Designing Against BGA Joint Cracks: Trade-offs in Board Stiffness, Pads, Placement, Assembly Support and Underfill, in 5 Stages from Measuring Strain to Verification

BGA solder joints that crack from flexure usually crack at the station where the board is bent too much or too fast: depaneling, through-hole insertion, in-circuit test, screw fastening or handling can all do it. There are five things the design side can do. Measure with strain gauges which station bends the board most; make the board bend less (board thickness, placement, bending direction); remember that pad design changes where it cracks, and that multiple reflows and more brittle laminates carry higher risk; support and calculate the loads from test, assembly and heat sinks; and know that underfill helps, but has limited effect where the board center bends too much, and makes rework harder. After changes, measure strain again at the same station and the same location.

This article covers how to reduce, from the design side, the chance of BGA joints and pads cracking from flexure during and after board assembly. How to judge the cause once a crack has occurred, thermal cycling life assessment, and comparisons of adhesives are outside its scope. It has five stages:

• Stage 1 | Measure which station bends the board most: put strain gauges beside the BGA corners and run through depaneling, assembly, test and handling.
• Stage 2 | Make the board bend less: board thickness and stiffness, placement, bending direction.
• Stage 3 | Pads and laminate: how the pad is defined changes where it cracks; multiple reflows and more brittle laminates carry higher risk.
• Stage 4 | Support in assembly and mechanics: loads from test, depaneling, screw fastening and heat sinks; supporting the board in fixtures, enclosures and handling.
• Stage 5 | Underfill and verification: when to add it, the benefit and the cost; measure again after changes.

Each stage covers how to do it and what public sources say, plus how JIAHON can help where relevant. Sources, exceptions and this article’s synthesis are in the notes at the end. This describes the general case; individual situations may differ.

Figure 1. Five stages of designing against BGA cracks (illustration). (1) Measure strain — gauge at the corners, all four; run every station: depaneling, test, handling; watch two numbers: strain and strain rate. Goal: which station bends most. (2) Bend the board less — thickness: stiffer bends less; placement: away from board edges and holes; direction: avoid the diagonal. Goal: the board layout. (3) Pads and laminate — pad definition changes where it cracks; multiple reflows raise risk; ask whether the laminate is brittle. Goal: pad and laminate specifications. (4) Support — test and depaneling: support underneath; heat sinks: load per ball; enclosure and handling: support and no stacking. Goal: fixture and mechanical drawings. (5) Underfill and verification — underfill: big improvement in thermal cycling; board center: little effect when bending is too large; measure again after changes: same station, same location. Goal: measured improvement.

Stage 1: Measure which station bends the board most

How to do it:

• Choose which components to measure first: large, flexure-sensitive BGAs; when there are many components, a measurement instrument maker’s presentation suggests choosing the three weakest and bringing in past experience (Source 4).
• Strain gauges: use rosette gauges with three directions (0°, 45°, 90°) stacked together, so the peak principal strain can be calculated (Source 4).
• Where to place them: measure all four corners of the BGA, with the gauge 6 to 10 mm from the joints and the grid lines parallel to the ball rows (Source 4).
• How to attach them: mill off the surface coating first, roughen with sandpaper if needed, clean, then bond (Source 4).
• Which stations to run: depaneling, all kinds of insertion and assembly, in-circuit test (ICT) probe pressing, and shipping are all times when the board is stressed (Source 4); also measure screw fastening, press-fit connectors and packing (this article’s synthesis).
• Sample fast enough: at least 500 Hz, with 2 kHz recommended (Source 4).
• Watch two numbers: peak principal strain, and how fast strain changes (strain rate); both affect life (Source 4).
• Who sets the limit: usually agreed between user and supplier; some makers of critical components publish their own limits (Source 4).
• Do not rely on electrical tests alone: cracking of the laminate under the pad cannot be seen by electrical or non-destructive testing, and limits set from resistance changes will be overestimated (Source 1).

