GPU Reballing:
Complete Rework Procedure

Board Preparation & BGA Removal

GPU reballing demands meticulous substrate cleanup before any rework begins. Start by identifying the GPU package size—most discrete GPUs use BGA-1024 (1.0 mm pitch) or BGA-1156 (1.27 mm pitch). RTX 40-series cards, Apple M-series chips, and AMD RDNA processors typically fall into the smaller pitch category.

Desoldering the GPU requires a controlled heating profile to avoid substrate warping and lamination separation. Set your rework station BGA rework oven (or IR preheater if available) to reach the solder glass transition temperature (Tg) gradually:

• Preheat zone: 140–160°C for 60–90 seconds
• Thermal soak: 180–200°C for 90–120 seconds
• Reflow peak: 245–260°C for 10–15 seconds (lead-free SAC305)
• Ramp rate: 2–4°C per second (avoid >5°C/s to prevent thermal shock)

Once the GPU reaches reflow, use a hot air pencil or vacuum tweezers to carefully lift the component. Verify all solder joints have melted by checking for wetted contact pads. If resistance persists, apply additional flux and ramp to peak again rather than forcing the part.

Reballing Application & Ball Placement

BGA reballing leverages solder sphere stencils and flux to rebuild the solder ball array. Select solder balls matching the original composition—lead-free SAC305 (96.5% tin, 3% silver, 0.5% copper) is industry standard for modern GPUs.

Materials & Setup

• Solder balls: 0.45–0.76 mm diameter (depending on pitch) with ≤0.3 mm diameter variation
• Flux: Water-soluble flux or no-clean (Kester #186 RMA or equivalent). Avoid acid-core flux on RF-sensitive substrates.
• Stencil: Laser-cut 0.1–0.125 mm thickness, aligned to ±0.1 mm on substrate pads

Apply flux to the entire pad array using a syringe or brush. The flux volume is critical: too little causes ball rolling; too much creates shorts and bridges. Target 0.25–0.35 mm³ per pad (roughly a 0.5 mm diameter paste volume).

Position your stencil and secure with registration pins. Using fine-tipped tweezers (fine enough to grip individual solder balls), place balls onto each flux-wetted pad. Work in organized rows from center outward to avoid ball displacement from neighboring placements.

Visual Inspection Checkpoints

After placement, inspect under magnification:

• All pads must have exactly one solder ball per pad—no doubles, no empties
• Balls must rest in flux without rolling or sliding
• No ball-to-ball contact (minimum 0.15 mm clearance at BGA-1024 pitch)
• Flux coverage extends slightly beyond each ball perimeter

Reflow Cycle & Joint Quality Validation

GPU reballing reflow follows a tighter thermal profile than standard BGA rework, since solder balls must merge with substrate pads without balling up or rolling. Use a thermal profiler with thermocouple on the substrate board during your first three reballing cycles.

Recommended Lead-Free Reflow Profile

Allow the board to cool naturally in still air to 60°C before handling. Forced cooling or ambient air fans can introduce thermal shock and cold-joint formation.

Post-Reflow Validation

After reflow cooldown, inspect all solder joints under magnification (at minimum 20×):

• Visual appearance: Each ball must form a cone or hemispherical shape, shiny and smooth (not dull or granular)
• Wetting: Solder must wet >50% of the pad perimeter; <25% wetting indicates insufficient flux or pad oxide contamination
• Bridging: Adjacent balls must have clear separation; solder bridges >0.05 mm require rework
• Voids: Subsurface voids >20% of joint volume indicate flux entrapment or inadequate thermal profile

Implement X-ray inspection (2D or preferably 3D cross-section) for critical arrays. GPU power delivery domains (PPBUS, PPVIO, PPCORE) demand zero-tolerance on opens and bridges—intermittent connection will cause thermal cycling failures.

Electrical Verification & Thermal Load Testing

Reballed GPUs require progressive electrical testing before return to service. Power measurements on PPBUS_GPU rail typically deliver 12–15V under full load; PPCORE and PPVIO rails operate at 0.7–1.1V. Use a precision multimeter (minimum 3.5-digit) at test points near the GPU to verify voltage stability under idle and 50% load conditions.

Perform continuity testing on all critical power rails using the PCB schematic to identify test points. Verify resistance <0.1 Ω between adjacent pads on same voltage net to confirm no open circuits at solder interface.

Once electrical baseline passes, load test the GPU at 50% utilization for 15 minutes using a synthetic workload (gaming benchmark, stress test, or compute kernel). Monitor die temperature using available thermal sensor firmware (if GPU firmware is functional) or infrared thermography. Target temperature should remain stable; thermal drift >5°C per minute during load indicates poor thermal interface or junction defects.

Document all test results—thermocouple traces, voltage readings, continuity verification—before releasing the board. Maintain reflow profile data and X-ray cross-section images for failure analysis if issues emerge post-repair.