EEVblog 1512 - Why Bypass Your PCB Like THIS?
ฝัง
- เผยแพร่เมื่อ 31 ต.ค. 2022
- Why add bypass capacitors on a PCB like THIS?
Answering a Twitter question:
/ 1586889993234038786
Are bypass capacitors really needed? • EEVblog #1081 - Are By...
Bypass capacitors visualized: • EEVblog #1085 - Bypass...
Bypass capacitor tutorial: • EEVblog #859 - Bypass ...
www.vgamuseum.info/index.php/...
docs.xilinx.com/v/u/en-US/wp4...
Forum: www.eevblog.com/forum/blog/ee...
Support the EEVblog on:
Patreon: / eevblog
Odysee: odysee.com/@eevblog:7
Web Site: www.eevblog.com
EEVblog2: / eevblog2
EEVdiscover: / eevdiscover
AliExpress Affiliate: s.click.aliexpress.com/e/c2LRpe8g
Buy anything through that link and Dave gets a commission at no cost to you.
Donate With Bitcoin & Other Crypto Currencies!
www.eevblog.com/crypto-currency/
T-Shirts: teespring.com/stores/eevblog
#ElectronicsCreators #Bypassing #pcb - วิทยาศาสตร์และเทคโนโลยี
I’ve been watching your videos Dave for 8 years. Ever since I was a young Marine working as an aviation electrician. Through the knowledge you taught me and applying that knowledge in the field I became Marine of the Year and I was able to pass the screening for a EE R&D job while grossing 6 figures without any college education. So I’m working and studying biophysics living the dream in my new home. Thank you Dave for being a crucial pillar in my professional development.
Nice to hear, thanks for sharing.
Well done you! And for crediting Dave. Somewhat hard-boiled though Dave sometimes seems, I have to think he gets a lot of satisfaction out of comments like this! By the way, have you also seen channels like Robert Feranec, FesZ, SDG Electronics and Phil's Lab? All thoroughly educational in different sectors of EE.
@@Graham_Wideman any of those for RF?
@@Graham_Wideman Yes I have Graham! All great and super informative channels.
@@TimPaddy "RF?" Yes and no. Yes in the sense that high-frequency digital is at RF frequencies. And in the sense of wireless and radio, I recall several videos across those channels on incorporating wifi, bluetooth and so on, including PCB and chip antennas, and low power wireless telemetry. But I don't recall for example design of radio transceiver gear for audio or video.
"2004 is almost 20 years ago"
Shit sure didn't feel that way
we're all going to die sooner than we expect!
This card was made by Integral Technologies as a video capture card for 16 or 32 channels of analog CCTV feeds. It would have connected to an external input box via the large connector on the card. A lot of the people that worked for them now work at Exacqvision.
a large box or a large pigtail??
Thanks for the info, it's definitely interesting. Another question is which of these caps are filters etc. in analog circuits, and which are there for DC rail filtering.
@@KeritechElectronics Or which are there to couple split planes together so that signals can cross the gap without the electrons flying off. I think Dave way underestimates the sophistication of the ASIC when presuming the power scheme for this board.
@@abitofabitofabit4404 "electrons flying off" LOL, I love that phrase!
Well, knowing that it's for cctv changes EVERYTHING imo. Throwing every possible bypass cap at it means it will be more reliable, particularly in hostile environments. And maybe helps to mitigate some sort of an RF attack from someone trying to disable it.
Not sure how expensive these boards would've been back in the day, but if they were pricey enough, they may have wanted to leave clearance around the BGA chip for manual rework. Like if you've got $100 worth of chips on the board and the only thing stopping it from working was a dodgy BGA joint, it could be worth a manual reflow, or possibly even removing the chip reballing and resoldering. That would save you from eating a $100 loss (plus the margin of that board).
The extra space around the larger BGA chip might have been provided for access for an inspection microscope. Back in the early days of using BGA packages some assemblers used a microscope that had a prism on the end which could look under the BGA package to check for unsoldered, deformed or missing solder balls.
I figured space for traces. If you cram the passives too close, there's no room left for routing.
EDIT: Actually, now that I'm thinking more about it, it's probably just adhering to a grid, there wasn't enough room for the ASIC if you cram in one more row, and so they just put the IC right in the middle of the void.
@@stephenbell9257 Cool
this is very likely to have been really expensive as VMware and Virtual Box both emulate an S3 card like this by default.