Figure 2. Strain gauge placement (illustration). Place gauges beside all four corners of the BGA, 6 to 10 mm from the joints, with the grid lines parallel to the ball rows; pay special attention near depaneling lines, screw holes and connectors (gold = strain gauges; red = stress concentrations needing attention). How to attach and measure: three-direction stacked rosette gauges; grid lines parallel to the ball rows; with many parts, choose the three weakest; sample fast enough.

What public sources say: a presentation by a measurement instrument maker introducing IPC/JEDEC-9704A says that during board manufacture, bending during installation, all kinds of insertion, transient strain peaks during depaneling, and in-circuit test probes pressing too hard can all crack or detach BGA joints; measurement points are chosen at fixed corners, stiffer areas of the board (for example near capacitors, as larger components stiffen the board) and near joints; all four BGA corners are measured, 6 to 10 mm from the joints, with grid lines parallel to the ball rows; sampling is at least 500 Hz, with 2 kHz recommended; life depends on strain rate as well as peak principal strain; and limits are agreed between user and supplier, with the standard giving boundaries by strain rate and board thickness (Notes, Table 1). A paper by engineers at a networking equipment company (the page does not say which conference it appeared at) says fine cracks in the laminate under the pad can form under excessive bending or shock in assembly, test and handling, cannot be detected by electrical or non-destructive testing, and may grow later and cause failure.

Stage 2: Make the board bend less

How to do it:

• Board thickness and stiffness: the stiffer the board, the less the same force bends it; in a university vibration test, ball life on a 1.6 mm board was more than three times that on a 1 mm board (Source 3).
• Check the test conditions: that was a low-amplitude vibration resonance test, not bending or drop (Source 3); after thickening, weight, cost, and the design of holes and connectors all need recalculating (this article’s synthesis).
• Placement: keep large BGAs away from depaneling lines, board edges, screw holes and connectors, which bend most or concentrate stress during assembly (this article’s synthesis; Source 4 lists fixed corners and stiffer areas near large components as high-risk measurement points).
• Bending direction: the networking equipment company paper found that with NSMD pads throughout, bending along the component diagonal made the laminate under the pads more likely to crack; with SMD at the corners, the effect of direction was negligible (Source 1). When laying out panels, keep the main bending direction, such as depaneling lines, from running along the diagonal of large BGAs (this article’s synthesis).
• The package itself: the same paper compared a wire-bond BGA of similar size the authors had tested before; its strain limit was clearly higher than that of this stiffer flip-chip BGA. The authors therefore believe that, besides package size and ball pitch, package stiffness strongly affects the strain limit (Source 1; this is a comparison of two packages).
• Fixing points: the positions of screws and supports decide how the board bends; provide support near the BGA, avoiding one side clamped tight and the other unsupported (this article’s synthesis).

What public sources say: a university study (2022) compared two board thicknesses in a vibration resonance test (experiments on two samples of 1 mm and 1.6 mm, with a 2 mm board added in simulation). After stiffening the board, ball life rose more than threefold in both simulation and experiment. The authors state that the conclusion comes from two samples at a very low excitation amplitude (0.5 G), cite other studies indicating the failure mode may differ at larger amplitudes, and suggest further study of failure modes at different amplitudes (Notes, Table 2). The networking equipment company paper says that with NSMD pads throughout, bending along the component diagonal increases the tendency for laminate cracking under the pad (with SMD at the corners, the effect of direction was negligible); comparing a wire-bond BGA the authors had tested before, it concludes package stiffness strongly affects the strain limit.

How JIAHON can help:

• For whether boards are flat and carriers are warped, JIAHON’s published article “White-Light Structured-Light Thickness and Deformation Measurement | Non-Contact Inspection of Carrier Warpage, Board Warpage and Paste Thickness” describes a non-contact method for measuring height differences across whole carriers and boards (the page says the system currently measures at room temperature).

