Assembly: Soldering, Reflow, and Rework
Solder is not glue. It is a metallurgical joint with a temperature profile, and the profile is not optional.
🎯 By the end you'll understand
What happens inside a solder joint — wetting, the intermetallic layer, and why it grows too thick if you are generous with heat. Why a reflow profile has four zones and what each one prevents, with typical SAC305 numbers and the component limits that cap them. How to read tombstones, bridges, cold joints, and voids under a loupe and trace each back to its cause. How to hand-solder and hot-air rework without lifting pads or cracking capacitors, and how Beacon's rotated WS2812B-2020 was fixed on five boards — and avoided on the next twenty-five without touching the KiCad footprint.
KEY INSIGHTHotter is not safer
A solder joint is a reaction between tin and copper that needs a specific temperature for a specific time: enough to form the intermetallic layer, not enough to make it thick and brittle. Every bad instinct at the bench comes from treating heat as a quantity where more is more.
Watch someone rework a board for the first time and you will see the instinct. The hot-air station goes to 450 °C "so it flows quickly." The joint does flow — instantly — and so does the plastic of the connector next to it, the LED's silicone lens goes cloudy, and a ceramic capacitor two millimetres away takes a thermal shock its datasheet never contemplated. The joint that results looks fine. It fails in three weeks, because the intermetallic layer underneath the shiny surface grew four times thicker than it should have and cracked the first time the board flexed.
The opposite instinct is just as common: an oven or a hot plate that never quite reaches the alloy's liquidus, producing joints that are grey, grainy, and electrically connected on the bench — until the first temperature cycle in the field. Both failures share a cause: treating soldering as a heating problem rather than a profile problem. Temperature versus time is the specification. The paste datasheet gives you a window; the components give you ceilings; your job is to fit a curve between them.
Three numbers define a joint's fate: the peak temperature (did the alloy fully melt and wet everything, including the part with the most thermal mass?), the time above liquidus (long enough for the intermetallic to form, short enough not to overgrow), and the ramp rates on the way up and down (slow enough that flux does its work and nothing cracks, fast enough that flux is not exhausted). Everything else in this lesson is about how to hit those three on a real board with parts that have different masses and different limits.
HOW IT WORKSWhat a joint actually is: wetting, flux, and the intermetallic
Molten solder does not stick to copper the way glue sticks to wood. It wets it: the liquid alloy spreads over the metal because the surface energy of the tin–copper interface is lower than that of the bare oxidised copper, and as it spreads, tin reacts with copper to form a thin layer of intermetallic compounds — Cu6Sn5 next to the solder, Cu3Sn next to the copper. That layer is the joint. Without it you have solder sitting on copper; with it you have a metallurgical bond. The layer forms in seconds once the alloy is above liquidus, and it keeps growing by diffusion as long as the joint is hot: roughly proportional to the square root of time, faster at higher temperature. A healthy layer is a micrometre or two. A layer several micrometres thick is brittle, and the joint cracks under thermal cycling and vibration.
Wetting cannot happen through an oxide, and copper and tin both oxidise on contact with air. That is what flux is for: an acid (rosin-based, water-soluble, or a "no-clean" synthetic) that becomes active above a certain temperature, dissolves the oxides, and forms a film that keeps oxygen away while the solder flows. Flux has a budget. Activated too early or held too long at temperature, it is consumed before the reflow zone and the joint forms on a re-oxidising surface — dull, grainy, poorly spread. That budget is the reason a profile has a soak zone with a defined length, not just a ramp to peak.
