PCB Mechanical Stress and Dimensional Stability: What Manufacturing Owns

The conclusion first: stress on a board is not one thing, it is three. Fabrication locks a portion of it into the board before it ships. Depaneling, screwing it down, mating connectors and clamping heatsinks add a second layer on the assembly floor. Once the unit runs, thermal cycling loads both of those repeatedly. The ownership boundaries are different for each — the first is something a fabricator can genuinely reduce, the second can only be avoided through mechanical design and assembly sequence, and the third is a long-term behaviour set by materials and structure together. Collapsing all three into "the board warped, have the fab sort it out" usually produces a request nobody can act on.

So this article does one thing: separate the three, state where each is handled, and give a diagnosis order for when a warped or cracked board lands on your desk.

1. Source one: stress locked in during manufacturing

A multilayer board's shape is not machined, it is fixed in place during lamination. Resin softens under heat, flows under pressure, then cures as the stack cools. The internal state at the moment of cure is the stress distribution the board carries out the door. Every later process step only adds to or subtracts from that baseline.

What pushes locked-in stress higher:

• Asymmetric copper distribution. When the copper on one side differs substantially from the other, resin shrinkage pulls unevenly and the board acquires a bias to bow. It is the most common contributor and the easiest to prevent at design time; the mechanism is covered separately in multilayer copper balance and warpage control and is not repeated here.
• Asymmetric stackup. Dielectric thicknesses, core placement and copper layer sequence that are not symmetric about the neutral axis produce the same effect as copper imbalance — the origin just sits in the stackup table rather than in the copper pour.
• Mixed material systems. Different resin systems shrink differently. Press them together and each layer wants to settle at a different dimension, leaving residual stress at the interfaces. Our selection criteria are in base material selection criteria.
• Non-uniform lamination itself. Uneven pressure or temperature means different zones of the same board cure in different states. Taken far enough this stops being a stress question and becomes voids and delamination — see lamination void and delamination prevention.

This source is the fabricator's responsibility, and it is where the fabricator actually has levers. Those levers are stackup symmetrisation, copper balancing, lamination profile adjustment, and scheduling stress relief between process steps so the board releases part of its internal stress before it is cut, drilled and etched. What they share is that they all happen before the board leaves the plant. Once it reaches the assembly line, the locked-in state is final.

2. Source two: depanel and assembly loads

The second source is not inside the board, it is applied from outside. Its signature is that both the location and the moment are specific, and a single event is often enough to cause irreversible damage.

Where it gets applied:

• Depaneling. Separating a board from its array bends or shears the region near the board edge. If the separation method was not coordinated with component placement, ceramic parts and BGA corner joints near that edge are the first to feel it. Array constraints are covered in panelization and array design constraints, and the tolerance language for outline work in outline routing and V-cut tolerance.
• Screwing it down. Tightening at the mounting holes presses the board against the chassis or standoffs. If the chassis bearing surfaces are not coplanar, the screws are using the board to absorb the chassis error — the board is forced flat and keeps the stress.
• Mating connectors. Press-fit and high-insertion-force connectors concentrate load in a local region during mating, and the board flexes there.
• Heatsinks and heavy parts. Spring-clip heatsinks, bulk capacitors and transformers apply a standing load after assembly and become stress concentrators under vibration.

None of this is something the fabricator can solve. The fabricator can do two things only: deliver dimensions and flatness within the agreed window, and avoid introducing extra stress on its side. The rest belongs to mechanical design — support point layout, keepouts around mounting holes, assembly sequence (locate first, torque second, rather than driving each screw home in turn), and moving sensitive parts away from separation edges.

A premise that routinely gets skipped: if chassis or standoff flatness is not itself controlled, then no matter how flat the board is, mounting will force it into the chassis's shape. Tightening the board's flatness requirement further buys nothing in that situation, because the problem is not on the board.

3. Source three: thermal cycling in service

The third source lives on the time axis. Power up, warm up, power down, cool down — each round makes every material in the board move by a different amount according to its own expansion behaviour. Copper, resin, glass fibre, solder and component bodies do not agree with each other, and that disagreement becomes repeated loading at the interfaces.

Its relationship to the first two is additive, not independent:

• Regions with high locked-in stress reach a failure threshold sooner under cycling.
• Regions held in place at assembly get cycled while already constrained.
• So thermal cycling rarely invents new failure locations. It turns the weak points left by the first two sources into cracks.

Typical locations: hole walls and interconnect structures, solder joints (particularly the corners of stiff packages with no compliance underneath), areas near board-to-chassis constraint points, and rigid-flex transition zones. Construction boundaries for rigid-flex are covered in rigid-flex vs flex construction boundaries.

Ownership: material system and mechanical design set the long-term endurance; manufacturing sets how clean the starting point is. A fabricator cannot promise a thermal cycling life, but it can guarantee the board does not start out already in debt.

