Critical Path Components

Hello and welcome to the macro fab engineering podcast. We're your host, Parker, Dolman.

And Steven Gregg.

This is episode 261. So the last one, okay, no, no, no, no, I know that. We're gonna keep talking. Okay, what's going on?

The idle? Nevermind. I think it's a next week thing. Forget forget any of this.

What's the next week? The Azure?

No like a milestone?

1. What's special about 262?

Not the number directly. I think next week is five years of the map.

Let's see map episode one.

It was It was February of whatever five You're

certainly met, according to Google is a member of the European Parliament's I don't think we are a member of the European Parliament. Ah, though. Result number two is the Mac five engineer podcast.

When did episode one come out? Jesus not on our

February 12. February two. Okay,

so So two podcasts from now is five years,

February 12 2016. Where we talked about the op op amp optimal power supply?

Let's not even bring that topic up. Don't even say it.

Oh, man, this was back when? Microchip bought Atmel?

Yeah, they've had them for five years now. Five years. And that

was when FTDI was the FTDI gate.

Oh, yeah. for that. Yeah.

And Fred's building a huge solar roadway this again.

Okay, enough reminiscing we do that like every week. Do we almost

that's represents versus about the future, then?

I do that. All right, what's up, Parker?

So two weeks ago, one week from now? We talked about automotive components shortages. True. See, I made that

in the future. But yeah, yeah, you're reminiscing now.

So we talked about automatic purchase orders. And one of the things I wanted to bring up, there was a whole segment about how you can alleviate those problems, like things you can think about as you're designing your product. And we just ran out of time, we talked about like 30 minutes of like the bad stuff, and never got to like as we normally do, yes, we normally do. So what can you do to alleviate component shortages? And like component lead times and your design? Because this is something that Steve and I do like, as we're picking our parts is probably like, this is like the number two thing I'm thinking in my brain when I'm picking parts out.

Let me let me take a stab here. I haven't looked at your notes. So I don't know what's in your notes here. Just buy all your parts like way, way, way in advance and buy like the world stock of them.

That's one way. That's how like Tesla, and like Apple do it.

Okay, not joking about this. WMD did that with one thing. Like there's a there's a very specific as a in surface mount. And there was one manufacturer of it. And they came out with a note saying, Hey, we're no longer supporting this part, which basically means the world will no longer have this part. And this was years ago, and yeah, we bought them all.

Do you still have stock? Yeah.

Yeah, for sure.

So on that note, is back at pre macro fab days, when I was working with still with Chris church at Dynamic perception. We had a product that it was a product before I came on board that you had had an opto coupler on it. It was a Joe Schmo optic coupler except it was a really weird four pin DIP, which actually for an optic couplers not weird that's actually kind of a normal thing for a optic coupler, except it's pin out was reversed. So you couldn't buy any other dip for Opto coupler for some reason this one was picked. And basically like when I came on board, like at the time, like through a broker, dynamic reception bought like the world's supply of this components. And then we did a read, I did a reskin of the board that basically just like SMT all the parts and replace that optocoupler.

Yeah, makes sense. So, no, that

is not what we're gonna talk about, though it's not by everything out and leave everyone else hung dry. And so yeah, so that's this is the number two thing I'm thinking about when I'm buying and picking parts. Number one thing is, is this part going to work for me? Right? Because that's obviously the most important thing is if it's gonna work or not. Number two is, can I get this part for the quantity and when to build it this thing. So the main thing I look at is like, I kind of separate my, my processes by components as well, it's like passive components, the majority of your passive, especially for digital stuff, is you could swap out a 10k resistor, that's an o 603. For all the time, like, because you're only running like five volts or 3.3 volts at a couple mil amps doesn't really matter. There's billions of these things out there, right, never have to worry about them. But the thing you have to watch out for is when you start doing weird stuff with passive components, especially if you were component has extra ordinary specifications, in particular to the package size. So like, this is like voltage ratings, wattage ratings. ppm, it could be what other like or capacitors like high high capacitance or really weird coefficients. And like, a small small packages and weird coefficients. So well, I basically when you pick apart that's weird like that in quotes, right? So let's say you need a 500 volt resistor, okay, which is for s&t part that's pretty high rated, most are like 200 volts, is look around in the part you pick. So let's say I picked an o 805. That's 500 volt rated. Look around and see how many parts on the market fit that specification, it is just yours might not use that part. And usually the case is, depending on the package size in for passive components is that dictates a lot of the specifications. And then when the manufacturer does something really weird, that's when you can do the weird stuff like 500 volts, and oh, 805. But if you go to, let's say, a 2010 package, you can get like a whole sheet worth DigiKey Mouser of like, once I hit your voltage rating. So if you can avoid using, let's say, let's say you have any space for a 2010 resistor, because that's a lot larger. But let's say you have plenty of space on your board, you don't have to have an o 805. Size, then change it to do package that has your availabilities go through the roof. Now, if you have to have an O 805, because of space reasons. I wonder how Look, what's your isolation on 500 volts? Like jumping the pads on? Oh five, I have no idea. I don't know what the calculation is their

