A Noise-Free DIY Switching Power Supply - How Hard Can It Be?
ฝัง
- เผยแพร่เมื่อ 8 พ.ค. 2024
- Switch Mode Power Supplies (SMPSs) need a printed circuit board (PCB), and James was wondering how hard it could be to design his own. Turns out, it can be pretty challenging! He does a deep dive into different options for this power supply design. Hint: You need copper zones to get a low-noise SMPS. bit.ly/3CN6zhv
Free routing resource: freerouting.org/
Engage with the element14 presents team on the element14 Community - suggest builds, find project files and behind the scenes video: bit.ly/3tmdewv
Visit the element14 Community for more great activities and free hardware: bit.ly/3q6YMpu
Tech Spotlights: bit.ly/3qPrDhM
RoadTest and Reviews: bit.ly/3pV5Bux
Project14: bit.ly/31wbnJY
#0:00 Welcome to element14 presents
#0:12 Overview
#1:06 Attempt 1: Breadboard
#1:47 Attempt 2: Auto Router
#4:36 Attempt 3: 6 mil Traces
#7:03 Attempt 4: 6 mil Trace ... With GND
#8:03 Attempt 5: Copper Pours FTW!
#10:03 Give your Feedback
#SMPS #powersupply #PCBdesign #SMPSdesign #PCB - วิทยาศาสตร์และเทคโนโลยี
This is an unbelievably great video. The early mistakes and making incremental changes to show what each change does.. Nice. For new EEs this is invaluable. Great video!
Great video. I love the way you show how there’s more to the project than just connecting the pins.
Pours are the means to an end. The end is low impedances between nodes in the power path where switching currents flow *and* small loop areas. The best-performing layout can still be improved. Ultimately, with a 2-sided board, it’s possible to close the current loops with loop area projecting sideways through the board, and it’s possible to get the on and off state current loops at right angles to each other. For some hints as to how that may work, look into multiphase converter app notes where they really emphasize small loop areas on the layout. Components on two sides of the board can make a big difference. The resulting volume is still the same or better.
Great Video - You explain stuff in such an easy to understand way James! Love your work!
Helpful to see the comparison between each successive improvement, particularly when the 2nd gen looked to be worse. Good job
It would be really interesting to see what difference a 4 layer PCB would make, with the ground plane shifted up to layer 1 (just below the top layer). Rick Hartley often makes the point that the bottom layer of a 1.6mm PCB is just too far away to be truly effective, and that there's almost never a good time to use 2 layer PCBs for EMC management. The prepreg separating the top two layers would typically be less than 0.2mm thick.
why not 10 layers?
I quite like those PCBs with clear solder mask. They're helpful for seeing traces, but they also look great.
Just imagine them installed into a Japanese copper-coated case. That’s must be something!
Call me old fashioned, but i don't
Green or Brown for me
Blue is acceptable i don't like Clear , White, Pink is fucking ridiculous and is why i have a personal issue with JTAGULATOR
Red is a maybe
But Green or Brown is what i like to see
@@martinkuliza green for reverse engineering; white, black and pink for the looks
@@king_james_official
i can live with black at times but never white or pink (not even for looks)
@@martinkuliza i loooove my white pc motherboard :)
Great video. Switch mode design is a specialty all it's own. Infact i know someone who has spent his whole career designing switch mode and he is at the top of his profession and still struggles with emi problems. He's almost had to dedicate most of his ongoing learning to that. Hes told me about testing the design in different attitudes, temps, humidity; controlled and relative. It's definitely an artform.
Another great video and this was very informative. Thank you.
Now this is a VERY good video for explaining how this works. So many new makers makes these mistakes on power supplies. I have done it myself. Several times.
This video randomly pop up on my recommendation, but im very grateful it pop out since im always confused why all smps design use pour instead of trace and how to design one, you answer all the questions i have about smps, thank you, gonna subscribe to this channel
Great video and thanks for the reference files
That was very educational! Thank you for this!
Incredible. Thank you.
Excellent lesson, thanks.
Amazing, as an ameteur I learned tons of stuff in 10 minutes. Thank you.
If there's a suggested layout in the datasheet, it's there because reasons. The manufacturer did the hard work for you.
I'd still probably add a bit of LC filtering on the output to clean up the remaining noise.