| Alloy | Melting behaviour | Where it is used | What it costs you |
|---|---|---|---|
| Sn63 / Pb37 (leaded, eutectic) | Sharp melt at 183 °C | Prototypes, hobby work, exempt equipment | Restricted for commercial electronics in the EU under RoHS (Pb ≤ 0.1 % by weight in homogeneous materials); many export markets follow. Easier to work, wider process window. |
| SAC305 (96.5 Sn / 3.0 Ag / 0.5 Cu) | Solidus ≈ 217 °C, liquidus ≈ 220 °C | The default lead-free alloy for contract assembly; what Beacon's boards were built with | Higher temperatures stress components and connectors; narrower window; joints look duller than leaded even when perfect. |
| Low-temperature SnBi (e.g. Sn42 / Bi58) | ≈ 138 °C | Heat-sensitive assemblies, some consumer products | Brittle; poor with leaded finishes; not a drop-in for a design reflowed at SAC temperatures. |
Reworking a SAC305 joint with leaded wire produces a mixed alloy with a lower, smeared melting range and a microstructure nobody characterised. It will usually work. It is also the joint most likely to fail first, and the mix makes later rework unpredictable because you no longer know the liquidus. Pick one alloy per board. Beacon's boards are SAC305, so the rework wire is SAC305 or a lead-free equivalent, and the iron runs hotter to match.
CORE CONCEPTThe profile, zone by zone
A reflow profile is temperature against time, measured at the joints — not at the oven's thermostat, which is why serious assemblers run a thermocouple taped to a real board. The four zones exist for four physical reasons, and the numbers below are typical for a SAC305 no-clean paste; the paste datasheet is the authority for any specific product and will differ by tens of seconds and a few degrees.
Preheat — ramp at 1–3 °C/s to about 150 °C
Drives off the paste's solvents gently and brings the board up without thermal shock. Ramp too fast and solvent boils explosively, scattering tiny solder balls around the pads and disturbing the paste under small parts; ceramic capacitors also dislike fast ramps. Ramp too slowly and you spend flux budget before the soak has begun.
Soak — 150–190 °C for 60–120 s
Two jobs. The flux activates and cleans the surfaces. And the board equalises: a 0402 resistor and the ESP32-C3-MINI-1's large ground pad have wildly different thermal mass, and without a soak the small parts reach liquidus while the module's pad is still 30 °C cooler. That gradient is the mechanism behind tombstoning on small parts and cold joints on big ones. A soak too long or too hot exhausts the flux before reflow, which shows as dull, grainy, poorly wetted joints and more voids.
Reflow — above 217 °C for 45–90 s, peak 235–245 °C
The alloy melts, wets, and forms the intermetallic. The peak must be high enough that the coldest joint on the board (the module's ground pad, the USB-C shell) clearly exceeds liquidus — a joint that only touches 220 °C at the thermocouple has not reflowed in its interior. The time above liquidus (TAL) is the intermetallic's growth window: too short and the layer is incomplete, too long and it thickens into brittleness. The ceiling comes from the components: J-STD-020 assigns small packages a 260 °C peak, connector housings and LEDs are often lower, and the paste's own maximum is usually around 250 °C.
Cool — controlled, roughly 2–4 °C/s
Faster cooling produces a finer grain and a stronger joint, up to the point where the rate shocks the ceramics: a multilayer capacitor pulled from 240 °C by a blast of room air can crack internally, and the crack opens weeks later as a short. Slow cooling coarsens the grain and lengthens the profile. Most ovens cool passively at about the right rate; hot plates are the problem, because the board sits on the heat source until you lift it.
Plastic-encapsulated parts absorb moisture from the air. Heated fast to reflow temperature, that moisture flashes to steam inside the package and delaminates or cracks it — "popcorning." J-STD-020 assigns a Moisture Sensitivity Level; the ESP32-C3-MINI-1 is typically MSL 3, meaning a floor life of about 168 hours at ≤ 30 °C / 60 % RH once the dry bag is opened, after which the part should be baked before reflow (typical guidance is 125 °C for 24 hours, per J-STD-033, adjusted for package thickness). Contract assemblers manage this for you. If you reflow a module yourself from a bag opened last month, bake it first.
