It's incredible how cheap it is to get custom PCBs made at prototype scale these days, even ones with advanced features. It feels like we're living in a golden era of custom manufacturing.
JLCPCB is often the manufacturer of choice with DIY PCBs, and the reputation is well deserved. I recommend them as well, have never had an issue with what they've made. It's a bit pricier for a one-off board given the MOC, so it's preferable to do some sort of group buy if you can.
OK, so people are ordering a custom PCB from China for a one-off? Whatever happened to making them at home? I remember my brothers doing this in the bathroom sink with a board coated with copper¹, some special markers to draw the circuit on the board and chemicals to wash off the extra copper. I’m guessing they might have used our dad’s electric drill to put some holes in the board as well, but looking at the article, this doesn’t seem like an order from China need.
⸻
1. Some details may be incorrect as I was 10, this was the late 70s and I only saw the results, not the process.
I still do mine at home, though I mostly only build one-offs with older style thru-hole components where tolerances are not as sensitive.
Usually I can get away with using a simple Sharpie marker mask, though I've also had good luck in the past with the toner transfer method for more intricate designs:
That method works for creating single sided PCBs with through-hole components or rough SMD parts. Double sided PCBs are more effort but still possible.
But everything more complex than that - multiple layers, blind or buried vias, fine-pitch SMD, anything involving clock frequencies higher than say 50-100 MHz or depending on specific impedances or capacitances, or requiring stuff like vias to groundplanes to dissipate heat in copper? No way even for a dedicated hobbyist. And you need special equipment to make a proper stencil if you can't use a soldering iron due to feature size or because it's impossible otherwise (=BGA parts).
And where it gets really spicy is sandgrain sized SMD parts (0201 and smaller). Even for professional repair shops, that size requires experienced experts...
I have a small CNC mill and can carve out a board if I don't have time to wait for ordered boards. It's a pain to set up to get it to cut properly, the boards are either one sided or with no vias, and the lack of solder mask makes it harder to solder, especially for SMD components that rely on the solder blob being a certain shape. Additionally, I find that the traces tend to peel off a little too easily, though that may be down to the Ea-Nasir quality copper clad boards I'm using.
It is pretty fast though, and worth learning if you need to do prototyping quickly. You can have a finished board in just hours instead of a week.
I've been trying to make a PCB but it's just so annoying having to sift through pages and pages of listings in order to find each individual tiny component without minimum order quantities of hundreds or more. The process of finding each component is so tedious that I gave up. (I was using EasyEDA hoping it would be easy to find components I could actually order -- nope)
Wow, very very similar to my first PCB, a breakout board for the TSL4531. I ordered it from OshPark back in 2016. Used the exact same soldering iron, and a Quick hot air station, though it's the less fancy one than you have.
From there on I've made various PCBs, and thankfully they've pretty much all worked on the first try. I do take a lot of extra time to verify the design and double check all the datasheets.
Nice board, looking forward to seeing your next one. A little feedback, helpful I hope:
A "via" is basically just a wire stuck right through the whole board top-to-bottom. (Okay, it's not really a wire. They drill a hole and then chemically grow a layer of copper that fills the hole.) You can connect the copper pattern on any layer(s) to that "wire", so it's often used to route a signal from one layer to the other. But you already have a real wire going through the board: the GND header pin! So no via is necessary; just connect directly to that pin on each layer (kicad probably did this for you automatically). This trick works with all thru-hole pins.
For RF or high-current applications, sometimes you cover a board with a grid of vias, just making redundant connections between the planes all over the place, "stitching" them together. But careful, add too many vias and the PCB shop will bill you extra.
Putting a GND pour on the top layer is a good idea. It's lower-impedance than individual skinny traces, and takes less/zero effort to route. The GND trace you manually routed isn't necessary; you can see by the transparent-red shape that kicad already has copper there. However, you ended up with a little "island" of dead copper between R2 and C2, which is the real reason you needed the via.
A better approach would be to use the bottom layer for +3.3V power instead of a redundant ground pour. This gets rid of the +3.3V traces (BTW, best to use a single, thicker one instead of 2x side-by-side) and unifies the island into the ground pour. Even though this is a micro-power application, playing the traveling-salesman game with long scraggly power traces is never a good idea. You would still need vias to connect each top-layer +3.3V SMD pad to the bottom-layer power plane, but the signal-integrity benefits of a plane make this worth it. Maybe the absolute best is an uninterrupted GND plane on the bottom and a +3.3V pour on the top.
