Back-drilling exists because the unused copper barrel left on a through-hole via — the "stub" — resonates at 25G/56G SerDes rates and wrecks signal integrity. Before you re-spin the stack-up, ask why back drill PCB practice is mandatory above 10 Gbps: your channel can fail the eye diagram even when the laminate is low-loss and the traces are impedance-controlled. The real culprit is usually not the stack-up — it is that dead stub dangling off the via below the last signal connection. In this article we break down what a via stub is, the quarter-wave resonance that closes the eye, the rate and thickness thresholds that make back-drilling non-negotiable, and the part most SI guides skip: the entry sheet above the panel quietly decides how much stub you leave behind.
Key Takeaways - A via stub is the unused plated barrel left below the last signal connection; at 25G/56G it resonates at f = c / (4 · L_stub · √εr_eff) and carves a deep insertion-loss notch that compresses the eye. - Back-drilling removes the stub, cutting residual length to roughly 0.15–0.25 mm and pushing the first resonance above 100 GHz — recovering about 3–6 dB of loss at 12.5 GHz. - Rule of thumb: back-drill once links reach ~10 Gbps NRZ or 56G PAM4, and on any board thicker than ~1.6 mm with signals running through deep vias. - The back-drill depth is measured from the top of the entry sheet, so sheet thickness variation becomes stub variation — the hidden, controllable link most specs ignore. - Specifying the right back-drill entry material (one-piece coated aluminum) is what turns a good depth program into a tight, repeatable stub budget.
What a Via Stub Actually Is
A plated through-hole via is drilled through the entire board and plated with copper so any layer can connect to it. But a given high-speed signal usually touches that barrel on only two or three layers. Everything below the last connection — the copper that keeps going to the far surface — is electrically dead. That dead appendage is the via stub.
Picture a 16-layer server board. A 25G pair drops from layer 1 to layer 3 through a through-hole via; the barrel then continues unused to layer 16. The signal only cares about the first few layers. The copper below it is a stub dangling off the trace — and the longer the board and the shallower the connection, the longer that stub.
Two numbers describe a stub, and they are not the same:
- Stub mean length — where the back-drill typically stops, on average.
- Stub length variation — how much that stop point scatters across holes, panels, and lots.
Most real-world failures are variation failures. A drill that hits target on average but wanders by ±0.08 mm still leaves stubs long enough to break the eye on the fastest lanes. We will come back to why that wander often starts at the sheet above the panel, not the drill. This is the heart of via stub removal pcb work: you are killing a resonant parasite, not polishing a via.
Illustrative scenario — a 25G backplane that "failed SI review" for the wrong reason. (Illustrative; drawn from common back-drill symptoms, not a named customer.) A contract manufacturer built a 24-layer 25G backplane and watched the eye diagram collapse on several lanes. The team re-spun the stack-up twice, swapping laminates and re-tuning traces, with no recovery. A cross-section told the real story: signals routed to mid-board layers left stubs of roughly 2.5–2.8 mm. At FR-4 (εr ≈ 4.0) that puts the first resonance near 13–15 GHz — inside the 25G Nyquist band. The stack-up was never the problem. Once they back-drilled the critical vias to a sub-0.25 mm residual, the eyes reopened without a laminate change.
Why Stubs Wreck 25G/56G SerDes Signals
A stub is not just wasted copper. It is an unterminated transmission-line section. When its electrical length hits one quarter of the signal wavelength, it presents a near-short at the via junction and swallows energy at that frequency — and at the odd harmonics above it.
The first resonant frequency is:
f_res = c / (4 · L_stub · √εr_eff)
where: - c = speed of light ≈ 3 × 10⁸ m/s - L_stub = stub length in meters - εr_eff = effective dielectric constant of the via environment (≈ 3.5–4.2 for FR-4 with mixed copper and dielectric)
At 25G NRZ the Nyquist content sits around 12.5 GHz; at 56G PAM4 it climbs past 14 GHz, and 112G PAM4 pushes beyond 28 GHz. The moment your stub's quarter-wave notch lands in that band, the channel develops a deep null exactly where your signal lives — the core of pcb back drilling signal integrity failures on every fast serial link.
Independent published stub-resonance tables lays out the practical mapping for FR-4 at εr ≈ 4.0:
| Stub length | First resonant notch | Effect on data rate |
|---|---|---|
| 3.0 mm | 12.5 GHz | Kills 25G+ and 56G/112G channels |
| 2.0 mm | 18.8 GHz | Marginal at 25G; kills 56G PAM4 |
| 1.0 mm | 37.5 GHz | Acceptable for 25G NRZ; marginal at 56G |
| 0.5 mm | 75.0 GHz | Safe even for 112G PAM4 |
| 0.2 mm (back-drilled) | 187.5 GHz | Safe for all current protocols |
The damage is not only at the notch. A stub degrades insertion loss across the whole bandwidth. Published measurements show a 2 mm stub adds roughly –3 to –6 dB at 12.5 GHz, while a 1 mm stub adds about –1 to –2 dB; a back-drilled 0.2 mm stub contributes almost nothing. On a channel spending its loss budget, that few-dB hit is the difference between an open eye and a closed one.
