CM Chokes & Impedance Matching: Open the Differential Eye
← Back to:Interfaces, PHY & SerDes
CM chokes and impedance matching open the differential eye by removing reflections and mode conversion at real discontinuities—measured by loss, balance, and pass/fail thresholds.
Follow a closed loop: rules → TDR → S-parameters → eye/BER; fix return-path continuity and symmetry first, then tune CMC/termination with data.
Thesis & Scope: What this page solves
Open the differential eye by removing reflections and mode conversion first—then (and only then) consider EQ/retimers for extra margin.
A common-mode choke (CMC) can reduce common-mode noise and help EMI, but imbalance and parasitics can turn it into an eye-closure source.
Outcomes (engineering-grade)
- Decide whether a CMC is appropriate (and when it should be avoided) using measurable criteria.
- Separate “loss-limited” from “reflection/mode-conversion-limited” failures with a minimal measurement loop.
- Apply termination/matching changes that translate directly into larger eye height and lower BER risk.
In-scope (covered deeply)
- CMC behavior that affects eye margin: Zcm/Zdm, imbalance, mode conversion, SRF effects.
- Differential impedance and termination topologies to reduce reflections and stabilize eye shape.
- Bench workflow: TDR → S-params → eye/BER with pass-criteria placeholders (threshold X).
Out-of-scope (only decision hooks)
- ESD/TVS part-number deep dives and surge compliance—only capacitance/placement decision hooks are mentioned.
- Protocol compliance rulebooks (PCI-SIG / USB-IF / HDMI / DP)—only “how matching impacts margin” is addressed.
- Retimer/CDR internal architecture—only “CDR presence changes tolerance to reflections” is noted.
Minimal workflow (fast discrimination)
- TDR: locate the largest impedance steps (connector/transition/termination distance).
- S-params: check Sdd (loss/return) and Sdc (mode conversion) around the critical frequency range.
- Eye/BER: run an A/B change (termination position, CMC swap, or bypass footprint) and compare margin against threshold X.
CM Choke Fundamentals: common-mode vs differential-mode
Concept: what the choke is supposed to do
A differential pair carries information as differential-mode (DM) current (equal and opposite on the two lines). Noise that rides on both lines in the same direction is common-mode (CM).
The ideal CM choke creates high impedance for CM (blocking CM current) while remaining nearly transparent for DM. Real parts are not ideal: parasitics and imbalance can couple DM↔CM, changing eye shape and jitter.
4-port view: the three behaviors that matter
1) CM impedance (Zcm vs frequency)
Should be high in the frequency band where CM noise/EMI is problematic. “Bigger Zcm” alone is not sufficient if DM loss or imbalance increases.
2) DM transparency (Zdm / Sdd21 insertion loss)
Excess DM impedance or loss reduces eye height directly. In practice, the goal is “enough CM blocking” with DM loss kept within the link budget.
3) Mode conversion (Sdc): the silent eye killer
Imbalance (layout asymmetry, mismatched parasitics, non-ideal choke symmetry) converts DM energy into CM and vice versa. This can produce asymmetric eyes, unexpected BER spikes, and “EMI improved but link got worse” outcomes.
Self-resonant frequency (SRF) warning: near SRF, the choke can behave like a resonator rather than a smooth impedance element, creating narrowband trouble inside the channel spectrum.
Common misconceptions (and the correct engineering lens)
Misconception: “Higher Zcm is always better.”
Correct lens: optimize across Zcm window, DM loss, and mode conversion. A high-Zcm part that increases Sdc can close the eye faster than it reduces EMI.
Misconception: “Insertion loss looks small, so it is safe.”
Correct lens: narrowband peaks (near SRF) and imbalance-driven conversion can cause large eye/BER damage even when average loss appears modest.
Misconception: “A CMC fixes EMI by itself.”
Correct lens: EMI is often driven by return-path discontinuity and asymmetry. A choke placed across a broken return path can increase CM current or convert energy into the band that radiates more efficiently.
Differential Impedance 101: what “100Ω diff” really means on PCB
“100Ω differential” is a system property: geometry + dielectric + return path + transitions. Correct line width/spacing alone does not guarantee a clean eye.
When the return path is broken or transitions are asymmetric, differential energy can convert into common-mode (Sdc), increasing radiation, reflections, and BER risk.
Rules & targets (what impedance control really controls)
Zdiff vs Zodd / Zeven (engineering meaning)
- Zodd tracks the differential energy path and is closely tied to reflections and eye height.
- Zeven describes the even/common-mode environment and becomes critical when return paths are disrupted or structures become asymmetric.
- If odd/even modes couple, mode conversion (Sdc) rises—often the root of “looks matched but still unstable.”
Control knobs that matter (beyond width/spacing)
- Stackup: dielectric height, copper thickness/roughness, soldermask effects (budget placeholders: X).
- Reference continuity: uninterrupted return plane under the pair; avoid plane splits/slots across the route.
- Transitions: via stubs, anti-pad geometry, pads, connector launches; keep them symmetric and short.