creepage would be kind of crazy to Yeah,

but that's just an example. I just pulled up my, my, my, my brain hole. Capacitors the same way, like you can buy a 10 microphone Oh, six or three. There's not a lot of them on the market, though. And they also have really crappy capacitance voltage curves, as well.

You know, we talked about that exact same thing with a bolt engineer. Where basically, if you take, let's say the three major passive sizes. Oh, 4020603 and, oh, 805 Let's just take those as examples. There's sort of like a bell curve for each one of those have values that each one of those fits in, and and show like, can you find a 10? micro farad? Oh, 603? Probably, but it's probably better to go with a 10 micro farad Oh, 805? Because that's right, right, because that probably fits on like the edge of the bell curve of values. And it's not an actual bell curve. I'm just using that as a. Yeah. Visualize

the, how many components are offered in that package for that value?

Right, right. Yeah, at the same time, like, yeah, like like 100, Pico farad, you can find that 1210 100 pico farad they exist, but why not use an o 402 100? Pico farad especially because I bet that's actually voltage rating. It could be it could be but but most of the time, if I'm using 100, Pico farad I I want it to be, it's likely in a situation where it's going to be like tight in the circuit. And I don't want this monster cap out there. Yeah,

that and you want it. You know, usually you're picking that up because you're preventing, like ringing on like a crystal circuit or an oscillation circuit or something like that.

Yeah, I took them real close into op amp stages. So to keep things under control. Yeah. Yeah.

You talked about that last week. So yeah, check, basically, for your passive components check for extra ordinary specifications that and just he's like, okay, is this component how many of these parts meet my specification? So that's passives. Usually, you're okay with passes unless you have weird stuff. So then you move to active components. And this is where stuff gets a little bit harder. But there's, you can separate active components into like two different categories. One are kind of like jelly bean style active components, like your discreet op amps, and transistors, and MOSFETs. And diodes, which this we talked about this logo for the podcasts and kind of argued about it, but I consider a diode an active component, but it's also kind of like a passive because you don't really do anything to it. And circuits, like you leave it. Like, it's got two pins, right? Input and Output.

Sure, but I think they are traditionally considered active component. Yeah.

So that's why what is an active component?

That's a difference. I was looking at lists of things that that constitute active versus passive. Before this, and one of the one of the items on there was, if it's active, gosh, okay, so if it's active, it requires like particular voltage to work. That's, uh, I stripped that, that down. Which that's semi true, or that is true of diodes, I should say. And then the other argument is, active components are nonlinear. We by nature, and passive components, in their truest form, are linear by nature. Yeah, cuz

I brought up capacitors are technically like in real life aren't technically linear, because changing the voltage changes how much capacitance they have.

Well, that also depends on the type of capacitor

that ceramic capacitors, which is like that's 99%. But that's again, in the ideal sense, then yeah, capacitor a, an ideal capacitor is a linear device. And a ideal diode is a nonlinear device. Right? That's right.

Yeah. It's got plenty, but I made the marker, like, depending on how like, wherever you look at the curve, you can find linear portions of the curve of a diode. You mean if

you look at a diode, like a digital device? Yeah, right. Instead of an analog device, would you have infinite scale? Infinite resolution?

Oh, yeah. Geez, that was a

long time ago. We found that data sheet for a potentiometer that it was like a 5k. Just you know, guitar knob trim pot. It wasn't borns. Yeah, but like it says under resolution it said theoretically infinite.

They're not incorrect.

If you if you move the wiper atom by atom

discretes the argument remains the same.

For years, five years ago.

Oh, that's great. So you know about like the four fundamental electronic components right? There's a funny joke about the Four Horsemen but okay, I think could be the resistor that capacitor the meme, the inductor and then there's the fourth the memristor Yeah, the memristor. So, resistor, capacitor inductor are all linear, but the memristor is not interest. So, it's a fundamental component. I don't think it's I think it's considered a passive component.

Yeah, it's just a Oh man. It's just remember, it remembers how much current it's a current device. It remembers how much current has flowed through it,

right. It changed its resistance base level of how much current has flowed through it. Yeah, yeah. Okay, sorry. Back anyways, okay. Yeah, active components.

