01 — Start With the Channel Requirement
Don't Start With the Differential Pair Rule
Before routing, define:
Protocol
PCIe?
CXL?
Ethernet?
DDR?
JESD204?
Custom SerDes?
Data Rate
What signaling rate?
Modulation
NRZ?
PAM4?
Other architecture?
Reach
Chip-to-chip?
Board-to-board?
Backplane?
Cable?
BER Requirement
How much statistical margin is required?
Package / Connector
What non-PCB structures are already consuming channel budget?
Equalization
What TX/RX capabilities exist?
The Protocol Defines the Channel Requirement.
The PCB Must Preserve It.
02 — Think in Channels, Not Nets
A CAD database shows: nets.
A signal-integrity engineer sees: channels.
A channel includes:
Driver
Package
PCB
Via
Connector
Cable
Receiver.
A Net Is a Logical Object. A Channel Is a Physical System.
03 — High Speed Is Defined by Edge Rate
A common mistake is:"This clock is only 100 MHz, so it is not high speed."
But if rise/fall time is very fast, the waveform contains much higher-frequency spectral content.
High-Speed Behavior Is Driven by Transition Time — Not Only Fundamental Frequency.
04 — Electrical Length
A trace becomes electrically significant when propagation delay is no longer negligible compared with signal transition time.
At that point: Lumped-Circuit Thinking Begins to Fail.
The interconnect must increasingly be treated as: A Transmission Line.
05 — Transmission-Line Behavior
A transmission line has distributed: L and C
along its length.
Its behavior depends on:
geometry
dielectric
reference plane
conductor properties
The signal does not instantly appear at the other end. Information Propagates as an Electromagnetic Wave.
06 — Characteristic Impedance
Characteristic impedance is approximately related to the distributed: L / C relationship of the interconnect.
It is therefore created by:
Trace Geometry
Reference Geometry
Dielectric. Impedance Is Geometry Expressed Electrically.
07 — 50 Ω / 85 Ω / 100 Ω Are Requirements — Not Trace Widths
A target impedance must be translated into manufacturable PCB geometry.
There is no universal:
100-ohm differential pair width and spacing. because stack-up changes everything. Impedance Target → Field Geometry → PCB Dimensions.
08 — Source Impedance
The transmitter has an effective output impedance.
If it does not relate appropriately to the channel and termination environment, reflected energy can occur.
Signal Integrity Begins at the Driver.
09 — Load Impedance
The receiver input is not an ideal infinite impedance.
It contains:
package parasitics
input capacitance
termination
ESD structures
The Receiver Is Part of the Channel.
10 — Reflection
When an electromagnetic wave encounters an impedance discontinuity, part of its energy can reflect.
Conceptually: Incident Wave → Transmitted Wave + Reflected Wave
The magnitude/polarity depend on the impedance change.
11 — Reflection Coefficient
A reflection coefficient describes the relationship between incident and reflected wave amplitudes.
The practical lesson is more important than the equation: Every Impedance Discontinuity Creates an Opportunity for Reflection.
12 — Why Reflections Matter
Reflections can create:
At lower speeds, the waveform may settle before sampling.
At higher speeds: There May Be No Time to Settle.
13 — Time Budget Shrinks as Data Rate Rises
As signaling rate increases, the symbol interval becomes shorter.
At very high speed:
Picoseconds Become Timing Budget.
This is why microscopic PCB geometry eventually becomes system-level performance.
14 — Unit Interval — UI
The Unit Interval represents one signaling interval.
Many SI requirements are usefully normalized in: UI.
This allows timing margin to scale with data rate.
At Higher Data Rates, the Same Picosecond Error Consumes More UI.
15 — NRZ Signaling
NRZ uses two primary amplitude levels.
Conceptually: 0 and 1. It provides one bit per symbol.
16 — PAM4
PAM4 uses four amplitude levels.
Each symbol carries two bits of information.
This allows higher bit throughput without doubling the fundamental symbol rate in the same way as binary signaling.
PCIe 6.0/7.0 use PAM4, and PCIe 7.0 reaches 128 GT/s.
But PAM4 creates: Three Eyes Instead of One. And substantially reduced voltage spacing between adjacent levels.
17 — Why PAM4 Is Harder
PAM4 trades: bandwidth efficiency for lower vertical signal margin.
This makes the system more sensitive to:
noise
linearity
reflections
crosstalk
jitter
Higher Throughput Per Symbol Comes With Smaller Eye Margin.
18 — Channel Budget
The complete system has finite tolerance for impairment.
Conceptually: TX Capability − Package Penalty − PCB Loss − Via Discontinuity − Connector Loss − Crosstalk − Jitter − Noise = Remaining Receiver Margin
This is why: Every Discontinuity Spends Part of the Channel Budget.
19 — Margin Engineering
The purpose of SI is not necessarily:
create a perfect waveform.
There is no perfect real-world waveform.
The objective is: Preserve Enough Margin for Reliable Communication Across Variation.
20 — Nominal Performance vs Margin
A nominal simulation may show a large eye.
But production adds:
process variation
temperature
voltage variation
material variation
silicon variation
Nominal Eye Opening Is Not Production Margin.
