Both pcbway and jlcpcb will cleanly cut your entire board outline with no panelization tabs. You can draw any arbitrary shape in your edge.cuts and get it in one clean continuous cut.
oshpark leaves rat bites that you have to sand down and clean up if the board outline matters to you. Worse, if you have castellated half holes, the rat bites can land right in the middle of what was supposed to be edge contacts.
I have boards that are small like the size of a dip28, where the entire length of the two long sides are castellated, and they are edge contacts to fit in a socket, not for soldering, and the entire board snaps into a carrier so essentially all 4 sides need to be clean and match the drawing. I have to sand and clean up oshpark boards, and live with a little bit of defect in the castellated holes. I don't have to do anything with jlc or pcbway boards. They arrive perfectly clean edged and ready to use.
I have one board that is only about 1cm square, but not square, it has a complicated outline, and both jlc and pcbway just pop it out with no fuss the same as any other order for the same nothing cost. I don't know how they even hold on to the tiny board to cut the outline. It does appear to be traditionally routered, not laser or water jet.
I guess I shouldn't be so amazed at the size, given how common tiny pcbs are, for example the little page counter chips on my printers toner cartridges are only about 1/3 the size of these, although simple square edged.
Here's a step-by-step video about how JLCPCB makes PCB's. The link goes directly to that step at 21:09: https://youtu.be/ljOoGyCso8s?t=1269 It looks like in this case they are using a CNC router to make the cuts, and a vacuum table to keep the PCB's in place (it's a table with a grid of tiny holes that has vacuum inside to suck the PCB's onto the table).
I don't know exactly why but I can't watch that guy for more than a few seconds. But those machines are sure cool. I'm criticizing him, or myself, not you just to be clear.
I've orderdered several times from oshpark because over the last 5 or so years I've iterated the design slightly many times. From oshpark I have received different results at different times from the same board outline. One time was a single rat bite in the middle of the short ends, leaving the castellations fully clear. Another time was two rat bites on each end at the corners, but on the short sides, again leaving the long sides clear. But most of the time the tabs are placed randomly other than being spaced apart.
Even the buggered ones were usable, it's just that one was usable after cleaning up, and still slightly uncosmetic even after cleaning up because you can only sand down not fill in, while the other was not only usable but fully cosmetic right out of the box.
I give all my boards nice rounded corners now just because apparently it's free so why not.
It's even functional not just cosmetic in this case. Any time a part has to fit into another part, there has to be some relief either on the inside or outside part, they can't both have perfectly sharp inside and outside corners and fit into each other. It's often not possible, and usually not desirable because of stress riser reasons, to manufacture perfectly sharp inside corners, and so you need the outside corner of the inside part to be cut down. Or, if the outside of the inside part must have a sharp corner, then you need to cut out extra relief in the inside corner of the outside part, like adding a round hole centered on the corner, both for stress-riser reasons and just to ensure the part will always be able to fit without interferance.
Like the inside corner on the left edge of this pic:
https://github.com/bkw777/WP-2_IC-Card/blob/master/COVER/WP-...
There's a connector body with square edges that fits into that part, so there's a little extra cutaway right in the inside corner.
In the carrier part above, the drawing for the carrier has simple imaginary perfect inside corners, and in real life the printer doesn't quite make a knife-edge inside corner of course. And the pcb has slightly rounded outside corners. If the fab couldn't do it for me with the router, I'd do it by sanding.
I would say those two particular projects (rex and teeprom) are a special case with needs that most people won't need to worry about, so I'm not saying oshpark is a bad choice. It's just a difference, and the whole point of the article was to compare and show differences.
I've placed probably 20-50 orders through JLC. Very consistently decent quality, and surprisingly good customer service, they manually check every design within a few hours and will email with issues and/or their adjustment.
Tip for smallish boards: "Panel by JLC" can further drop the price. It's often cheaper to buy say 10 units of a 2x2 panel than to buy 40 individual boards.
I was stunned to see they added teflon substrates recently (for RF layouts. I've paid $1000s for teflon fab); very excited to try that out.
Side note: JLC will make PCBs out of sheet aluminum, intended for use as heat sinks for LEDs. Design an aluminum board with nothing but silkscreen and drill holes and you have an incredibly cheap, very precisely made front panel! This is even nicer now that KiCad supports fonts.
I've used them for keyboard plates. They charge a small penalty fee (it was $8 on $60 worth of plates last time) because you're punching a huge number of tiny holes in, but it's still dramatically cheaper ($70 for five versus $100+ for one) than calling out a laser/water jet cutting service.
