Is the AI Machine Learning based DeepPCB autorouter better than Dave at routing a simple double sided PCB? Let's try it! TLDR; It sucks. Humans for the win.
If you find my videos useful you may consider supporting the EEVblog on Patreon: http://www.patreon.com/eevblog
00:00 - DeepPCB AI AutoRouting?
03:29 - Autorouting is 90% placement, human powered!
03:53 - The marketing claims
05:22 - Ironically, Dave doesn't have an Altium license.
05:39 - Let's start
06:39 - Live update routing
07:41 - This is not looking good
08:13 - What would Dave do?
09:14 - Rip up and Retry
11:11 - AI is failing
12:20 - WTF is it doing!
14:39 - Choose your crappy solution
15:24 - After 24 hours, it FAILED, hard.
16:34 - This thing has no idea how to route a PCB
19:26 - This is just awful
20:27 - Human tips for routing
22:16 - To be fair, Altium isn't much better.
23:24 - I can't see any point in using this
Human vs Autorouter: https://www.youtube.com/watch?v=sffuvnGhano
Nixie tube project Playlist: https://www.youtube.com/playlist?list=PLvOlSehNtuHutdg1kZkG7aAYhjoJnk2fc
Forum: https://www.eevblog.com/forum/blog/eevblog-1535-deeppcb-ai-autorouting-tested!/
Web Site: http://www.eevblog.com
Other channels:
EEVblog2: http://www.youtube.com/EEVblog2
EEVdiscover: https://www.youtube.com/eevdiscover
T-Shirts: http://teespring.com/stores/eevblog
#ElectronicsCreators #ai #fail
If you find my videos useful you may consider supporting the EEVblog on Patreon: http://www.patreon.com/eevblog
00:00 - DeepPCB AI AutoRouting?
03:29 - Autorouting is 90% placement, human powered!
03:53 - The marketing claims
05:22 - Ironically, Dave doesn't have an Altium license.
05:39 - Let's start
06:39 - Live update routing
07:41 - This is not looking good
08:13 - What would Dave do?
09:14 - Rip up and Retry
11:11 - AI is failing
12:20 - WTF is it doing!
14:39 - Choose your crappy solution
15:24 - After 24 hours, it FAILED, hard.
16:34 - This thing has no idea how to route a PCB
19:26 - This is just awful
20:27 - Human tips for routing
22:16 - To be fair, Altium isn't much better.
23:24 - I can't see any point in using this
Human vs Autorouter: https://www.youtube.com/watch?v=sffuvnGhano
Nixie tube project Playlist: https://www.youtube.com/playlist?list=PLvOlSehNtuHutdg1kZkG7aAYhjoJnk2fc
Forum: https://www.eevblog.com/forum/blog/eevblog-1535-deeppcb-ai-autorouting-tested!/
Web Site: http://www.eevblog.com
Other channels:
EEVblog2: http://www.youtube.com/EEVblog2
EEVdiscover: https://www.youtube.com/eevdiscover
T-Shirts: http://teespring.com/stores/eevblog
#ElectronicsCreators #ai #fail
Hi back in video number 975 six years ago. Whoa! Um I did a very popular Uh video where I pitted my Uh PCB routing skills against Altium's Auto router to see which one could do a better result on my Nixie 2 bar project uh which was like a most like five or six part video series. If you haven't seen it. where I uh did designed this Nixie tube project for my YouTube Silver award thing anyway LinkedIn down both LinkedIn down below if you haven't seen it and uh, tldr is that nah the auto router was yeah, it was No.
I was you know I don't want to brag, but I was much better than the order router. Now as a former professional PCB design engineer who did route in every single day like for decades. Yeah, Auto routing is actually useful. but as I've said many times before when discussing Auto routers is that it actually it sometimes takes more time to set up the auto router properly.