But in discrete, the style, which are diodes, MOSFETs, stuff like that. Usually you can get alternatives for those two So when you're picking those components is also pick parts that you can get, you know, multiple, like, different manufacturers of building them in different in the same package, right, like a to 220, that's a three year old part, but you can, you can buy a MOSFET, N channel MOSFET, that's probably going to work from like eight different manufacturers, and all those manufacturers probably actually sell, like 10 Different MOSFETs that will work for your application,

you know, great example above and beyond this. Okay, so say you got you got fancy, I don't know, let's say, I'm just going to say motor controller or something like that, you need some MOSFETs to drive, or let's just say an H bridge kind of thing. And, and you're designing the next whiz bang H bridge out there, and it's going to be the best H bridge ever. And you know, you need a lifespan of 10 years that this product is going to run right. So you got these two MOSFETs that you're looking at. One comes in this like, patented Uber Power Pack 9000 package that's super cool by this place, and it lock on it's gonna solve all your problems. And then this other one, it's not as flashy and it's not as cool but it comes in a regular Deepak and it'll work just fine for your application. Which one do you pick? Probably the Deepak one right. Because it down because I

have a package already designed my EDA tool.

Well, okay, it makes it a little bit but the Deepak one because it's a lot easier to find another Deepak MOSFET down the road, as opposed to Ultra Power Pack 9000 from whatever company that might go out of business, right?

Yeah, I really wanted a Ultra Power Pack 9000 package, though.

I mean, just go look at every MOSFET manufacturer, they have they're super like Mega frozen, cool package, or whatever they call it. Like they always had some kind of buzz word around

it, there's those new style MOSFET packages of the head, like the big thermal pad that come out to like four fingers, and they're really flat. And they're really flat. Yeah, yeah. So you can design a chunk of aluminum on them. Yeah, those are designed to go right on the heatsink. And back in five years ago, six years ago, when we started the podcast, those were kind of new, right now, a lot of manufacturers make those. So those are probably completely fine to make, to use now in your designs, in terms of supply chain stuff. So yeah, you can diversify that way, basically looking at different packages, diodes, same way. Now, then you get to the other part of active components that are, it's very difficult to find alternatives, if at all, and that's like integrated components, like microcontrollers. ADCs, that kind of stuff. Anything that's I was gonna say like multi lead, like, like QFN or qf P or, but that was like your friends don't have leads.

Anything over single function, right?

Yeah, over single function is good way to put it. That gets really difficult a lot. Sometimes you might be able find like a different family and Micucci or st family, my controller, but it's got like a different RAM setup or like a different amount of RAM or ROM, there are a different function if you're not using. Or you might be able to find, oh, this qf P and QFN share the same pin out and they're the same number of pins. And you can fit the QFN inside the QFP package on your board. So you can use either one. You know, I've seen that before. I've seen like actually with BGA is done that way too. Like they had the BGA. And then they had a q if P package. And they were just nested inside of each other. Which helps you out on on your supply chain, you know, O Q, if p is out of stock, let's use the BGA. So the BGA is out of stock used to QPS. And then sometimes you can get different. That's like one kind of different packaging. And then there's the other kind of packaging, which is how you actually get it from the manufacturer or distributor is it on real or in tray or in tubes, because sometimes the manufacturer will spec that as a different part number, like FTDI is famous for doing like dash TR at the end of their part numbers for tape reel. Some manufacturers don't specify that. But most most of them seem to have different part numbers for how the packaging from like how it's physically packaged, versus electrically packaged.

Oh yeah, we've all seen that big spreadsheet at the bottom of every Texas Instruments data sheet. Oh, the bazillion parts in all their different packaging cells. It's

like pz our package and you're like What is PCRs you have to go back up and find the mechanical drawing of it. You know, okay, that's that's decent, right. So sometimes you can get Lucky there is basically like, okay, the TR is gone, oh, the dash t the tuba version is available. So you can, you can spec that also as well. So you can go TR and then you can put t as your alternative. Yeah, I'd like to hear more about what other people had to say about like the integrated components swapping is like how to make that better. Because I don't besides like that there's not really any good solutions, I don't think.

I mean, if you're looking at the hierarchy of parts that you're talking about here, these are sort of the top of the pyramid, as in like, you're going to have the fewest amount of these, most of the time in your design. And so if one goes obsolete, it's easier to put your effort into replacing one of these than it is saying like, Oh, damn, that. Oh, 402 3000 micro farad cap that I was looking for what obsolete, I'm gonna have to redesign everything.

