that by a thousand. That's ten milli ohms per meter.

So we're looking at two millions for 20 centimeters. Multiply that by 8 amps. We look at it. 16 milli volts drop on one of those wires.

That's not two of them. so you might have to double that cut. That could be take up to 32 milli Volts drop just forgetting you're eight amps through there like that. So yeah, so yeah, whether or not they're doing that voltage drop.

Reasons: I think it's just like a rest our routing thing I Think they're just trying to avoid all the stuff in here by just manually routing around there because they probably couldn't have I don't know why they couldn't have put their planes in here like I showed before, but anyway I wasn't there when this was laid out I don't know and I'd have to get the CAD files. you have to look at the actual layout on here. So after what 30 40 minutes we finally you can answer the Opie's question. What with trace do you need to be equivalent to at the wire? Well, as let's say it's 15 a WGY which is looks roughly what we've got here.

10 ohms per kilometer. We're talking one milli ohm for a hundred millimetres. So that's not much, but it looks like we can do that with say 25. Solve: Let's say just over 20 millimeters.

So I was roughly right. Something like that. there you go. 22 millimeters.

Answer the Opie's question and the thing is like this is a fairly simple board. Like, as far as FPGAs go, this is yeah, we've got a quite a large margin. the 125 millivolts on our rail which as I said, you can extend it if you tweak that and tweak the voltage of the converters up here, you put them on the high side. so that's a wide often.

Way back in the day, a 5 volt rail wouldn't be 5 volts. They'd actually set it to five point Two Five volts so that all the circuitry near the connector ended. Get Five Point Two Five Volts. We should be on the upper side of your 5 volts tolerance on you 5.

Everything's fine. But by the time I got all the way to the other side of the board with your hundreds and hundreds and hundreds of chips on there, as the old boards were, it might drop down to Four Point Seven Five and you're still right. You were still within the margins. But modern devices within naught point Nine volts, Point eight, Even lower core voltages.

And if you put, you know, you might have half a dozen of these large. FPAA PGA's on a big complex board. You can be talking tens of amps, 50 amps, even a hundred amps. You can go into triple figures on the amps for a really complex board like this, and that's a big deal.

And and that's when you might go in here and go bugger it. I Need to ounce copper if you're desperate? I Need to ounce copper in at least a couple of your internal layers. You wouldn't use two ounce copper for example, if you had real a mix of high power stuff on your board with lots of other signal stuff, you wouldn't have two ounce copper on every one of your eight layers. Should tell the PCB manufacturer and you stack up the chart.

You'd tell them that hey, III need these? You know layers three and four, they they're my power rails I want those two ounce copper please plus two ounce copper on your ground as well. And you might have multiple ground layers. but then your shield ground layers for example I don't care I can use point Two five. Now it's copper.

It's not carrying any of the current. for example, it's just like using for switching. it's it just might have lower dynamic requirements switching AC dynamic current requirements rather than the big bulky DC type stuff. But because the problem with two-ounce copper as I mentioned in the previous video on this IBM monitor.

If you've got two ounce copper on all your layers that actually retains a lot of heat when it goes through the reflow oven and then the balls on your BGA may not read, it gets much trickier. gotta hold your tongue at the right angle you mind if you're calling the Graybeard to operate your reflow machine to get that vapor phase setting just right so that the balls like a so that you get reflow on your balls and doesn't. These boys can come out piping hot like a fresh pizza from an oven. and those two ounce copper really retains a lot of heat.

If you've got multiple layers in there, Wow Can stay hot forever and that slows your cooldown time because a lot of your A that's a lot of art and science to actually setting the temperature profile. It's not just ramping up the temperature people think, oh, it's just all about ramping it up like this and then the solder melts. it's also about cooling it down. You can't cool it down too slowly either because then the solder can enter the plastic region and cause all sorts of dry joints and all sorts.

A You know, then you get into material science and I Just met friends but on a real complex board like this with thousand pin or 1500 P and BGA is drawn like I'm with two ounce copper it's kind of ruin your day. I've had boards which you've used and really thick to ounce copper and then all of our components like tombstone all of our resistors Tim Kent Stone because we didn't glue them down and just the slight imbalance between the pads was enough to flip them all up. and yeah, it's not fun. So have I waffled on enough? Have I explained why they put those wires in there? We can't be exactly sure, but hopefully you're still with me and down this rabbit hole deep dive and I could go further than this as well, but I I might leave it there.

hopefully. I've given you a taste of what it's like to design these and layout these high-end boards with these large current FPGA devices. There's so many things involved, you know people just think about bypassing, but they often forget about that DC characteristic and the voltage drop on the traces. and and how when you have big round planes you can't really? Well, you can kind of.

You can sort of simulate this like on a spreadsheet. You can do it crudely. You can do your own poor-man's our finite element analysis and and try and calculate and stuff like that. but you know, generally that's kind of why I like to separate my powers and grounds on the board not only for very good technical Starr grounding system reasons, but also then you can start calculating stuff like this because I've got a block like that.

it's going to have X amount of resistor and then a block goes up like this. As I said, like this will actually be the shape of some of these ground planes. In fact, I might pull up a board. I'll show you a real board example.

It's not as complex as this, but oh, I think I've got one. Okay, I found an example with a Virtex 5 FPGA It's a fairly large once eleven hundred and thirty six pins and and as you can see, we're fanned out. you know a good majority of those this: I don't know 50 not used or something like that one. This is actually a ten layer PCB So let's have a look.

This is the top layer. Okay, we can actually. and if we turn on all the layers there, doesn't that look funky? You can just see all the various layers. Anyway, let's go back to a single layer mode.