21 — Insertion Loss
Insertion loss describes how much signal transmission decreases through the channel versus frequency.
Major PCB sources include:
Conductor Loss
Dielectric Loss
Discontinuities. High-Frequency Energy Does Not Travel for Free.
22 — Frequency-Dependent Loss
A trace is not:
2 dB loss
at all frequencies.
Loss rises with frequency through several mechanisms. A Digital Channel Has an Analog Frequency Response.
23 — Conductor Loss
Copper has finite resistance.
At higher frequencies, current distribution changes because of: Skin Effect.
Current concentrates increasingly near conductor surfaces.
24 — Skin Depth
Skin depth decreases as frequency rises.
Therefore increasingly less conductor cross-sectional area carries much of the high-frequency current. High Frequency Changes the Effective Copper Geometry.
25 — Copper Roughness
Copper surfaces are not perfectly smooth.
At high enough frequency, surface roughness can increase effective conductor path length and loss. Microscopic Copper Texture Can Become Macroscopic Channel Loss.
26 — Roughness Modeling
High-speed channel models may use conductor-roughness models to better correlate simulation and measurement.
The exact model should match:
Ideal Smooth Copper Can Overestimate Real Channel Margin.
27 — Dielectric Loss
PCB dielectric absorbs electromagnetic energy.
Loss is influenced by properties including:
and frequency. Long high-speed channels can therefore require lower-loss laminates.
28 — Dk
Dielectric constant influences:
propagation velocity
impedance
wavelength
But a single datasheet Dk should not blindly be treated as universally correct. Dk Depends on Frequency, Material Construction, Test Method and Effective Field Distribution.
29 — Design Dk
For high-performance channel modeling, engineers may use an effective/design Dk that correlates better with actual manufactured transmission structures.
Material Characterization Should Serve Electrical Prediction.
30 — Material Selection
Material selection should balance:
Loss
Dk Stability
Manufacturability
Reliability
Cost
Availability.
Lowest Loss Is Not Automatically Best Product Design.
31 — Channel Reach
Higher loss limits practical electrical reach.
A system may compensate through:
equalization
retimers
lower-loss PCB
shorter routing
optical architecture
Reach Is a System Architecture Variable.
32 — Return Loss
Return loss characterizes reflected energy.
Poor return loss can originate from:
vias
connectors
package escape
neckdowns
pad transitions
A Channel Can Have Acceptable Loss and Still Fail From Reflection.
33 — Localized Discontinuities
A short discontinuity may look physically insignificant.
But at sufficiently high speed it can create:
reflection
resonance
mode conversion
Physical Length Is Not the Same as Electrical Significance.
34 — Trace Width Changes
Necking a transmission line changes its impedance.
If short, the penalty may be acceptable.
If long or severe: The Discontinuity Consumes Channel Margin.
35 — Reference-Plane Changes
If the signal changes layers or reference planes, its electromagnetic environment changes.
A Layer Transition Is Not Only a Signal-Via Event.
It is: Signal + Return-Current Transition.
36 — Return Path
High-frequency current flows as a complete loop.
Forward signal current has associated return current. Signal Integrity Is Loop Integrity.
37 — Return-Path Discontinuity
If return current cannot follow the signal closely, the loop expands.
This can increase:
inductance
radiation
common-mode energy
impedance disturbance
Many "Signal Problems" Are Actually Return-Path Problems.
38 — Ground Via Placement
A nearby ground via can provide return-current continuity during layer transitions.
The optimal geometry depends on:
stack-up
frequency
via structure
Signal Via and Return Via Should Be Designed as One Electromagnetic Transition.
39 — Multiple Return Vias
Higher-frequency structures may benefit from multiple return connections where they materially improve field confinement and symmetry.
More Ground Vias Are Useful Only If They Improve the Actual Field Structure.
40 — Via as a 3D Structure
A signal via consists of:
Pad
Antipad
Barrel
Stub
Reference Planes
Nearby Ground Vias.
A Via Cannot Be Reduced to "0.2 mm Drill."
41 — Via Inductance
The barrel and return environment introduce inductive behavior.
At high edge rates, even small inductance matters. Nanohenries Can Become Significant Impedance.
42 — Via Capacitance
Pad-to-plane geometry contributes capacitance.
Larger pads/smaller antipads can increase capacitive loading. Via Design Is an L-C Optimization Problem.
43 — Antipad Engineering
Antipad diameter strongly influences:
Sometimes Removing Ground Copper Improves the High-Speed Transition.
44 — Via Stub
Unused via barrel forms a stub.
A stub behaves as a resonant transmission structure.
At a critical frequency: It Can Become a Serious Channel Impairment.
45 — Stub Resonance
Stub resonance depends on:
physical length
dielectric propagation velocity
As data rates rise: Shorter Stubs Become Relevant.
46 — Back Drilling
Back drilling removes unused via barrel.
This can materially improve:
insertion loss
return loss
resonance behavior
for demanding channels.
Remove the Stub Before Trying to Equalize Around It.
47 — Residual Stub
Back drilling does not create zero residual stub.
Manufacturing tolerances determine the remaining barrel length. Residual Stub Is a Manufacturing-Controlled Electrical Parameter.