Yeah. What did I call those kind of? I can't remember why I was calling those parts. Oh, like critical path components? Basically, parts, you just can't have a sub four.

Oh, right. You know, and that brings up an interesting point about like, the beginnings of design, like there's so many different design philosophies. But before you even put pen to paper, or you drop parts onto a schematic, there's, there's like an exploratory I don't know phase, where you're looking at like, is this product even worth looking at? And and even if you don't think you do that, you probably still do it. Because like, if you have this idea in your mind word, like, Oh, I'm gonna go make a thing. Like, you just pass that phase by saying you're gonna go make that thing because it like it clicked in your mind as being like, Oh, this is good to do. They pass the sniff test, right? But but I'm, like I say a more grand scale. Like, if you have to make a presentation to I don't know, the the executive team and say, hey, I want to make this product. And here's what it'll do. And this is why it's whiz bang. At that point, one thing that's that's something really important to kind of keep in mind is, if you're designing your product around, say, a chip, or an IC, that does that super awesome, cool function that you want. And it's the only way to execute it, you're building yourself kind of a trap in a way, that's not to say that it's not worth doing, it's just, you need to build a good reason as to why you need to do it. You know, I deal with that a lot in the audio industry, where like, somebody will come up with a chip where it's like, Oh, it does this audio thing that's really, really cool. And you can only do it if you buy this one chip. And a lot of people buy that one chip. And what ends up happening a lot, I've seen a lot of times where like that chip goes, that company goes out of business, and now that product cannot be built anymore. Or you ended up getting the whole market filled with a bunch of people who also had your exact same idea of using that same exact chip. And they just everyone just used the example in the datasheet as to like here's how to execute it. So you get 5500 products that are all doing the exact same thing with the exact same circuit inside. So like, keep that in mind when developing something like that kind of goes into that exploratory phase I was talking about.

One more thing for me tonight. Or whenever you're listening to this podcast is connectors. And connectors. We've said on the podcasts are like the bane of your existence when trying to find because you got to find that right connector. But now you got to find another alternative part connector that works with that part with that footprint right? Sometimes you can skirt that by having different coatings or different plating. Like Molex is like in my mind famous for that. Because when you go on their debt, the data sheets they'll have like pages of like these are the alternative parts that fit in this family and then hold up these are all the plating options. So we have like we have gold but we have gold that's over brass, we have gold that's over 10 We have gold that's over bronze phosphorus phosphate pins, like all that kind of materials in there. And then we have like, you know, tin we have bear connect connectors, so so make sure like if you're specking it out, you might not need fine gold connectors and you might be able to get away with tin, right? Sure. So take a look at that too is like You know, if you're using connectors that have really low volume, like because they're really expensive or are that's, that's actually another thing is look at is also don't just look at what distributors have, because why just said they're expensive components, like if your components like 30 to $40, like let's say FPGA, Mouser, and DigiKey is not going to keep 10,000 of those on their shelves, they're going to keep like, a tray of them maybe. And so you might have to actually contact like Mouser DigiKey, or your contract manufacturer directly and be like, hey, I need like, 10,000 of these potentially, what does that look like? What is my lead time look like to actually get 10,000 of those, like, so we can build with this? Because, you know, if you try to call up, let's say Altera, which is now Intel, right? Was it silence? I can't, I can't remember who by who? You can't they if you call them up, and you're just like Joe Schmo, they're probably not gonna talk to you.

Oh, I can tell you from experience, like you're not getting an answer. Yeah.

So you're better off probably working through another distributor like Mouser or DigiKey, or with a contract manufacturer directly, and being like, Hey, I'm designing new product. And there's only a couple of these, like, there's like 10 that Mauser. But if I can use this part, I need like, 10,000 What's that look like? And they can give you like, okay, that's, we can actually get you 10,000 In a couple of months, or never, or next week.