Conclusion = Introduction to Power integrity
Very nice video thanks for the educational content
Great explanations, thank you.
Great detail! Similar, if not the same I'm thinking generally, for RF designs... with more details to consider the higher in frequency you go I'm thinking since every trace is like an antenna with LCR characteristics to consider.
Really great structure to this video. I almost shouted at the screen at your first attempt, then I realized what you were doing. ;) Excellent tutorial method. A board I would have liked to see is one that uses the final parts placement, but rather than fills, using VERY LARGE traces. Like 5 mm, and following the same "route" as the final copper pour method. I suspect it would perform somewhere in the middle, but just how bad? I also suspect that may be another path a beginner might take, not knowing about or understanding how to do copper pours.
Good job, explained well without the usual hype
Impressive improvement
Thanks for sharing.
Very Helpful
This is gold, any more PCB design tips videos comming soon?
such nice video, congrats!
Great video!
amazing video, thank you very much
i have this same issue and you just saved my board!
I've been designing power supplies for 30 years .. WELL DONE ! Too many scammers wiring SMPSs on perfboard.. that won't ever work unless you are switching at 60 hz haha
great video!
I once built a flyback converter on a breadboard. It worked. Until the output capacitor became loose. Then the USB unplug sound sounded a bunch of times as USB connections half a meter away were killed due to interference.
But other than that, if you try to minimize parasitic inductance, resistance and capacitance in the critical sections, it should be fine to test a SMPS design on a breadboard before making a permanent PCB. Just make sure the connections are solid.
It would be interesting to actually see a breadboard version of the circuit from this video.
Great video, one of the best i have ever seen on this subject ! but If I can't trust my dmm, it would have been interesting to compare the results with a good one, Fluke 289 or UT181A multimeter to.
awesome video, simply said..
James is in the zone today.
Nice video, good pacing! I usually need to speed it up, but this was good at 1x. It felt aimed at noobs a little too much here and there. This was a missed chance to explicitly say "never, ever use autorouting".
I'd like to see how a 4 layer board compares. I use SIG/GND/GND/SIG with power routed on 1 and 4. I use JLCPCB's stackup that has 0.0994mm prepreg between 1&2 and 3&4 with a thick 1.265mm core, since it doesn't cost more. It'd be interesting to see how that compares to 2 layers and to the typical SIG/GND/PWR/GND. I use 4 layers on all my boards, as the cost isn't much more and I think JLCPCB does a better job than they do on their $2 2 layer boards.
FWIW, I use 0.3mm/12mil traces only where I need small traces, otherwise I use 0.6mm/24mil everywhere and 0.8-1mm/32-39mil for routing power. It's doesn't really make routing any harder, so I don't see any point to using smaller traces unless there is really no room.
Wow ... this is amazing. Like and subscribe. I was searching for a power supply for a DIY amplifier that doesn't make any noise nor interference ...
Great video and project! 🙂
Designing an amp[s] capable filter for the output of this would be another nice challenge... simple low pass, pi filter, how much current a simple op amp design could provide
Very interesting
I've designed switchers successfully. It's a PITA. The board you show is still likely to have EMC problems. You really want a 4 layer board, with high currents or fast-switching pulses separated from their return paths by only the thickness of the prepreg, not all the FR-4. But (inless you're Rick Hartley), you can seemingly do everything right and still struggle to tame the things.
Why not a 8 layer or 20 layer board? Or just don't build anything at all, just order it pre-made in China like everyone does. Man, the DIY scene is completely dead.
To be honest I think using pours for power and ground should be the default. Especially ground. There's no cost to using as much copper as you can!
I agree. Separating signal types and return paths is something I spend alot of time on but still have trouble with. I have yet to see or read it explained in a way saying this is how it is for a given usecase. Look at audio, mix up the return path on an elaborate circuit and a simple oversight turns into a tail chase where in the end you want to go mirc on yourself with a trout to the face. Haha. Great comment man, thanks.
4 PCB design in under 11min? wow!
subscribed
How do you best get started with designing your own pcbs and circuits?
1:20 COMMON! You could have *at least tried* We want to know what happens if you do it anyway!
My prediction is it would """work""" but not very well. Be noisy AF, and possibly unstable with horrible ringing and stuff around the switching node.