COMPARISONDefects and how to read them
Every assembly defect has a signature under a 10× loupe and a cause in the profile, the stencil, or the footprint. The point of learning them is diagnostic: a bridge on the module's castellations tells you something different from a bridge under the USB-C receptacle, and a tombstone on one specific 0402 usually indicts its pads rather than the oven.
| Defect | What it looks like | Mechanism | Usual root cause |
|---|---|---|---|
| Tombstoning | A chip part standing on one end like a gravestone | One end's solder melts first and its surface tension pulls the part upright before the other end melts | Unequal pad thermal mass (one pad tied into a plane without thermal relief), unequal pad sizes, part misplaced off-centre, soak too short |
| Bridging | Solder spanning adjacent pads or pins | Too much paste for the gap, or paste smeared between pads | Stencil aperture too large, stencil too thick, poor stencil release, no solder-mask dam between pads (Lesson 5.8) |
| Cold / insufficient joint | Dull, grainy, lumpy; solder not spread up the termination | Alloy never fully wetted: too little heat, flux exhausted, or oxidised surface | Peak too low at that joint, soak too long, old paste, contaminated pads; classic on a big ground pad in a weak oven |
| Voids | Cavities inside the joint, visible only on X-ray (or as a hollow ring when the joint is probed) | Flux volatiles trapped under a large pad as the solder solidifies | Big thermal pads with a single stencil aperture; too-fast ramp; a few percent void area is normal, large voids reduce thermal and mechanical performance |
| Solder balls / spatter | Tiny spheres near pads or under parts | Solvent boiled off explosively, scattering paste | Preheat ramp too fast; paste too cold or too wet; too much paste |
| Popcorning / delamination | Cracked or bulged package, sometimes audible during reflow | Absorbed moisture flashing to steam | MSL floor life exceeded without baking |
| Lifted pad | Copper pad detached from the laminate, often during rework | Adhesive between copper and FR-4 weakened by heat, then mechanically pulled | Rework too hot or too long, no preheat, tugging on a part before the solder is fully molten |
| Head-in-pillow | BGA ball sitting in the paste without merging | Oxide skin prevents the two solder masses from coalescing | Warpage separating ball and paste during reflow; exhausted flux. Not a Beacon risk (no BGAs), but you will meet it |
What counts as acceptable is not a matter of taste. IPC-A-610, the acceptability standard for electronic assemblies, defines three classes: Class 1 for general products where function is all that matters, Class 2 for dedicated-service products where extended life is required but uninterrupted service is not (most commercial and industrial electronics — Beacon's category), and Class 3 for high-performance products where failure is not acceptable (medical, aerospace, military). The classes differ in numbers, not concepts. For a chip component, for instance, the solder-side overhang of the termination beyond its pad is allowed up to 50 % of the termination width (or pad width, whichever is less) in Classes 1 and 2, but only 25 % in Class 3. The standard is what you and a contract assembler point at when you disagree about whether a board is good.
Paste volume is set by the stencil: a stainless foil, typically 100–120 µm (4–5 mil) thick for a board with 0402s and 0.5 mm-pitch parts, with apertures cut to the pads or slightly reduced. IPC-7525 guidance for reliable paste release is an aperture area ratio (aperture area over aperture wall area) of at least 0.66 and an aspect ratio above 1.5 — below that, paste sticks in the aperture instead of on the pad, and the fine-pitch joints starve while the big ones bridge. When your assembler's DFM report mentions "aperture reduction" on a thermal pad, this is why: a windowpane pattern lets volatiles escape and prevents the part floating on a lake of solder.
SEE ITDesign a profile — and watch each zone earn or lose its keep
Shape a profile for a SAC305 board carrying Beacon's parts: a large-mass module, 0402 ceramics, a WS2812B-2020 whose datasheet family often specifies a lower peak than the generic 260 °C, and the through-hole JST-PH header, hand-soldered after reflow — but its nylon housing sets a limit of its own the moment anyone suggests pin-in-paste. The predictor scores each zone against typical paste and component windows (illustrative thresholds, labelled as such) and estimates how many of five boards come out with a defect. Then try the instinct: push the peak up and the time above liquidus out "to be sure" and see what it costs.
Two experiments worth running. First, keep everything nominal and shorten the soak to zero: the module's ground pad and the 0402s now reach liquidus at different times, and the predictor reports tombstones and a cold module joint from the same cause. Second, from nominal, raise the peak to 262 °C and the time above liquidus to 120 s. The joints are unquestionably molten — and the LED, the connector housing, and the intermetallic thickness all cross their lines. "Hotter to be safe" is the profile that fails inspection in three places at once.