You know, two more, two more quick design things that that I tried to employ that I think is worthwhile. The note on diodes here reminds me of zingers and things Zener diodes come in a variety like you can get, you can get almost any voltage you want in a zener. And the thing about Zener is, is 18.33 volts, that's a great example of something you probably shouldn't pick, unless it is like the magic zener that you must absolutely have, what's the lead zener will pick like if if 13.33 is like 370, sorry, three, seven, would a 12 volt zener work in that situation, or what a 15 volt zener or something that will last a lot longer. 12 and 15 are going to be around for a long time, let's put it this way. But 1337 is probably not. So it gives you that right tone, that elite tone, you know, and another thing to resistor values, they they're not always fixed. There are as you know, there's the different categories of resistor like e 24, and E with 96, or whatever they are. A lot of times it is actually cheaper to just series and parallel resistors to get a more unique value than to go spec a single unique value that you know has your secrets or is like in your approved vendor list. You know, so consider, like, great a great example with this that goes through my mind all the time. A 50k resistor doesn't actually exist, I'd say 5k is the same thing. Fuck it. Yeah, cuz you get 49.9k or 4.99k or whatnot. Nine, nine. Yeah. Okay, so you could spec of 49.9k and deal with whatever error there is, if you're using like a gain circuit, or whatever, divider, or just put 200 K's in parallel, and then you actually get 50k. And I I'm using quotes here actually get 50k. But you get so and on. The beautiful thing about that is it's one less skew. If that's the only place you're, you're putting it, I usually go and sanitize all of my circuits. Now, whenever I'm done, like designing them, and I look at, I look at my list of resistors I actually print them out and I see what how many times have I used this value and this value this value. And I look at the very bottom of that list, and I see how many resistors I have where they're like single values to use, because single use means that somebody has to go through the effort of putting a real on a feeder for it to put one resistor down. Could I use any of those resistors at the top of my list to account for those ones down at the bottom? And most of the time I can. And most of the time whenever I scrub circuits like that, I'll get rid of 567 unique line items by just series and parallel. Yeah, it means you spend a very minimal amount more per board, but it costs more to have a guy put parts on a reel than it does for you to put two extra resistors on a board. So it's worth considering and for the future. If you have like a weird value. It's Always ask yourself, do you really need a weird value like Is it super required that you have 365.24 ohms in in your LED lighting circuit, you know,

so not with a circuit board design, but with a a Jeep project I've been working on this, this came up and I'm working on the the rear defroster replacing the rear defroster on the wagon. And when you I found out the whole grid, it's like point one ohm. Okay, well put 12 volts across that. And yeah, that's, that's a lot of power. And so you're only supposed to pull about 12 amps. So your grid needs to be like one ohm. Okay, so I'm like, Okay, I need put it in series resistor. Okay. Calculate it out. It's like 150 watts. Trying to find 150 Watt, one ohm. It's pretty hard to find that on like, you know, Amazon, you can probably go to Mauser and get it. But I'm like, I was being lazy. But I can get lazy Mauser. Yeah, I can get to 100 Watt, two ohm resistors. And just put it in parallel. Yeah, you go. And it was that ended up being cheaper than getting 150. Water. Sure. And I'm like, Oh, good. Now wants to dissipation. Yeah. Yeah.

And we proved a long time ago, that if you put parts in, in parallel, like you get, you get a tolerance bonus, right? Sure. Two parts of parallel is not going to give you much of it. My

defroster grid really needs that.

You want to hit 12.000 amps, right?

Sure. Somewhere in that range is good enough.

Plus or minus an amp?

Yeah. I think it's like a 20 amp fuse. So

okay. Yeah, you're good, fine.

So that's all I have on that. That's a big, meaty, juicy. Yeah,

that was a 30 minute topic. Yeah, that's beefcake. Okay, so let's follow that up with a really short one. So I was thinking about what what I talked about last week about like, search features in in in the big vendors and stuff. And I want to add one thing that I think would be really cool. And this is totally like, no learning AI. Oh, God, wouldn't that be amazing? Well, actually, that that could that could apply to this in a way. Now. Now, this is totally like pie in the sky would absolutely love this. And this would be like, the robots are more intelligent than us. We don't need designers at this point. But no, okay.

We just get armor, heat those and drink on the beach.

So I ran into a situation actually, earlier this week, where I had a transistor, or I needed a circuit that had an NPN and PNP transistor. I didn't have really strict requirements about those transistors, but I did kind of want them to be similar. They didn't have to be, but and in similar, um, meaning more like package than, like, characteristic, they just have to be able to handle an easy task. So I went, I we have one of these in stock right now at the, at the factory. So I went and grabbed that and I was like, cool. This one will work and it wasn't an NPN. And I was like, I want to know, what is the complement of this? NPN. And most not most of the some of the time, if you find an NPN there is a PNP complement to it. And I guessed the PNP number and got it right, like just flat out guessed it. So the transistor I was using is the is the BCP, 54, which is an NPN, moderate, small power transistor. And I just guessed, well, if I put in BCP, 52, will that be the NPN? Or the the PNP version? And lo and behold, it was add, like, how did you figure that out, though? Because I've dealt with enough transistors that typically the last digit, you just change it?