FYI for anyone interested: PCBWay, PCBgogo & ALLPCB (last I checked a couple of years ago) sent their stuff to the same factory.
It's easy to test since they let you know exactly in which step the manufacturing is in. So you can order a fairly unique combination of solder mask color, board thickness, etc. and see that they all report the board is in the same step, down to the minute.
I've had pretty good experiences with Aisler[1]. They're based in Germany and produce prototype PCBs (2, 4 and 6 layers) ridiculously quickly. For simple 2 layer boards they are often sent in the post within 2 days and arrive the next day. They're also quite cheap [2] and have excellent customer service.
They also do turnkey board assembly these days. Just put in an order for 6 assembled prototype battery management system boards yesterday. Can't recommend them enough.
Maybe I'm making different types of PCBs but they are consistently the most expensive compared to all other fabricators [0] even with shipping cost savings.
According to PCBshopper JLCPCB would make 3x 2 layer 10cm*10cm boards for 31.59€ and delivery in 7 days, while Aisler would make them for 37.20€ (with delivery often within 3 business days). In my experience there's often a 20€-ish customs charge and sometimes additional delays when receiving packages outside the EU, so in the end Aisler ended up being cheaper in my case.
* JLCPCB Green 5 €1.86 total €0.37 each (Minimum order qty 5)
* AISLER Green 3 €37.20 total €12.40 each
Even with shipping & customs (20EUR extra only for JLC), AISLER is still 76% (1.76x) more expensive. Now I do see that for some reason AISLER's shipping is more expensive which is dumb because it's significantly less distance, but it seems like I can't replicate your JLCPCB data; did you further customize it?
32EUR from JLCPCB for priority 7d and 26EUR for standard 7d (not sure what the difference is to be honest). The claimed price from AISLER is 57EUR. Although I have ordered from JLC before I don't have any receipts (and usually can afford slow shipping); to BE/NL/LU/DE it is generally delivered within the claimed timeframe (although 2 working days deliver is not an option).
> Actual example from a 33x55mm board
For 3x 33x55mm it does seem like AISLER comes out on top, probably because they charge by the cm^2, although in the higher order quantities that scaling makes it quite expensive.
I think the price per cm^2 of AISLER is about 5-28x JLC (JLC doesn't charge by cm^2 for <100cm^2, so a lot of variance; for larger boards it's back to about 5x). The shipping costs is (for me) about 0.5x, so when when your cm^2 is high (which for us it often is) it can be more expensive. Also something to note is that the JLC price per board decreases as you go to higher order qty, about x/|_log2(n)_|.
So my takeaway is that for <100cm^2 & >2000mm^2 and medium-high order qty JLC will probably give significant savings, at <2000mm^2 AISLER & <10 boards costs about the same as JLC everything considered (if your shipping costs is lower it will probably be at a higher mm^2 cut-off).
The only other criteria missed in the otherwise excellent article would be commentary on the design rules for those services... If they tell you not to do something and you ask them to do it anyway and the result is 'meh' its hard to say who is in the wrong.
For example if they tell you they do 0.5mm vias and you ask for 0.4mm vias and they ship you 0.5mm vias... who's in the wrong?
But it would be nice if we had some DRC rule format that the PCB vendors could just publish and all PCB programs use.
Or step further and expand it to DRC + any other rules for generating gerbers and rest of the files so it would be just single operation to get the PCB program ready to generate what is needed
There are more things missing. I just commented about the quality of the board outline.
pcbway and jlc both way better than oshpark, no panelization rat bites, just magic perfect unbroken outline matching the drawing all the way around.
It particularly matters to me because I have small boards where most of the outline is castellated edges that oshpark buggers up because they have to stick the panel tabs somewhere. But somehow it's no problem for jlc or pcbway. It's like they have some way of holding the pcb from the center, letting them cut the entire outline in one continuous cut.
I've been learning Kicad and have ordered a few batches of PCBs from JLCPCB in the last few months. Some of my thoughts:
Getting PCBs made is shockingly cheap. For a single PCB for an 8hp module to go into a modular synthesizer, it was about $8 for 5, or about $9 for 10. For 100 it would be around $20. (That's for 2 layers, lead-free HASL.) Included in the price is someone does a manual review of the board and will email you with questions and/or manually do small fixes to the file if they spot something wrong.