48 — Blind / Microvia Transitions
HDI structures can shorten vertical transitions.
Potential benefits include:
lower parasitics
reduced stub
but they increase:
High-Speed Benefit Must Be Balanced Against Manufacturing Risk.
49 — Coaxial / Shielded Via Concepts
For extreme high-speed or RF transitions, advanced structures can attempt to make the signal and ground environment more coaxial.
The goal is:
field confinement
return-path continuity
impedance control. The Vertical Interconnect Can Itself Be Engineered as a Transmission Line.
50 — Via Arrays
High-density SerDes packages may contain many adjacent signal vias.
Individual via optimization is not enough.
Engineers must also consider: Via-to-Via Crosstalk.
51 — Crosstalk
Electromagnetic energy from an aggressor channel can couple into another channel.
Coupling arises through: Electric Field and Magnetic Field.
High-Speed Nets Communicate Even When the Schematic Says They Are Not Connected.
52 — NEXT
Near-End Crosstalk is observed near the aggressor source side. Its magnitude depends on the transmission structure and coupling.
53 — FEXT
Far-End Crosstalk appears toward the far end.
Different PCB transmission structures can produce different FEXT characteristics. Crosstalk Is Distributed Electromagnetic Coupling.
54 — Spacing Rules
Simple rules such as:
3W
can provide useful first-order guidance.
But they are not universal physics limits. Required Spacing Depends on Stack-Up, Parallel Length, Edge Rate and Allowed Noise.
55 — Parallel Coupling Length
Two lines may be close briefly without significant total coupling.
Long parallel runs increase interaction. Crosstalk Is a Function of Both Distance and Length.
56 — Layer-to-Layer Crosstalk
Adjacent routing layers without an intervening reference plane can couple strongly.
Stack-Up Can Solve Crosstalk Before Routing Begins.
57 — Broadside Coupling
Signals on neighboring layers can couple vertically.
This may be worse than obvious same-layer coupling. What You Cannot See in 2D Can Still Couple in 3D.
58 — Differential Signaling
Differential signaling uses two related conductors carrying complementary information.
Ideally, external noise couples similarly to both and is rejected by the receiver.
But this depends on: Symmetry.
59 — Differential Mode
Useful signal energy ideally occupies the differential mode.
60 — Common Mode
Asymmetry can convert some differential energy into: Common-Mode Energy.
Common-mode energy can:
Differential-to-Common-Mode Conversion Is Both SI and EMC.
61 — Mode Conversion
Sources include:
Differential Pair Symmetry Must Extend Through Every Transition.
62 — Pair Skew
If P and N paths have different propagation delay, the receiver sees reduced temporal alignment.
This can create:
Intra-Pair Delay Mismatch Is More Important Than Pretty Equal Trace Length.
63 — Glass-Weave Skew
At sufficiently high speed, one trace of a pair may travel over different glass/resin composition than the other.
Because effective dielectric properties differ: Differential Delay Can Change.
64 — Glass-Weave Mitigation
Possible strategies can include:
depending on platform.
Material Microstructure Can Become Timing Architecture.
65 — Differential Pair Coupling
The desired amount of coupling depends on:
geometry
stack-up
routing constraints
A pair does not need to be as tightly coupled as physically possible in every design.
Differential Behavior Comes From the Complete Transmission Structure.
66 — Pair Separation Through BGA Escape
BGA breakout sometimes forces temporary geometry changes.
Those transitions should be analyzed where margin is limited. The Channel Begins at the Package Escape — Not After Fanout Is Finished.
67 — Connectors
A connector contains:
contacts
dielectric
ground structures
launches
and has its own S-parameters.
A Connector Is a Multi-Port Electromagnetic Component.
68 — Connector Launch
The PCB footprint/launch can dominate the performance of an otherwise excellent connector.
Variables include:
reference structure.
A Good Connector With a Bad Launch Is a Bad Channel.
69 — Breakout Footprint Optimization
Footprints optimized only for assembly may not be electrically optimal at extreme rates.
The challenge is to satisfy:
simultaneously.
70 — Cable Channels
High-speed cable links add:
cable loss
connector transitions
mode conversion
The PCB launch into the cable can be just as important as the cable itself.
71 — Backplanes
Backplane channels can include:
daughter card
↓
connector
↓
backplane
↓
connector
↓
daughter card.
At high speeds, every section needs a shared channel budget.
72 — Package Effects
The signal begins inside the IC package.
Package loss and discontinuity can consume meaningful margin before the signal reaches the PCB. The PCB Does Not Receive an Ideal Signal at the BGA Ball.
73 — Package Models
High-end channel analysis may include vendor-provided:
S-parameters
package models
IBIS-related models
where available.
Package + PCB Co-Simulation Can Be Essential When Margin Is Tight.
74 — Die-to-Package Interaction
At cutting-edge speeds, semiconductor I/O, package escape and PCB can no longer be optimized completely independently.
Silicon and PCB Meet Through Electromagnetic Structures.
75 — SerDes
A modern SerDes contains far more than: TX → Wire → RX.
The PHY can include:
equalization
clock recovery
training
FEC
depending on standard. Modern High-Speed Links Are Communication Systems.
76 — TX Equalization
Transmitters may use: Pre-Emphasis or De-Emphasis to reshape spectral content.