Stage 3: Pads and laminate

How to do it:

• Two kinds of pads: when the solder mask opening is larger than the copper pad and the pad edge is exposed, it is NSMD (non-solder-mask-defined); when the solder mask covers the edge of the copper pad, it is SMD (solder-mask-defined).
• The benefit of NSMD: a semiconductor company’s application note recommends NSMD, because the solder can grip the pad area plus its sides (Source 5).
• The cost of NSMD: in the networking equipment company paper’s tests, with NSMD throughout, all bending failures were cracking of the laminate under the pad (Source 1).
• Changing the corners to SMD is not necessarily better: in the same paper, changing the corners to SMD turned failures into a mix of intermetallic fracture and laminate cracking, and the strain at electrical failure was clearly lower (Source 1).
• Multiple reflows: in the same paper, after three reflows, both laminate cracking and intermetallic fracture occurred more easily (Source 1); double-sided boards and rework both add reflows (this article’s synthesis).
• Laminate: phenolic-cured laminates used for lead-free processes are more brittle than the dicyandiamide-cured laminates used for traditional tin-lead processes (data cited by Source 1); ask your board maker about this when choosing a laminate (this article’s synthesis).
• Corner balls: some packages make the corner balls non-functional, so a crack starting at a corner does not cause immediate failure; you can include this in your comparison when choosing components (this article’s synthesis).

What public sources say: a semiconductor company’s application note on BGA board design recommends NSMD pads, because etching copper pads is easier to control than solder mask openings and the solder can attach to the pad area and its sides. The networking equipment company paper ran four-point bend tests on a 40 × 40 mm flip-chip BGA with 1 mm ball pitch and found that pad design determined the failure mode: with NSMD throughout, all were laminate cracking; with SMD at the corners, a mix of intermetallic fracture and laminate cracking, with clearly lower electrical failure strain. With NSMD throughout, three reflows and bending along the diagonal both increased the tendency for laminate cracking; with SMD at the corners, the effect of direction was negligible, and multiple reflows also lowered failure strain. It also fitted a Weibull distribution to failure strains detected by acoustic emission on samples after three reflows and estimated at a 1% failure rate: for the three groups of NSMD throughout (along the edge and along the diagonal) and SMD at the corners along the edge, strain limits were all 1250 to 1350 microstrain; just the group with SMD at the corners along the diagonal was about 850 microstrain, which the authors suggest was dominated by brittle intermetallic fracture (Notes, Table 2).

Stage 4: Support in assembly and mechanics

How to do it:

• Test and depaneling: when the in-circuit test press comes down and during depaneling, support the board under the BGA so the load does not turn into bending (Source 4 lists these two stations as times of stress; the support approach is this article’s synthesis).
• Screw fastening and press-fit: support the underside of the board during screw fastening and press-fitting connectors too (this article’s synthesis).
• Heat sink loads: the semiconductor company’s application note says that for its own BGA packages, a one-time compressive load (such as fitting a heat sink) should not exceed about 20 g per ball, and a sustained load about 4 g per ball, provided the board is well supported and does not bend or bow (Source 5).
• Support inside the enclosure: once the product is in its enclosure, support posts or ribs near the BGA can make the board bend less when pressed or dropped; but if they push on one side alone, local bending may get worse, so measure again after changes (this article’s synthesis).
• Handling and transport: when boards sit in trays, carriers or carts, keep them flat and supported and do not stack them under load (this article’s synthesis; Source 4 lists shipping as a stress condition).
• Measure again after changes: after changing fixtures or mechanics, measure again at the same station with the same gauge locations (this article’s synthesis).

What public sources say: the measurement instrument maker’s presentation lists transient strain peaks during depaneling, in-circuit test probes pressing too hard, and static loads and vibration in shipping as conditions that stress the board. The semiconductor company’s application note says that for its own BGA packages, one-time loads such as heat sinks should be within about 20 g per ball and sustained loads within about 4 g, provided the board is well supported, and notes that these are reference values that may need adjustment (Notes, Table 3).