IN PRACTICEHand soldering and hot-air rework without collateral damage
The oven's profile is the reference; hand tools approximate it locally, and the approximation is where damage happens. An iron tip at 340 °C touching a joint is a ramp of hundreds of degrees per second at that spot, with the rest of the board at room temperature. The rules below exist to make that approximation survivable.
Iron work
For SAC305, a tip temperature of 320–350 °C, a chisel tip sized to the pad (a fine conical tip transfers heat badly and tempts you to raise the temperature), and flux on every joint — the flux core in wire solder is not enough for rework because it has burned off by the time you reposition. Heat the pad and the termination together, feed solder to the joint rather than to the tip, and leave within two to three seconds. For the module's castellations, drag soldering along the row with a well-tinned chisel and generous liquid flux gives cleaner joints than dabbing each one. Remove excess with braid (wick) and flux, never by reheating and hoping. Clean no-clean flux residue near the SHT40 anyway: Sensirion warns that contamination and outgassing affect humidity accuracy, and a flux-coated sensor is a contaminated sensor.
Hot air
A hot-air station replaces the oven's reflow zone in a spot a few millimetres wide. The two settings that matter are nozzle temperature and airflow; a typical starting point for lead-free rework is 330–380 °C at the nozzle with low flow, and the number on the dial is not the temperature at the joint — measure once with a thermocouple if you plan to do this often. The non-negotiable step is preheating the board from below on a preheater or hot plate at 100–150 °C. It cuts the gradient the nozzle has to bridge, which is what protects pads from lifting and ceramics from cracking, and it stops the neighbouring 0402s from floating off when your air hits them. Kapton tape masks parts you want to protect; a small nozzle keeps the heat where you want it.
Verify the fix before touching the iron
Under the microscope, identify pin 1 on the part itself from the package's corner mark, and on the datasheet drawing. Decide the correct orientation from the schematic, not from the silkscreen — the silkscreen is what was wrong.
Preheat, flux, remove
Board on the preheater at ≈ 120 °C for two minutes. Liquid flux on the part. Hot air at ≈ 350 °C, small nozzle, low flow, circling the part. When all pads are molten the part floats; lift it with tweezers without pulling. If it resists, it is not molten — more time, not more force.
Clean the pads, place, reflow
Wick the pads flat with flux and braid. Fresh flux. Place the new part, oriented per step 1, centred on the pads. Hot air again until every joint flows and the part self-centres; do not press on it. Let the board cool on the preheater as it ramps down, not on the bench.
Clean, inspect, re-test the rung
Isopropyl alcohol and a brush; dry. Inspect every joint on the part and every neighbour you might have disturbed. Then re-run the bring-up rung for that subsystem with the same limit and prediction as before — the fix is proven by the current signature, not by the appearance of the joint.
The ESP32-C3-MINI-1 has a large ground pad under the body plus a ring of castellations, and its shield can soaks up heat. Removing it takes a large nozzle from above, a preheater at ≈ 150 °C, and 60–90 s of patience; in that time every 0402 within a centimetre is at risk. The pads under it are easily lifted by an impatient tweezer. If a defect can be fixed around the module — a bodge wire, a lifted pin, a swapped passive — do that. If the module itself is dead, it is often cheaper to scrap a prototype board than to rework it.
BEACONBeacon's LED, fixed by hand on five boards — and avoided on twenty-five
Lesson 6.1's bring-up log recorded LED-1: the WS2812B-2020 placed rotated on all five Rev A boards, because the project footprint's pin-1 mark was on the wrong corner (the defect Lesson 5.3 warned about and the review in 5.9 did not catch). Under the microscope the part's corner mark sat diagonally opposite where the schematic needed it. With the LED rail enabled, the rail drew about 15 mA more than a lit LED should and the LED stayed dark — a supply landing on the wrong pads, conducting through the part's internal protection diode.