Real-world deviation sources (why “calculated 100Ω” still fails)
Broken return path → CM radiation + reflections
Plane splits/slots force return currents to detour, coupling DM into CM (Sdc). Symptoms include asymmetric eyes, unstable margin, and EMI peaks that change with grounding/cabling.
Via/pad/connector transitions → impedance steps + stub resonance
Discontinuities are often the dominant reflection source. Even a short stub can create narrowband trouble that collapses margin near the channel’s critical spectrum.
Asymmetry → mode conversion spikes
Any “not-mirrored” transition (one-line detour, unequal via fields, uneven reference stitching) increases Sdc and can cause BER spikes even when average insertion loss looks fine.
When to Use a CM Choke: decision tree (CMC vs no CMC)
A CM choke is a boundary tool for common-mode problems. It is appropriate only when CM evidence exists and the differential margin can tolerate the added parasitics.
The decision must include a verification loop: CM improvement is not sufficient if mode conversion (Sdc) or DM loss collapses eye margin.
Decision tree (symptom → action → verify)
- Evidence of CM/EMI issue? Examples: radiated peak fails, cable radiation obvious, CM current probe reads high (threshold X).
- If YES: consider CMC near the boundary (connector) only after return-path continuity and symmetry are verified. Verify: CM current ↓ ≥ X%, while eye margin ≥ X and Sdc peak does not worsen beyond X dB.
- If NO (or margin is already tight): prioritize matching/termination and transition cleanup first. Verify: TDR steps reduced and return loss improves; eye height ≥ X and BER meets target.
Risk list (why adding a CMC can make it worse)
Imbalance → Sdc ↑
Asymmetric pads/vias or unequal coupling makes the choke a DM↔CM converter. First check: mirrored geometry and reference continuity around the footprint.
Zdm / DM loss too high → eye height ↓
A part optimized for CM impedance may add unacceptable DM insertion loss. First check: Sdd21 / insertion loss delta vs the link budget (threshold X).
Resonance near Nyquist band → narrowband margin collapse
If SRF/peaks land near the channel’s critical spectrum, BER spikes can appear even when average loss is small. First check: frequency sweep for sharp peaks around the Nyquist region (threshold X).
CM Choke Selection: the parameters that actually matter
CM choke selection must be engineered as a multi-objective trade: balance, Zcm window, DM loss, and SRF/parasitics. A single “Zcm @ 100 MHz” number is not sufficient.
The priority order for link safety is typically: balance > Zcm(f) window > Zdm/DM insertion loss > DCR.
Selection priority (why this order works)
1) Balance / symmetry (mode conversion risk)
Imbalance drives Sdc peaks and asymmetric eyes; it is a failure amplifier even when average loss looks acceptable.
2) Zcm(f) window (match the EMI/CM band)
Choose Zcm that is high and smooth in the band of concern (f₁–f₂), not just at one marketing frequency.
3) Zdm / DM insertion loss (eye-height cost)
Extra DM loss directly reduces eye height and jitter tolerance; evaluate ΔIL against the link budget (threshold X).
4) DCR / rated current / bias behavior
Bias and current can shift curves and create narrowband artifacts; treat “rated current” as a functional constraint, not a footnote.
Parameter cards (impact → quick check → common traps)
Zcm(f) curve
Impact: determines how strongly common-mode current is suppressed in the band where EMI/CM noise dominates.
Quick check: identify the band of concern (f₁–f₂) using near-field/CM current probe, then choose parts with high, smooth Zcm across f₁–f₂.
Common traps: selecting by “Zcm @ 100 MHz” while the real EMI peak is elsewhere; ignoring sharp peaks that land inside the channel’s sensitive spectrum.
Zdm(f) / DM insertion loss (Sdd21)
Impact: the direct eye-height cost; excessive DM loss reduces jitter tolerance and closes the eye even if CM improves.
Quick check: compare DM insertion loss curves or mixed-mode Sdd21; enforce a budget: ΔIL ≤ X dB in the critical band.
Common traps: average loss looks small but narrowband features (SRF/parasitics) collapse margin near Nyquist-related content.
Balance / mismatch (ΔR/ΔL, amplitude-phase mismatch)
Impact: sets the mode conversion ceiling. Imbalance creates Sdc peaks that distort the eye and increase BER spikes.
Quick check: prefer parts with explicit balance metrics; verify with mixed-mode S-params if available (Sdc peak ≤ X dB).
Common traps: assuming symmetry by default; ignoring footprint asymmetry and local return-path breaks that dominate conversion.
DCR, rated current, package parasitics, SRF
Impact: bias and parasitics can move resonance and reshape curves; SRF near the sensitive band can create narrowband failures.
Quick check: use bias-aware curves when provided; otherwise enforce A/B validation (bypass vs CMC) with eye/BER thresholds X.
Common traps: missing bias conditions in datasheets; selecting a part whose SRF/peaks align with Nyquist-related content.
Placement & Layout: where CMC belongs and how it breaks return paths
A CMC is a boundary component: it should sit near the interface where common-mode energy must be contained (connector / cable exit), while preserving continuous reference planes and symmetry.