The goal is to compensate predictable channel loss. The Transmitter Can Pre-Distort the Signal to Help the Receiver Recover It.
77 — FFE
Feed-Forward Equalization can use several taps to shape the transmitted waveform.
At high rates: TX Settings Become Channel-Dependent.
78 — CTLE
A Continuous-Time Linear Equalizer can boost higher-frequency components relative to lower frequencies at the receiver.
Equalization Trades Noise and Gain for Channel Compensation.
79 — DFE
Decision Feedback Equalization uses previously detected symbols to cancel predictable intersymbol interference.
The Receiver Can Use Past Decisions to Improve the Present Decision.
80 — Equalization Limits
Equalization cannot repair every problem.
It cannot magically eliminate:
deep notches
excessive reflections
severe crosstalk
nonlinearity
Equalization Is Margin Recovery — Not Permission for Bad PCB Design.
81 — Channel Training
Modern protocols may automatically tune:
TX equalization
RX equalization
during link startup.
Training Optimizes the PHY Around the Actual Channel. But the channel still must remain inside a trainable operating region.
82 — Retimers
A retimer recovers data and timing, then retransmits the signal. CXL 4.0, for example, supports up to four retimers to extend channel reach.
Retimers Reset Channel Margin — at a Cost.
That cost includes:
thermal complexity.
83 — Redrivers
A linear/redriver architecture can amplify/equalize without fully terminating and recreating the protocol link.
Its benefit/limitations are architecture-specific. Redriver and Retimer Are Not Equivalent Solutions.
84 — Jitter
Jitter is variation in signal transition timing.
At the receiver it consumes: Horizontal Eye Margin.
85 — Random Jitter
Random jitter is often modeled statistically.
Sources can include:
thermal noise
oscillator noise
Its tails matter at very low BER targets.
86 — Deterministic Jitter
Deterministic jitter may arise from:
data-dependent effects
periodic interference
duty-cycle distortion
Different Jitter Mechanisms Require Different Solutions.
87 — Data-Dependent Jitter
Channel bandwidth limitation creates intersymbol interference.
Earlier bits influence later transition timing. The Previous Bit Pattern Can Move the Present Edge.
88 — Periodic Jitter
Power supplies, clocks or external interferers can modulate timing periodically.
Power Noise Can Become Timing Noise.
89 — Clock Jitter
Clock and PLL quality can influence both:
transmitted timing
receiver sampling
A Perfect PCB Cannot Repair a Fundamentally Poor Clock Architecture.
90 — Jitter Budget
Total timing uncertainty must fit inside the available UI.
Conceptually:
Channel ISI
TX Jitter
Clock Jitter
Crosstalk
RX Effects
must preserve sufficient sampling margin. Timing Margin Is a Budget.
91 — Eye Diagram
An eye diagram overlays many symbol intervals.
It exposes combined effects of:
jitter
noise
ISI
reflections
The Eye Is the Visual Summary of Channel Margin.
92 — Eye Height
Eye height represents vertical voltage margin.
It can shrink from:
noise
crosstalk
attenuation
93 — Eye Width
Eye width represents timing margin.
It can shrink from:
94 — PAM4 Eye Analysis
PAM4 creates three eyes.
Each eye may have different:
PAM4 Must Preserve Multiple Decision Thresholds Simultaneously.
95 — Eye Diagram Is Not BER
An apparently open eye does not necessarily guarantee the required extremely low error probability.
Statistical Tail Behavior Matters.
96 — Bit Error Rate — BER
BER is the proportion of incorrect bits.
High-performance systems may require extremely low failure probabilities. Testing every possible bit directly can become impractical.
This drives: Statistical SI Analysis.
97 — Bathtub Curve
A bathtub curve characterizes BER sensitivity versus sampling time.
It can reveal timing margin at defined error probability. BER Turns Eye Margin Into Probability.
98 — Q-Scale / Statistical Eye
Statistical tools can estimate extremely low BER regions without brute-force simulation of every bit.
High-Speed SI Is Increasingly Statistical Engineering.
99 — PRBS Patterns
Pseudo-random bit sequences are commonly used to exercise channel behavior.
Different patterns stress different:
run lengths
spectral content
equalization behaviors.
Test Pattern Matters.
100 — Pattern-Dependent Behavior
A channel can behave differently for: 101010
versus: long repeated runs.
This is due to finite channel bandwidth and memory. Digital Channels Have Memory.
101 — Intersymbol Interference — ISI
Energy from one symbol extends into neighboring symbols.
This creates:
data-dependent jitter.
Loss Becomes Timing and Voltage Error Through ISI.
102 — Pulse Response
The channel pulse response reveals how one transmitted pulse spreads over time.
If One Bit Persists Into Many Future Bits, the Channel Has Significant Memory.
103 — Impulse Response
Frequency-domain S-parameters can be transformed into time-domain behavior.
This links: S-Parameters to Eye Simulation.
Frequency and Time Domains Describe the Same Physical Channel.
104 — S-Parameters
S-parameters represent network behavior versus frequency.
For a 2-port channel: S11 → input reflection
S21 → forward transmission.
But real differential channels often need: Multi-Port S-Parameters.