Stub Resonance vs Data Rate — the quick reference
Read the table the other way — the failure threshold by channel:
- 10G NRZ / 25G NRZ — a stub near 3.0 mm (≈12.5 GHz notch) lands inside the band and breaks the link.
- 56G PAM4 — even a 2.0 mm stub (≈18.8 GHz) is dangerous; you need to be well under 1.0 mm.
- 112G PAM4 — anything above ~0.5 mm starts to bite; only a back-drilled residual stays safe.
That is the whole argument for back drill 25g serdes and 56G work: move the resonance out of every bandwidth your product will ever use, not just the one you tested.
When Back-Drilling Becomes Mandatory
There is no single magic line, but the industry has converged on clear triggers. Back-drilling is the cost-effective default once you cross them; blind or buried vias cost 50–200% more to achieve the same stub removal (per independent industry cost studies).
Rate thresholds - ≥ 10 Gbps NRZ — back-drilling typically becomes necessary. - ≥ 25 Gbps NRZ — effectively mandatory on through-hole vias. - ≥ 56G PAM4 — definitely required; stubs must be driven small. - 112G PAM4 — only a back-drilled residual stub keeps the resonance out of band.
A practical design rule: if your via stub is longer than λ/10 at the Nyquist frequency of your fastest signal, back-drill it. That single check answers when to use back drilling pcb better than any fixed number.
Board and stack-up triggers - Board thickness > ~1.6 mm with signals on outer layers passing through deep vias — the stub is inherently long. - Tight loss budgets (less than ~5 dB of margin) — even 1–2 dB of stub loss is unaffordable. - High layer count (server, backplane, 16–68 layers) where fast signals connect to mid or deep inner layers.
When back-drilling is rarely needed: data rates under ~5 Gbps, thin boards under ~1.6 mm where stubs are naturally short, or designs that already use HDI blind/buried vias.
One more spec point: the back-drill bit is typically 0.2–0.3 mm larger than the original through-hole so it fully clears the copper barrel even with registration tolerance, and the standard production depth tolerance is about ±0.10 mm (±4 mil), leaving a residual stub target of roughly 0.15–0.25 mm (5–10 mil). For trigger-to-material mapping, see our back-drill PCB material selection guide.
Illustrative scenario — depth was fine, stub still scattered. (Illustrative; drawn from common back-drill symptoms, not a named customer.) A telecom-line-card shop held back-drill depth to ±0.05 mm on the machine and still saw marginal lanes slip through lot to lot. The defect was not the drill. Their entry sheet for the reverse pass was a repurposed first-pass foil whose thickness varied across the panel; because the back-drill depth is counted from the top of that sheet, every micron of sheet variation became a micron of stub variation. Switching to a one-piece coated aluminum entry held to a tight gauge narrowed the stub scatter enough that the marginal-lane scrap dropped and the depth program stopped needing per-lot re-tuning. The drill never changed. The sheet above it did.
The Hidden Link — Your Entry Sheet Controls the Stub Too
Most "why back-drill" posts stop at the resonance formula. Here's what they miss: back-drilling runs from the opposite side of the panel, so the drill's Z-depth counter starts at the top of the entry sheet on the flipped panel — that sheet is your Z reference plane. We break down the full tolerance chain and the one-piece design that controls it in our back-drill depth control guide.
The stub you leave is set by two things, and only one gets spec'd: the drill's depth setting, and the sheet's thickness and its variation. A uniform, dimensionally stable, one-piece coated sheet gives a repeatable Z-zero; a variable or laminated sheet quietly adds scatter you won't see until the cross-section. That is the real differentiator between a back-drill entry sheet and a standard aluminum entry board used on the first pass. Our back-drill aluminum entry sheets are built exactly for that second job: thin (0.16–0.19 mm), one-piece, coated, and held to a tight gauge.
Ready to see how the sheet above the stack moves your stub budget? Learn how entry-sheet thickness controls back-drill depth tolerance — and why "one-piece + margin-reserved" aluminum beats a laminated sheet on variation. For the material choice itself, see choosing between aluminum and phenolic back-drill entry material.
Back-Drilling in 5G, Servers & Comms — Why Material Choice Scales
Back-drilling volume is climbing because the boards that need it are everywhere — 5G radios, server motherboards, switch and router line cards, data-center backplanes all run links fast enough that stubs must be controlled. That demand, documented in our 5G back-drill demand growth insight, is why a purpose-built entry sheet is now a procurement item, not an afterthought.