The most common failure is placing a CMC at a broken return path; this creates mode conversion and increases common-mode current instead of reducing it.
Placement principles (where it belongs)
- Place the CMC at the boundary that must not leak CM energy (connector / cable exit), not deep inside the SoC region.
- Ensure reference continuity across the CMC footprint; maintain a stable return plane and use stitching where needed.
- Keep transitions short and symmetric: CMC is not a substitute for termination/launch quality; it must not add avoidable stubs.
Routing & via rules (checkable, not subjective)
Continuous reference before/after the CMC
Avoid plane swaps or splits at the footprint. If layer changes are required, provide tight stitching to keep return currents local (threshold X).
Symmetric entry/exit routing
Both lines must be mirrored through pads/vias and neighboring copper. Any one-line detour increases ΔC/ΔL and raises Sdc peaks.
Minimize vias; if unavoidable, mirror and stitch
Vias create discontinuities and stubs. Use paired/identical via fields and nearby stitching to avoid odd/even coupling (threshold X).
Don’ts (high-probability failure patterns)
- Do not place the CMC over a plane slot/split (creates CM instead of blocking it).
- Do not route one line around obstacles while the other stays straight (imbalance → Sdc ↑).
- Do not add unnecessary stubs/branches near the footprint (narrowband artifacts near SRF).
- Do not assume “EMI improved” means “link safe”; verify eye/BER against threshold X.
- Do not cluster high-parasitic protection parts without controlling symmetry; keep the transition region clean.
Termination Topologies: differential, split, AC, source vs load
Termination is the other half of opening the eye: it eliminates reflections by controlling where energy is absorbed. The correct choice depends on DC/common-mode constraints, allowed power, and placement error tolerance.
A perfect topology can still fail if the termination is not placed at the boundary: distance becomes an effective stub and reflections return.
Reusable topology library (use when / trade-offs / verify)
Differential termination (Rdiff across the pair, at load)
Use when: a clear load endpoint exists and power budget allows continuous absorption; long-run reflections dominate.
Trade-offs: adds DC power; requires tight placement and symmetric routing to avoid introducing mode conversion.
Verify: TDR end-step minimized; Sdd11 improves (target RL ≥ X dB); eye reflection ripples reduced and BER meets X.
Split termination (R to Vcm/ground + optional C for CM control)
Use when: common-mode noise or bias stability matters; a defined CM reference is needed at the boundary.
Trade-offs: extra components add parasitics; asymmetry can increase Sdc; capacitor placement strongly affects HF behavior.
Verify: CM probe improves ≥ X%; Sdc does not worsen beyond X dB; eye remains symmetric and margin ≥ X.
AC termination (series C with R path to absorb HF; blocks DC/low-frequency)
Use when: DC isolation is required while controlling reflections in the operating band.
Trade-offs: low-frequency reflection is not absorbed; capacitor ESR/ESL can introduce narrowband artifacts.
Verify: in-band Sdd11 improves; eye/BER pass in the operating mode; no new peaking near the sensitive band (threshold X).
Source series termination (Rseries at source; absorbs reflections at the source)
Use when: driver is strong/low-Z and ringing/overshoot is the main issue; power must be minimized versus load termination.
Trade-offs: slows edges and reduces delivered amplitude; may not suppress far-end reflection if the receiver is very sensitive.
Verify: near-end overshoot reduces; TDR signature smooths; eye stays within amplitude and timing limits (threshold X).
Placement error = effective stub (why distance matters)
If the termination is not at the endpoint, the trace segment between the endpoint and the termination behaves like a stub. This creates a visible reflection step at the stub distance and can re-introduce eye ripples.
Quick check: use TDR to confirm the reflection location matches the physical offset; enforce a placement limit (distance ≤ X) and re-check the step amplitude (ΔZ ≤ X).
Matching Workflow: from stackup rules → TDR → S-params → eye
Use a closed-loop workflow to avoid guessing: rules prevent common errors, TDR locates discontinuities, S-parameters quantify loss/reflection/mode conversion, and eye/BER validates the result.
Each iteration should change one primary factor and run an A/B comparison with logged thresholds (X placeholders).
Step cards (goal → what to look at → interpret → next)
Step 1 — Rules first (stackup / routing / stub control)
Goal: prevent high-probability impedance and return-path failures before measurement.
Look at: Zdiff target (XΩ), return plane continuity, via stub length ≤ X, symmetric transitions and stitching density ≤ X.
Next: run TDR to locate discontinuities instead of guessing from the eye alone.
Step 2 — TDR (where the reflection is)
Goal: map impedance steps to physical locations (connector / via / termination offset).
Look at: step location and amplitude (ΔZ ≤ XΩ), signatures of stub resonance, and whether the end-step aligns with the termination position.
Next: use S-parameters to quantify in-band loss, return loss, and mode conversion.
Step 3 — S-parameters (Sdd21 / Sdd11 / Sdc)
Goal: separate loss (eye height) from reflection (ripples) and mode conversion (asymmetry + EMI coupling).