105 — 4-Port Differential Channels
A differential pair can be represented as a four-port single-ended network and transformed into mixed-mode parameters.
This enables analysis of:
mode conversion.
106 — Mixed-Mode S-Parameters
Examples include:
SDD21 Differential-to-differential transmission.
SDD11 Differential reflection.
SCD21 Differential-to-common-mode conversion.
Mixed-Mode Parameters Reveal What Ordinary Insertion Loss Can Hide.
107 — Touchstone
Touchstone is a widely used format for storing network S-parameters.
The IBIS Open Forum's current Touchstone 2.1 specification was ratified in 2024.
A Channel Model Can Become Portable Measurement/Simulation Data.
108 — Passivity
A passive interconnect model should not mathematically generate energy.
Poorly processed S-parameter data can violate passivity.
Model Quality Matters.
109 — Causality
A physical channel cannot respond before excitation arrives.
Model processing should preserve causal behavior.
A Non-Causal Model Can Produce Beautiful but Physically Impossible Simulation Results.
110 — Frequency Range
S-parameter measurement/simulation must cover enough bandwidth for the channel problem.
If the Model Stops Before Important Harmonics Do, Time-Domain Results Can Become Misleading.
111 — Port Definition
The location and impedance of measurement/simulation ports matter.
S-Parameters Describe the Network Between Defined Reference Planes.
Changing the reference plane changes the model.
112 — De-Embedding
Fixtures and launches can be mathematically removed to move the reference plane closer to the structure of interest.
Measure the DUT — Not the Fixture.
113 — Fixture Design
Poor fixtures can create more discontinuity than the PCB structure being measured.
High-Speed Measurement Hardware Is Part of SI Engineering.
114 — TDR
Time-Domain Reflectometry launches a fast edge and observes reflections versus time.
It can identify:
impedance changes
discontinuity locations
TDR Shows Where the Channel Changes.
115 — TDR Impedance Profile
A channel might show:
100 Ω
↓
85 Ω via
↓
100 Ω trace
↓
110 Ω connector.
That spatial information is extremely useful.
TDR Converts Geometry Problems Into Time-Domain Location.
116 — TDR Resolution
Measurement rise time limits spatial resolution.
You Cannot Resolve a Tiny Discontinuity With an Extremely Slow TDR Edge.
117 — Differential TDR
Differential TDR allows characterization of the complete pair structure rather than only each conductor independently.
118 — TDT
Time-Domain Transmission observes the waveform transmitted through the structure.
Together with reflection data it helps characterize the channel.
119 — VNA
Vector Network Analyzers measure magnitude and phase of S-parameters over frequency.
For high-speed channels they can characterize:
insertion loss
return loss
crosstalk
mode conversion.
VNA Tells the Channel Story in Frequency.
TDR Tells It in Time.
120 — Measurement Bandwidth
A measurement must support the required:
frequency
dynamic range
port count
of the channel. Measurement Capability Must Scale With Channel Speed.
121 — Calibration
VNA measurements require appropriate calibration.
The goal is to move measurement accuracy to a known reference plane. Calibration Defines Where Measurement Reality Begins.
122 — Repeatability
At extreme speeds:
cable movement
connector torque
probe placement
can change results.
Measurement Repeatability Is Part of Measurement Accuracy.
123 — Simulation-to-Measurement Correlation
One of the highest levels of SI engineering maturity is:
CAD Geometry
↓
EM Extraction
↓
Simulation
↓
Fabrication
↓
VNA / TDR
↓
Compare
↓
Model Correction
Build a Model That Predicts the Board the Factory Actually Makes.
124 — Correlation Error
If simulation and measurement do not agree, possible reasons include:
wrong material data
wrong copper roughness
geometry deviation
connector model
fixture error
Disagreement Is Engineering Information.
125 — Material Characterization Through Correlation
Test structures can help determine effective:
for a real fabrication process. The Factory Can Build Electrical Knowledge About Its Own Materials and Processes.
126 — Coupon Design
Test coupons can represent:
transmission line
via
material
impedance
characteristics. A Coupon Is a Controlled Experiment Fabricated Beside the Product.
127 — Product vs Coupon
A coupon does not automatically prove every on-board channel.
Differences can exist in:
location
copper distribution
local stack-up/process behavior.
Coupon Data Is Evidence — Not a Substitute for Channel Engineering.
128 — IBIS Modeling
IBIS models describe digital I/O electrical behavior without revealing proprietary transistor-level details.
Current IBIS development remains active: IBIS 8.0 was ratified on December 5, 2025, and the current Open Forum also maintains Touchstone and interconnect-model standards.
IBIS Connects Silicon I/O Behavior to Board-Level SI Simulation.
129 — Why IBIS Matters
A PCB channel alone does not create the final waveform.
The driver has:
rise/fall behavior
impedance
nonlinearity
The receiver has:
threshold behavior. SI Simulation Needs the Electronics at Both Ends.
130 — IBIS Model Quality
Not every model is equally accurate.
A good workflow may check:
completeness
consistency
correlation
where possible.
The Simulation Is Only as Reliable as the Models Feeding It.
