The design side of controlled-depth drilling — which parameters the designer owns, which belong to the fabricator, and the seven callouts that decide whether the stub actually lands inside tolerance. Each callout maps onto the entry material you select and to the depth the process can hold.
Back-drilling fails in two places: at the drill, and on the drawing. The drill side is a process problem a fabricator can solve. The drawing side cannot be solved downstream — if the antipad is too small, the keepout is occupied, or the depth is referenced from the wrong surface, no amount of process control recovers it. This page covers the design half, in the order a CAM engineer has to check it.
Most back-drill arguments start because two teams assume different owners for the same number. Separating them makes the drawing review short.
| Parameter | Owned by | Why it cannot move to the other side |
|---|---|---|
| Stub length target | Designer (signal integrity) | It is a function of the data rate and the layer the signal leaves, not of the machine. |
| Antipad diameter on the drilled side | Designer (stackup and plane) | The fabricator can only drill what the copper artwork allows. |
| Keepout radius | Designer | Anything placed inside it gets cut, smeared or lifted. |
| Depth reference surface | Designer must state it; fabricator must honour it | A depth is meaningless without the surface it is measured from. |
| Depth tolerance and capability | Fabricator | It comes from the machine, the pressure foot and the stack, not from the drawing. |
| Drill diameter for the back-drill pass | Joint | It has to clear the finished hole plus registration error, and still leave the pad. |
| Inspection method and sampling | Joint | The designer needs a number they can trust; only the fabricator knows what they can measure. |
If a drawing leaves any row above unstated, the fabricator substitutes a default. That default is almost always the cheapest one that passes their own electrical test, which is not the same as the one your channel model assumes.
An antipad is the copper clearance in a reference plane around a via. On a normal via it exists so the barrel does not short to the plane. On a back-drilled via it does a second job: it gives the back-drill tool somewhere to go that is not metal.
That second job is where designs break. A plane antipad sized for electrical clearance alone is often only 0.05–0.10 mm larger than the finished hole on each side. A back-drill tool cutting into a plane at that margin will:
A working rule for the drilled side is that the antipad must be at least the back-drill tool diameter plus twice the registration allowance, and the registration allowance on a mechanical back-drill should be taken as the machine's positional capability, not the drill's nominal tolerance. In practice that lands most designs between 0.15 and 0.30 mm clearance per side, larger for thicker boards and for high layer counts where stack-up error accumulates.
The same clearance has to hold on every plane the tool passes through, not only the plane nearest the drill entry. A stack with a generous antipad at the top plane and a tight one three layers down will stop the tool at the tight one.
The keepout is the region on the back-drill side that must contain nothing the tool could damage: no pads, no test points, no exposed copper, no components, no mask openings that matter, and no silkscreen that is used for alignment.
It is left out of drawings more often than any other callout, and the consequences are specific:
| What sits in the keepout | What happens | How it is usually found |
|---|---|---|
| Test point | Drilled through or skinned; probe later reads open | At electrical test, after assembly is committed |
| SMD pad | Pad lifted or solderability destroyed by smear and burr | At first-article assembly, or never if the pad is a spare |
| Exposed copper / finish | Burr and shavings welded to the surface | Visual, if anyone looks; otherwise as intermittent leakage |
| Another via's antipad | Two barrels now share one clearance; wall thickness drops | Cross-section, or as a reliability return months later |
| Component body | Physical collision with the pressure foot or the spindle | Immediately, and loudly |
Sizing follows from the tool and the machine: the keepout radius is the back-drill tool radius plus the positional capability plus a margin for the pressure foot, which is usually much larger than the tool. Ask for the pressure-foot footprint before you finalise a tight layout, because on some machines it is the constraint, not the drill.
A back-drill depth is not a property of the hole. It is a property of the hole plus the surface the Z axis was zeroed on. Three conventions exist and all three are in use:
State the convention in words on the drawing, not only as a dimension. A dimension without a datum gets reinterpreted at CAM, and a 0.10 mm misinterpretation is enough to remove the stub entirely on one code and leave 0.3 mm on the next.
Board thickness tolerance is part of the depth budget. If the finished thickness varies by ±0.10 mm and the depth is referenced from the drilled surface, the stub varies by the same amount, even with a perfect machine. Designs that need a stub under 0.15 mm usually have to tighten laminate thickness or reference the depth to the target layer.
The stub is a transmission-line stub. Its first quarter-wave resonance sits at roughly f = c / (4 · L · √εeff). Keep that resonance well above the highest harmonic you care about and the stub stops being the limiting defect.
| Interface | Common stub target | Why that number |
|---|---|---|
| Up to 3–5 Gb/s | 0.5 mm or less | Resonance is far above the band; the stub mainly adds capacitance |
| 10 Gb/s class | 0.25 mm or less | Resonance starts to land near the third harmonic |
| 25 Gb/s class | 0.15 mm or less | Insertion-loss dip becomes visible in a TDR |
| 56 Gb/s PAM4 and above | 0.10 mm or less | Residual stub cannot be separated from via impedance by tuning alone |
These are starting points, not commitments. The number that matters is the one your channel simulation used, and it has to be written on the drawing, because the fabricator will otherwise aim at whatever their process holds comfortably.
Two designs commonly get this wrong: backplanes, where the stub on a 4 mm board is long enough that a single pass cannot reach the target, and HDI boards, where the stub is short but the depth tolerance is a large fraction of it.