Look at: Sdd21 insertion loss (ΔIL ≤ X dB), Sdd11 return loss (RL ≥ X dB), Sdc peaks (≤ X dB) in the band of concern.
Next: tune termination or symmetry/CMC/layout based on which metric is dominant, then validate with eye/BER.
Step 4 — Eye / BER (A/B compare and accept)
Goal: confirm that the change improved margin without breaking compatibility or introducing narrowband failures.
Look at: eye height/width ≥ X, jitter margin ≥ X, BER ≤ X (or PRBS errors within X) under worst-case conditions.
Rule: change one primary factor per iteration and keep a bypass/0Ω option for clean A/B comparisons.
Failure Modes: why CMC + “almost matched” still closes the eye
Real failures are rarely “one big mistake.” A slightly imbalanced footprint plus a CMC resonance or a small termination offset can turn common-mode energy into differential disturbance and close the eye.
Use symptom-driven cards to identify the dominant mechanism first: differential loss (Sdd21), reflection/stub (TDR/Sdd11), or mode conversion (Sdc).
Symptom cards (symptom → likely causes → first probe → fix → pass criteria)
Symptom: eye height drops after adding the CMC
Likely causes: Zdm insertion loss too high; resonance near SRF in the operating band; termination effective impedance shifted by added parasitics.
First probe: A/B bypass the CMC footprint (0Ω) and compare eye height; if available, compare Sdd21 (ΔIL ≤ X dB).
Fix: choose lower-Zdm / higher-balance CMC; shorten transitions and reduce stub; keep footprint symmetric and reference plane continuous.
Pass criteria: eye height ≥ X; ΔIL ≤ X dB; no new narrowband peaking near the sensitive band (X).
Symptom: occasional BER spikes (mostly OK, sometimes fails)
Likely causes: temperature or DC bias shifts CMC characteristics; small asymmetry creates frequency-dependent Sdc peaks; termination offset creates a small stub with narrowband sensitivity.
First probe: correlate errors with temperature/voltage/load; swap to a higher-balance CMC and re-run BER; check for eye asymmetry and narrowband artifacts (threshold X).
Fix: enforce symmetry (routing/footprint/planes); move termination to the boundary; select CMC with better balance over the full temperature/bias range.
Pass criteria: BER ≤ X across temperature; Sdc peak ≤ X dB; eye symmetry within X.
Symptom: EMI improves, but link becomes less stable
Likely causes: common-mode energy is reduced externally, but increased mode conversion injects differential disturbance into the receiver; imbalance dominates over Zcm benefit.
First probe: measure CM current near the connector (should improve) and eye/BER (should not worsen); if instability rises, suspect Sdc increase (ΔSdc ≤ X dB).
Fix: prioritize balance and symmetry; re-place the CMC at the boundary; ensure planes and return path are continuous; reduce skew and asymmetric stubs.
Pass criteria: CM current improves ≥ X% while BER ≤ X and eye margin ≥ X; Sdc does not increase beyond X.
Symptom: “almost matched” still shows ripples / overshoot
Likely causes: termination is not at the endpoint (effective stub); connector/via transitions still create impedance steps; split/AC termination components are too far from the boundary.
First probe: TDR to locate the dominant step and map it to the physical feature; verify termination offset distance ≤ X.
Fix: move termination to the boundary; reduce via stub; improve connector/launch geometry; tighten symmetry around the transition.
Pass criteria: ΔZ ≤ XΩ (TDR); return loss RL ≥ X dB; ripple amplitude ≤ X and eye margin ≥ X.
Symptom: eye becomes asymmetric (left/right slopes differ)
Likely causes: routing or footprint imbalance; unequal via count or reference-plane switching; split termination not symmetric; nearby copper/keepout differences around the CMC.
First probe: compare near-end and far-end eye symmetry; inspect physical symmetry around CMC and termination; check for narrowband artifacts (X).
Fix: enforce mirrored placement and routing; equalize via transitions; keep plane reference consistent across the CMC and termination region.
Pass criteria: eye symmetry within X; Sdc peak ≤ X dB; BER ≤ X with stable margin.
Measurement & Quick Checks: what to probe first (bench-friendly)
Start with probes that split the problem quickly: locate reflections (TDR), separate loss vs reflection (near/far eye), and verify whether the CMC is acting at the boundary (CM current / near-field).
Use minimal A/B experiments (CMC bypass, swap CMC, move termination) to confirm the dominant mechanism before committing layout changes.
Toolbox (what it tells → where to probe → fast signature → pass criteria)
TDR
Tells: where the discontinuity is (connector / via / termination offset) and how large the impedance step is.
Probe: launch at connector side and compare to receiver-side launch when possible; mark step location versus layout.
Fast signature: a strong step at a distance matching the termination offset indicates an effective stub.
Pass criteria: ΔZ ≤ XΩ; dominant step reduced by ≥ X% after the fix.
Near-end vs Far-end eye
Tells: whether loss/ISI dominates (far-end collapses) or reflection dominates (near-end already shows ringing).
Probe: measure near the source and near the receiver under the same pattern and conditions.
