After reviewing 500+ RF PCB orders over the past 18 months, we've identified the 12 most common DFM handoff failures that cause respins. 42% trace back to a single issue: specifying Dk at the wrong frequency.
Here's the complete checklist for RF designs using Rogers 4350B, RO4003C, PTFE, and hybrid stackups.
The 12-Item RF DFM Handoff Checklist
| # | Item | What to Specify | Failure Rate |
|---|---|---|---|
| 1 | Material MPN | Exact vendor P/N per layer | 18% |
| 2 | Dk at operating freq | Value + frequency + source | 42% |
| 3 | Df at operating freq | Value + frequency | 12% |
| 4 | Core thickness + tol | Nominal + acceptable range | 8% |
| 5 | Copper weight/type | Weight + ED/RA/LoPro treatment | 6% |
| 6 | Impedance per layer | Target + tolerance + ref layer | 28% |
| 7 | Via fence spec | Pitch + diameter + connection | 12% |
| 8 | Bonding material | Prepreg type for hybrid builds | 18% |
| 9 | Backdrilling | Depth + stub length tolerance | 5% |
| 10 | Test coupon plan | Location + structure type | 15% |
| 11 | Edge plating | Where + thickness + continuity | 3% |
| 12 | Transition notes | Via anti-pad, clearance | 4% |
Note: Failure rates overlap (one order can have multiple issues).
The #1 Problem: Dk at Wrong Frequency
Rogers RO4350B has Dk ≈ 3.66 at 1 MHz (IPC test method) but Dk ≈ 3.48 at 10 GHz (clamped stripline resonator). If you design using Dk = 3.66, your trace will be narrower than optimal. The manufactured board sees Dk = 3.48 at your operating frequency, producing impedance 3-4 ohms higher than target.
The fix: Specify Dk at or near your operating frequency. Cite the measurement method. State what Dk value your impedance calculation assumed.
Impedance: Per-Layer, Not Blanket
Don't specify "All traces: 50 ohm ±10%." Instead, provide a layer-by-layer table:
Layer | Net Class | Target | Tolerance | Trace W | Material
L1 | RF_50 | 50Ω SE | ±5% | 12.8mil | RO4350B
L3 | USB3 | 90Ω DF | ±10% | 4.5mil | FR-4
L5 | DDR4 | 40Ω SE | ±10% | 5.0mil | FR-4
This tells the manufacturer where to focus process control, potentially reducing cost while improving RF performance.
Hybrid Stackup Bonding: The Missing Specification
For Rogers/FR-4 hybrid builds, 18% of designers completely omit the bonding material specification. This matters because:
- CTE mismatch: Rogers Z-axis CTE (32 ppm/°C) vs FR-4 prepreg (50-70 ppm/°C) creates stress during thermal cycling
- Dk at boundary: Standard FR-4 prepreg (Dk ≈ 4.0) vs Rogers bondply (Dk ≈ 3.54) — which one does your model assume?
- Press compatibility: RO4350B works with standard FR-4 press profiles, but PTFE materials require modified profiles
Copper Roughness: The Hidden Variable
Above 10 GHz, copper surface roughness changes effective Dk by 3-5% and adds 0.5-1.5 dB/inch loss. Standard ED copper (RMS ≈ 5-7 μm) vs LoPro/HVLP (RMS ≈ 1.5-2 μm) is a meaningful difference.
Always specify copper type. For designs >10 GHz, call out "LoPro" or "HVLP" explicitly.
Via Fence Pitch: Lambda/4 Rule
Via fence spacing must be < λ/4 at highest frequency:
- At 28 GHz (in RO4350B): max pitch ≈ 2.5mm
- At 77 GHz: max pitch ≈ 0.9mm (requires laser vias)
Don't reuse your digital ground stitching pitch (1-1.5mm) and assume it works for RF isolation at mmWave.
Test Coupon Strategy
For RF boards, generic impedance coupons aren't enough. Consider:
- Standard impedance coupon — per controlled net class, same panel location
- Through-loss coupon (TRL) — for >10 GHz insertion loss verification
- Isolation coupon — two traces with via fencing, measure S21
Key rule: specify "on production panel, within 2 inches of board array" — not on a separate panel with different press conditions.
The Complete Handoff Package
A professional RF PCB handoff includes:
- Gerbers (RS-274X or X2)
- Drill files (separate for mechanical and laser)
- Stackup drawing with material MPNs, Dk/Df at frequency
- Layer-by-layer impedance table
- IPC netlist
- Fab notes (substitution policy, via fill, surface finish)
- Test requirements (coupons, cross-section locations)
- Critical dimension callout
Missing any of these for an RF board means you'll get questions at best, incorrect assumptions at worst.
Originally published at AtlasPCB Engineering Blog — our RF process engineers review every submission for DFM completeness before manufacturing starts.
Top comments (0)