Two process orders are in commercial use, and the choice is the fabricator's. It is worth knowing what you are buying either way.
| Order | What it buys | What it costs |
|---|---|---|
| Back-drill after solder mask | The mask is never inside the drilled depth; the barrel interior stays clean of resist | Mask around the hole is cut and can lift; shavings land on finished mask and are harder to remove |
| Back-drill before solder mask | Cleaner mask edge and no mask damage | Resist can enter the back-drilled cavity; removal needs a step that is easy to under-run |
If your design fills or caps the via after back-drilling, the resist-in-cavity case is the one that shows up as trapped material under the fill. Ask which order the shop uses, and if the answer is “whatever the panel needs”, get that on record before the first lot.
Yes — that is the normal case. Back-drilling removes the unused plated barrel below the target layer, and the remaining barrel above the target layer keeps plating continuity. What cannot be back-drilled is a non-plated hole: there is no barrel to trim, and the operation would only enlarge a mechanical hole.
The related design question is what happens at the transition. The tool leaves a short conical or stepped profile at the bottom of the cut, and the plating at that step is thin and easy to damage. Designs that fill the via afterwards are more tolerant of this than designs that leave it open and reflow over it.
If the layer you are drilling to is close to the surface, the practical limit is the tool: below a certain remaining depth the drill cannot be guided reliably, and the operation turns into a countersink. That is why shallow back-drills are often replaced by blind or buried via construction instead.
No single method proves both depth and remaining stub on a production panel. The useful approach is to combine a fast proxy with an occasional destructive check.
| Method | Proves | Does not prove |
|---|---|---|
| Cross-section | Residual stub, barrel wall, resin smear at the target layer — the reference measurement | Anything about the rest of the panel; it is destructive and sampled |
| Optical depth measurement (OCT / confocal) | Depth from the drilled surface to the bottom of the cut, fast, per hole | Where the target layer actually is; depth is not stub |
| 2D X-ray | Gross depth error, missing or double-drilled holes | Small residual stubs — contrast is too low |
| 3D X-ray / CT | Residual stub non-destructively, including inside the barrel | Resin smear and plating quality at the interface |
| TDR on a coupon | The electrical effect the stub actually has in the band you care about | Per-hole depth on the production panel |
If the drawing names only “back-drill to 0.25 mm stub”, the supplier can satisfy it on a cross-section and be out of tolerance on the panel. Naming the proxy method, the sampling rate and the acceptance number is what makes the requirement testable.
The back-drill entry is a fresh cut through mask, copper clearance and laminate, and it produces a burr ring at the surface plus aluminum shavings from the entry sheet. Both have to be removed before via fill or final finish.
Entry material affects all three. A flat, thin, coated entry sheet is used on the back-drill pass for the same reason it is used on the first pass: it holds the tool steady at the surface, which is where the depth reference lives. See the back-drill aluminum entry range for the sheet spec, and the back-drilling overview if you need the process sequence first.
Seven lines on the drawing remove most of the ambiguity. Add the last three if the design uses fill or a tight stub.
| # | Callout | Format that works |
|---|---|---|
| 1 | Back-drill side | Secondary side, stated in words and marked on the layer drawing |
| 2 | Vias in scope | Explicit via list or drill-table reference, not “all vias of class X” |
| 3 | Depth reference surface | Named surface plus the datum used for Z zero |
| 4 | Depth and tolerance | Nominal plus ±, with the tolerance the machine can hold |
| 5 | Stub target and maximum | Both numbers; a target alone cannot be inspected |
| 6 | Antipad on the drilled side | Diameter per via class, on every plane the tool passes |
| 7 | Keepout | Radius, plus a note that no pads or test points are permitted inside it |
| 8 | Inspection | Method, sampling rate, acceptance number |
| 9 | Post-drill treatment | Chamfer, burr limit, cleaning requirement |
| 10 | Fill or cap | Material, and whether the cavity is included |
Two practical notes. First, a depth tolerance tighter than the fabricator's documented capability will be quoted as a capability review, not as a price — better to ask for the number before you write it. Second, if the design references the depth to a named layer, say how that layer is to be located, because the answer changes the inspection cost.
The depth arithmetic itself is in the back-drill depth calculator, which converts a stackup into a depth and a residual stub. The process sequence and the reasons the operation exists at all are in the back-drilling process guide. If the defect you are chasing is a burr rather than a stub, exit burr control covers the four levers; if it is positional drift after back-drilling, hole position accuracy covers the stack side.
The entry sheet used on the back-drill pass is specified like any other entry material — alloy, temper, thickness, flatness, coating — and the material-level data is in the AA1100 H18 material profile. Procurement-side questions about how that sheet is priced are in price, MOQ and landed cost.
The antipad is copper clearance inside the plane, around the via, and it is what the tool cuts through. The keepout is the region on the drilled surface that must contain no pads, test points or components. The antipad is a hole in copper; the keepout is a rule about the surface.
By the machine, the pressure foot and the stack, not by the drawing. A mechanical back-drill on a well-supported stack typically holds a few hundredths of a millimetre; on a thin panel with poor support the same machine holds much less. Ask for the documented capability before you write a tolerance.
Yes, with 3D X-ray or CT, which resolves the remaining barrel. Optical depth measurement is faster but measures depth from the surface, not the residual stub, so it cannot replace a cross-section on its own.
Not always. Both orders are in commercial use. Drilling after mask keeps resist out of the cavity but can lift mask around the hole; drilling before mask gives a cleaner mask edge but risks resist inside the cavity. Ask the fabricator which order they run and record the answer.
Usually finished thickness variation. When the depth is referenced from the drilled surface, any change in board thickness becomes a change in stub. Referencing the depth to the target layer removes most of that variation.