DEV Community

Cover image for USB4 PCB Fabrication: What 80 Gbps Demands From Your Board Manufacturer
AtlasPCBEngineering
AtlasPCBEngineering

Posted on Originally published at atlaspcb.com

USB4 PCB Fabrication: What 80 Gbps Demands From Your Board Manufacturer

USB4 Version 2.0 at 80 Gbps is pushing commercial PCB requirements into territory that was previously reserved for telecommunications backplanes and military hardware. If you're designing a USB4 product in 2026, the PCB is no longer a commodity — it's a critical engineering component that determines whether your device passes compliance testing.

Here's what we've learned manufacturing USB4 boards for docking stations, Thunderbolt 4 host controllers, and multi-port hubs over the past year.

The Core Problem: Standard FR-4 Cannot Physically Support USB4

USB4 Gen 3 operates at 20 Gbps per lane (10 GHz fundamental). USB4 Version 2.0 uses PAM-3 modulation at 26.7 Gbaud (13.3 GHz fundamental). At these frequencies, standard FR-4 with its dissipation factor of 0.020 introduces approximately 1 dB of insertion loss per inch of trace.

For a typical USB4 channel with 4 inches of trace routing, that's 20 dB of loss at the fundamental frequency alone. The USB4 specification budgets only 8-12 dB for the PCB portion. Standard FR-4 is physically impossible.

What You Actually Need: Material Tiers

Tier 1 — Mid-Loss (USB4 Gen 3 at 40 Gbps, traces under 4 inches):

  • Panasonic Megtron 4 (R-5725): Df ~0.005 at 10 GHz
  • Isola I-Tera MT40: Df ~0.005
  • ITEQ IT-170GRA: Df ~0.007

Tier 2 — Ultra-Low-Loss (USB4 v2.0 at 80 Gbps, or longer traces):

  • Panasonic Megtron 6 (R-5575): Df ~0.002
  • Isola Astra MT77: Df ~0.002

From a cost perspective, Megtron 4 adds approximately 40-60% to total board cost versus standard FR-4. Megtron 6 adds 60-100%. As of mid-2026, Megtron 6 lead times are running 8-12 weeks due to AI server demand consuming available supply.

Impedance: 85 Ohm ±5% Is Harder Than It Sounds

USB4 specifies 85 ohm differential impedance (not USB 3.2's 90 ohm). The tolerance is ±5% for Version 2.0, compared to ±10% for USB 3.2. That's a total acceptable window of only 8.5 ohm.

Achieving this in production requires:

  • Trace width control: ±0.3 mil (direct-write laser imaging)
  • Dielectric height control: ±0.4 mil (controlled prepreg lay-up)
  • Mandatory TDR verification on every production panel
  • Internal process limits tighter than spec (we run ±4% to provide measurement margin)

One consideration designers miss: solder mask on outer-layer traces reduces impedance by 2-5 ohm. If you're routing USB4 on outer layers, your impedance model must account for the mask overlay. Our recommendation: route USB4 pairs on inner stripline layers whenever possible.

Via Stubs Kill USB4 Signals

At USB4 frequencies, through-hole via stubs create quarter-wave resonances directly in the signal band. A 55-mil stub (typical for L1-to-L2 transition on a 62-mil board) resonates at approximately 13-15 GHz — right where USB4 lives.

Three solutions, in order of cost:

  1. Back-drilling — Removes unused barrel mechanically. Adds ~$0.50-1.50 per via. Residual stub: 8-10 mils (moves resonance above the signal band).

  2. Blind vias — Zero stub, but requires sequential lamination. Adds 30-50% to board cost.

  3. HDI microvias (VIPPO) — Best electrical performance, highest cost (60-100% premium).

For most designs, back-drilling provides the best cost-performance balance. We recommend it for every via in the USB4 signal path regardless of stub length — even a 20-mil stub causes measurable reflection at Version 2.0 frequencies.

Stackup: Minimum 6 Layers, Practically 8

For a single USB4 port, 6 layers can work:

  • L1: USB4 TX high-speed pairs
  • L2: Solid ground plane (reference)
  • L3: USB 2.0 + control signals
  • L4: Power distribution
  • L5: Solid ground plane (reference)
  • L6: USB4 RX high-speed pairs

For multi-port designs (hubs, docking stations), 8 layers become the practical minimum. The critical constraint: every USB4 differential pair needs an unbroken ground reference plane on the immediately adjacent layer. No splits, no routing underneath.

Real Cost: 40-80% Premium Over Standard FR-4

Breaking down the USB4 PCB cost premium:

  • Material: Low-loss laminate is 3-5x standard FR-4 price (but material is only 30-40% of total board cost)
  • Process complexity: Tighter tolerance = slower etch, more inspection, mandatory TDR testing (+15-25%)
  • Yield impact: Narrower acceptance window reduces first-pass yield from ~95% to 85-90% (+5-15%)
  • Back-drilling: Per-hole cost + additional process step

Practical example: An 8-layer USB4 hub board (4×5 inch, Megtron 4, ±7% impedance, HVLP copper, back-drilled, ENIG finish) costs approximately $18-25 per panel versus $10-14 for equivalent standard FR-4.

Common Mistakes We See (Manufacturer's Perspective)

Based on our production data from USB4 boards in the past year:

  1. Missing Dk specification — Designer provides impedance target but not the Dk value used in simulation. Different field solvers report different Dk for the same material at different frequencies, causing systematic offset.

  2. Unspecified back-drilling in connector fanout — Short stubs (20-30 mils) considered 'short enough' that still fail compliance at Version 2.0 frequencies.

  3. Incompatible hybrid stackups — Specifying low-loss material only for signal layers with standard FR-4 for others. Different lamination temperatures, flow characteristics, and shrinkage rates create manufacturing challenges.

Key Takeaways

  • Standard FR-4 physically cannot support USB4. Period.
  • Budget 40-80% cost premium over standard boards
  • Plan for 8-12 week material lead times (Megtron 6)
  • Back-drill every via in the USB4 path
  • Specify Dk value, not just impedance target
  • Engage your manufacturer during layout, not after Gerbers

At AtlasPCB, we maintain certified processes for all major low-loss material systems and provide complimentary stackup design review for USB4 projects. If you're starting a USB4 design, reach out for a stackup recommendation before finalizing your layout — it eliminates the iteration cycle that costs weeks.

Originally published on the AtlasPCB Engineering Blog.

Top comments (0)