Fast signature: if near-end looks clean but far-end collapses, suspect insertion loss (Sdd21 / Zdm); if near-end rings, suspect reflection (Sdd11 / termination/transition).
Pass criteria: eye height/width ≥ X at far-end; ripple/overshoot ≤ X.
CM current probe / Near-field probe
Tells: whether the CMC is acting at the correct boundary and whether common-mode energy is being contained before the cable/connector.
Probe: measure on the connector side and board side of the CMC; scan near the CMC and connector launch region.
Fast signature: CM current decreases but BER worsens → suspect mode conversion (Sdc) and imbalance, not Zcm benefit.
Pass criteria: CM current improves ≥ X% while eye/BER do not degrade; no new hot spots near transitions (X).
Quick A/B checks (minimal experiments that confirm direction)
A/B #1 — Bypass the CMC footprint (0Ω)
Observe: eye height / far-end margin. Interpret: immediate improvement points to Zdm loss or SRF-related artifacts. Next: select lower-Zdm / better-balanced CMC and shorten transitions.
A/B #2 — Swap CMCs (same Zcm class, different balance / Zdm)
Observe: BER stability and eye symmetry. Interpret: large BER change implies imbalance/mode conversion dominates. Next: fix symmetry and placement; prioritize balance over “Zcm @ 100 MHz”.
A/B #3 — Move termination location / toggle termination option
Observe: TDR step location/amplitude and eye ripples. Interpret: strong sensitivity indicates stub/endpoint placement is the main lever. Next: enforce distance ≤ X and re-check RL ≥ X dB.
Engineering Checklist: design → bring-up → production gates
This section turns best-practice into gates. Each item must have a measurable check and a pass criterion (X) before moving forward.
Scope: impedance continuity, CMC balance/loss, termination placement/topology, and closed-loop verification (TDR → S-params → eye/BER).
Gate A — Design (rules and layout readiness)
Item: define impedance targets and discontinuity budget
Why: “100Ω diff” must include launches, vias, pads, and connector transitions; small steps can dominate reflections.
How to check: document Zdiff target, ΔZ budget for each transition, and maximum allowed termination offset (stub).
Pass criteria: Zdiff = X Ω; transition ΔZ ≤ X Ω; termination offset ≤ X (length); pair skew ≤ X.
Item: via / stub control plan (and backdrill rule if needed)
Why: stub resonances can land near sensitive bands and cause narrowband eye/BER failures.
How to check: define maximum stub length per layer transition; specify whether backdrill/via-in-pad is mandatory at boundaries.
Pass criteria: via stub length ≤ X; no uncontrolled stubs at connector/CMC/termination region.
Item: CMC selection priority locked (balance → loss → Zcm band-fit → SRF risk)
Why: balance (mismatch) drives mode conversion (Sdc), which can hurt stability even when EMI improves.
How to check: record chosen CMC family + rationale and the frequency band of concern; require a bypass option for A/B.
Pass criteria: balance metric ≥ X; Zdm / ΔIL ≤ X dB in band; SRF not in sensitive band (X); Zcm fits EMI band (X).
Example CMC material numbers (verify exact suffix/package):
- TDK: ACM2012D-900-2P (common high-speed diff CMC example)
- Murata: DLW21SH101XQ2 (DLW series example ordering code)
- Würth Elektronik: 744231091 (WE-CNSW / CMC example ordering code)
- TDK: ACM2012-900-2P (family variant example; confirm D-suffix needs)
Note: these are representative material numbers; always confirm insertion loss, balance, SRF behavior, and package footprint in the datasheet.
Item: termination topology + placement rule (endpoint means endpoint)
Why: a correct resistor value placed at the wrong location becomes a stub and re-introduces reflections.
How to check: choose topology (diff / split / AC / source series) based on DC/common-mode constraints; enforce symmetry and distance rule; use matched parts if split.
Pass criteria: termination offset ≤ X; ΔR (split/matched) ≤ X; tolerance ≤ X; tempco ≤ X.
Example termination material numbers (verify value/tolerance/package):
- Yageo: RC0603FR-0749R9L (49.9Ω, thick film family example)
- Yageo: RC0603FR-07100RL (100Ω, thick film family example)
- Vishay: CRCW060349R9FKEA (49.9Ω family example)
- Bourns resistor array: CAY16-499J4LF (matched network family example; confirm tolerance/ratio)
- Yageo resistor array: YC164-JR-0749R9L (network family example; confirm array size/ratio)
Note: for split termination and symmetry-sensitive nodes, prioritize matched networks (ΔR tracking) and stable tempco over “lowest cost” parts.
Gate B — Bring-up (measurements and closed-loop confirmation)
Item: TDR location + magnitude must match physical features
How to check: locate the dominant step and map distance to connector/via/termination; apply one change and re-measure.
Pass criteria: ΔZ ≤ XΩ; dominant step reduced ≥ X%; termination-offset signature eliminated (X).
Item: S-params window (if available): loss, reflection, mode conversion
How to check: evaluate mixed-mode band windows (Sdd21, Sdd11, Sdc). Focus on the band that correlates with eye closure/EMI.