All the parameters that go into you know all the constraints and parameters behind order routing. Before you hit the auto router button, you can't just place your components and just hit the auto router and Magic happens. Um, that just no, it doesn't work. Auto routers are actually incredibly useful for lots of uh, especially the more complex a PCB gets.
Like you know, motherboards and things like that. you know, incredibly complex things usually not going to 100 hand route those you'll probably like. Set up a lot of constraints, but you'll spend. You could spend a week setting up constraints before you hit that auto router button and then it can actually produce really good results.
But anyway, in this particular example which was a double-sided uh board here, um yeah, the one on the top is the auto router result. the one on the bottom is uh, mine. So yeah, anyway, you can check that out. there's the bottom, there's the top.
I've put in a nice ground fill and everything else. Anyway, yeah, the Italian Auto router didn't do a very good job, but it can if you set it up and constrained it correctly. But a lot of people for quite a few years now. In fact, I checked my emails.
Someone sent me this in 2019. There's a there's an AI because Ai's all the rage. There's this company called Deep PCB who have an AI Auto router. Now there's nothing new about supposedly Magic Auto Routing.
you know AI technology. It's been going back many, many decades. Uh, Neural Net routing was all the rage in the 80s and 90s and stuff like that. And there was.
They had fuzzy logic and they had all sorts of right. Every manufacturer had their own you know, trademarked flavor of Auto router that's sort of like you know, pretended to be machine learning Ai and all that sort of stuff. but of course AR these days it's pretty Advanced as I've done videos on so I thought I'd try this So this deep PCB people have been asking me to give this a go. so I thought I would take the exact same file out of this that I use for this example so that will have a comparison with my route in with the Altium routing and with the Altium Auto routing and then try this. AI it's still beta version 1.000 So I I don't know what they've been doing for like four years or whatever. but anyway, um, some people say it's pretty good. others go. Yeah, now of course the big thing with Uh Auto routing.
one of the big things is placement. As I've said before, PCB layout is like 90 placement. get in the placement right now. Um, as I've already placed these components, what I'm going to do is take the existing board file.
I'm just going to unroute it so it's reasonably good. placement done by a verified human and I thought we'd just load that in there and see if it can order out Pure AI powered Cloud Native PCB Routing Design: Complex boards in less than 24 hours So I can't do this. Live Live Right now I think I have to like wait a day and it's free for two layer uh boards I believe because it's still in beta or whatever. upload your board.
Um, our fully automated, no human in the loop Cloud platform gets to work as soon as it receives your DSN PCB file. In no time or wasted distributed machine learning Now Beta Release covers the following features: Routing for printer circuit boards up to 150 pairs two layers is currently available I'm not sure what the 150 pairs is. is that like controlled impedance pairs? Um, like I'm not sure what the deal is. Um, all our solar boards are DRC checked fully compatible with Keycad? Uh, leverages Innovative Insta-deep AI Technology combined with powerful Gpus and Cloud infrastructure to offer the first pure.
AI PCB Routing engine results in 24 hours, not weeks or months. Uh, save time accelerates well. If it's a two layer board like, it's not going to take you weeks or months to lay out the board usually unless it's an absolutely massive board. your questions answered I you are you doing this? Madeline I Would bet you have Humans in the loop actually deep PCB is fully automated tour currently in beta automation possible thanks for reinforcement learning and AI technology particularly suited to decision making problems.
Board games such as Chess or Go Logistics Mobility or PCB Routing insta-deep so there you go anyway. I will I'll link it in down below Apparently there's no humans in the loop, so apparently all it supports is a DOT DSN file. So that's an Orcad Spectra file. So unfortunately my Altium license and yes, I know the irony of this.
It has expired. and yeah, they won't give me another one just for like, educational use or whatever. So anyway, I can't export my original Uh project here. but thank you very much.