ENIG (gold) plating adds a lot to the cost in small runs, but for bigger runs it's maybe less of an issue.
The shipping costs can easily exceed production costs for small runs. If a U.S. company wants to compete with JLPCB for U.S. customers, this is the one area where they could gain a huge advantage.
I just ordered a board that uses their pick-and-place service, which should show up in about a week. Parts are surprisingly cheap. I'm not sure, but it seems like their prices are lower than equivalent parts from Digikey or Mouser. Probably some of this is attributable to being located close to where many of those parts are manufactured. The medium and high-value parts, though, aren't necessarily all made in Shenzhen.
BOM costs add up fast. Designing around low cost parts really pays off.
The existence of something like JLCPCB seems like a huge boon for everyone, especially China for whom it's a major economic strategic asset. It leaves me wondering if all that CHIPS act funding might have been better spent on lowly domestic board manufacturing services rather than high-end fabs. It seems like China is good at recognizing that if you make something that's cheap and good enough you end up with most of the customers while the other advanced economies are off chasing shiny things (which is what they've done with solar panels and batteries).
Kicad + JLCPCB is a powerful combination, but the user experience could be smoother for trying to get Kicad to work properly with their assembly service.
I have used macrofab[1] for several orders over the years and I found them to be great for handling PCB manufacturing and assembling. Highly recommend checking them out.
This is good when you need a few like 10 or 50 boards and you can wait a bit for boards to ship from China.
It would be nice to have a local fast service for when you are in hurry, want to make a prototype or a one off project, or you are just tinkering, send the design and receive the board the next day.
Yes, you can fab the board at home but it takes a bit of skill and time.
Up till about 5 years ago it was cheaper and faster to get a turn out of Asia than the US. I have no idea now. This was true for anything more than 1-2 panels (24"x18") with 4layer 4mil trace/space and no blind/buried vias. You could easily get 4 day turn (2 day hot lot was double).
We really wanted to use local or US manufacture, but it had multiple problems:
1 US worked 9-5 so you had to deliver by 9 whereas China you could deliver at 5pm.
2 US didn't run equipment on weekends, and often better equipment only 1 time a week.
3 In US the equipment wasn't constantly running so calibrations went off and sometimes multiple runs were required to dial it in.
You could order a couple of panels form Asia on Friday at 5pm and have boards available by courier Monday before noon for <$4k. All of that assumed that there were no problems or questions on the gerbers so you needed to do a really good review and know their capabilities. There were places in Vietnam, Thailand, and Shenzhen that did the work. We stuffed (SMT) the PCBs locally, but that was moving to Asia as well since availability of some parts was easier there. I'd worry about counterfeits more now.
That was mostly a concern with the CMs. We had the advantage of owning the Main ASIC FW and the Driver so a serial# challenge response could detect fakes. Difficulty was doing that without implementing public key on a tiny micro and keeping reverse engineering difficult.
We have these in the UK, but they are prohibitively expensive. Like you are looking at £500 for a simple board overnight. For something more complex could easily be multiplies of that.
Even if you are not in the rush, most are stuck in the 90s when it comes to ordering, so before you agree everything, you'll probably get your boards delivered from China at a fraction of the price.
There used to be a great PCB manufacturer in the EU that had very short lead times, although still expensive, it was a good alternative if you were in a rush.
Now often you get your PCBs quicker in the UK from China than from the EU.
It's a shame such (one would think) crucial services are so underdeveloped locally.
I'd say one thing is that equipment is very expensive, the other is that nobody believes in engineering culture anymore here. Engineering has become just another blue collar profession and people's attitude has become hostile - population see an engineer as someone who sits by the computer all day (meaning not actually "working"), makes "a lot" of money and engages in tax avoidance.
> It's a shame such (one would think) crucial services are so underdeveloped locally.
Well, the domestic PCB shops are in a catch-22 local minimum. They don't move enough volume which means they can't optimize and automate every step which then means that they can't move the same volume as China which means ... This one infuriates me as it could be fixed but nobody does.
However ... PCBs have a lot of toxic glop and emissions. How do you contain and dispose of it all? How do you prevent people from breathing it? China doesn't really care all that much about those considerations and that adds quite a bit to prices.
I you are located close to a DHL hub, then the order from China can come in 2 days, which make the international order a moot point as disadvantage. JLCPCB provide 24-hour run on 2-layer board.
PCBway has an option for same-day fabbing and DHL takes about 3 days. That's not too bad for less than $100. The same service from Sierra Circuits or Imagineering would be $1000.