Now I'm curious... could you link up a large ground plane, like a separate PCB, to an existing board and get an improvement?
Sure, yes, but it would be tedious. In the case of retrofits, I would strategically add ground bus wires in parallel to the high-current and high-frequency tracks on the PCB.
The layout is equally important. It's critical to reduce the area of the switch node, and also reduce the area of the feedback node, in order to achieve good EMI performance.
Lots of ground vias could also be added near the controller IC in order to provide better heatsinking.
Wow, this is really scary, I never expected SMPS designs requires extremely careful PCB designs as well for best efficiencies. I was assuming just learning the math and SMPS theory will make great high efficient SMPS devices, well this video just showed its not the case since proper PCB deign also needs to be considered which is also complex. Excellent video in explaining to get best efficiencies and why it's crucial to properly design the PCB layouts and implementations for the SMPS design.
You can make it either efficient or have low EMI. To make it efficient you have to drive it on and off hard, and that causes massive voltage transients that ring hard and long. You have a spice simulation or something already right? Throw a 10nH inductor right before the diodes to represent some parasitic inductance, and watch it go nuts.
@@douggale5962 Thanks for the technical reply. No I have never done any spice simulations. I'll start learning about them later. What do you mean that to get high efficiency, switching the voltage needs to be driven "hard"?
@@ShopperPlug Each time you switch a MOSFET or transistor, there is a huge pulse of heat and power dissipation when it is crossing the partially conducting region (between the points where it is all the way on or off). You would like to slam the MOSFET on and off instantly if you could, but you can't. You can control how fast it switches by how much current you drive in and out of the gate when it switches. The gate of a high current MOSFET feels like a capacitor to the gate driver circuit. You have to charge and discharge the gate of the mosfet rapidly. To make it rapid, you need to apply and drain out large pulses of current, like 500mA for tens of nanoseconds. This is what I mean by "driving it hard", you are putting huge pulses of current in and out of the gate.
Sounds great right? You just slam it on and of really really hard and profit, right? No, the problem now, is when you slam the MOSFET off really hard, all that current flowing through the inductor has to go somewhere, and it can't right away. There is inductance between the inductor output and the diodes and output capacitors. When the MOSFET turns off, all that current can't go anywhere, briefly, because the current isn't flowing through the diodes yet. This causes a huge brief voltage spike. Now you are threatening to blow the mosfet from exceeding its voltage rating.
You have to balance how hard you slam the MOSFET off and on, with how low you can get your board layout inductances.
@@ShopperPlug Here's an analogy that might help. If you instantly switch the mosfet on and off, it is like slamming a car into another gear without the clutch. There is a gigantic spike of force. Switching the mosfet slowly is like smoothly engaging a clutch. The clutches burn off energy as heat but they make it really soft and smooth. High efficiency power supplies are doing the moral equivalent of neutral drops (or popping the clutch), at a high rate of speed.
@@douggale5962 Thanks for the detailed explanation. Makes sense now. It would be great if you can share your vast knowledge in SMPS, maybe make an entire TH-cam video series about. Thanks.
Great one as long as it's DC/DC, AC/DC on the other hand will take a hole lot more than just layout best practices, specially if you bring up the inductive spike subject. I enjoyed this one and looking forward for more 👍
So in summary make traces as short and as thick as possible, and also make a ground plane, I think this is a good advice for any PCB, but with high frequencies is a must. Does long and thin traces work as small inductors in high frecuency circuits?
Yes. Those are also called Antennas. :)
Larger traces act as small capacitors at high frequency too. This is how microstrip filters work is using large and small traces to act as capacitors and inductors
@@Cracked1ce Only if they have a low impedance return path. Otherwise, they’re inductors.
@@bald_engineer Good point. and low impedance means more than just DC resistance at high frequency. Managing return paths, proper drive strength, correct termination and trace impedance are the most important considerations when dealing with high frequency. Jeronimo's comment about making the traces as thick as possible being important for high frequency is not necessarily true. For high frequency transmission traces you should use a trace thickness that matches the target trace impedance. And in SMPS design, your switching node should be as small as possible while still being large enough to pass the needed current. The plane you have in your design on the switching node is just adding additional capacitance to the switching node and coupling more noise into the ground plane. Also with it being close to the edge of the board it is going to radiate more noise into the environment. However, these are just small nitpicks and the video did a good job of demonstrating the importance of a good ground plane and proper layout.