How JIAHON can help:

• When test, depaneling or assembly fixtures need to be custom-made, JIAHON’s “Process & Fixture Design” page says: “If a standard product can’t solve it, let us evaluate, design and build it.” Its “How to start” section lists: “Start with one sentence or a phone photo,” “Tell us a few numbers” (size, quantity, operating environment, where you are stuck), “Engineering replies whether it can be done and how,” and “If needed, validate on a small scale before scaling up.”
• For trays used in handling, JIAHON’s “Tray Doesn’t Fit the Spec or Isn’t Deep Enough? JIAHON’s Custom Design Improvements” says “a JIAHON affiliated company’s engineering department can get involved from the structural design stage.”
• Designs JIAHON has made for customers are collected under “Design Cases”: “Things JIAHON designed and built for customers: each article explains the background and outcome, with measured figures.”

Stage 5: Underfill and verification

How to do it:

• Reduce bending first: an article co-written by an underfill maker and a semiconductor company recommends in its conclusion that the board’s mechanical design and product structure be considered in detail to keep bending low (Source 2; putting this step before adding underfill is this article’s synthesis).
• What underfill does: in the same article, under thermal cycling (−40 to 125 °C), the first failure was extended from 358 to 2,254 cycles; drop improved clearly in the two zones other than the board center, and bending in one zone went from thousands to tens of thousands of cycles (Source 2).
• Where the effect is limited: at the board center, where bending amplitude is highest, adding underfill did not clearly improve drop and bending; the article says beyond a certain stress, underfill helps almost nothing (Source 2).
• The cost: with underfill, the component is very hard to rework; the process gains a dispensing and curing step, and you need to confirm that the adhesive and flux residue do not interact (this article’s synthesis).
• Corner or edge bonding: some apply adhesive at the corners or along the four edges alone, which is easier to inspect and rework than full underfill; verify the effect yourself (this article’s synthesis).
• Verification: after changes, measure strain again at the same stations and locations as in Stage 1 and compare with the agreed limit; run drop, bending and thermal cycling reliability tests per the product specification (this article’s synthesis).

Figure 3. Four countermeasures, each with benefits and costs (illustration, compiled from Sources 1 to 5). Board thickness and stiffness — benefit: bends less under the same force; cost: weight, cost, and holes need recalculating. Pad definition — benefit: you can choose where it cracks; cost: a different failure, not necessarily better. Assembly and mechanical support — benefit: loads do not become bending; cost: supporting one side alone may bend more, so measure again. Underfill — benefit: big improvement in thermal cycling; cost: hard to rework, and little effect at the board center.

What public sources say: a trade magazine article co-written by an underfill maker and a semiconductor company (URL dated February 2005) compared small BGAs with 0.5 mm pitch and 84 balls (10 per board) with and without underfill: thermal cycling used 3 boards per condition, and drop and bending 1 board per condition. Under thermal cycling at −40 to 125 °C, first failure was extended from 358 to 2,254 cycles and characteristic life from 556 to 11,001 cycles. In drop and bending, improvement in the board center zone was not clear; drop improved clearly in the other two zones, and bending in the second zone went from thousands to tens of thousands of cycles. The conclusion recommends considering the board’s mechanical design and product structure in detail to keep bending low, and reducing thermal expansion differences between joints, components and board (Notes, Table 3).

How JIAHON can help:

• For underfill pretreatment, voids and release criteria, JIAHON’s published article “Flip-Chip Underfill in Production | Plasma Pretreatment, Void Countermeasures and Release Criteria” covers this using flip-chip packages as the example; for underfilling BGAs on boards, follow the adhesive maker’s data.