This is also the difference between JLCPCB and many others.
I go to JLC and they show me a price. First thing, in my face. Including the (very reasonable) constraints that I have to meet to get that specific price.
I go to PCBway... price, in my face.
I go to Aisler... "upload your Gerber file here". This is fine for returning hobbyists who already have a file, but if you're considering whether you want to invest the time into the project, not knowing whether it will be $2 or $200 puts a serious damper on even wanting to start.
With hobbyists, this isn't just a deciding factor to get customers choose the competitors, it creates customers: Seeing the JLCPCB prices actually motivated me to learn KiCAD and make my own PCB.
Of those choices, JLC is the only one that makes sense for practical purposes (Along with some others mentioned at the end like PCBWay): OSHPark and (for now, but maybe changing?) Digikey don't do assembly.
I'd love a future where you can order assembled PCBs from somewhere other than Shenzen at a reasonable price. Shenzen will do for now. Maybe Digikey will change this.
Of note, JLC is consistently 10 days (Using the ~$20 DHL shipping) to east coast USA, order to receiving the PCB.
IIRC, JLC has a bit of a first-mover advantage, in that they designed their facilities and systems around the idea that the whole process should be vertically integrated and use as much automation as possible. For the steps that do require manual labor, that labor is very inexpensive compared to other tech hubs.
That's the vertical integration aspect - if they own each step in the process, they can avoid supply shocks and have a more efficient business. It is also a consolidation of power which can create risks for consumers, though. As explained by 30 Rock:
Jack: “Imagine that your favorite corn chip manufacturer also owned the number one diarrhea medication.”
Liz: “That’d be great, because then they could put a little sample of the medicine in each bag.”
Jack: “Keep thinking.”
Liz: “Except then, they might be tempted to make the corn chips give you –”
Jack: “Vertical integration.”
In this case, JLC could decide to steer their customers towards copycat parts which they have a commercial interest in, knowing that the marginal cost of doing your own assembly or switching providers would be fairly high.
I figure there's tyo opportunities to steer design decisions:
* When the schematic is being developed
* When the order picking is being prepped.
The second choice is an easy win for them, but probably limited to near-commodity parts-- I don't care which brand of 330 ohm resistor I use, as long as it fits the footprint. I'm not sure if there's that much margin though-- unless you're swapping out some absurd diamond-plated milspec parts, how much kickback can you fit?
The first choice seems the more lucrative one, but it may be off-limits for a lot of hobbyist and short-run products though. You'd have to tap into the design life cycle very early, and I suspect that's not as widely available as you'd think. A lot of projects still come out of things like datasheet reference implementations, canned designs, or transcribing a prototype breadboard, and these customers would need a lot more than a marginally lower price to resolve the risk factor of swapping parts out.
For Germany I would mention betalayout and eurocircuits. 5-10x more expensive than production in China. But I don’t mind producing my things in Europe.
I would recommend you to check out Aisler. Boards made in Germany, cheap and fast! And pretty good quality as far as I can tell. I haven't noticed any defects in many boards ordered.
Jlc’s preferred (?) circuit making program is EastEDA.com and it has pretty tight integration to parts on lcsc.com. You can of course use any circuit program like kicad. Jlc will assemble boards inexpensively for you if you select the right components (which easyeda will show you if a component can be picked from jlc). I transitioned away from through hole parts to surface mount parts with these tools. pretty awesome that they offer this service for a home hobbyist. They now offer double sided assembly, too.
Or, you can order the components from lcsc and they’ll give you a shipping discount. It’s pretty affordable overall, although prices have increased in the past year.
Pay a little more, get a better board. I've been using Sunstone for many years and the results are consistently good. 6"x6" 4-layer board with a 1-week turn marginal cost $50 per board. Compared to how much time you will spend figuring out that the PCB shop blew out one of your plated holes, it's a bargain.
Sunstone is still exorbitantly priced for what you get. U&I in Korea can do that spec faster for cheaper ($35 each at qty 10, with 5 working day turn, 7 days delivered), and the quality is better to boot.
I really don't see any technical reason to prototype with US fabs anymore. People who insist on doing so are just wasting time and money. I understand not wanting to do to China, I really do... but there are other options than just China.
Both OSH Park (and BatchPCB before it) and OSH Stencil are great! I have not used JLCPCB but I have used PCBWAY and they're pretty good, although it usually takes 2 weeks to get an order assuming no customs release delays which sometimes happens (best used for bulk).
