Fine-Pitch BGA Rework: When to Reball, When to Replace

When a fine-pitch BGA fails post-assembly, the first question is usually treated as a technical one and it shouldn't be. It's an economic one: is the part more valuable than the rework labour plus the residual reliability risk? For a $2 MCU the answer is almost always replace. For a $400 FPGA the answer is almost always rework. The interesting boards are the ones in the middle — the $30–80 SoCs that show up in handsets, set-top boxes, and industrial controllers.

Our rework decision matrix:

• Part cost < $5: always replace. The part is the cheap input; rework labour is the expensive one.
• Part cost $5–40: case-by-case. If the part is in stock and the board is easy to access, replace. If the part is on allocation or the board is dense, rework the existing part.
• Part cost $40–200: default to reball-and-reattach for any single-event failure. Replace only if X-ray shows die-level damage.
• Part cost > $200: always reball. The economics overwhelmingly favour rework, and we've reballed FPGAs three times on the same board successfully when programme logistics required it.

"A $400 FPGA failed by a single open ball is a rework job. A $4 MCU failed by anything is a replace job. The middle ground is where most of our rework hours go." — Pioneer Horizon rework cell lead

The reliability question runs alongside the cost question. A reballed BGA, attached with the right profile, can match new-attach reliability — but only with disciplined process control. The rest of this article walks through how we hold that discipline on our Madurai rework cell, and how we keep void rates on reballed parts under 8%.

The board is usually more expensive than the BGA. A board with a $40 SoC on it might have $300 worth of other parts and 80% of a build's labour invested. Removal that damages the board is removal that turns a $40 rework into a $400 scrap.

The removal sequence on our PDR rework station

1. Bake the board. 4–8 hours at 110°C to drive out moisture. Skip this and the BGA can pop on heat-up, lifting pads and damaging surrounding components.
2. Preheat from below. Bottom-side heater ramps the board to 130–150°C. The whole-board preheat reduces thermal gradient when the top heater applies localised energy.
3. Top-side localised heat. Convection nozzle sized to the package — typically 1–2mm clearance around the package edge. Profile mirrors the original reflow: ramp 1.5°C/s to soak (180°C, 60s), then ramp to peak (235–240°C) with thermocouple monitoring at the package corner.
4. Vacuum lift at peak. Vacuum pickup tool engages the package as soon as the thermocouple confirms TAL. Lift vertically without tilt — even a few degrees of tilt drags balls and lifts pads.
5. Site cleanup. Residual solder removed with a desoldering wick and low-residue flux. Final pad finish flat and bright.

What we monitor in real time

• Package corner temperature via thermocouple.
• Adjacent-component temperature via second thermocouple on the closest 0402 within 3mm.
• Pickup-tool vacuum integrity (loss of vacuum means we abort — the package may have shifted).

The adjacent-component temperature is the one most rework operators ignore. Pulling a 0805 cap off a board next to a BGA because the second thermocouple wasn't watched is a rework into a rework — exactly the kind of cascading damage that turns a routine BGA replacement into a half-day scrap-or-save discussion.

Reballing is the step that decides whether the reworked joint is as good as the original, better than the original, or worse than the original. Done correctly with a controlled profile and clean prep, reballing produces joints that outperform original-attach reliability in our thermal-cycling data — the second pass cleans up pad irregularities that the original attach missed.

Reballing toolchain

• Reball stencil: stainless-steel reball stencil matched to the BGA ball-out, with apertures sized 0.05mm under nominal ball diameter to control deposit.
• Reball spheres: SAC305 spheres matched to original ball diameter. We buy 99.5% sphericity grade from a tier-1 vendor — generics save $20 per package and cost 3–5 percentage points on voiding.
• Flux: tacky low-residue flux, applied via stencil. The flux holds the spheres in place during placement and provides the activation needed during reflow.

Reball profile

Our reball reflow profile is gentler than the production profile because the package is exposed and the thermal mass is much smaller:

• Soak: 150–180°C, 90 seconds (longer than production — gives flux more time to clean residue from used pads).
• Ramp to peak: 1.5°C/s.
• Peak: 230°C, 25–35 seconds TAL.
• Cool-down: 2–3°C/s, no forced cooling.

