The Reflow Profile Audit: Why Joints Look Right but Fail Thermal Cycling

The most expensive failure mode in SMT is the one that passes AOI, passes ICT, passes functional test, and then fails six weeks into customer thermal cycling. We see it most often on power-electronics boards and on automotive programmes where the field profile sees a real day-night temperature swing of 60°C or more. The visual inspection signature on these joints is unremarkable — a normal fillet, a normal toe-and-heel — and yet the metallurgy underneath has been compromised by an out-of-spec reflow profile that no one logged.

The four ways profile drift produces invisible weakness:

• Intermetallic compound (IMC) growth. Too long above liquidus and the Cu-Sn IMC layer at the pad interface grows past 5µm. The joint looks fine; it cracks under cyclic stress within months.
• Cold joints from insufficient TAL. Too short above liquidus and the joint hasn't fully homogenised. Wetting looks correct; bulk grain structure is wrong.
• Trapped flux residue. Ramp-rate too fast through the soak region means flux activator didn't have time to do its work. The joint looks bright; corrosion seeds are inside.
• Whisker-prone tin layers. Profile peaks that are too low for SAC305 (under 230°C peak) leave the joint with poor grain refinement, accelerating tin-whisker formation on lead-free finishes.

"Every joint that fails in the field passed inspection in the factory. The profile is what decides whether 'looks right' actually is right." — Pioneer Horizon reliability engineer

A reflow profile audit is the structured way to catch these failures before the field does. We run one quarterly on every production oven on the Madurai floor, and on every new programme at the validation gate.

A profile is only as good as the measurement that proves it. We use Datapaq Q18 profilers with seven thermocouples on a standardised test board, plus one floating couple on the production panel itself for spot checks.

Thermocouple placement

1. TC1: Smallest signal ball on the densest BGA — coldest joint, worst-case wetting risk.
2. TC2: Centre of the largest thermal pad — hottest, worst-case IMC growth risk.
3. TC3: 0201 capacitor at the panel edge — fastest-heating, worst-case tombstoning risk.
4. TC4: Largest electrolytic can or thermal-mass component — slowest to heat, worst-case cold-joint risk.
5. TC5: Centre of the panel, top side — reference profile.
6. TC6: Centre of the panel, bottom side — checks for ΔT across the board thickness.
7. TC7: Floating air probe at oven exit — captures actual exit temperature for downstream handling.

Measurements we extract from each run

• Peak temperature at every TC — target 235–240°C for SAC305, with ΔT across the board no greater than 10°C.
• Time above liquidus (TAL) at every TC — target 60–90 seconds, ΔT across the board no greater than 15 seconds.
• Ramp rate through soak and through reflow — target 1.5–2.5°C/s, with no spikes above 3.0°C/s anywhere on the curve.
• Soak time in 150–200°C band — target 60–90 seconds for normal builds, 75–120 seconds for thermal-pad-heavy builds.
• Cooling rate from peak — target 2–4°C/s; faster than 6°C/s induces brittle grain structure, slower than 1°C/s grows IMC excessively.

Each measurement gets a pass/marginal/fail classification. A single marginal at TC2 with everything else green means a paste or paste-mask tweak. A pattern of marginals across multiple TCs means an oven calibration is overdue.

Of every twenty audit findings we file, twelve relate to ramp rate. It's the easiest parameter to let drift — convection ovens warm up, conveyor speeds get adjusted for cycle time without re-profiling, and operators sometimes "speed up the soak" because it's running long. Every one of those changes shifts the ramp profile.

The pre-heat ramp (ambient to 150°C)

Target: 1.0–2.0°C/s. Faster than 2.5°C/s and you risk thermal shock on ceramic components — capacitors crack, MLCC barium-titanate develops microfractures that don't show up until humidity exposure. We've seen 0805 X7R caps from a reputable vendor with crack-rate jumping from 0.02% to 0.18% on a single oven that had drifted to 2.8°C/s pre-heat ramp. The board passed AOI 100%. The customer found it three months later.

The soak transition (150°C to 200°C)

Target: 0.5–1.0°C/s. This is the band where flux activator does its work and where solder paste outgasses. Faster than 1.5°C/s here is the dominant cause of constellation-pattern voiding under BGAs.