I'm glad to see he calls out thin soldermask. This is the bane of my existence with every single PCB manufacturer--even the expensive ones.
I have had innumerable problems because the soldermask doesn't mask. If you have some big pad (QFN, DPAK, etc.) and you route a line requiring soldermask isolation next to it you can expect to have some percentage of the boards fail. It's infuriating, but at this point I just have to assume that it simply will happen and design around it.
I worked 25 years until 1998 in a prototype PCB shop near Washington DC. Anyone have links to factory tours? I'd love to see current costs of labor, materials and equipment. It's amazing to see such low prices.
I've always used protoboard and hand-wiring for all my home projects. Is it worth the time and effort to layout a PCB and have several fabricated for a one-off project? Do the aesthetics of a PCB win out?
One more factor to consider is time. If you have all the parts already on hand and want to make it now you will probably choose the protoboard. But if you will have wait a week for parts to arrive anyway and the cost of board (for the specific project) is negligible might as well order the pcb in parallel. For a simple board that you could do on protoboard, the layout process on a computer shouldn't take much effort either.
The comment about more easily mounting certain parts applies not only to SMD parts, but can also apply through hole components. You protoboard will probably have 0.1 inch spacing between holes, just like most DIP ic packages, and you can bend resistors and capacitor legs however you want. But there are occasionally some through hole components like buttons, switches or specialized connectors which might have a different requirements. Trying to workaround those can result in a bit janky result. And if you are making something practical you want to use daily, having something more robust and less janky can be nice even for a one off project.
At the end of day it's one more tool in your toolbox. And different people make different kind of hobby project with different needs. I recommend you to try doing it a few times so that you are familiar with the process and overcome the stage where you avoid it just because you have never done it. Afterwards for your future projects you can choose the method that's best suited for specific project.
Being familiar with more tools can also open you doors to new projects that you wouldn't try otherwise.
Hand built boards are great for getting something done right now. It's also a fun sort of challenge sometimes.
Just drawing up a PCB can take as much time as building out the same circuit on perfboard. Manufacturing and shipping obviously adds even more time.
Take this next part with a mountain of salt. Things vary wildly based on circumstance, but in general it's easier to make mistakes on perfboard. They're less resistant to physical damage from handling, impact, vibration. They're harder to repair, especially if any time has passed between design and repair.
You should choose a PCB if your circuit:
- needs to last a long time
- must be repairable by another person
- will be produced more than once
- will be anywhere other than a box that never leaves the bench
- has physical or mechanical constraints from the enclosure
Personally, I find joy in building quick and dirty circuits on perfboard. But I get the same joy from a well-designed PCB.
You don't really have a choice if you wanna use high quality SMD parts, like a SOT-3 and a LQFP 32 or something.
------
That being said, there is a lot you can do with hand wiring and DIP parts. And Microchip still makes PDIP uC like AVR DD, ATtiny, MCP6002, and other important parts for projects.
So PDIP / through hole is still alive, but you gotta set your expectations way low. All the MOSFETs and OpAmps and such are just worse, all good parts these days are seemingly surface mount only.
That is essentially going out of your way. It's like forgetting why you were using THT and perfboard in the first place. Maybe not technically, because sure you could have a stock of common adapter blanks already on hand, allowing you to build right now, which was the point, but it's getting very close to blowing that point if you're using smt parts on adapter pcbs on tht perf boards.
I guess it's still practical while still developing on solderless protoboards.
I'm not sure what you're saying... there is a clear benefit to perfboard and protoboards: they let you prototype instantly, without waiting for PCBs to be manufactured. An earlier critique of that approach was that some parts are no longer available as through-hole. I'm pointing out that there is a simple workaround that still lets you prototype quickly. Generic breakout boards are around $1 a piece and you don't need them custom-tailored to a project.
There are other benefits: A PCB design can be shared so that others can fabricate it themselves. (Or modify it if they wish). A PCB design will likely have a schematic, too, which can be studied.
I'd also bet that PCB projects are more likely to inspire better PCB projects. -- Or at least, the domain I've tried my hand at has been PCBs for custom mechanical keyboards.
If a project calls for 3+ of the same board, then PCBs become worth it quickly, in my opinion.
Even more than aesthetics, I've had a few projects where an unpopulated board outlives any other remnants of a project and so becomes the last remaining document/example of how it was done.