4. How to specify dimensional stability on a drawing

"Dimensional stability" is not a single number in engineering language. It has to be broken into several separately measurable requirements — and the thing most often missed is that each of them is measured in a different state.

What to state separately:

1. Outline and hole position tolerance. Standard practice, but state the datum — which reference edge, which reference hole. Without a declared datum the same drawing yields different measurements. Tolerance notation is covered in manufacturing tolerances in drawing notation.
2. Flatness requirement, with the measurement state. Free-state or constrained measurement give different answers. A flatness number with no stated measurement condition hands the verdict over to whatever the inspector habitually does.
3. Finished thickness. Thickness ties back to stackup and resin flow; our approach is in finished thickness tolerance control.
4. Layer-to-layer registration. Dimensional stability on a multilayer board is not only the outline — it includes where the layers sit relative to each other. See multilayer lamination registration control.
5. Identify the critical regions. If only certain areas are dimensionally sensitive (a mounting hole group, the connector zone, optical fiducials), mark them — that is both more achievable and cheaper than tightening the whole board.
6. Permit stress relief. If the process needs a bake or equivalent treatment, the drawing or spec should allow it. Otherwise a fabricator may skip it rather than risk violating the documentation.

Overall tolerance language and our controlled items can be aligned on the tolerance page; inspection and release practice sits on the quality page.

5. Diagnosing a warped or cracked board

The order matters: state first, then pattern, then process. Reverse it and you end up in the "we changed the material and it still warps" loop.

Step 1: in what state did it warp. Did it come out of the packaging warped, or warp after depaneling, after assembly, or after thermal cycling. This single answer decides which of the three sources you investigate.

Step 2: what shape is the deformation. A bow in one direction, a diagonal twist, or a local bulge or dent — these three point at different origins. Overall bow usually relates to symmetry; twist more often to material behaviour or uneven lamination; local deformation more likely to applied load or local structure.

Step 3: is the whole lot like this. Every board of the same part number showing the same shape in the same direction points to a systematic design or process factor. A scattered few point to process variation or a one-off handling event.

Step 4: are the cracks near constraint points. Cracking concentrated around mounting holes, chassis support points or separation edges is the signature of source two and has little to do with locked-in stress.

Step 5: only now look at the process. With the scope narrowed by the first four steps, review stackup symmetry, copper balance, lamination records, and whether the material system changed for that lot. Material implications are summarised on the materials page.

One decisive test: if the same part number stops warping after the assembly scheme changes, it was never a manufacturing problem. Conversely, if boards stay deformed after being removed from the chassis and the whole lot agrees, the investigation belongs on the fabrication side.

6. What only design can fix

A few items cannot be recovered by any amount of process adjustment and have to be settled at design time:

• Stackup and copper symmetry. Once the board is released, the copper distribution is fixed.
• Mounting hole location and surrounding keepouts. A hole in a stress-sensitive area can only be mitigated later, never relocated.
• Distance from sensitive components to separation edges. The magnitude and location of depanel stress is a layout question, not a process question.
• Rigid-flex transition treatment. The construction at the transition is decided in design.
• How heavy parts are supported. Carried by the board or carried by the chassis is a mechanical decision.

The process-side boundaries we can offer are listed on the capabilities page. Reviewing these items together during design removes a lot of what would otherwise surface at assembly.

FAQ

Can the fabricator fix a warped board?

It depends which source. Stress locked in during manufacturing can genuinely be reduced through stackup symmetrisation, copper balancing, lamination profile work and stress relief. Loads applied during assembly cannot — those need support point layout, mounting hole keepouts and assembly sequence. Identify the source before deciding whose problem it is.

Is a tighter flatness requirement always better?

No. Without the measurement state declared (free or constrained), a strict number will still be measured differently by different methods; and if the chassis bearing surfaces are uncontrolled, a flatter board will still be forced into the chassis's shape. Marking the genuinely sensitive regions is more effective than tightening the whole board.

Are bow and twist the same problem?

No, they point at different origins. A board bowing in one direction relates more to top-to-bottom symmetry; a diagonal twist relates more often to material behaviour or non-uniform lamination. Reading the deformation shape first eliminates a lot of wasted investigation.

Does stress relief need to appear on the drawing?

It should at least be permitted. If the specification leaves no room for it, the fabricator may omit the step to stay compliant — and that step is exactly how part of the locked-in stress is released before cutting and drilling. Write it as permitted rather than prescribing parameters, and leave the method to process engineering.

What does cracking near mounting holes tell me?

Usually that this is applied stress at a constraint point rather than the board's own internal stress. The attention belongs on bearing surface flatness, torque sequence and structural keepouts around the hole, not on adjusting fabrication steps.

Should the same board carry identical dimensional requirements across different machines?

Not necessarily. The real requirement comes from the assembly interface — in a different chassis with a different support layout, the sensitive regions may be completely different. Marking critical regions per assembly scheme beats applying one blanket tight requirement to the whole board.