131 — IBIS-AMI
Traditional bit-by-bit transistor-level simulation becomes computationally difficult for extremely high-speed SerDes.
IBIS-AMI enables statistical/algorithmic modeling of advanced SerDes behavior.
Its importance continues to grow as equalization and signal-processing complexity increase; IBIS 8.0 added further AMI-related support and 2026 IBIS technical work includes AMI modeling of advanced D2D and high-speed systems.
At Modern SerDes Speeds, the PHY Algorithm Is Part of the Channel Simulation.
132 — Statistical Simulation
AMI-style workflows can evaluate enormous virtual bit populations efficiently.
This is useful for:
channel comparison.
Statistical SI Makes Trillion-Bit Questions Computationally Practical.
133 — SerDes Algorithm Models
TX/RX models may include:
depending on the provided model.
The Receiver Is Increasingly a Signal-Processing Engine.
134 — Eye Prediction With AMI
A physical channel S-parameter model can be combined with algorithmic SerDes behavior to predict system eye/BER performance.
PCB SI Is Moving From Interconnect Simulation Toward Link Simulation.
135 — COM — Channel Operating Margin
Some standards/ecosystems use Channel Operating Margin-style methods to combine channel impairments into a standardized metric.
The broader value is: Convert Complex Channel Behavior Into a Compliance-Oriented Margin Metric.
IBIS technical sessions in 2026 continue discussing the relationship between COM and IBIS-AMI for advanced SerDes analysis.
136 — Compliance vs Engineering Margin
Passing a standardized channel metric is important.
But product design may need additional margin for:
production variation
environmental variation
interoperability.
Compliance Is a Boundary.
Engineering Margin Determines Robustness.
137 — PCIe Signal Integrity
PCIe channels must satisfy generation-specific electrical requirements involving:
loss
reflections
jitter
crosstalk
and increasingly: PAM4 behavior.
PCIe 7.0 is currently the approved top-generation PCIe specification at 128 GT/s, released June 11, 2025.
PCIe Generation Changes the Channel Budget Dramatically.
138 — PCIe 7.0
PCIe 7.0 continues PAM4 signaling and doubles PCIe 6.x's rate from 64 to: 128 GT/s.
At this level, small discontinuities that were insignificant generations earlier can consume substantial margin.
139 — PCIe Lane Architecture
A PCIe link may use:
x16.
Each lane is a high-speed serial channel.
Multi-lane PCB architecture must also consider:
lane mapping
routing density
crosstalk
140 — PCIe Clocking
PCIe systems can use different supported clock architectures depending on generation/platform.
The exact implementation must follow the applicable specification and devices.
Clock Architecture Changes Jitter Requirements.
141 — PCIe Retimers
Retimers can extend reach by terminating and regenerating the link.
But every retimer adds:
system complexity. Strong PCB SI Can Reduce Dependence on Active Channel Repair.
142 — CXL Signal Integrity
CXL builds on the PCIe physical infrastructure while adding coherent/cache/memory protocols.
CXL 4.0 doubled data rate to 128 GT/s, building on PCIe 7.0.
Memory Coherency Still Depends on Analog Channel Physics.
143 — Why CXL Margin Matters
CXL increasingly connects:
CPUs
accelerators
memory devices
switches.
An unstable electrical link becomes a: System Architecture Failure. Not just a PCB inconvenience.
144 — CXL Retimer Reach
CXL 4.0's support for up to four retimers illustrates how the industry is balancing extreme signaling rate against real system reach.
Channel Architecture Is Becoming a Platform-Level Design Decision.
145 — 224G Electrical Interfaces
OIF's CEI-224G work targets several reach classes:
with MR targeting electrical links through up to approximately 500 mm PCB and a connector, and LR up to approximately 1000 mm backplane and up to two connectors within the project's defined signaling range.
This illustrates a critical principle: There Is No Single "224G PCB Channel."
Reach defines the electrical challenge.
146 — XSR
Extra Short Reach targets extremely short electrical paths such as:
die-to-die
die-to-optical-engine
architectures.
Short Reach Trades Channel Loss for Extreme Density and Power Efficiency.
147 — VSR
Very Short Reach commonly addresses:
architectures.
This becomes especially important for:
networking platforms.
148 — MR
Medium Reach must tolerate more PCB path and physical transitions.
Each Additional Millimeter and Connector Requires More Channel Engineering at 224G-Class Rates.
149 — LR
Long Reach electrical links place the greatest stress on:
material loss
connectors
equalization
crosstalk.
Long Copper Reach at Extreme Data Rate Is a System-Level Achievement.
150 — The 448G Direction
As of 2026, OIF lists: CEI-448G Framework
and active:
projects in current work; 448G was also part of OIF's OFC 2026 interoperability showcase.
This should be presented on the 365PCB website as: Technology Direction.
Not as: "365PCB has already qualified 448G production channels." unless evidence exists.
151 — Why 448G Matters to PCB Engineering
It tells us where the industry is going:
Tomorrow's PCB Requirements Are Being Defined by Today's SerDes Research.
152 — Electrical vs Optical Reach
As electrical channel difficulty rises, systems increasingly evaluate optical conversion closer to compute/switch silicon.
PCIe 7.0 itself includes support goals around optical interconnection and extended reach architectures.