As someone designing a product with a high end modern processor, I WISH the manufacturer had supplied us with a 50 page document on power timing. We're lucky to have a couple pages with 2-3 diagrams.
What are you designing for. Intel has always delivered lots of docs. Nxp has been ok, I've never done an AMD processor so maybe one of their parts?
I’ve also had some bad experiences. Learned my lesson and technical documentation is now a selection criteria point
@@enginerdy Nice! Some DataSheets are amazing, and others are garbage. Right now I'm in software, and it's the same. Except StackOverflow means community is extremely important. It's why Open Source dominates. I don't care if the company can afford whatever commercial product that's "better", if I can't find the answers I need to do my job it's worthless.
Lying in bed 2am in the morning in Germany. Can’t sleep. Enjoying your videos.
Sleep is more important.
Heh, one night many years ago I couldn't sleep. I finally got pissed off enough that I got up to mess around on the computer. I ended up finding and following a tutorial for EAGLE PCB designer. I am glad for that sleepless night as that was my introduction into PCB design.
so what u wearin? ;)
Great explanation Dave! As a repair technician these types of videos do massively help to understand why things work the way they do. Keep it up bud
I dig the new lecture style setup
I would assume from all the I/O it's probably a broadcast grade card and if it's like my broadcast audio cards it has a lot of extra shielding and noise reduction. Working in studios that share close proximity with the transmitters noise and interference issues a constant. I haven't quite seen it quite like that but I wouldn't be surprised if I did.
Thanks for the board tour! The pen/tablet seemed to work well!
I love the concept of using a drawing pad overlay on the circuit board. Once you get it all figured out, it will be a really great tool.
like the format! tanks for the video as always!
It's a pretty full layout. So well done to the designer for getting it all to fit top side. The crystal routing is a bit 'how ya doin?' though :D New setup looks great Dave.
Where do you see the crystal?
@@666Tomato666 i assume they are looking at the silver package at the top of the image, near the center
@@666Tomato666 There are five crystals on this board. It looks like nearly every chip has it's own clock. They are the black plastic devices with four pins.
@@royhenderson4085 the three ones clumped together in the middle, and one on the top?
I'm sceptical, usually quartz oscillators have just two contacts... can't make up the etched text so can't say they definitely aren't...
Kudos to whomever for squishing all those caps on one side though.
Must have been satisfying when the last track snapped in place and they zoomed out to admire the rows and columns.
Perhaps all those caps were put there to reduce RFI into the analog video inputs keeping in mind that not everyone correctly terminated their video feeds. The other thing is those caps would have been pretty cheap and it does make the board look cool. Given the choice potential customers are likely to go for the board that looks full.
Hard to say without looking at the schematic, but if there are a lot of channels of video with parallel busses, depending on the format and bit depth, you could have a lot of simultaneous switching currents. Especially if the channels are genlocked. With a mixed signal board like this, you'd want to keep things like ground bounce to a minimum. I could see wanting a lot of caps on board, although not sure that many...
I could tell a lot more by looking closely at the board and component selection, but I'll guess it was a brand-new start-up that hired a young design engineer right out of school or with limited experience and put them in charge of the board design without an experienced supervisor. It clearly has a bunch of I/O, so in order to cover their butt the engineer told the layout person to decouple absolutely everything. With the quantity buying power they got from all of the caps on this board they may not have even lost much money on the deal. There are often back-stories on board designs like this one. Also, it looks like the company may be a home-monitor and control company, and I've seen some boards from such companies that were definitely belt and suspender designs, practically aerospace grade, and priced accordingly.
That layout is pretty satisfying to look at. wrt the format, in some shots Dave's noisy background kinda mixes in with the noisy board layout. I wonder if floating Dave head would work better here?
Probably didn't read all of the manufacturers decoupling guidelines. Maybe read the required number/type of caps but not how to arrange them. Happens a lot when the circuit designer and the board layout designer don't talk to each other.
A couple thoughts on this :
a) The system designer started out the schematic capture phase by dedicating one page to caps (decoupling and bulk) and just cut/paste a metric tonne of decoupling caps on it... maybe a post-pub edit session?