Putting SMDs on the bottom side makes the board cost more, so good call leaving them all on top, but putting traces on the bottom layer is free. So you can even move signal traces between layers to avoid cutting up your planes too much. It's "fun" with big complex boards, like untying a giant knot...
You are very miserly with your +3.3V global net symbols in the schematic! You can place as many as you want to optimize the schematic's readability. Especially near the CSB pin, the 4-way solder dot looks like some intermediate signal in a voltage divider, but it's actually just +3.3V. Same suggestion with GND -- basically, it's more informative to read "this pin is GND, and this pin is GND" than "these two pins are connected, I wonder what they're doing...oh, it's all GND".
Pull-up resistors are usually oriented vertically, too, so they graphically pull "up"!
Anyway...I hope you don't mind all my constructive criticism. It's nice to see something on HN I know about!
For a simple-enough design, I'd just route the signals as on a 2-layer board. Play around with placement & orientation of parts to minimize overall trace length. If you have some A-B duo with many connections between them: place those first & then put the other parts.
Then fill the remaining space with power planes. GND on one side, most-used supply voltage on the other side.
That way you reduce/avoid the need for blind (or worse: buried) vias.
Another trick: route signals roughly east-west on one side, roughly north-south on the other side, and use a via where you change direction. Many early (complex!) computer mainboards were done that way on 2-layer boards (often using through-hole IC pins or passives as via).
You shouldn't really need 3+ layers unless you get into high density stuff like many-ball BGA parts & the like.
> So no via is necessary; just connect directly to that pin on each layer (kicad probably did this for you automatically). This trick works with all thru-hole pins.
Note that without a plated through hole (vias are small plated through holes, those designed for component legs are bigger) the solder will only adhere to one side. For mounting large components where vibration may be a problem this will make the pad weaker & more likely to lift off. A plated through hole lets the solder flow through & bond to pads on both sides. So you sometimes do need plated through holes.
Otherwise I agree. These days 4-layer boards are so cheap it's often worth just dedicating an entire layer to ground, and one to power. E.g. Signal, ground, power, signal. All your ground connections become vias. Most of your power connections are vias to a lower inside plane. Your signal traces are available to probe & rework easily. You get a bit of distributed capacitance from the power & ground planes.
This guy, with his air quality sensor, should get together with the guy yesterday that built a fan controller. Connected, fans could turn on when really needed. That's the basis of intelligent HVAC.
Did you find out that designing, manufacturing and testing PCBs are very interesting, but also very exhausting. You have quite a lot of repetitive jobs, especially paper work to do.
It's always cool to see people discover KiCad and the amazing world we live in now where a hobbyist can come up with something that can masquerade as a commercial offering.
The transition to SMD and having the "fab" do the assembly was another hurdle for me.
Now I tend to consult LCSC Electronics at the same time I am designing my circuit: checking availability and price of the various components I am proposing to use in my design.
Maybe, but that's a good use of Claude IMO. How else would one learn--by googling? Claude does the annoying aggregation of data and distills into a list.
I can tell you where Claude wouldn't have been much use: if it didn't fire right up due to a short or voltage drop that might occur during reset propagating through circuit. You're on your own at that point and gotta break out the multimeter.
25 comments
[ 8.7 ms ] story [ 45.2 ms ] thread⸻
1. Some details may be incorrect as I was 10, this was the late 70s and I only saw the results, not the process.
Usually I can get away with using a simple Sharpie marker mask, though I've also had good luck in the past with the toner transfer method for more intricate designs:
https://www.thompdale.com/pcb_etch/gotee_board_production/Ea...
I've got a cat and live in a carpeted apartment. I prefer to keep most projects "dry" for that reason.
That method works for creating single sided PCBs with through-hole components or rough SMD parts. Double sided PCBs are more effort but still possible.
But everything more complex than that - multiple layers, blind or buried vias, fine-pitch SMD, anything involving clock frequencies higher than say 50-100 MHz or depending on specific impedances or capacitances, or requiring stuff like vias to groundplanes to dissipate heat in copper? No way even for a dedicated hobbyist. And you need special equipment to make a proper stencil if you can't use a soldering iron due to feature size or because it's impossible otherwise (=BGA parts).
And where it gets really spicy is sandgrain sized SMD parts (0201 and smaller). Even for professional repair shops, that size requires experienced experts...