Copper and Optics Are Becoming Complementary Interconnect Technologies.
153 — Co-Packaged Optics
Moving optical conversion closer to high-bandwidth silicon can reduce long electrical reach.
But it creates:
thermal
optical
packaging
power
serviceability
challenges. Moving the Electrical Boundary Moves the Engineering Boundary.
154 — DDR Signal Integrity
DDR differs from SerDes.
It combines:
Parallel Timing
Strobes
Command / Address
Power / References. DDR Is Timing Integrity as Much as Signal Integrity.
155 — DDR Timing Budget
Important relationships exist among:
DQ
DQS
clock
command/address
depending on memory generation.
Match What the Controller Samples Together.
156 — DDR Fly-By
Command/address topology may intentionally introduce propagation differences.
Training then compensates expected behavior.
Routing Topology and Controller Training Are Designed Together.
157 — DDR DQ / DQS
The relationship within a byte lane matters more than making every data line on the entire board identical.
Timing Groups Should Drive Length Groups.
158 — DDR VREF
Reference voltages directly influence receiver decision thresholds.
DDR Is a Digital Bus With Analog Reference Sensitivity.
159 — DDR Power Integrity
Rapid simultaneous switching can modulate:
ground.
DDR SI Without DDR PI Is Incomplete Engineering.
160 — Read / Write Eye
Memory interfaces may have different margins for:
controller → DRAM
DRAM → controller.
Direction Matters Because Drivers and Loading Change.
161 — Package / Board Timing
Controller and DRAM package delays contribute to the complete timing relationship.
PCB Length Matching Alone Does Not Define Total Path Delay.
162 — SI / PI Coupling
Power supply noise can shift:
driver edge timing
receiver thresholds
clock phase.
Power Noise Becomes Signal Error.
163 — Simultaneous Switching Noise
Multiple I/Os switching together produce transient supply demand.
Effects can include:
jitter.
The Digital Data Pattern Can Become a Power-Integrity Excitation Pattern.
164 — PDN-Induced Jitter
PLL/transceiver supply noise may modulate timing.
Jitter Can Enter Through the Power Pins.
165 — Reference Noise
A receiver may make decisions relative to:
common mode.
If the reference moves: The Threshold Moves Even If the Signal Does Not.
166 — SI / PI Co-Simulation
Advanced systems may combine channel analysis with:
supply-noise effects. High-Speed Links Operate Inside a Power-Distribution Environment.
167 — EMI and SI
Many SI failures also create EMC symptoms.
Example:
Differential imbalance
↓
common-mode conversion
↓
cable current
↓
radiation.
A Better Channel Can Also Be a Quieter Channel.
168 — Common-Mode Radiation
A tiny common-mode voltage driving a large external cable can create significant emissions.
EMC Is Often Driven by Small Electrical Imbalances Connected to Large Antennas.
169 — Spread Spectrum Clocking
Some interfaces use spread-spectrum clocking to reduce concentrated spectral energy.
The SI analysis should still account for the applicable clock behavior.
Frequency Modulation Can Affect Both EMI and Timing Analysis.
170 — PCB Geometry Tolerance
Fabrication introduces variations in:
trace width
trace thickness
dielectric thickness
registration.
The CAD Channel Is Only the Mean of a Manufacturing Distribution.
171 — Impedance Distribution
Suppose nominal impedance is:
100 Ω.
Production may create a distribution around that value.
The key question is: Does the Entire Acceptable Manufacturing Distribution Preserve Channel Margin?
172 — Loss Distribution
Copper roughness, material variation and geometry can also change insertion loss.
Production Channel Loss Is a Distribution Too.
173 — Dk Variation
Material dielectric properties vary between:
constructions
lots
frequency
environmental conditions.
Material Tolerance Becomes Timing Tolerance.
174 — Etch Variation
Finished trace geometry differs from nominal artwork.
Manufacturing Etch Becomes Impedance.
175 — Registration Variation
Layer alignment can alter:
via antipad geometry
coupling
return structures.
Registration Can Become SI at High Density.
176 — Backdrill Tolerance
Residual stub length varies.
Therefore post-layout SI should not always model: perfect zero stub.
Model the Manufacturing Process — Not an Impossible Ideal.
177 — Connector Tolerance
Connector manufacturing also introduces:
pin geometry
material
assembly
variation.
Every Channel Segment Has Its Own Statistical Distribution.
178 — Statistical Corner Analysis
A stronger design can evaluate combinations of:
fast/slow silicon
material variation
temperature
manufacturing geometry
to understand worst-case margin.
Worst Case Is Rarely One Single Parameter at Its Maximum.
179 — Monte Carlo SI
Where required, statistical analysis can evaluate populations of geometry/model variation.
Production Reliability Is a Probability Problem.
180 — Sensitivity Analysis
Instead of asking only: Does this channel pass?
ask: Which variable causes the most margin loss?
Potential answers:
via stub
connector
trace loss
crosstalk
material.
Sensitivity Analysis Tells Engineers Where Improvement Creates the Most Value.
181 — Design Margin Allocation
If the system has 10 units of margin, spending 8 on the PCB leaves very little for:
temperature.
Don't Consume the Entire System Budget in One Discipline.