A couple of people I asked on Twitter A couple of people did that, they unrouted it for me, saved it as a DSN file and well, we're going to give it a bowl. All right. So let's drag over. open the DSN file, submit and apparently it will display like a preview image or something. Price free Thank you. A number of layers, a number of air wires 238, number of pins 363. no I can't even get bored. information here: I can't get anything anyway.
Um, the I presume it'll do the full thing I hope it's within the constraints. Um, it's not a huge board so let's confirm. Okay, free. Uh, done.
or to play successfully my board so I can view it presumably board processing. Okay, um, I'll get back to you. it's it's running, it's doing its business. Will it? actually? Will this be like a live update? That'd be really cool.
And of course you don't want to sit there watching the AI do it? Do you? Uh, you know you just want to like simply Nemo when it's done, please routing from empty board. Yeah, it knows there's no uh traces there. I Haven't told it anything about ground planes or anything like that. Whether or not it actually is smart enough to like whack down a ground plane under all that, not that it's needed I'm something like this.
it's just. you know it. it's just a done thing. It's the vibe.
Think it's This is a Nixie tube project that you know written, you know, a reasonably high voltage DC stuff. So clearance is important. So once again, all that clearance sort of stuff would be part of your manual. uh, constraints.
When you're we're set up constraints for your auto. Where are you? Auto Router There you go. Completed in four minutes. What's for? No, it's still running I Was excited there.
It looks like it's done 117. Yeah, there you go. It changed from 117. Oh look, look it.
it's doing it. It's doing it live. I Thought this would be like a 24-hour thing. Look, it's got these.
Uh yeah yeah, you can see my cursor there. It's got like non right angle tracers coming out. that's just a No-No Okay, that's with the Um Altium Auto router did. but I don't know if this is like an intermediate route or it's a final route or anything like that.
Now look I'm going to forgive anything to do with like, uh, clearance, uh, constraints for uh, for the high voltage. you know, uh, part of it. So I'm you know I'm just going to ignore that. But look at this.
look at these traces up here. Okay, it's just it's just why why not continue that across there and go down? So let's go to my board over here and it was obvious that this like common, right high voltage. uh anode. I think it is like a line.
that's a common thing for all of those displays so any human right would would know like you kind of route that first right? a It's a high voltage line so you like snake it around the outside so it's got a clearance from everything else and it goes to these uh, dropper resistors down in here like this. Okay, and then it's common to all of them. So like that was one of the first things that I routed as a human right and and then it's no, but this thing right? it's got It's dropped. via's down there, right? Oh no. Oh, this is no. this is terrible. Muriel Whether or not I assume it's a rip up retry like uh, algorithm as it's called. that's the industry term for when if you get sort of like locked in.
oh, it's going to change. When you get like locked in. um sort of like constrained in you might have to rip up some traces and it can't figure out. Anyway, inside has to rip up some traces and then retry again.
Yeah, it is. Definitely rip up and retry. It definitely changed. I was just explaining how it put a Veer in there and uh and jumped over that uh, that high voltage line over there and it's just ripped up that and completely redone it.
See, But it doesn't know because we haven't told it. But yeah, this is what these are. the things humans know, right? And the and the AI can't possibly know that because it doesn't have any schematic information. It's got no knowledge of what the circuit's doing or anything like that.
It's just routing, so you know once again, this is where. like, you could, you could utilize a tool like this and you could partially route a board. Like all the important stuff you could route. and then the unimportant stuff, you just let the auto router rip.
And as a professional PCB designer, that's what I've done countless times before. It's like, oh, I've got all this digital crap and I've got a route. You know, a thousand traces and like. and they're just like low frequency stuff.
It doesn't matter, they're just all you know or whatever. or you're routing. like big buses or something like that. and um, and you can just let it do all that.
sort of like unimportant, um, stuff. But the important stuff you've got to do yourself and set as a priority. So I I guess I can't blame it I I Can't blame it for not getting that, but it's doing this live. This is pretty cool.