Pass criteria: Sdd21 loss ≤ X dB; Sdd11 return loss ≥ X dB; Sdc peak ≤ X dB in-band.
Item: eye / BER gate with before/after evidence (not a single screenshot)
How to check: compare near-end vs far-end, and record the margin change after each A/B experiment.
Pass criteria: eye mask/margin ≥ X; BER ≤ X; margin does not regress after adding CMC (Δmargin ≥ X).
Item: mandatory minimal A/B experiments for root-cause confirmation
How to check: (1) CMC bypass, (2) swap CMC for balance/Zdm contrast, (3) termination position/topology toggle; record deltas.
Pass criteria: each A/B yields a clear directional result; ambiguous outcomes require revisiting symmetry/return path before further tuning.
Gate C — Production (consistency, sampling, regression)
Item: alternate-part / lot equivalency checks (CMC + termination)
How to check: require equivalency evidence on balance and insertion-loss window; verify resistor network ΔR tracking and tempco.
Pass criteria: CMC balance ≥ X and ΔIL ≤ X dB vs baseline; termination ΔR ≤ X and tolerance/tempco within X.
Item: sampling strategy (what to sample, how often, and triggers)
How to check: define quick tests that correlate with failures (TDR step, simplified far-end eye, CM current at boundary).
Pass criteria: sampling ratio ≥ X; trigger escalation on supplier change, lot change, yield drop, or new EMI anomalies (X).
Item: failure regression fields (so root causes do not reset to zero)
How to check: log minimum fields for every failure: board revision, lot codes, CMC/termination PNs, reflow profile, symptom type, first-probe evidence, A/B outcomes.
Pass criteria: 100% failures have complete regression fields; corrective action links to gate item and re-test evidence (X).
Applications & Part Selection Notes (placed at the end)
Use scenario buckets to choose CMC and termination traits without turning this page into a protocol-specific guide.
Material numbers below are examples. Always verify insertion loss (Zdm), balance, SRF behavior, and footprint compatibility.
Scenario: Cable exit / connector boundary
Typical symptoms
- EMI peaks tied to cable orientation or chassis coupling
- CM current is high outside the connector, yet differential eye looks “almost OK”
Recommended traits
- Placement: CMC close to the connector boundary; keep planes continuous through the CMC region
- Balance first: minimize Sdc; avoid asymmetry around pads/vias
- Zdm / loss window: keep insertion loss low in the data band (ΔIL ≤ X dB)
- Zcm band-fit: choose Zcm curve matching the EMI band (X)
Example material numbers (verify)
- CMC examples: TDK ACM2012D-900-2P, Murata DLW21SH101XQ2, Würth 744231091
- Termination examples: Yageo RC0603FR-0749R9L (49.9Ω), Vishay CRCW060349R9FKEA (49.9Ω)
- Matched network examples (split termination): Bourns CAY16-499J4LF, Yageo YC164-JR-0749R9L
Pass criteria: CM current improves ≥ X% at the connector side, while BER ≤ X and eye margin ≥ X.
Scenario: Board-to-board / transition-dominated channels
Typical symptoms
- Ripples/overshoot tied to connector/via transitions
- Eye sensitivity to small termination placement changes
Recommended traits
- Fix transitions first: TDR map discontinuities before adding a CMC
- Termination discipline: endpoint placement rule; keep stubs under X
- CMC is not default: only add if CM/EMI evidence exists and margin allows
Example material numbers (verify)
- CMC (only if justified): TDK ACM2012D-900-2P (low-loss candidate example)
- Termination: Yageo RC0603FR-07100RL (100Ω), Vishay CRCW0603100RFKEA (100Ω family example)
Pass criteria: ΔZ ≤ XΩ at the dominant transition; RL ≥ X dB; eye ripple ≤ X.
Scenario: Noisy ground / strong common-mode disturbance
Typical symptoms
- CM current and near-field hotspots near boundaries
- EMI improves, but BER worsens (mode conversion dominating)
Recommended traits
- Balance dominates: minimize Sdc peaks; enforce symmetric pads/vias/copper
- Return path continuity: plane stitching around boundary; avoid slots under CMC/termination
- Matched termination: if split termination is used, require tight ΔR tracking
Example material numbers (verify)
- CMC (balance-focused): Murata DLW21SH101XQ2, TDK ACM2012D-900-2P
- Matched networks: Yageo YC164-JR-0749R9L, Bourns CAY16-499J4LF
Pass criteria: Sdc peak ≤ X dB while CM current improves ≥ X% and BER ≤ X.
Scenario: High-speed diff pairs (margin is tight)
Typical symptoms
- Eye height drops after adding components near the boundary
- Narrowband failures suggest resonance near the sensitive band
Recommended traits
- Zdm / insertion loss is critical: choose the lowest-loss candidate that still meets CM goals
- Avoid SRF in-band: do not introduce peaking near the sensitive band (X)
- CMC optionality: keep a bypass path and validate benefit with A/B evidence
Example material numbers (verify)
- CMC low-loss candidates (examples): TDK ACM2012D-900-2P, Würth 744231091
- Termination baseline (examples): Yageo RC0402FR-0749R9L (49.9Ω family), Yageo RC0402FR-07100RL (100Ω family)
Pass criteria: ΔIL ≤ X dB in-band; eye margin ≥ X; no new narrowband peaking near the sensitive band (X).