Scale changes the math. On a pilot run you can hand-tune depth and live with scatter. Across thousands of panels a month, stub variation becomes scrap, then cost. The entry sheet is the one variable you can spec before the drill runs — uniform thickness, documented coating CpK, one-piece structure with no bond line. Those are exactly the properties our coated aluminum entry sheets and the one-piece coated back-drill aluminum entry sheet are built around.
Illustrative scenario — scaling exposed the sheet, not the drill. (Illustrative; drawn from common back-drill symptoms, not a named customer.) A server-PCB line ramping a 56G PAM4 card passed first-article with clean cross-sections, then watched yield drift as volume climbed. The drill program never changed — but batch-to-batch entry-sheet supply did, mixing laminated foils with inconsistent gauge. Locking the line to one one-piece coated aluminum grade with documented coating CpK stopped the stub scatter from tracking supplier batch. At scale, the sheet above the stub is a process control, not a commodity.
If your line is fighting exit burrs or copper nubs on the reverse pass, the cause is often the same sheet story from the other side — see how entry-sheet specs decide back-drill exit burr and yield.
Frequently Asked Questions
At what data rate do I need back-drilling? Once serial links reach about 10 Gbps NRZ, back-drilling typically becomes necessary; at 25G NRZ it is effectively mandatory on through-hole vias, and at 56G PAM4 it is required. A practical rule is to back-drill any via whose stub exceeds λ/10 at the Nyquist frequency of your fastest signal. The real trigger is "stub resonance inside your bandwidth," not a single number on a datasheet.
Is back-drilling required for PCIe Gen5? Yes, in practice. PCIe Gen5 runs at 32 GT/s with Nyquist content near 16 GHz, and a long through-hole stub on a thick board resonates right in that band. The same independent analysis that flags 24-layer boards with ~2.8 mm stubs at ~13.4 GHz notes this falls inside the PCIe Gen5 Nyquist window. Unless your design uses HDI blind/buried vias or routes Gen5 on very shallow layers with naturally short stubs, plan to back-drill.
What stub length is acceptable after back-drilling? Production back-drilling typically leaves a residual stub of about 0.15–0.25 mm (6–10 mil). That pushes the first resonance above ~100 GHz, safely out of band for 25G/56G and even 112G. For the tightest links some shops target the lower end (≈0.10 mm or below) but stay above ~0.05 mm to avoid drilling into the signal layer. The number that matters is variation, not just the mean — a stable entry sheet is what keeps every panel inside that window.
Does back-drilling add much cost? It adds roughly 5–15% to bare-board cost (per published industry estimates), from the second controlled-depth pass, verification, and slightly lower yield. That is far cheaper than the HDI blind/buried-via alternative, which can cost 50–200% more for the same stub removal — which is why back-drilling is the default mitigation on high-speed backplanes and servers.
Conclusion — Stub Control Starts Above the Stub
If your 25G/56G channel is failing SI review, do not assume the stack-up is wrong. Check the stub first. A via stub is an unterminated transmission line that resonates at f = c / (4 · L_stub · √εr_eff) and carves the insertion-loss notch that closes your eye — and back-drilling is how you move that resonance above 100 GHz and recover 3–6 dB of loss. The rule is simple: back-drill above ~10 Gbps NRZ or 56G PAM4, and on any board thicker than ~1.6 mm with deep vias. And remember the hidden link — the entry sheet above the panel is your Z reference, so its thickness variation is your stub variation.
YUESHAN — Your PCB Drilling Material Partner — supplies one-piece coated back-drill aluminum entry sheets (0.16–0.19 mm) built to hold that reference plane tight, with full TDS/MSDS and ISO 9001:2015 manufacturing.
Message Mr. Lucky Xu on WhatsApp +86 13433290964 · Email: sales@www.yueshanpcb.com — Mr. Lucky Xu handles back-drill inquiries. Request a free back-drill sample, the full TDS/MSDS, and a no-cost comparison of your current stub-control setup — we will map the right sheet to your layer count, fastest link rate, and current back-drill process.
Meta Title: Why Back-Drill PCBs? Stub Removal for 25G/56G SerDes | YUESHAN Meta Description: Back-drilling removes via stubs that wreck 25G/56G SerDes signal integrity. Learn why, when to spec it, and how entry-sheet choice controls stub length. Primary Keyword: why back drill pcb Secondary Keywords: via stub removal pcb, pcb back drilling signal integrity, back drill 25g serdes, back drill 56g pcb, when to use back drilling pcb URL Slug: /blog/why-back-drill-stub-signal-integrity.html Word Count: 2,588