It's much easier and so this is our top layer. Here's all our pads. You can see how our vias have been like. we found out a via from each one I've done a video on BGA Fan Out and stuff like that.

Don't know if I use. This was an example, but you can see about the different rails in here. We've got 1 volt, 2.5 volts 3. There should be 3.3 volts for I/o Oh yes, our wire 3.3 volt iOS on the outside.

That's quite common because the I/o pins are on the outside so your VCC pins are on the VCC your VCC I/o pins are on the output. I Believe Yeah, we've only got the three voltage rails there, so not that many. Okay, but this required a 10 layer board to fan this out and get the power in. So let's go have a look at the second layer.

Second layer is a GP so that's ground powered I Don't care what P stands for, but that's just one bigger ground plane right over the whole thing so didn't split the ground plane up layer three. It's just got some high speed differential gigabit Ethernet traces going out and just some signal layers going out. There's no power happening on there, but once again, you could have used. this layer is an extra one to get power into there because all this spare space in there.

and then there's another ground layer because of the high speed differential pair attending. I Don't know. Was it 18 bits per second? I can't remember. Anyway, so we had multiple ground layers in there.

That's we needed the ten layers and then we've got one layer dedicated to the 3.3 volt rail. So that's just flood fill right over. So this was like and I'm pretty generous on this board if you want to get your cost down. we could have.

So this one's just start signal And here is where we start to get into our power. So this one in here is. Note: that's that. That's ground fill.

That's just ground fill. nothing. nothing special there. Just decide to put some extra flood fill in there.

and here we go. Here's where we do that block. I Didn't put a block across the whole thing, but you can see that there's an internal. so there you go.

That's our 2.5 volt. So that's coming in from the yeah, it's coming in from the bottom here. So yep, there we go is flowing in up. and there you go is flowing in from over here.

So 2.5 volts. So it's a reasonably small trace that wasn't a particularly high current thing, but you can just that's an example of how that you can just put the big block in there on one layers. but you know, of course like I had to get that trace out there. Bugger it I couldn't do it on some other layer I didn't want to.

Whatever. But yeah, that's an example of having the power block and we'll see that again. Yet we see that again. You can see the split down here.

so this one is for the 1 volt. This is the core voltage, so this one's going to be higher power. so hence why Look, it's huge. You can see the split all around here like this.

Yep, there we go. It's that big pink lock. like that huge big low impedance block like that and that's coming from. Yeah, I Don't think we had a regulator on here I think it was coming from elsewhere and I could have anyway.

so that's coming from up here and that's hugely low impedance going right down to the Fpga core. So that's a good example there of doing that. that power block and big low impedance running into it. And as you can see, we don't really have many tracers coming out on that layer, just a few on the outer pins and stuff like that.

But of course when you're laying out something like this, you would make sure because of that DC requirement for FPGA that's fairly critical. You want to ensure that you do route that first so that you don't try and sneak it in later. You know you really want to take care of that right up front and then figure out how you do your signal traces later and then the bottom. I Don't think we had anything special on the bottom note that was just ground and there we go.

We had various, just the bypass in caps and stuff on the bottom. so there you go. That's eleven hundred and thirty six pin Virtex 5 FPGA but not particularly major high current requirements. Apart from that, that one volt rail did take a bit, but not much on the 2.5 or the 3.3 but every FPGA is different and I was actually going to get a power estimator for you because all the FPGA companies have these power estimator programs, spreadsheets, whatever they are or tools to, you're putting your number, engage your specific family using your switching speeds, What proof rules You're using what IO and all that sort of stuff you're doing and it will simulate fairly accurately.

depends on you know, garbage in garbage out of course. but if you put enough stuff in there ill of you or if you see me or if you finish your design and then just run the calculator on it putting on all the various factors, it'll give you a very accurate estimation the power consumption I Wanted to do that for the Apex FPGA but I Went to the Altaira Tool Power Estimator Tool website and it has everything but the Apex series. So I don't know what happened to it. Nothing.

It's probably out there somewhere anyway. and here's actually a better example of a larger board. once again, a big 600 and something pin BGA over here, but let's have a look at the different layers. This won't tell you much, but if we go, that's the ground.

Oh sorry, you couldn't see that. Yeah, my face is a bit. If you go to the power rail, you can see how they're actually split on the bottom here. that just the you know shows how you can snake things around and just avoid various things.

It's not the best example, but it shows you just how you would might split up by your power planes on the board to win to avoid voltage drops on one high power section causing issues on the lower power section. stuff like that. So I think we'll call it quits there. Sorry for the link for this video, but as you can see, there's a whole bunch of stuff involved in this and we're not gonna say we're scratched the surface, but we're probably only scratched the surface of what you can deep dive into on this sort of thing.

There's just so many different permutations and combinations of different scenarios that you can get on boards like this, and hopefully you can appreciate. What a PCB designer! What a professional PCB Designer and professional design engineers have to go through for some of these and consider for some of these complex designs that have all these you know, high-end FPGAs and things like that and this one from 2000 in this IBM one is not that complex, but they decided we need it and the wires in there for insert might perhaps multiple reasons in there, perhaps at the design stage, perhaps isn't afterthought? we don't know exactly. but anyway, it's really interesting I Hope you found that useful. And if you did, please don't forget to give it a thumbs up and subscribe and notify as well.

Hit that little bell icon wherever the hell it is. And to make sure you get notifications if YouTube will be gracious enough to send notifications to everyone when I release a new video Anyway, hope you enjoyed it. Discuss down below. Catch you next time.