It doesn't look like it's going to be any better than the Altium Auto router which is the one up the top here. But yeah, once again, like all the traces up the top here, it's done a similar sort of thing. like it's just no, yeah, it's just like taking them through the middle of the pins instead of like taking it right around the outside. And that's what a human would have done.
that's why mine looks much cleaner and and then the top side of mine just looks cleaner. It just looks like a more human result. Anyway, we can actually like, just accept current solution. I Guess if you're happy with All right, stop I'm happy with that.
Don't rip up anymore once again. I I Didn't expect this to work. It's exactly It produced the exact or it's producing the exact result I expected. which is kind of like just all over the shop.
It just looks like it's Auto routed. Nah, Ai's not helping here once again. like keep out layers like I didn't do any keep out layers anyway. you can see I've turned the wires off there and you can see that it shows the net. There we go and you can see how you know. like the placement is ideal for this. Like each uh segment has its own chip or or two uh chips actually. Uh, which? go to those two.
Those two. those two like you know, so nothing's not much is like crisscrossing across the board and stuff like that. There are a few things like there's a few common control lines down here. It's a big uh shift register.
If you don't know, it's basically you know data in and data out and just uh drivers for the high voltage Nixie uh common uh drain uh drivers for the high voltage Nixie displays here and uh Bob's your uncle but uh, unless it's absolutely uncritical and you really don't care. Um, you know, but have some pride in your PCB layout please From an old school PCB designer I Have some pride and don't do like have traces coming out of it. Look at this. look at this one coming back in and just going and it's coming along and then it's going and going up at this angle.
No no Oh God Look at this one up here with this big jaggy like that. oh this is terrible. It's having a red hot go but hey yeah yeah no that's it. Our situation has not improved.
This is saying traces goes up and away. Oh look at this. it's going all the way this this. Trace Here it's going around here.
it's dropping down a Via There it's going on the red layer up the top. Hopefully you can see that my curse is bigger on the screen and then this goes around around here and then all the way back down to here like it's going all the way up around this side of the board and then all the way back. Now it's it's gotten it's got rid of that. Now this is its rip up retry thing and obviously realize that was dumb.
This this jaggy on the bottom still down here and I just yeah yeah and uh I'm just gonna have to wait. Now you can see up the uh top right here. This is actually Solution 16. So is it going through like 16 complete? No, it wouldn't be complete in like 16 16 rip up retry efforts because some I know some tracers like this jaggy over here.
This is not changed in several iterations now. So yeah, um, it looks like it's just it's going through all these Solutions how it comes upon the best one I don't the AI determines what is the best one? Anyway, let's let's uh, cast final judgment over the final board once it's done. but we have actually verified that there's no point sitting here watching it do its business. Although, you can get a good idea that uh oh no I should have constrained it better.
For example, I don't want to do in that sort of stuff. No I forgot to constrain it here and here and you know, I probably should have put my ground plane down first. I should have laid down these high uh priority traces first and things like that. and yeah, you can get an idea of where it's uh, goofing up. So this is actually a really good interface. I Like it. Okay, it's been going two hours and nine minutes Uh now and it's still only up to a hundred. and whoa.
Oh oh look, you can actually choose a solution. Oh that's pretty cool. Just discovered that so it's routed 196 out of 238 connections. so it's it's creeping towards there.
and yeah, yeah, nah, it's still a year now, All right. I left it running overnight. It looks like it's locked up. It hasn't gone anywhere.
It's exactly the same as it was you saw in the previous uh clip. Solution 48 at two hours and nine minutes, it's just it's just died there. so it looks like it actually died before I Actually uh, recorded the previous clip because I came back at night and recorded that one and it I think it had already Frozen Ah, it's a fail. I've emailed support.
see what happens? Well, it's 24 hours later and it did. actually. well it says it succeeded here. It took one day and three minutes.
So I guess they're you know, pretty spot on with their 24-hour claim. but uh yeah. Look, the solution here is uh, sorry, you can't see it. It's only gotten down to 48 there.