Ah, well, because NPN minus PNP is not too

bad, right? Of course, well, like the classic 3906 and 3904. They're two digits away from each other. The small signal like everyone on earth has used those. So like, I just guessed, and I got it, right. But I was thinking like, how cool would it be if I could type in blah, blah, blah NPN. And it would be you know, it's sort of like, I get that not everyone needs to know the compliment to their transistors, but it would be awesome if my search results would be like, Oh, if you need to know the compliment, it's this part and the end by complement, I mean, it's just the NPN version in the exact same package, exact same pinout everything is identical so and I actually purchased the transistor and tested in some circuits and worked out really well for me. But it would just be really nice to know what the complement of a product is. Now, at the same time, if you take that and you make it way more difficult for an AI as if it's not already difficult to be able to detect, and a compliment, I would love it if I found a connector. And my search results would be like, Oh, this is the meeting male or female part to it. Or all of these parts will connect in tune the other

seen some connector companies start doing that. What was it? JST? Does that? Yeah, they're family connectors, they'll have, you know, they'll have this huge grid of like, this is like the connectors and then this is like the terminals you might you can use with each one. And they're, they're the stuff they hook up into and all that good stuff.

Yep. Yep. Which, that's great if you if you're going and searching from all the manufacturers, websites themselves, because a lot of them provide that information. But it would be cool if my big vendors would do that. No, I understand what I'm asking for. Right. There's ridiculous but that's like, super cool future stuff. Cyber search. Okay, another topic. We want to go over I ran into a video this week. I think al Williams actually wrote an article on Hackaday about this. It's it's about PCB crosstalk. So the video is, is made by a guy named Robert for INEC. I probably pronounced that incorrectly. I've watched a handful of his videos, and he is fantastic. He takes electrical engineering concepts, and he breaks them down and he makes them he breaks them down in a way that they're still like for electrical engineers. It's not like breaking them down for just like anyone. But he takes really difficult stuff. And he explores it, and these videos are fantastic. So he has a video with a with a guy named Eric bogatin, where they talk about PCB crosstalk. And they they don't go into like the nitty gritty math behind things. So it's not like that awful electromagnetics class that you had to take. But it's way more practical stuff. It's practical in terms of like, okay, visualizing what the problem is, how it manifests and, and ways to mitigate the problem. And so we'll post up the link to this. It's like a 50 minute video talking about PCB crosstalk. And it is excellent if he asked me. So I wanted to talk about these big crosstalk in this segment real quick, and talk about the two kinds of things that really contribute to PCB crosstalk crosstalk and then list out some things on some methods on how to mitigate it. So there's the two main aspects that contribute to it is capacitive coupling and inductive coupling, and those words are loaded there. Because it's not exactly capacitively coupling, and not exactly inductive coupling, but just like, those are ways of visualizing it.

And the Yeah, I was gonna say visualize it, but like, the way to understand it is to think of it like that.

Exactly, exactly. And they go through that in this video. So capacitive coupling is will basically couple voltage spikes from one track to another, and inductive coupling will mutual mutually conduct current from one trace to another. So you have two things that you're battling when you're talking about PCB crosstalk. And PCB crosstalk is mainly due to traces running in parallel to each other, close on a PCB, which if you look at PCBs, or if you like conceptualize a PCB, a lot of times you get those like, really cool images in your mind. Or if you go to Google even even type in PCB, where there's like 50 tracks like running parallel to each other. And then they all make a cool 45 Look, and then they all like turn. And if you think about like, a ton of different tracks running next to each other, very close in proximity, and running exactly parallel like that, in a lot of ways that can spell disaster for your board. Especially some of the situations that they were mentioning in this video is, you know, one one trace has large voltage spikes on it, and it was causing a microcontroller to go into interrupt loops when it shouldn't, because it was because two tracks were running next to each other and one had an interrupt trigger on it. So another track was causing the PCB to like basically go into like an infinite loop where it was like trying to spit something out and then interrupting itself. So Okay, so basically, I'm not going to try to recreate this video on this podcast, I think you should just go and listen to it. It's a it's like I said, it's super great. But what I want to do is just like between the two of us kind of like walk through, what are some ways that we know right now that we can mitigate? crosstalk between two lines on a PCB? I actually wrote down a list. So I have some things, but I kind of want to get your thoughts on it, Parker.

So one, make your PCB really, really big and put them on opposite sides of the board.

Well, I said, I said mitigate, I didn't say like, completely get rid of it.