The fix on the five boards followed the four steps above. Two details from the record are worth keeping. First, the JST-PH header is 4 mm from the LED on Rev A; it was masked with Kapton and the nozzle kept pointed away from it, because a reflow-grade nylon housing is rated for the oven's profile, not for a 350 °C nozzle held nearby for a minute. Second, the first board took 35 minutes including the setup and the nerves; boards two to five took about 12 minutes each. All five then passed rung 6a: +8 mA on the rail, and light. The reworked joints went into the log as a known reliability watch-item — a hand-reworked 2020-package joint is the least-characterised joint on the board, and Lesson 6.6 treats it that way.
The pilot needed 25 units, which meant a second Rev A batch. Editing the KiCad footprint and regenerating Gerbers would have been correct — and would have meant a fresh DRC/DFM cycle on files that had already been built once, for a change that touched only one part's orientation. The cheaper, equally correct path was to leave the fab files alone and correct the LED's rotation in the placement (CPL) file the assembler uses, then confirm it in the assembler's placement preview. The pads are symmetric; only the rotation was wrong. Twenty-five boards arrived with the LED correct. The footprint in the project library is still wrong, which is why "fix the WS2812B-2020 footprint's pin-1" sits on Rev B's change list in Lesson 6.9 with the evidence from both batches attached.
One more thing the second batch made visible: it was ordered with the same profile the first batch was built with, and both batches show the same joint quality under the loupe. That is the assembler's process being stable across a month — the same kind of evidence Lesson 6.1 wanted from five boards agreeing with each other, now from thirty.
WEAKNESSWhere this bites
A joint that has flowed has a shiny (leaded) or satin (lead-free) surface regardless of what happened underneath. An overcooked joint with a thick, brittle intermetallic looks identical to a good one on the day it is made. An undercooked joint on a large pad can have a perfect fillet at the edge and an unwetted interior. The only ways to know are process control — the profile measured at the joint, inside the paste's window — and, for high-reliability work, cross-sections and X-ray. On a prototype bench, the honest substitute is discipline about temperature and time, and a functional re-test after every rework.
Skipping the preheater because "it's just one part" is the single most common cause of lifted pads and cracked ceramics in rework. A 350 °C nozzle on a 25 °C board creates a gradient of hundreds of degrees across a few millimetres: the laminate under the pad softens and its adhesion drops while the copper is being tugged, and the multilayer capacitor next door sees a ramp its datasheet forbids. The crack in that capacitor does not show up now. It shows up as a short, in the field, after a few humidity cycles. Preheat is not for the part you are fixing — it is for everything around it.
The number on a hot plate, a toaster-oven controller, or a hot-air station is a setpoint at a sensor that is not your joint. Hot plates in particular overshoot, have cold spots, and hold the board on the heat source during what should be the cooling zone. A thermocouple taped to a pad on a scrap board, logged once, tells you what your tools actually do; without that, every profile you "run" is a guess, and the defects you get will look random because you never saw the variable that caused them.
KEY INSIGHTHow this connects
- Board Bring-Up: Smoke Test to First Blink (6.1) — the LED-1 defect was found there by inspection and confirmed by current; every rework here is proven by re-running that rung.
- Symbols, Footprints, and the Lie in Every Library (5.3) — the rotated pin-1 is a footprint defect, and the CPL workaround is a placement fix, not a library fix. The library gets corrected in Rev B.
- Test Jigs, Pogo Pins, and Production Test Firmware (6.3) — next: the assembly defects in this lesson's table are exactly what the jig's cheap tests must catch, and the intermittent cold joint is the one they mostly cannot.
- Reliability and Field Failure Modes (6.6) — hand-reworked joints and thermally shocked ceramics are infant-mortality candidates; the pilot watches them accordingly.
Assembly is where the abstractions of Stage 5 — pad geometry, mask dams, thermal relief, courtyards — turn into physical outcomes. A footprint you drew in Lesson 5.3 is a tombstone or a clean joint here, and nothing in between.
GLOSSARYKey terms introduced
- Liquidus / solidus
- The temperatures above which an alloy is fully liquid and below which it is fully solid. SAC305 is solid below ≈ 217 °C and fully liquid above ≈ 220 °C; eutectic Sn63/Pb37 melts sharply at 183 °C.