It's exactly where we were before and it's only routed 196 of 238 connections now. I haven't heard back from Deep uh PCB about what happened here, but obviously from throws on that 48th solution there and it just it looks like it's ever refused to go any further because I don't know it backed itself into a corner or something or whether or not we reach the limitation of how many Nets it can do or whatever. But anyway. um, we've got the results here.
and yeah, nah, it's crap. All right. So what I've got here is both the top layer and the bottom layer. Uh, with the order deep PCB Auto rounded up the top and my manual.
uh, route down the bottom. So this is uh, and this is Dave CAD versus AI up here and no, no, no, no, no, no, no. this thing has no clue how to route a double-sided PCB one of the golden rules for or even multi-layer Uh, Pcbs. Uh, it could take any of the old school boards I've shown this before.
Let me get it again. Okay, here's an old school. PCB Okay with through-hole parts, but stick with me. Okay, you'll notice all the traces on the bottom there.
Most of them are going left and right horizontal like that. and if you have a look at the top they're going. they'll actually be going vertical. I Know it's not.
you know, clear in there. but that's how you route aboard because you don't want to be caught. The first thing when you're laying out a double-sided board like this generally is that you don't want things going. Piggly Okay, this way, this way you know crisscross because then you just run out of room completely and you have to add a ton of beers.
Now look at my board here right? What I tried to do is I tried to Route almost everything as a first pass on the top layer or the bottom layer just on one layer. and in this particular case, I made sure that the two chips associated with the display are placed near so that and in fact I even rotated the part so that they were in the correct orientation. You can watch them. I've done a full layout video of how I laid out this board. You can see all these traces feed in nicely into here and you'll notice it's the same thing here. It's the same thing here. same thing here here, right? Every display like this. This is the what a human would do.
It prioritizes, uh, groupings of components. That's another layout technique here. Place components near each other that are supposed to be part of the group and this is how your your Synergy with your schematic. Uh, when you draw your schematic parts that are supposed to be close together, you make sure they're physically close together on the board as well.
and then you're going to have the shortest route in path like this. But you'll notice that they you know. Look, they've done a couple of lines here which are prioritized, but look at how. look at how this was easily routed like this: I just manually routed all these around and it was the I look.
They're almost lined up perfectly to Route them through like that and they just they just haven't done that right. Oh look, they managed to get one which went all the way over there, but that's like no and look in the end, it actually didn't route these nets at all. You'll notice that these were part of the Nets that were un routed and like that was my first priority when I routed this board is to make sure those segments are there and then I could do the digi. You know, the um, interconnections between the chips like later.
maybe on like a horizontal layer like this, but they've just gone higgledy-piggledy everywhere. Okay, and next up, just just look at some of the stupid stuff they've done here. This is just nutso, right? Go in here. Look what is this right? What is this? Why have we got two V like like why is that going uh like and what is this going around to there just routed into there for goodness sake, right? It's an absolute joke.
Another one just come in 45 degrees out of here. It's just that's just stupid. No you drop that straight down to there and having these just like go into the corner and that no, you have it coming out and then you Branch up like that out in your branch over like that. Oh come on.
seriously and what the heck's going on over here? Look, it's like that's going into there. It's like they haven't treated it as like the one physical net like they've just, you know, treated as different routing pasta like it's just terrible. Absolute Amateur Hour right? I No, it's got no idea how to prioritize. and Route A double A simple double-sided PCB And look at the bottom here.
Right as a human laying out a board, right? You try and like get it all on the one layer and then the stuff that you can't put on the one layer you would put on the bottom because then you'll have them. You'll be able to maximize your ground planes, You'll be able to get let your power around and you know stuff like that so you notice. I Don't actually have that many like traces on here. really. it's just actually joining some chips together and going horizontal. Um, like that basically. And as you saw all my top layer is just all vertical stuff. mostly vertical stuff.
