I've assembled many dense(-ish) PCBs with small components (down to 0201) during the last three years. A summary.
Already have an account? Log in.
To make the experience fit your profile, pick a username and tell us what interests you.
Table of contents:
I'll add more and rearrange them when I have more, of course. Things that will definitely come:
I got this tip from someone else when I had to assemble my first 0201 board. It seems a bit weird at first glance, but it works very well!
Your solder might come in paste form, but it doesn't have to be in that state when you place the components. You can pick up paste with a soldering iron and pre-tin pad by pad for all components you want to populate. No stencil involved, and you get immediate feedback and control over the applied amount of solder.
The pads might look cold and bad after this, but we just need the right amount in the right places for now. Like so:
The footprint at the lower edge is for a FET in DFN1006B-3 (transistor package, about as large as the 0402 next to it), and the solder even bridges two pads (gate and source). All other pads look pretty much like inacceptable cold joints. That's not a problem though, because we'll level that all out with hot air:
Now the board is ready for components:
This sounds very tedious, and somehow it is, but it's worth it and doesn't really take that long with some practice. Now place your components and reflow - the tacky flux will hold them in place:
Some more remarks:
A little test with the multimeter in diode mode confirmed that at least the data lines (ZOE-M8Q's Rx and Tx, LIS3MDL's SCL and SDA) do have contact with the PCB. The rest I'll have to find out in some actual test runs.
Here's a mixed alloy board with a switch in the top right corner, a little BGA, and a u.Fl connector (not visible, but it's there! I promise). Those have high temp solder, the rest is low temp and not even completely done. I'll save that for later so that I can take some appropriate pictures.
Remember that when building a prototype, we don't have to assemble everything in one go. We are allowed to
These two points are related via the use of low temp solder. It's awesome in the right place. When I want an easy and relaxing assembly experience, i try to use low temp solder in as many places as possible. That's a hard to write list, so let's flip it around:
Use high temp solder where you have to, and low temp everywhere else.
Where to use high temp solder:
I usually try to limit the number of high temp reflow cycles for two reasons:
So the first two reflow cycles are for a limited set of parts as outlined above; one for each side of the board. A stencil doesn't really help, because only a limited set of pads will get high temp paste. Of course you can design a stencil just for high temp parts though, but it's probably not worth the effort.
Example for a weighed down BGA (STM32WLE5):
This was one of my earlier boards and I forgot to add some temperature indicator. Like a blob of solder paste (no available space for that) or one or two passives. This is better (a voltage regulator with caps, to the bottom left in the picture at the top of this log):
Here I had applied high temp paste for two capacitors (0201 and 0402) right next to the 4-BGA to tell me when the paste reflowed. If the caps are done, the BGA is done, too. I didn't have an extra weight for this one, but it worked nonetheless (again on a hot plate). You can also see that the BGA's silkscreen outline is completely useless.
Spoiler: You can also add extra weight to LGAs:
But more on that later when we get to the low-temp tips. LGAs don't come with the right amount of solder in the right place, so we have to do that ourselves - but at least we get to choose.
Why do you have 0201 or similarly sized components on your PCB? Probably to save space, or because of some more occult reasons like stray capacitance or whatnot.
Courtyards create sufficient space between components to ensure somewhat minimal insulation, allow for placement errors, support the reflow process and provide some extra space where the pick and place machine can maneuver. All those make sense. However, we usually encounter large, rectangular courtyards - and that leaves room for improvement. Most of the time we don't need them to achieve any of the above, because most components are rather flat (reflow, maneuvering), the signals require only little insulation and the pnp machine will be able do its job just fine even when the layout is very dense.
So to allow for denser layouts we can hide the courtyard layer and ignore courtyard violations (or configure them to be warnings or just messages), or redesign the courtyards. Both make sense:
Example: 0603 capacitors, KiCad stock footprint, with and without courtyards. The caps to the right are moved as close to each other as the design rules (0.1 mm / 0.1 mm) allow.
With good tweezers that's no problem. 0402 and 0201 are similar, because the design rules stay the same so there's always a minimum gap between the components.
One more thing to note about this style of passives: If their pads don't overlap with other parts, you cannot have a physical collision between parts and the design rule check will make sure that you don't have these overlaps. The pads are larger than the actual package.
Example: QFN with non-rectangular courtyard. QFNs have corners that occupy extend between the pad rows:
Also, if your EDA software has a 3D preview, make sure your 3D models are sufficiently accurate. Then the 3D view can tell you if it really fits even if you violate the courtyards:
Some even say that courtyards are luxury. I wouldn't go that far, but using stock rectangular courtyards feels a bit excessive.
Vias occupy space if they're placed outside of pads. And quite often we're tempted to place them inside pads, which is absolutely fine when the PCB is made with proper via-in-pad tech. When we now head over to, say, electronics stack exchange, we'll quickly realize that placing an "open" via in a pad will quickly result in great-great-grandma raising from her grave to grab us by the ear and pull us to the table because it's time for dinner before we can even release the mouse button. And for a production board that would probably be correct. But we're building a prototype and proper via-in-pad is expensive.
So - again - what can we get away with? Turns out, A LOT. Here's a bunch of 0201 caps on 0402 footprints with open vias in pads:
Honestly that still needed a little touchup and cleaning, but it works!
So if it's really an advantage for your layout: just do it. A via in an 1206 pad is certainly not a problem at all, and even down to 0603 it's not really a problem. Below that, you'll want to fill the via with solder first. Then it's business as usual. Vias in pads are ok.