Ok, so lets discuss how a Surface power supply would go about eliminating switching noise. Looking at the attached drawing, I've captured a generic switching regulator on the left, and have it's output [ Vin ] feed into an EZ81 vacuum tube rectifier [ pins 1 and 7 ]. Output of the EZ81, nice clean DC, as shown [ Vout ]. How is possible ? Let me disclose what is not seen here.
What is not seen here, the vacuum gap between plates and cathode of the tube. The DC voltage will forward bias the rectifier, thus plate current will flow. However, the switching noise does not jump to the cathode, due to the free-air-space between plate(s) and cathode.
What does it cost ? First, you need to supply the filaments 6.3 volts, at about 1 amp. Also, in this configuration, series plate resistance on the order of 180/2 ohms. So there is some insertion loss to factor in. However, if the application does not require current sourcing up to 1 amp, other tubes could used, and you would save on filament power. Also, some care is required to set this up, with the dual inputs, or you can red-plate these tubes. I would also recommend some type of simple BITE circuits to keep an eye on tube health and performance. You set this up right, these tube last almost forever. . . . . ..... ... . . ......
Side notes. I show output filter cap at 100uF. The data sheet for the EZ-81 states 50 uF max. So, why do I run output filter cap at 100uF. It goes back to understanding how data sheets are written. You see, the EZ-81 come out in 1956. At that time, the biggest filter capacitors back then, were these wax and paper rolled jobs, max value was 47uF. So, this is just a guess, but I would say they were saying this tube could handle the biggest cap available. The higher capacity electrolytics did not come out until the mid 1970's. Secondly, there is no issue with surge current for two reasons. 1. This tube rated for 0.5 amp continuous, per plate. 2. It takes about 12 seconds for this tube to warm up, and gently ramp up to full conduction. I even ran this tube feeding a few 1000uF caps over a few days. It did not care. Why? Because a vacuum tube is not an active electronic device. It is a mechanical device, with - certain - electrical - properties. It does not generate any waste heat internal to the device, because is a vacuum between plate and cathode.
. . . . .... . .. .... . . .... . . . ... . ...
"autorouted (lol)"
"baldengineer"
i chortled
Good introductory video. Thank you.
Viewers should be aware that designing an efficient and quiet SMPS, which passes EMC testing, is quite a bit more involved and nuanced than what is shown in this brief video.
Sorry, how can you create PCB? Do you use a website? Thanks
nice... nice...
Very good video with very little views.
What if I use a breadboard BUT shove the critical pins together in the same holes to avoid capacitive coupling?
Just solder components together as close as possible avoiding breadboard at all.
For ground part of the circuit loop use thick solid copper wire.
It will work at up to some point in kHz range.
How much inductance did you end up with at the switch node? I almost did a project like this, but I was deterred by FUD about the transients blowing the mosfet. I did some simulations and, you'll blow up an 80V mosfet with 40nH of parasitic inductance at 4A. How huge were the transients, really, at the drain of the mosfet? I have doubts that you probed it where it would have shown the true transient when the mosfet turns off.
They make many switch mode controllers with on chip switch FETs do you don't have to add one and they run several amps and use 20uH+ all the time without issue. The range on inductors I've been see are between 2-30uH depending on switching frequency used, max current etc.
"you'll blow up an 80V mosfet with 40nH of parasitic inductance at 4A" your simulations are wrong
Second failed board is a classic example of Murphy's law; you should be expected not to be too surprised, apparently.
one of the biggest keys to a decent layout of a switcher, is to visualize the size and paths for the current flow. A 3 amp switcher often has nodes with 10 A plus spikes
It would have been even better if you had taken the step to optimize the PCB layout to get the best performance out of it.
Damn… and her I thought Kicad was pronounced “keycad” for the longest time 😮
That's the pronunciation the creator uses but to each their own
In the old days i designed SMPS from literally components. Just getting to the point of just making an output measurment was a big deal. The battle started by just trying to keep the switching FETs from blowing up. LOL dont ever want to do that again thank goodness for modern day and fun video BTW glad i stopped.