Minimal adjacent component examples (ESD arrays) — keep placement logic here, detailed selection belongs to the ESD/TVS page
- TI: TPD4E05U06 (ESD protection array example)
- Nexperia: PESD5V0S1UL (ESD diode example)
Note: confirm capacitance and bandwidth impact for high-speed differential nets.
Recommended topics you might also need
Request a Quote
FAQs (CM Chokes & Impedance Matching)
These FAQs close out long-tail debugging without expanding the main body. Every answer uses the same 4-line, measurable format.
Notes: “X” is a placeholder for your project thresholds (band, data-rate, mask, BER target, and lab limits).
Added a CMC and EMI improved, but eye height dropped — what parameter was wrong first?
Likely cause: differential-mode loss (Zdm / ΔIL) too high, or SRF-related peaking/notching near the sensitive band; balance-induced mode conversion can amplify the damage.
Quick check: A/B the CMC footprint (bypass vs CMC) and compare far-end eye height + BER; if available, compare Sdd21 (ΔIL) and Sdc peak in-band.
Fix: prioritize a lower-loss / better-balance CMC; keep the bypass option; if a narrowband resonance is suspected, reduce stubs and add minimal damping (layout-first, then part change).
Pass criteria: added ΔIL ≤ X dB over [f1–f2]; Sdc peak ≤ X dB over [f1–f2]; eye margin ≥ X; BER ≤ X.
Same CMC footprint, two vendors behave very differently — what’s the first correlation check (balance vs Zcm curve)?
Likely cause: balance / asymmetry (ΔL/ΔR, amplitude/phase mismatch) causing mode conversion; Zcm@100 MHz can look “better” yet still lose on Sdc and Zdm loss.
Quick check: correlate vendor A vs B by Sdc peak and in-band ΔIL (Sdd21); if no VNA, compare eye symmetry + BER and whether CM current reduction comes with eye regression.
Fix: select by “balance-first” (low Sdc) and low in-band Zdm loss; keep pad/via/keepout symmetry tight around the footprint; validate with a controlled A/B swap.
Pass criteria: vendor-to-vendor ΔIL difference ≤ X dB; Sdc peak ≤ X dB; eye asymmetry ≤ X%; BER ≤ X.
Link fails only at high temperature after adding CMC — SRF shift or DC bias effect? what to log?
Likely cause: temperature-dependent parasitics shifting resonance/peaking, or bias/current effects changing the effective impedance; DCR-driven self-heating can amplify drift.
Quick check: log temperature, common-mode voltage, any DC current through the pair, and compare eye/BER across temperature; if available, capture in-band ΔIL and Sdc at hot vs room.
Fix: ensure no unintended DC bias path through the CMC; choose a part with adequate current rating / lower DCR and stable in-band loss; keep SRF away from sensitive bands with margin.
Pass criteria: BER ≤ X across [Tmin–Tmax]; in-band ΔIL drift ≤ X dB; Sdc drift ≤ X dB; package temperature rise ≤ X °C at worst case.
TDR looks fine, but BER worsened after CMC — how to detect mode conversion quickly?
Likely cause: mode conversion (Sdc spikes) or narrowband loss/peaking that TDR does not reveal; imbalance can turn reduced CM into extra DM jitter/noise at the receiver.
Quick check: compare eye symmetry (left/right or upper/lower) and far-end BER with CMC bypass; if available, measure Sdc peak in-band; also check CM current reduction vs BER regression.
Fix: tighten symmetry around pads/vias/planes, reduce discontinuities near the CMC, and select a better-balance CMC; only after symmetry is fixed, re-optimize termination if needed.
Pass criteria: Sdc peak ≤ X dB; eye asymmetry ≤ X%; BER ≤ X (or errors ≤ X per test window); CM improvement ≥ X% without BER regression.
Split termination reduced common-mode noise but increased jitter — what’s the first placement mistake?
Likely cause: split termination is not symmetric (ΔR/unequal routing) or it is not at the true endpoint, creating a stub; the split reference (cap/return) may inject noise if grounded poorly.
Quick check: verify placement distance to the receiver pins and symmetry (same via count, same copper environment); TDR for a “termination offset” signature; measure ΔR or use a matched network.
Fix: move termination to the endpoint; use a matched resistor array (tight tracking) and keep split return short to a quiet reference; enforce mirrored routing and identical via transitions.
Pass criteria: ΔR ≤ X%; termination offset ≤ X mm; jitter increase ≤ X ps; BER ≤ X; eye margin ≥ X.
Moving the termination 10 mm changed the eye a lot — how to estimate “stub budget” fast?
Likely cause: the termination is no longer at the endpoint; the extra trace becomes a stub that reflects energy back into the main line, especially with fast edges.