- Wetting
- The spreading of molten solder over a metal surface, driven by surface energy once oxides are removed. Wetting is what turns solder on copper into a joint.
- Intermetallic compound (IMC)
- The reaction layer (Cu6Sn5, Cu3Sn) between solder and copper that constitutes the metallurgical bond. Micrometre-thin is strong; several micrometres is brittle. It grows with time above liquidus.
- Soak
- The profile zone held between roughly 150 and 190 °C where flux activates and the board's parts equalise in temperature before reflow. Too short causes tombstones and cold joints on high-mass parts; too long exhausts the flux.
- Time above liquidus (TAL)
- How long the joint stays above the alloy's liquidus. Typically 45–90 s for SAC305: enough to form the IMC, not enough to overgrow it.
- Tombstoning
- A chip component pulled upright on one pad by the surface tension of the first end to melt, when the other end lags. A thermal-gradient or pad-imbalance defect.
- Moisture Sensitivity Level (MSL)
- The J-STD-020 classification of how long a plastic package may be exposed to ambient humidity before reflow without risk of popcorning. MSL 3 gives about 168 h; beyond that, bake per J-STD-033.
- IPC-A-610
- The acceptability standard for electronic assemblies. Its Classes 1, 2, and 3 set progressively stricter numeric criteria for joints, overhang, fillets, and cleanliness. Beacon is a Class 2 product.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1Why does a reflow profile have a soak zone at all? What two different defects appear if you remove it, and why are they the same failure?›
The soak lets the flux activate and, more importantly, lets parts of very different thermal mass equalise before the alloy melts. Remove it and the small 0402s reach liquidus while the module's large ground pad is still tens of degrees cooler. On the small parts, one pad melts before the other and surface tension stands the part up — a tombstone. On the module, the pad may never fully exceed liquidus before the profile moves on — a cold joint. Both are the same failure: a thermal gradient across the board at the moment the alloy melts. That is why the soak's duration is a process parameter rather than a delay.
2A colleague reworks a joint at 450 °C "so it's fast and flows well." The joint looks perfect. What is wrong with it, and when will you find out?›
The joint's surface tells you nothing about the intermetallic layer beneath it. At 450 °C the tin–copper reaction runs far faster than at a 240 °C peak, and the IMC grows thick — several micrometres instead of one or two. Thick IMC is brittle. The joint holds on the bench and cracks under the first thermal cycling or mechanical flex it meets, typically weeks later in the field, where it presents as an intermittent that no bench test reproduces. The collateral is just as real: the nylon connector housing softens, the LED's silicone lens clouds, and the ceramic capacitor beside it takes a thermal shock that may crack it internally. Fast and hot is a loan against the joint's lifetime.
3Beacon's LED footprint was wrong on all boards. Why was the second batch fixed in the CPL file rather than in KiCad, and what debt does that leave?›
Because the pads were correct and only the rotation was wrong. Correcting the part's angle in the placement file the assembler uses fixes the built board without regenerating Gerbers, which would have forced a fresh DRC/DFM cycle on files that had already been validated once — for a change that touches no copper. The debt is that the project library still contains a footprint with pin 1 on the wrong corner: anyone who reorders from the repository, or reuses the footprint in another design, repeats the defect. That is why the library fix is on Rev B's change list with evidence from both batches, rather than being considered closed by the workaround.
4You are about to hot-air a single 0402 on a finished board. Name the step people skip, and explain the two failure mechanisms it prevents.›
Preheating the board from below to 100–150 °C. First, it reduces the temperature gradient the nozzle must bridge, so the laminate under the target pads and their neighbours does not go from cold to well above its glass-transition temperature in seconds; that is what keeps the copper's adhesion intact and stops pads lifting when the part is touched. Second, it keeps the nearby multilayer ceramic capacitors from seeing a ramp rate far beyond what they tolerate; thermal-shock cracks in ceramics are latent, and they surface later as shorts. Preheat is for the parts you are not reworking.