It is pretty fast though, and worth learning if you need to do prototyping quickly. You can have a finished board in just hours instead of a week.
https://github.com/bschwind/tsl4531-module
https://imgur.com/ozk7UuS
(sorry for imgur, not sure what people use for image hosts these days)
I then went on to also make a BME280 breakout:
https://github.com/bschwind/bme280-module
From there on I've made various PCBs, and thankfully they've pretty much all worked on the first try. I do take a lot of extra time to verify the design and double check all the datasheets.
A "via" is basically just a wire stuck right through the whole board top-to-bottom. (Okay, it's not really a wire. They drill a hole and then chemically grow a layer of copper that fills the hole.) You can connect the copper pattern on any layer(s) to that "wire", so it's often used to route a signal from one layer to the other. But you already have a real wire going through the board: the GND header pin! So no via is necessary; just connect directly to that pin on each layer (kicad probably did this for you automatically). This trick works with all thru-hole pins.
For RF or high-current applications, sometimes you cover a board with a grid of vias, just making redundant connections between the planes all over the place, "stitching" them together. But careful, add too many vias and the PCB shop will bill you extra.
Putting a GND pour on the top layer is a good idea. It's lower-impedance than individual skinny traces, and takes less/zero effort to route. The GND trace you manually routed isn't necessary; you can see by the transparent-red shape that kicad already has copper there. However, you ended up with a little "island" of dead copper between R2 and C2, which is the real reason you needed the via.
A better approach would be to use the bottom layer for +3.3V power instead of a redundant ground pour. This gets rid of the +3.3V traces (BTW, best to use a single, thicker one instead of 2x side-by-side) and unifies the island into the ground pour. Even though this is a micro-power application, playing the traveling-salesman game with long scraggly power traces is never a good idea. You would still need vias to connect each top-layer +3.3V SMD pad to the bottom-layer power plane, but the signal-integrity benefits of a plane make this worth it. Maybe the absolute best is an uninterrupted GND plane on the bottom and a +3.3V pour on the top.
Putting SMDs on the bottom side makes the board cost more, so good call leaving them all on top, but putting traces on the bottom layer is free. So you can even move signal traces between layers to avoid cutting up your planes too much. It's "fun" with big complex boards, like untying a giant knot...
You are very miserly with your +3.3V global net symbols in the schematic! You can place as many as you want to optimize the schematic's readability. Especially near the CSB pin, the 4-way solder dot looks like some intermediate signal in a voltage divider, but it's actually just +3.3V. Same suggestion with GND -- basically, it's more informative to read "this pin is GND, and this pin is GND" than "these two pins are connected, I wonder what they're doing...oh, it's all GND".
Pull-up resistors are usually oriented vertically, too, so they graphically pull "up"!
Anyway...I hope you don't mind all my constructive criticism. It's nice to see something on HN I know about!
Then fill the remaining space with power planes. GND on one side, most-used supply voltage on the other side.
That way you reduce/avoid the need for blind (or worse: buried) vias.
Another trick: route signals roughly east-west on one side, roughly north-south on the other side, and use a via where you change direction. Many early (complex!) computer mainboards were done that way on 2-layer boards (often using through-hole IC pins or passives as via).
You shouldn't really need 3+ layers unless you get into high density stuff like many-ball BGA parts & the like.
Note that without a plated through hole (vias are small plated through holes, those designed for component legs are bigger) the solder will only adhere to one side. For mounting large components where vibration may be a problem this will make the pad weaker & more likely to lift off. A plated through hole lets the solder flow through & bond to pads on both sides. So you sometimes do need plated through holes.
Otherwise I agree. These days 4-layer boards are so cheap it's often worth just dedicating an entire layer to ground, and one to power. E.g. Signal, ground, power, signal. All your ground connections become vias. Most of your power connections are vias to a lower inside plane. Your signal traces are available to probe & rework easily. You get a bit of distributed capacitance from the power & ground planes.
I wish I could 3d print a board with all its components - instant gratification (which incidentally was one of my half baked projects)
The transition to SMD and having the "fab" do the assembly was another hurdle for me.
Now I tend to consult LCSC Electronics at the same time I am designing my circuit: checking availability and price of the various components I am proposing to use in my design.
I can tell you where Claude wouldn't have been much use: if it didn't fire right up due to a short or voltage drop that might occur during reset propagating through circuit. You're on your own at that point and gotta break out the multimeter.