Since all the challenging stuff happens on the OTHER schematic pages (and who brings up the cap page???) the over-abundance slipped through.
b) They used some rather new-for-the-time chip and were unsure how it would behave, so they over-populated the board with decoupling caps.... since you CAN leave some off and suffer little penalty (at those frequencies and transition currents) they were going to test how much difference they'd have between the over-populated board vs one with 3-4 caps per BGA.
c) Are all the small yellow-bodes the caps, and the grey-bodies resistors? If so, much of what we are looking at may belong to some series termination scheme. I'm not a video designer, but it is (or was) a fairly common thing to do, and if they have a bit of an odd stackup for trace impedance (e.g. their systems engineer is a repurposed software guy who isn't the most pcb-cognizant of people) they might have mad to do a little catchup impedance matching.
d) This could be an example of off-site/off-country PCB design where the wording of the pcb spec was a little ambiguous, and the pcb-house wasn't too keen on asking questions... they just follow a spec.
e) some smart-child wanted to see what happens to a nice quite set of power planes edge-to-edge inductance if you blow it full of holes.
f) Is this driving a standard display family? This isn't for some old high-voltage plasma display, is it?
g) In review, I'm probably more in the camp with the folks who say this boards got multiple power rails because of a mixed-signal function. And they want to decouple away power-ground spikes caused by some noisy WIDE logic bank (> 64 simulanteous bit transitions) and a higher-impedance analog side (subject to pickup spikes). .. and an overzealous noise-budget engineer.
PCI cards that have two notches in the connector can accept both 3.3 and 5V, though most most desktop mainboards have 5V slots (indicated by the notch away from the bracket)
It is an analog input board. Having a row of bypass caps surround digital components could be an attempt at "fencing in" digital noise by providing a very low impedance path between ground and power planes around and between the chips.
A thing of beauty and a joy forever.
I could not tell, but do you think there were a couple of different cap values per pin for some of the parts (0.1 and 0.01uF for example)? That PCI bus to that one IC is running at 33mhz and probably faster internally so decoupling over a few frequency ranges might have been needed.
The layout designer used to design suburban cities
Love the Back to the Future reference!
most likely that super savage has multiple power domains and they may be staggered. so they will be alternating 5v and 3v3 decoupling caps. A lot of the other capacitors have resistor chains nect to them. look at the 25 pin connector, you see traces running from one resistor to the next. those are most likely impedance matching / low pass filters for tha analog i/o (video signal). the same goes aroudn the focus chip and between the focus and savage chip. those are most lilely matching networks / lowpass for video
I see another reason for the grid. When you have to untangle circuits that have many components it helps to set the snapping grid pretty coarse. That will speed things up.
i love the new video style!
Likely they took all the effort to keep capture quality clean. Any ripple in the video signal encoding. I’ve seen less decoupling in high end video equipment lol. They just went all in for this one. “If there is any interference it’s not my fault”
Maybe you can comment on the PCB design that IBM used in their 370 systems in the 70's and 80's?
That also looked very interesting...
Ah, good old days, where the Notebook Chips had the memory on chip or, in this case, in the package.
Seems like that was a Later Logo of S3...
I bet the decision for single sided was cost motivated due to the (likely pricey) FPGA up top. The designer must have had a tough time with managing to lay out this board to fit all the bypass caps on top, there's an awful lot of 0-ohm jumpers on it! Not sure how many layers are on that PCB but I bet there'd have been a lot of pressure to reduce that for cost reasons as well.
Thanks Dave,
it would be good to see those bypass capacitor computer model programs in action.
I think they are all very high end hugely expensive packages. Not aware of any free or low end ones.
@@EEVblog - OK thanks Dave,
I was wondering if Altium could do it?
The long tracks to the caps have inductance that is bad at microwave frequencies. e.g.
An FPGA working at 250 MHz would have big problems with that layout.
Look up Robert Feranec on youtube, he has some videos showing them.
It is a bit OTT from an electronics point of view, but it is a work of art Dave, if i had that i would have it in a glass case on the wall, alongside my 4MB Western Digital 5 1/4" HD
always nice videos
Surely a lot of those capacitors are for coupling and filtering? I see nearly as many resistors as capacitors; looks to me like there's a lot of bandpass business in there.
That's what I was thinking. If this captures 16 analog inputs, all the caps near the large IO connector would simply be analog coupling/band pass filtering to the A/D inputs.