Quick check: compute round-trip stub delay ≈ 2·L / vprop and compare to UI and edge rate; a quick FR-4 estimate uses vprop ≈ X mm/ns. Confirm with TDR: a post-step / ripple indicates stub behavior.
Fix: move the termination to the receiver pins (true endpoint), reduce via stubs (backdrill/via-in-pad if needed), and keep the termination network physically compact and symmetric.
Pass criteria: round-trip stub delay ≤ X% UI (or ≤ X% tr); TDR shows no significant post-step ripple; eye margin ≥ X; BER ≤ X.
CMC before TVS vs after TVS — what’s the quickest A/B to decide?
Likely cause: TVS/ESD capacitance and imbalance can interact with the CMC and create extra reflection or mode conversion; order changes the effective discontinuity “seen” by the channel.
Quick check: perform two controlled layouts or rework A/B (Order A vs Order B) and compare (1) far-end eye margin, (2) BER, and (3) CM current at the boundary; keep all other variables fixed.
Fix: choose the order that preserves symmetry and minimizes stubs; keep the ESD branch extremely short and matched; keep the CMC where reference planes remain continuous and routing is mirrored.
Pass criteria: eye margin ≥ X and BER ≤ X for the chosen order; CM current reduction ≥ X%; no added ΔIL > X dB in-band.
Eye is asymmetric (one side worse) after CMC — what layout asymmetry causes this most often?
Likely cause: non-mirrored routing around the footprint (different via count/antipad/return path), unequal copper pour/keepout, or asymmetrical loading (TVS branch length mismatch).
Quick check: run a symmetry audit: same layer transitions, same via geometry, same plane continuity and stitching; if available, correlate with Sdc peak and differential skew.
Fix: enforce mirrored pair geometry, identical transitions, and equal exposure to copper/keepouts; keep any protection branches short and symmetric; re-validate with an A/B bypass check.
Pass criteria: eye asymmetry ≤ X%; skew ≤ X ps; Sdc peak ≤ X dB; BER ≤ X.
EMI peak moved upward in frequency after CMC — resonance with package/trace? first damping step?
Likely cause: a new LC resonance formed by the CMC package parasitics and nearby trace/launch; above SRF the CMC can behave capacitively and shift peaks upward.
Quick check: confirm the peak frequency with a near-field scan and correlate to a narrowband notch/peak in the channel (eye sensitivity at that band); A/B bypass to confirm the CMC is the contributor.
Fix: reduce stubs and tighten symmetry near the footprint; if damping is needed, start with the smallest effective damping change that preserves eye (layout-first, then part-family change).
Pass criteria: EMI peak reduced ≥ X dB (or below limit by X); no new in-band ΔIL > X dB; BER ≤ X; eye margin ≥ X.
Compliance margin is tight: remove CMC or change to lower Zdm loss? first decision rule?
Likely cause: compliance is loss- or resonance-limited, and the current CMC adds too much in-band ΔIL or creates a narrowband defect; alternatively, balance issues raise Sdc and eat margin.
Quick check: compare compliance margin with CMC bypass vs CMC; if margin recovers mainly by bypass, prioritize lower-loss/higher-balance CMC or remove; confirm with ΔIL (Sdd21) and Sdc peaks.
Fix: if loss dominates, switch to a lower-loss CMC or depopulate; if EMI/CM is the gating issue, keep a better-balance CMC and reduce discontinuities first; preserve optionality via footprint/bypass.
Pass criteria: compliance mask margin ≥ X; CMC-added ΔIL ≤ X dB; Sdc peak ≤ X dB; EMI margin ≥ X dB (or CM current reduction ≥ X%).
Cable OK, board-to-board fails — is it connector discontinuity or CMC interaction? first discrimination step?
Likely cause: the board-to-board launch/connector transition dominates reflections or mode conversion; the CMC may be amplifying the sensitivity if placed near a weak transition.
Quick check: TDR to locate the dominant step at the connector/launch; then bypass the CMC to see if failures persist. If bypass doesn’t help, the connector transition is first priority.
Fix: fix the connector/launch first (geometry, stub control, termination placement). Only after reflection is controlled, re-evaluate whether a CMC is needed for CM/EMI.
Pass criteria: connector/launch ΔZ ≤ X Ω; return loss (Sdd11) ≥ X dB; eye margin ≥ X; BER ≤ X with and without CMC (no regression).
Production yield drift across lots — what CMC parameters are most lot-sensitive and must be incoming-tested?
Likely cause: lot-to-lot variation in balance (mismatch), in-band Zdm insertion loss, and resonance behavior (SRF/peaking); DCR and current/bias conditions can shift behavior under real load.
Quick check: define an incoming test window at key frequencies: measure in-band ΔIL and a mode-conversion proxy (Sdc peak if available); log lot code, supplier, and correlation to BER/yield.
Fix: lock a correlation-based incoming spec (not just Zcm@100 MHz); qualify second sources by Sdc + ΔIL windows; trigger escalated sampling on supplier/lot change or yield shift.
Pass criteria: incoming ΔIL within baseline ± X dB; Sdc peak ≤ X dB; BER on golden channel ≤ X; yield ≥ X% with lot traceability = 100%.