Hi. I need to drop 1.2V @1A. Doing this linearly is really easy. But the P-CH mosfet gets too hot to touch and I don't like that. Also it is not very efficient. All these SMPS buck converters need at least 2.5-3.0 V above target voltage to work. They can't regulate for example 6.2 down to 5 volts. Do you know of a solution for this? I also don't want to go the buck-boost route cause that's inefficient as well. Is there a very low voltage drop out buck converter?
Plenty of modern switchers handle such a low drop just fine. TI's WEBENCH is a great starting part.
Parallel two mosfets with low ohm ballast resistors to get better current sharing
1.2V x 1A is 1.2W, not really all that bad. 7805 is rated for 1A and drops at least 2V, often more. You just need to sink that heat.
@@bald_engineer Is it possible that TI Webench doesn't tell the whole truth? One solution it gives me is with TPS54202. But its datasheet shows that on the high side it uses an N-CH mosfet. Any solution in its datasheet says drop 8-24v down to 5. But Webench says you can drop 5.8v down to 5v. Datasheet says as long as I maintain a 2.1 Volts above the switching node. Of course cause it's a high side N-Channel. But where do I find that voltage? Charge pump? I bought 10 TPS54202 from china cause they are out of stock in North America. So far used two of them and nothing works. I wonder if chips from china are nothing but licorice tic-tacs with lazer prints on top of them.
@@ivolol yes true but I am working on a switch mode solution and I don't see one out there. They don't seem to make p-ch smps IC's. That's why they don't work when IN voltage is close to Out voltage?
Not bad but you stopped one step too soon: you didn't optimize the component placement to shrink the hot loops. You can also optimize trace placement so that the feedback signal isn't picking up noise from the inductor, and so that filter capacitors aren't sitting out at the end of stub traces, and so on.
where can I find the rest?
The color black is what? If not the solder mask
Clear solder mask, black substrate.
@@bald_engineer Got it thanks...those boards look like something Ferrari would use. Top-notch
10:03 how hard is it to DESIGN a SMPS PCB ?
Well, doing the layout is one thing. Choosing the components and their values is another one.
A lot of the datasheets have tables with recommended component values for common outputs plus formulas to figure out anything else. The manufacturers really do want to sell their parts do they try to make it as easy for you as possible.
Yes yes yes plz from Total scach house hold items?
It's also possible to integrate KiCad with freeroute without ever quitting kicad, no need for a 3rd party tool.
after dark
When is thicker traces or using couper pours a problem due to their higher capacitance?
I'm going to channel my inner Rick Hartley and ask you the question: where is the energy in a PCB? Answer: it is in the dielectric space. When current flows an electromagnetic field is formed. That field spreads through the dielectric space looking for a return path of the lowest inductance. If that return path isn't ground/power then it is another signal. In the analog audio world that might present as oscillation or crosstalk. Once you getting into high-ish speed switching all bets are off. Also, the manufacturer's datasheet is as good as toilet paper in a lot of cases. There are layout and stackup suggestions that are fundamentally wrong, but it just happened to work in their application. Unknowing designers use those suggestions as gospel and perpetuate bad design practices.
As Rick would say, "All datasheets are wrong, until proven otherwise!"
Noise-free? I think you forgot the bit where you add a ferrite bead pi filter to really drive the output ripple down ;)
Excuse me sir i want to ask, there is only R1 that without a value, can you tell me how much the proper value for R1 is?
It is clearly explained in the datasheet.
@@bald_engineer thank you sir
*"...noise free..."*
I thought that about my X wife so I doubt this is any different
do _not_ autoroute your spouse
hello sir. i dont get it..
this board is smps
220v Ac to 5vDC ?
because i need it so bad..thanks.
0:43 15V DC down to 5V DC
How much power and what size?
If you do not need a lot of power you can go with a stepdown transformer, a filtering step and a linear regulator.
It is really the best option for most cases.
how hard..? lol what could go wrong?
i think kicad v6 does have autorouting now
No, it does not.
why is this still on my wall ?
nice.... but we needed far better, so went to linear power supply
Good effort but yet another video that has absolutely no clue on how to correctly meassure a dc converter/use an oscilloscope.
You did not give any indication of the target noise that you would need to attain to pass the CE tests.
Now make full pc SMPS of 2000watts
bro i hate auto router
its supposed to be "key-cad" - thats the right way to say it, right ?
Audio too quiet