Inspection after reball

Every reballed package goes through X-ray inspection before reattach. We're looking for:

1. Ball-count completeness. Missing balls = redo the reball.
2. Ball-diameter consistency. ±5% diameter variation across the array. Outliers are remelted.
3. Co-planarity. All ball-bottom heights within 0.05mm. Failures are remelted.
4. Voiding. Target under 8% on signal balls, under 15% on any thermal centre area. Re-balls don't typically have central thermal pads, but if they do, our standard voiding signature analysis applies.

A reballed package that fails any of these gets remelted, not rejected. The cost of a remelt is minutes; the cost of a rejected $400 FPGA is real money.

Reattaching a reworked BGA to the board is the closest thing to the original assembly step, but it's done as a localised reflow with thermal-mass and temperature-gradient conditions very different from a production oven. The profile must reflect that.

Why the reattach profile is different

• Asymmetric heating. Top-side convection heats the package; the board absorbs heat from below at a different rate. Without compensation, the package reaches peak before the board is ready.
• Smaller thermal volume. Only one component is being reflowed. The local environment cools faster and heats faster than an oven build.
• Older paste/flux interaction. The paste is freshly applied on the board pads, but the reball flux is from the earlier step. Activator chemistry interactions need slightly more soak time.

Our reattach profile

• Bottom-side preheat: board to 130–140°C, held for 5 minutes to homogenise.
• Top-side ramp: 1.2–1.5°C/s through soak. Slower than production reflow to compensate for asymmetric heating.
• Soak: 75 seconds in 160–200°C band. Longer than production to allow flux to fully activate around the freshly applied paste.
• Ramp to peak: 1.5°C/s.
• Peak: 235–240°C, 60–75 seconds TAL. Same target window as production.
• Cool-down: natural convection plus low-speed fan, 2–3°C/s through 217–150°C.

Post-reattach inspection

X-ray every reattached BGA, full panel. Compare void rates to baseline production data — a reattached BGA should not show higher voiding than the same package on the original build. If it does, the reattach profile needs adjustment.

We keep a per-package reattach-profile library on the rework cell — we know the right soak time and peak for a 0.5mm-pitch 256-ball BGA, for a 0.4mm-pitch 484-ball FPGA, for a 1.0mm-pitch 196-ball SoC. Walking up to a fresh part and developing a profile from scratch is how rework cells get into trouble.

Reworking a Class 3 board doesn't automatically lose its Class 3 status — but it does require the same documentation discipline that the original build had. Anything less and the rework is a footnote in the audit trail that someone will eventually find and reject.

Documentation pack per reworked board

1. Failure mode at incoming (electrical, visual, X-ray).
2. Rework operator certification record — IPC-7711/7721 valid certificate.
3. Removal profile trace with adjacent-component temperature log.
4. Reball X-ray pre-attach (full ball array, voiding measurements).
5. Reattach profile trace with package corner thermocouple.
6. Post-reattach X-ray (joint void rates).
7. Functional test result, comparison to original spec.
8. Sign-off by rework cell lead and QA.

Test sequence post-rework

• X-ray of the reworked joint and surrounding joints (the bake/preheat cycle stresses adjacent solder joints too).
• ICT of all nets connected to the reworked part plus 2-hop adjacencies.
• Functional test against the same vector set used for original build.
• Thermal-cycle conditioning for Class 3 deliverables: 25 cycles between -10°C and +85°C (or the customer's spec). Re-test functional after conditioning.

When we won't sign off

• Visible pad lift adjacent to the rework site.
• Solder mask damage that exposes copper across more than 0.5mm.
• X-ray showing voiding above 12% on signal balls post-rework.
• Any rework that's a third pass on the same package without explicit customer authorisation.

"Class 3 means the rework looks like a build, not like a repair. If the paperwork wouldn't pass on a fresh build, it doesn't pass on a rework either." — Pioneer Horizon QA lead

If you're carrying a small inventory of failed boards from a previous build and weighing scrap vs. rework, send us the failure summary — we'll quote the rework cost per board and tell you which ones are economically worth saving.