But that is one is like, is keep? Well, I do this, like what I do is keep your high power, or high current traces away from data traces. Yeah, that's that's basically rule number one, I would say. So like, if you look at the big big one would be like the penetrator pinball board, the all the high power stuff for like the power input for like I say, the low voltage side, which is 12 volts and down, that's in it's one part of the board, that all the filtering all the fuses for that all the current monitoring for that is in one section of the board, all the 50 volt, high voltage stuff is in one section of the board. And then the low voltage signals are on its own separate part of the board. And then they just go out and and had the, you know, they just touch those areas to pick up signals and stuff, or send signals out in terms of like, triggering a MOSFET, or something like that. And so keeping stuff separated as much as you can, within your space constraints is like, probably the easiest way to mitigate crosstalk.

Right, yeah. So that's actually the first thing I put on the list is increasing trace separation. And that's just like, across the board. A across the board, keeping the separation just larger, everywhere, as much as you can. So in other words, to flip that around, say more like avoid running traces directly next to each other, especially over long spaces, if you have to, for a short period of short section, you know, you'll just have to accept the results of that. But the effective capacitive, or the effective capacitor that goes in between them becomes a lot smaller, the further your distances away from each other, and the, the smaller the actual parallel section of those traces are. So basically more space more better, right.

And then, the other thing is, when you have if you have to run traces, let's say on another side of the board, and to cross is basically minimize how much they are parallel. So like if you're crossing your traces. So like, let's say you have a couple of traces that are running left to right, and you need to have a trace that goes from top to bottom, well, don't run them parallel to each other for a bit and then kick up, right, is actually just trying to jump as fast as you can across them.

Right, right. And, and so the, it's funny, because a lot of times it's easy to think about a layer as like an individual thing and other layers as like completely individual separate things. And that's just totally not right. Because, yeah, if you run two traces parallel on one layer, you're creating a capacitor between those two orky, you're creating capacitance between those two traces. But if you run them parallel on two separate layers, you're doing the exact same thing in a different Axi. It's so you're not, you're not mitigating anything by running them parallel on two different trade layers. Now, it will be different, because you now have a different dielectric in between them, and you, you potentially have more space in between them, but you're not necessarily solving the problem by doing that. And I think one of the one of the key things is if you have to cross traces that you know, are potentially problematic, cross them at 90 degrees, and cross them at 90 degrees on separate layers. And that kind of gives you the the, your best situation there.

I'd like to see you try to cross a signal through different signals on the same layer.

Right, right. So so here's another one. That's, that's fun. Let's, let's consider we're not doing two layer boards, let's say four and above. Where do you put your ground plane actually matters, right? So a lot of times it's easy to say Oh, my top is signals and my bottom layer is ground plane that may not be optimal, for for reducing crosstalk So if, if take, for instance, another situation where you have your top plane as signals, and then the plane directly beneath that as ground, you've now placed your ground significantly closer to your actual signals. And what ground or a reference plane does is it actually kind of groups and gathers electro magnetic field lines around the trace. And it can actually, in a way, steer them away from adjacent traces. So it's super complex. And I'm not doing it justice by saying this, once again, watch the video, it gives a lot better i ideas on it. But you can, you can help mitigate crosstalk between two traces by putting your reference plane closer to the actual traces. And by the way, I didn't know this. I love this, by the way, that the whatever trace is cross talking to another trace, whichever one is the one that's the problem is called the aggressor. And the one that's like being crossed talked to is the victim. Oh, great. It's the aggressor. So, so yeah, where you placed your ground reference inside of your, your four layer board, whichever plane, you have to accept what the you know, the impact of doing that is, and, you know, whatever traces you have that are like very sensitive traces, it helps to put them on a layer that's closer to the ground plane, as opposed to one that's further separated from it. There's three other things that you can do right off the top of my head, that technically work, but you don't normally have control over these three things, but they're just, I guess it's like academic to consider them. If you change the voltage. In other words, your you know, your digital step doesn't have to go as far you reduce crosstalk. If you change the current, in other words, the load is higher, and you have less current than you'll have less current, sorry, mutual conductance, less crosstalk. And then the last one is if you lower the DV DT, so the change in voltage over change in time, so in other words, the speed or your frequency, Yeah, or like on a square wave the step if you if you make the steps slower, he slew it. That also helps with with changing it, but most of the time, you don't have choices on those three things. So but like I said, academic it's worth considering

I Oh, I'm actually thinking about completely wrong. Yeah, got the slew. I'm thinking about the actual frequency, slowing that down, which actually would that would reduce crosstalk, but you're talking about reducing the slew, which would reduce, especially on a square wave, we would reduce how many high frequency? events come off off that that square wave?