You know there's a couple of horizontal here that joins some of the uh like the common uh, clock pin or whatever it was uh down here. but you know that's about it. I Mean this thing has no clue. To be fair though, of course it.
It didn't know our intention, but even then some of the stuff it's doing is just. it's just awful. Look at this right? it's just jagging off there. Why not just go like that, right? It's just end.
this just branching off there like that. It's just, no, absolutely awful. And it knows right that all these resistors should be connected like this. Okay, and this is what I've done on my board down here.
You can see I've just joined them all together as a priority. not because it doesn't know, it's a high voltage trace and all that sort of stuff. right and then like it comes down here like this and then branches up and then whoop Wiggles back like that, down and around and it's like what and then they've like caught themselves in a corner here. no, no, no, no, no, no, no.
and there's more weirdness happening around here and here. What's going on there? That's just dumb. oh okay, and then this going up like that and then down, oh, it's just. it's awful.
Now to be fair, the Altium one. I Guess this is not a huge amount better. There's some you know stuff around here. you'll have to go watch my, uh, previous, uh video.
It's got some reverse jaggies in there and you know, some weird Branch offs and maybe a little overlap in there. Sorry I can't uh, zoom in on this. Don't have the original file to hand? Um, it's yeah, but no. like it is.
Certainly it's It's no better than a traditional Auto router like this. It's just no. so yeah. I'm very disappointed by this.
AI it is just really dumb. Auto router stuff. There's nothing AI about it at all. There's no like it.
like it learns from its own routes. That's why it doesn't like multiple times or something. I would have like expected multiple times for it to figure out. Oh look, there's some Association some you know grouping Associates some tied Association over here and if I go into the individual traces we might actually be able to see.
um you know so like some issues where like it's just plain only has no clue what it's doing, it hasn't learned I it looks like it hasn't learned a thing so I can't see any advantage to using this over. you know a traditional Auto router which is um, based on decades and Decades of a refinement for various algorithms and you can set it up with uh, tell it different algorithms, how many layer board you want do you want? You know a horizontal, vertical, uh, priority and you know all sorts of different algorithms you can actually select in. You know any good? Uh Auto router. You could actually select an optimal algorithm algorithm for it. but yeah, this thing's just it's it's terrible Muriel that's that's a complete and epic fail. and did it? Just give up and couldn't complete some of the routes even though they're like really obvious stuff. like you know going through here like this I mean like it's just you know it's they're taking one Trace through there and that just chopped off everything else right? And like a human's just going to know this stuff even if the human Auto router. and this is quite common in the industry actually where you just have a PCB layout person, they didn't design the schematic.
they've got no idea they might even have no idea what the thing does. I know a lot of superb old school PCB designers who come from a drafting background. they've got no idea about Electronics but they can see when you have the rat's nests they can see the groupings up here like this like oh so deep. PCB AI Huge thumbs down I Expected better.
It's it. No, no. I wouldn't bother using it. Yeah, just go back to your traditional Auto router and just let it loose on very specific requirements that you got.
You know, a board like this? You you take some pride in your work. Lay it out yourself and it can be. You know, quite nice like this with minimal you know number of uh traces on the bottom. uh.
go in here goody. Piggly and then you've got priority to your high voltage. Your higher priority traces like the high voltage one and other you know stuff you can set up as a priority. but anyway, no I I Expected better than that for a machine.
Learning System Um, that's supposed to learn from all those mold 48 passes. It did and it still couldn't figure out priority of stuff. So yeah, very poor. Anyway, if you found that video interesting, please give it a big thumbs up.
As always, discuss down below and over on the EV blog forum and catch me on all my alternative channels: I'm on the Odysseys I'm on the utreons I'm on the bit shoots I'm on the rumbles I'm on the other no no I think I stopped being on Dailymotion anyway I'm everywhere, even on the Raiavb blog.com Old school. You can get the RSS podcast feed of this thing which comes video comes directly from my own server. Been doing that forever. Catch you next time.
i love get in to making my own pcb is thear any free tool to make them i have a one time projet i like do for a phone line powerinjetor
Here we are… Cheering up for the thing to fail so our jobs are not threatened too soon 😂😂😂
My preferred primary routing software is Common Sense 2023.