It really depends on what your going to use it for. It's necessary if you want to use surface mount parts with a fine pitch. Or anything with a really wonky footprint. I find it's quite useful for the mechanical properties, being able to size it to the exact enclosure and put mounting holes where needed. And you can print cool art on them for the extra wow factor :) I particularly like OSHpark's after dark board for this, drawing in the copper layer looks fantastic with them.
In general I find it's not worth it for a weekend project but if you are going to be looking at it often and want it to last, definitely worth it.
It's so convenient to draw a perfect board in kicad and have it delivered at about the same time as the parts arrive from digikey, it's all I ever do. I have some protoboards, and have never actually used a single one.
Yes and no. I used protoboard to put my permanent WLED setup together. The complexity of the ESP32 is contained on its own board, the I2S microphone came from Ali in breadboard friendly form, the IR Receiver and LED terminal block were leaded components, etc. Protoboards are a good option for making anything you've built up around an ESP or Arduino on a breadboard more permanent as long is it doesn't have to stand up to rough handling. This covers a lot of ground in the maker community.
"Is it worth" is entirely subjective, I don't think someone who's happy with perf board particularly cares about RF performance or routing complex BGAs against the increased complexity of debugging a multi layer design. In general I would think that if you aren't sure you need a multilayer board, you probably don't.
The more layers, the easier to design IMO. You have to do more QC when placing a via, and there's more layers in general to flip between, but routing puzzles become easier.
I find having access to traces to cut and probe to be a lot more important for hobby projects (that usually don't go through a simulation first). Routing is a very simple exercise below RF. YMMV though, exact application and goal probably matters more for this discussion.
It will save you hands-on time overall for all but the simplest projects, but of course you have to wait a few days for delivery. It may be more or less fun for you, why don't you try it on your next project?
(1) Find someone you can pester with questions. (2) Understand the basic workflow. (3) Start by copying and modifying existing designs. (4) Build solutions using functional modules, and version these. (5) Bench test everything. (6) Be aware of startup and edge conditions when designing. (7) Take notes on all design changes or prospective changes, hypotheses and findings. This process will greatly accelerate your learning. This is easier if you have LXI-capable equipment and can capture test information easily. https://github.com/lxi-tools/lxi-tools
I'm also a hobbyist. It's daunting, but there are a lot of good videos on Youtube.
I think the most frustrating thing is just trying to find the correct symbol and footprint for the part you want. Sometimes they're hard to find because you have to understand Kicad's organization scheme. Sometimes what you want isn't there at all so you have to make your own symbol or footprint or both. (Which, to be fair, is reasonably easy to do.)
Exporting a Kicad project's gerbers and drill files to JLPCB is surprisingly easy, but to use their pick and place service is more work. (I just ordered an assembled board for the first time, and had to manually reposition all the parts in their web interface because the position file put all the parts offset way off to one side and with the wrong rotation. Not sure if I did something wrong or there's just a bug.)
I think Kicad just takes a long time to learn. There are a lot of very powerful features that you might not know about at first. Kicad also expects you to do things in a particular way, and if you don't it tends to cause problems.
Meta: I noticed that the date near the top of the linked page (under the title) briefly flashes "Mar 24" before switching to "Mar 23". What's happening here?
There's a zulu timestamp in the markup that is being formatted client side into the TZ being presented to your browser, which happens to be at least a few hours behind UTC, setting the post back a day.
oshpark leaves rat bites that you have to sand down and clean up if the board outline matters to you. Worse, if you have castellated half holes, the rat bites can land right in the middle of what was supposed to be edge contacts.
I have boards that are small like the size of a dip28, where the entire length of the two long sides are castellated, and they are edge contacts to fit in a socket, not for soldering, and the entire board snaps into a carrier so essentially all 4 sides need to be clean and match the drawing. I have to sand and clean up oshpark boards, and live with a little bit of defect in the castellated holes. I don't have to do anything with jlc or pcbway boards. They arrive perfectly clean edged and ready to use.
I have one board that is only about 1cm square, but not square, it has a complicated outline, and both jlc and pcbway just pop it out with no fuss the same as any other order for the same nothing cost. I don't know how they even hold on to the tiny board to cut the outline. It does appear to be traditionally routered, not laser or water jet.
https://github.com/bkw777/TPDD_Cable
pics of an earlier version with slightly different outline. https://photos.app.goo.gl/TdYxGhzK94KT9rS78
I guess I shouldn't be so amazed at the size, given how common tiny pcbs are, for example the little page counter chips on my printers toner cartridges are only about 1/3 the size of these, although simple square edged.