Related reading

To cover the related process steps, continue with these articles:

• “Tray Doesn’t Fit the Spec or Isn’t Deep Enough? JIAHON’s Custom Design Improvements”: a tray structural design improvement case
• “White-Light Structured-Light Thickness and Deformation Measurement | Non-Contact Inspection of Carrier Warpage, Board Warpage and Paste Thickness”: measuring carrier warpage, board warpage and paste thickness without contact
• “Flip-Chip Underfill in Production | Plasma Pretreatment, Void Countermeasures and Release Criteria”: pretreatment, voids and release criteria for flip-chip underfill

Notes: sources and exceptions

These notes are for readers who want to check the sources; you can follow the steps above without them. Where your product specifications, component and board makers’ specifications, or the strain limits agreed with your customer differ from this article, they take precedence.

Sources for each stage

Table 1 | Strain measurement

Sources: see each row.

Table 2 | Bending less, pads and laminate

Sources: see each row; Source 1’s figures apply to that 40 × 40 mm flip-chip BGA and that board alone.

Table 3 | Support and underfill

Sources: see each row.

Standards and references

Refer to the current version published by each issuing body for the standards cited here.

• IPC/JEDEC-9704A, printed circuit board strain measurement guideline (cited by Source 4; this article did not read the full standard). Issuers: IPC; JEDEC Solid State Technology Association
• IPC/JEDEC-9702, monotonic bend characterization of board-level interconnects (Source 1’s four-point bend test followed it; this article did not read the full standard). Issuers: IPC; JEDEC Solid State Technology Association

Exceptions and limitations

• What was and was not read: this article draws on 5 public sources (a paper by networking equipment company engineers, a trade magazine article, a university study, a measurement instrument maker’s presentation and a semiconductor company’s application note), plus JIAHON’s Process & Fixture Design page, Design Cases category and published articles. It did not read the full texts of IPC/JEDEC-9704A and IPC/JEDEC-9702; for strain limits and measurement methods, follow the full standards and the component makers’ and customers’ specifications.
• Perspective of the sources: some authors of Source 2 work for a company that sells underfill; Source 4 is a company that sells strain measurement equipment; Source 5 is a company that sells components.
• Figures represent those tests alone: Source 1’s strain limits apply to that 40 × 40 mm flip-chip BGA and that board alone; Source 2 used small BGAs with 0.5 mm pitch and 84 balls, with 1 board per condition for drop and bending; Source 3 is a vibration test on two samples at low amplitude and cannot be applied directly to bending or drop; Source 5’s load per ball is that semiconductor company’s reference value for its own BGA packages.
• Two sources look from different angles: Source 5 looks at solder attachment and recommends NSMD; Source 1 looks at laminate cracking — with NSMD throughout, all failures were laminate cracking, while SMD at the corners changed them to a mix of intermetallic fracture and laminate cracking with lower electrical failure strain — and does not recommend either. Confirm which to use with your package and board makers.
• This article’s synthesis: measuring screw fastening, press-fitting and packing in Stage 1; the cost of thickening, placement, bending direction and fixing points in Stage 2; reflow count, asking the board maker and corner balls in Stage 3; the support approach, support inside the enclosure, handling and measuring again in Stage 4; and in Stage 5, the order of reducing bending before adding underfill, the cost, corner or edge bonding and verification, are this article’s synthesis. Every other statement has a matching passage in the sources or on JIAHON’s pages.
• Scope: this article covers bending during and after assembly; how to judge the cause once a crack has occurred, thermal cycling life assessment and comparisons of adhesives are outside its scope.
• JIAHON’s pages: “How JIAHON can help” quotes published pages on the site.

This article summarizes the general case. It is not a restatement of any standard, not an endorsement of any component maker, adhesive maker or measurement equipment supplier, and not a promise about your product or process. Follow your product specifications, component and board makers’ specifications, the strain limits agreed with your customer, and your own validation; individual situations may differ.

Request a quote

Request a quote →