The other side is valuable area for components. It's a little trickier to assemble, of course. One thing I try to do is to put all components that require high-temp solder on the same same side. These are
Another thing you can try to do is to design the PCB such that one side has only very flat parts.
But both of these points are optional, so don't worry about it too much.
With appropriate visual feedback (= magnifying glasses or a stereo microscope) you can adjust a part's position by 0.1 to 0.05 mm. It takes patience and a bit of practice, but it's certainly doable. I use a stereo microscope with 10x magnification most of the time, good tweezers (to place parts) and those pointy dentist things (to poke parts):
Let's have a look at how that compares to the size of small passives. Here's KiCad's stock 0201 footprint with the corresponding 3D view:
The pads are 0.46 mm wide. Let's offset the part a bit, say, 0.15 mm:
That's not going to ruin your PCB (unless it overlaps with other footprints), and you can easily give it a little nudge in the right direction so that it sits almost perfectly. The solder will pull it towards the pad's center and do the rest. Well, almost, because the stock footprint is too long. If one pad is already at reflow temperature and the other isn't, it might pull the part too far - like so:
and the part could tombstone. Let's fix that with a shorter footprint (I'll add it to the project files):
Left: stock KiCad footprint. Right: What I use (iirc it comes from some AVX datasheet or appnote). I use this very footprint in #Effect of moon phase on tombstoning and it has been pretty robust so far.
Not only does it not mess with you, there's one more benefit: The pad clearance areas don't overlap (assuming 0.1 mm spacing) and there's enough room to route a 0.1 mm trace between them:
JLCPCB supports that width/spacing for 4 and 6 layer boards, so it's really not out of reach for hobbyists.
Let's also compare that with a larger QFN which might seem less intimidating than an 0201 resistor. Here are both next to each other, with the QFN (0.4 mm pitch) displaced by 0.1 mm:
If you have ever soldered a QFN, you've probably tried to correct such a displacement.
Bottom line: 0201s are not harder to place than large fine-pitch packages, and they'll behave alright with the right footprint.
To demonstrate that we can get away with considerable displacement, here's a before and after reflow picture of three 0201 resistors I deliberately placed slightly off.
Microscope with 30x magnification.
top: placed slightly to the left. It's hard to see, but the resistor barely touches the pad on the right side.
middle: That's at an angle.
bottom: about 0.15 mm off both pads towards the bottom
This is how it turned out:
You want to assemble a prototype. Maybe even 3 or 5. But you're not a PCB assembly line.
That's a simple but important fact to keep in mind. You may do things that an assembly line may not do, and some things that can be taken for granted in an industrial setting are pretty different at home.
This log covers:
Oshpark and Aisler sell PCBs in sets of 3, JLC and PCBWay start at 5. That's a handful of boards, and usually that's about as many as one might be willing to do by hand. You're not planning for 100, and even 10 can feel more like work than like a relaxing hobby. When the first prototype works, usually on of these things happen:
Your design will probably have between 10 and 100 components in it, definitely not 1000.
Be well rested. Don't rush. Slow down, take your time, and you'll have it done faster. Take a break in between when you need one.
So for a board with 25 different components that's half an hour of preparation before we even touch the PCB. Of course you can touch it earlier but that won't get you anywhere.
No, clean your board first!
If I have a stencil:
Usually the paste is on after 5 to 10 minutes.
If I don't have a stencil:
Manual application takes a lot of time. Also depends on the application technique (more on that in a later log). So that's a large unknown, and sometimes you just have a bad day.
I need 30 to 60 seconds per component. A lot of factors go into this. Most of my parts are on cut tape in zip lock bags. That means that for every BOM entry I need to open the bag, get the tape out and un-tape the required number of components. That's pretty much the same overhead no matter how many I need of a specific type.
If I need just a single part of a given type, that's one minute including placing. All additional parts take maybe 10 seconds.
On average, I need 40 seconds per component for a board with mixed component quantities (some singles, some multiples). So for a board with 100 parts, expect 67 to 100 minutes just for placing stuff.
Again, depends. Each side of a PCB is reflowed separately. If you build in steps, there will be even more reflow runs (unless each complete side is one of your build steps). Hot air gun, toaster oven and hot plate all have different timings - you know if it's done, when it's done. Let's say 10 minutes and another 5 to let it cool down.
I'll cover some techniques in a later project log, for low-temp, high-temp and mixed boards. Yes you can mix alloys, which might be quite expensive if you just went for contract assembly.
If I use a stencil and do each side in one go, no additional assembly steps:
For a single sided board with 50 components: 30 + 7.5 + 33 + 15 = 80 minutes or 1 hour 20 minutes.
For a double sided board with 100 components: 30 + 2*(7.5 + 33 + 15) = 141 minutes or 2 hours 20 minutes.
Again, take a break in between. Maybe even two or three for double sided.
Don't have or can't use a stencil? It will take longer then.
Build in steps? That will also take longer.
You'll need to correct your own work. In an assembly line, this would barely be acceptable - but at home it's absolutely ok. That applies to paste application, misaligned parts, shorted pads and every...
Read more »
As someone that's in the process of designing a small and dense PCB (for Tetrinsic), this writeup came at the right time.
Thanks for sharing what you're learning! I hope you will be able to continue this.
Christoph
tonyo
Alan Green
deʃhipu
Jean Simonet
Nice, I might want to do something small anytime and your tips are really helpful!