Without knowing why it stopped you are kind of just yelling at the sky here. It tries to optimize for the shortest circuit path so for each iteration it should get closer so looking at an incomplete result here isn't giving you much of information.
Wait… so David's Altium license expired? LOL sorry KiCad fan boys go craze now. Let's switch to it outright!
As I'm watching I think how all this information is being routed through the internet.
This is the part I always wanted to see in a Terminator movie. Robots making robots 🙂
Dave, you're missing the main point of the "AI" here!
Adding "AI" to any random system automatically construes a 4000% boost to the number of clueless venture capitalists who'll throw money at you hoping for ROI, regardless of if it actually benefits the user.
When a PCB has a 10+ layer PCB, are those likely to be auto routed or is a human routing boards like that?
Interesting to see that the 'AI' autorouter is no better than the Altium autorouter. It actually looks worse than the Altium autorouter! Great video Dave!
Dave raves about AI in adjacent fields, but in his field it's a "NO!" I actually see it a lot when folks are impressed with what AI does in the area they lack expertise (or even basics).
I've got a random router.
At least robots can't create new type of robots. For now…
Dave, there can't possibly EVER be a good autorouter until at schematic entry level we specify a LOT MORE properties for every net and for every component. Like voltage rating and current rating for every segment of a trace, and so on! I won't go into more detail as of now, but I think it is very obvious what I mean. As I am trying to pursue that route myself. Will see how it comes out!
To produce better results model would consume way more computing resources than would cost to hire someone to route board manually, but maybe someday, in a galaxy far away..
This is my hypothesis. The company probably expects the users to submit smaller designs. Then by letting the customers chose the best implementation out of many they might be using the information to implement RLHF. Even the information that one of the routings might be 0.1% better than the others could be amplified during reinforcement learning.
I've so far been less than impressed when it comes to most of these "professional" AI solutions. I tried integrating ChatGPT into my workflow at work, and it needs an incredible amount of hand holding in order to generate code that even somewhat resembles a complete solution. They seem to have trouble with 'breadth' and are only able to focus on tiny details. A.k.a a single function, or a single error in a cmake script, or in your case, a single trace.
The truth is, auto routers can be good, they are the lifeblood of the ASIC industry. In the PCB industry, the money just isn't there when compared to the ASIC industry. There was an article on chiplet designs where they mentioned they use PCB design tools to do most of the work for connecting chiplets together but really have to create custom automated tooling because the pin count is getting too high. The likes of Synopsys and Cadence has all the know how to do PCB automation tooling, it just doesn't make sense for them to.
do they route solar roadways too?
After 40+ years of doing PCB layout, I still do routes manually. Auto routing can be useful for fine tuning component placement and as a sanity check. A point to point one-net-at-a-time auto route could be useful.
In the mid-nineties, we used a $25,000 auto router that required incredible amounts of setup work to get routes. Then lots of work to clean up those routes for final gerber output.
AI has made the equivalent of the New York subway system on your board. XD
The thing that kills me is that it took over 2 hours to fail to route this extremely simple board.
Auto router is supposed to be significantly faster than me. Otherwise whats the point
LOL I started reading through your video section names and burst out laughing. I guess I'll put the dream back on the shelf for another few years then.
Muriel! Its terribruell!
Welcome to Non-Polynomial algorithms! It difficult to infer priorities based on examples of already routed boards. Without hints and priorities, these types of algorithms can go off the rails and never comeback. I suspect that is what happened. The iterations are likely cause due to backtracking when an iteration yields a non-optimal solution. E.g. it got TRONed into a corner.