If the ground and power layers are "high up" in the PCB stackup (which they should be), then mounting the bypass caps on the top side of the PCB results in lower loop inductance than mounting them on the bottom side. Also for lowest loop inductance, the power and ground vias should be placed to the side of the cap, close next to each other, rather than hanging off the ends.
Probably not, but looks a lot like the led wall capture/interface cards i used near 2010-2012. Ground, EMI interference is quite common in the use case scenarios, so everything is over engineered for that. The VGA captured a crop from the pc video card and translated it to the wall protocol.
Would the part placement not have been automatic for something like this? (I.e. not just the routing)
Nice presentation. Which tablet is that?
The fact it was a PCI card gives away it was not from the 1980s. PCI did not come around until 1992. I rarely saw it in PC's here in Australia until around 1994.
1:45 "It's a bit younger than I expected, I thought this was late '80s or something". The original PCI 1.0 specification was first released in June 1992, so it couldn't have been any older than that.
Non-symmetrical vias on passive pads is not only a problem because of ugliness, but mostly because of asymmetrical solder cooling, which can lead to tombstone effect on small packages.
That is a nice orderly pcb design
i have seen boards entirely covered in bypass caps in every empty spot that just went to the ground plane
i suppose if you have the room and a need to decouple large fast current gulps for fet drivers or something ... caps are cheap enough
Must've been a bored board layout intern that wrote a script to place bypass caps wherever there was room on a grid pattern. Maybe they were also trying to see how many 0.1uf caps it took to trip the power supply with the inrush current on power up... 394 - no, not enough, 398, no, still not enough - I've got a fever, and the only prescription is --more cowbell-- more bypass caps
Hey could the space bar be the pan tool?
"You can see down here in the bottom right that's entirely covered by my face cam....."
I have a question regarding the looks of modern vs older PCBs. I have noticed that many modern PCBs completely stopped using silkscreens. One well known example would be apple mainboards or modern GPUs. This is in contrast to a lot of PCBs in old devices - just like the one you have reviewed in this video. Now my question is: What made the change to frequently opt out of using silkscreen? My guess is, that the higher density of modern designs would simply not have the space for silkscreens or limitations in the printing machine's accuracy prohibit the use if silkscreen. Please enlighten me, Dave!
It’s because they don’t want people repairing their devices, they want you to throw it away and buy a new one, why help the repair tech with useful information.
they could just leave the silkscreen blank though
What tablet was that you were using Dave?
Are you sure about it being best to put decoupling caps on the other side of the chip? I remember from a Rick Hartley video, that decoupling on the same side with power/gnd planes closeby is even better, since you don't go all the way to the other side of the PCB with a via.
What's closer, "All they way" through a 1.6mm board or 2 to 3 cm on the same side?
A video on split grounds would be great if you felt like it 😉
It's very heplful content! Like!
That's one hell of a board lol
I dont really understand the question. How else would you do it?
"I thought it was a card from the 80's"
> shows card with PCI bus
ಠ_ಠ
Have a great day, love the vids.
BEAUTY
What software are you using to let you do the live drawing/annotation? Need something like that!
He mentions it in the video: Drawboard PDF: www.drawboard.com/pdf
DrawBoard PDF, mentioned in the video.
@@EEVblog Great, thanks! I missed it 🤦🤦
I dont know about the Timeline of when they where available, but this looks symmetric enough to come from an autoplacer.
15:38 is the feducial really on the silk screen or is it just hasl-ed over?
and he notices it right after damit
That FPGA cost $20 on its own back in 2000, but they couldn't spring for double-sided loadout 😂
I think it looks kind of good, gives a "serious bit of kit" look like they were not messing around
How does this beast stack up against the 4090?
0:32 - holy crap, I know Ian! haha, neat!
The audio in this sounded quite weirdly compressed, as in low bitrate.
4:20 I DONT THINK I CAN EXCUSE THE CRUDITY OF THE MODEL
why so many different pin sizes? does it have to do with current handling? or is it purely for inductive and impotence reasons? does having the staggard pin array with an uneven number of randomly sized pins help train the new ai overlords to hate us all for not using an autorouted design?
Is there an experiment to show a bypass cap in action?
Ah ha, I think you’ve already done the video. I’ll go and watch that
Hello Dave, this is an analog board and 90% of the caps are used for RC filters :)
my guess was transmission line RC terminators
The tablet setup is working peachy, but the instructor is a little wonky! ;)
The bypass caps wouldn't even raise a question for me, guessing the reasons is easy. But why so many 0 resistors? That's a multi-layer board, it wouldn't have been that hard to route most lines without a jumper bridge, would it?