The reflow ramp (200°C to peak)

Target: 1.5–2.5°C/s. Faster than 3.0°C/s induces tombstoning on chip components — the small pad wets and lifts the part before the large pad has reached wetting temperature.

The cool-down ramp

Target: 2–4°C/s through 217–150°C. Faster than 6°C/s and the SAC305 grain structure becomes brittle. Slower than 1°C/s and IMC layer growth resumes during the cooling phase.

We re-profile every oven monthly on a reference board, and after any maintenance event (heater replacement, conveyor belt change, blower service). The profiler sits in the oven for fifteen minutes including setup and recovery. The cost of doing it is 0.5% of one shift; the cost of not doing it is the field-failure ticket nobody wants to take.

Peak temperature and TAL together control the joint metallurgy. Get them right and the joint will outlive the warranty. Get them wrong and you've built a board that passes test and fails the field.

The peak temperature window for SAC305

• Below 230°C at the coldest joint: incomplete wetting, gritty fillets, high risk of cold joints on high-mass components. We never sign off a profile with TC1 (smallest signal ball) below 230°C.
• 230–240°C: target window. Full wetting, controlled IMC growth, repeatable grain structure.
• 240–245°C: acceptable for low-thermal-mass components but starts pushing MSL ratings on plastic BGAs. Watch your popcorn risk.
• Above 245°C: do not exceed. J-STD-020 component peak rating starts being violated, package delamination risk climbs sharply.

TAL — the IMC growth governor

IMC layer growth follows roughly t½ kinetics at SAC305 reflow temperatures. The first 30 seconds of TAL build a 2µm IMC layer — necessary for joint formation. The next 30 seconds add another 1µm. Past 90 seconds, every additional 30 seconds adds 0.5µm but disproportionately weakens the joint against thermal cycling because IMCs are brittle.

For Class 3 automotive and medical work we hold TAL between 60 and 75 seconds. For Class 2 consumer we'll accept 60 to 90 seconds. We don't sign off any profile with TAL exceeding 100 seconds.

The ΔT-across-board problem

The hardest part is keeping the spread between the hottest and coldest joints below 10°C. A 12mm × 12mm BGA with a 5mm thermal pad and an adjacent 0402 cap can see a 15°C swing on a poorly profiled oven. The fix is usually conveyor speed reduction plus mid-zone temperature trim — small adjustments that take ten minutes to validate. If the spread can't be brought under control with profile changes, the board layout needs help — talk to us in the DFM review.

The audit isn't a one-time event. It's a discipline. Here's the cadence we run on the Madurai floor and what we deliver to customers from it.

Cadence by trigger

1. Per oven, monthly: reference-board profile against documented golden curve. Pass/fail per parameter. Logged in the line MES.
2. Per programme, on first build: full seven-TC profile on actual production panel. Customer-signed validation report.
3. Per programme, every 5,000 boards: spot-check with floating TC on a production panel mid-run. Two-page report.
4. Per oven, after maintenance: full reference profile before next production run. No production resumes until pass.
5. Per programme, on yield drift: if joint-defect rate moves by more than 0.3 absolute points week-on-week, the oven is profiled within 24 hours.

What the customer gets

• The seven-TC trace as a PDF on first-build validation — peak, TAL, ramp rates highlighted against spec.
• A void-rate report (X-ray sample of 5 BGAs per lot for Class 3 work) — see our BGA voiding article for the signature reading methodology.
• Monthly oven calibration summary for any customer with a recurring programme — a single page per oven per month showing parameter drift over time.

What we look for when something drifts

Profile drift is usually one of four things: a partially blocked heater zone (cleaning fixes it), a worn-out conveyor drive belt (slipping under load — replacement), a sensor calibration drift (re-cal every twelve months minimum), or an HVAC change in the building affecting oven intake air. The last one is the sneaky one — we've traced two yield events on the line to monsoon-season ambient air changes upstream of the oven.

"You can't out-process a drifted oven. Profile audits aren't optional — they're how we keep Class 3 yield numbers that we'd sign with our name." — Pioneer Horizon line lead, Line 3

If your contract manufacturer doesn't ship a profile report with the build, that's a flag. Ask us how we'd profile your programme — we'll send a sample report from a comparable build before you commit a panel to us.