Well, okay. Yeah. So actually, so a square wave is a interesting case, right? Because if you reduce the frequency, then yeah, you'll reduce the number of events. Yeah, you'll reduce the number of events. But let's say that the edge on it is a one nanosecond rise or one nanosecond fall like that still can ring and cross through, even on a 10th of our hurt signal. It just happened. Let's say a crosstalk 1/10 of a hurt, you know,

yeah. No, you're right. Yeah, I was thinking about it wrong. Yeah, yeah.

Well, and here's the thing, like, we've actually run into this at work with some things where we kind of had an unavoidable situation where we had traces running in parallel. And even down into the audio range, we ran into some crosstalk situation where two channels on on one of our modules, like just slightly bleed to each other, and it kind of sucks, so we have to mitigate that, but that so, you know, take as many of these things as you can and apply them to, to your signals and you'll you'll likely get results right away. Now. There's a lot of other things that ripple through there too, right? Because like now, you still have to consider EMI, you still have to consider FCCS you have to consider all this other crap. This is more like before all that stuff. This is more about like, Well, my circuit just have to work first. Well,

yeah, but uh, doing a lot of this stuff helps you with those other things like FCC and EMI emissions. That's right. Like, like all what we just talked about is what I actually I don't know anything about it as for crosstalk I think about as I got past FCC.

Yeah, right.

But in the end is the same thing. What is crosstalk it is the aggressor emitting on the victim.

Yeah, he's effectively Right. Yeah, yeah, I guess FCC is just like they don't care if you have aggressors in victims on your board, they just want your aggressors not talking to the rest of the world. Yes, they don't care if your board works, they just don't want it polluting. So, there's one other thing that I think is worth talking about that appears to be crosstalk, but it actually ends up just being a design flaw that isn't necessarily crosstalk. And they don't talk about this in this video, this is something I've run into myself, I call it feed forward current, it probably has a correct name, but I don't really know what it is. So if you are running, unregulated power supply rails, or moderately regulated power supply rails, all of the items or items, all the circuits that pull power from that rail, can reflect whatever their current is up onto the power supply rail. And if you have any kind of resistance in your power supply rail, then the voltage can actually fluctuate on your power supply rail. And if your voltage fluctuates on your power supply rail, you're now feed forwarding whatever signals you have on your individual circuits to all the other circuits. And so it's different from like a ground lift, it's actually the power supply rails moving like oscillating, yeah, effectively, being modulated by whatever signal is going in. And that can feed forward, I actually ran into that in that situation not too long ago on a circuit where if, if you took the volume effectively down on on my circuits, the volume happened somewhere downstream in in my circuit from my input, which means that I have active circuitry between my input and my volume knob. So if you have your volume knob all the way down, that doesn't mean that the circuitry in between isn't fluctuating with your input signal. And so I could have my volume all the way down, and then turn stuff up later in the circuit. And I could detect my signal, it was actually bypassing my volume knob effectively, because my power supply rails were not regulated well enough. And that's not a situation we run into very often. Because we typically have voltage regulators that that handle that. And most of our active circuitry have PSR, or a power supply rejection ratio that is pretty damn good. Like you can get PSR RS and like the 100 120 140 DB range now. So like even if your power supply rails wiggle a little bit, it's not going to leak through your active circuits. But you know, take an old radio, for example, where you'll have discrete transistor stages, the power supply rejection ratio on a discrete transistor stage is awful. It's virtually non existent. So if you wiggle the power supply rails, the whole radio is gonna wiggle, right. So and the thing is like it that can very, very easily be interpreted as crosstalk and it's not technically crosstalk that's just feed forward current or whatever the official term is for that. So I don't know, if you're diagnosing a circuit and you see that don't immediately just go well, that's crosstalk like, it might not be.

Sounds good to me. I've never experienced that before. So experienced crosstalk not feed for current.

Yeah, like I said, in most modern circuits, we have well regulated power supply rails. So yeah, I don't know, like, I guess it could happen in a situation where you say you have switch mode power supply, and then you put some passive filtering after it to kind of get rid of some of the junk and noise on it. You could experience that situation because the passive filtering is buffering, your active it's it's buffering and it's an impedance in line. So that's a possibility.

Yeah. Cool. So that possibility was the macro fab engineering podcast. We are your host Parker, Dolman. Steven

Greg. Let everyone take it easy

Thank you. Yes, you are listener for downloading our podcasts and making our podcast the second best electronics podcasts on the entire internet. According to random people on Reddit. If you have a cool idea of project or topic, let's see and I know this is laughing like crazy. I just dropped that item. Tweet us at backfat at Longhorn engineer at analog EMG or equals that podcast at Mac fed.com. Also, check out our Slack channel the second best electronics Slack channel on the internet according to this person right here it's Mac fab.com/slack is how you get an invite