You kinda missed the point on the neat rows of discrete components… it’s for the “pick and place machine, efficiency” !
It minimizes the pick and place head rotations, faster placement of repetitive placements for mass production 👍🏻
Not really. The physical arrangement doesn't really impact the speed. The head has to go back and forth regardless. Rotation, yes maybe if the head need to slow down as the part rotates, but that's not a positional thing.
how many of all the resistors are 0 ohm :o
The RF effects of that board give me a headache just trying to imagine how I'd simulate it. They've almost created a metamaterial.
My guess is that someone designed the circuit assuming it will be a double sided load. Then the PCB designer was like "Eh, I can pull it off with a single sided load".
It's nice
Huh, that's weird; S3 Graphics was sold to Via in 1999 and they make terrible 00's GPU (early Savage & Virge) but the Super Savage & Pro Savage were only sold to OEM & Notebooks (Thinkpad once had this one too IIRC) so it's probably made by 3rd party contractors or some sort for... mil-spec requirement? either way that bypass seems really overkill for.. general usage.
Some broadcasting companies like Audioscience, Digigram, Lynx or RME do has a pretty similar "neat smd grid" board layout for their internal card, but not to this kind of "neat"-ness.
And no, i haven't heard of "Integral Technologies" either.
s3 is why virtualbox and vmware really become useless if you need any kind of virtualized hardware acceleration, even if we have the power to emulate something much much better.
Splitting hairs here.. but some polishing could "include": Less Jolting Zooming (if the software allows) and a solid background behind you to help "pop" you out of the the busy background image..... Just saying. Thanks.
I think I like lab as a background better, this one gives me university lecture vibes.
putting your caps on the bottom will not always give you better performance, if you have a tightly coupled plane pair close to the populated side of the board having your caps somewhat far away on the same side can be better than having long inductive vias, this is not what is happening with this board but i wanted to mention it anyway
Putting caps on the bottom usually ends up having way more impedance than caps on the same layer as the component, even if they're inches away. Rick Hartley has great videos on this concept.
@@asid61 it depends. We are talking about a bga package here. You can get a cap much closer to a pin in the middle of a part if its on the underside.
Ive seen that video by Rick, so i feel you. It just depends on your stackup i guess. I generally strive to have all decoupling on the top layer too, but it wont always be the best solution.
@@jeremyglover5541 still doesn't really matter, unless your BGA is more than 4" per side. The charts in the Rick Hartley videos about power show it well - impedance to the cap is way higher when the signal has to cross ground planes.
Somebody had fun
DB-25 and DE-15 ?
So.... it was laid out by the intern, is what you're saying?
Maybe!
Bean counter: "how much can we save only running board through smt machine 1 time? "
I had a few Intel motherboards, and they were totally chaotic in component arrangement.
someone spilled the box of birdseed on this one!
I'm starting to wonder if Bruce Dickinson moonlights as a PCB designer
Outwardly visible signs of dysfunctional management, and a labor crew that just doesn't have any 🤠s left to give.
I make everything in a grid except for one part that's laid out, intentionally, slightly skewed. I'm silly.
Those "old school" board designers likely had to put in every trace by hand and made excessive use of the copy & paste feature.
I'd love to see Robert Feranec review this board with Eric Bogatin or Rick Hartley for a Grey Beard chewing over :-)
You really need access to the board files to get a good look at what's going on.
@@EEVblog Yes, fair enough
Drink every time they'd say "ground bounce" or "loop inductance"
@@scottpelletier1370 Hahaha
@2:08 Super Suhvoj ?!? @ lol. its super SAVAGE.
Adderall shortage due to Y2K?
Dave I think your wrong about all those caps are bypass on the that board. I am willing to bet most of those caps & accompanying resistors are filtering for all the signal lines.
Maybe one of the engineers was getting paid by component count? 😁
Looks like a cemetery for electrons 😯😂
The decoupling is so good, simply touching the board will cause your girlfriend to break up with you.
The first suck of the Savage :)
11:00 No push-back? I reckon it was a bet.
You forgot one reason... either it was Friday and they wanted to get it done and get out of there or it was Monday and they had plenty of time to kill.
No? Oh you have a spice model? No? Cap cap cap cap cap cap......