182 — Channel Optimization
High-performance engineering should optimize the largest penalty first.
If: connector launch = 30% of problem and trace bend = 1%
then polishing bends is not the priority. Optimize by Contribution — Not by Visual Obsession.
183 — Material vs Geometry Trade-Off
An expensive low-loss laminate may recover margin.
But shortening the channel by changing placement may recover even more at lower cost.
Architecture Can Be Cheaper Than Exotic Material.
184 — Via vs Layer Trade-Off
Changing BGA escape layer can reduce:
stubs.
But it may require more PCB layers. SI Optimization Has Cost Consequences.
185 — Retimer vs Better PCB
One architecture might use: standard material + retimer.
Another: low-loss material + shorter routing.
The right answer depends on: Cost + Power + Thermal + Reliability + Performance.
High-Speed SI Is Product Architecture Economics.
186 — Pre-Layout SI
Before detailed PCB routing, engineers can answer high-value questions:
Is standard material sufficient?
How long can the channel be?
Is backdrilling needed?
Which connector?
Which stack-up?
Use Simulation to Choose the Architecture Before CAD Makes It Expensive to Change.
187 — Topology Exploration
Compare:
before committing. Simulation Is Most Valuable Before the Design Is Frozen.
188 — Stack-Up Exploration
Different:
dielectric heights
trace geometries
routing layers
can be compared. Stack-Up Is an SI Design Variable.
189 — Via Exploration
Simulate: through via vs backdrilled via vs blind via
where relevant. Know the Electrical Value Before Buying the Manufacturing Complexity.
190 — Connector Exploration
Connector selection can be compared using:
vendor models
PCB launch models
before mechanical design is finalized.
Select the Connector With the Channel — Not After the Enclosure.
191 — Post-Layout Extraction
After PCB routing, simulation should use: Actual Geometry.
Not the early idealized routing assumptions.
192 — Critical-Channel Extraction
The most important channels can be extracted with:
for final analysis. Sign Off the Board That Will Be Manufactured.
193 — Full-Wave EM
Complex 3D discontinuities may require solving Maxwell's equations numerically.
Examples:
via fields
BGA escape
connector launch
When Rules Stop Being Reliable, Solve the Fields.
194 — 2D Field Solvers
Transmission-line structures can often be efficiently modeled using 2D electromagnetic field solvers.
Useful for:
coupling. Use the Simplest Accurate Physics Model for the Problem.
195 — 3D EM Solvers
For structures where fields vary significantly in all three dimensions: 3D Modeling Becomes More Valuable.
196 — Model Complexity
More complex simulation is not automatically better.
A model should be: As Detailed as Necessary — and No More.
Because excessive complexity increases:
debugging.
197 — Simulation Verification
Always ask:
Are ports correct?
Are materials correct?
Are boundaries correct?
Is frequency range sufficient?
A Solver Will Precisely Solve the Model You Built — Even If the Model Is Wrong.
198 — Channel Sign-Off
A mature sign-off might include:
Impedance
Insertion Loss
Return Loss
Crosstalk
Mode Conversion
Eye / BER
Protocol Metric
depending on interface.
No Single SI Metric Defines Every Channel.
199 — Pass / Fail Criteria
Every analysis should connect to an explicit requirement.
Not: "The eye looks pretty good."
But: Does It Meet the Required Margin?
200 — Measurement Planning Before PCB Release
If high-speed performance is important, define before fabrication:
where to probe
what coupons
what connectors
what reference structures.
Design the Board So the Channel Can Be Verified.
201 — Probe Access
High-speed probe pads can disturb the very channel being measured.
Test Access Is Another Discontinuity.
Therefore it should be engineered.
202 — Breakout / Test Structures
Dedicated replicated structures may allow easier characterization without disturbing the real product channel.
Sometimes Measure a Representative Structure Instead of Damaging the Product Channel.
203 — Golden Channel
A characterized reference board/channel can support comparison across builds.
But: Golden Means Characterized — Not Simply "This One Worked."
204 — Production Impedance Testing
Controlled-impedance manufacturing may use TDR coupon testing. Manufacturing Should Verify the Transmission Structure It Was Asked to Build.
205 — Loss Testing
For sufficiently demanding platforms, selected production/test coupons can characterize frequency-dependent loss.
High-Speed Manufacturing Quality Can Extend Beyond DC Continuity.
206 — Lot-to-Lot Correlation
Compare electrical data across PCB lots.
This can reveal:
material variation
process shift
impedance drift.
SI Measurements Can Become Manufacturing SPC Data.
207 — Manufacturing Statistical Process Control
Imagine tracking:
impedance
Dk-related delay
insertion loss
over time.
That can move advanced PCB manufacturing from: Specification Checking
toward: Process Intelligence.
208 — Correlating Fabrication Data
Electrical results can be connected with:
laminate lot
press cycle
etch data
copper information
drill/backdrill.
The Future High-Speed PCB Factory Should Know Which Manufacturing Variables Move Electrical Performance.
209 — Root-Cause Analysis
A failing link can originate from:
Silicon
Package
PCB
Connector
Power
Clock
Firmware / Training.
"PCIe Link Failure" Is a Symptom — Not a Root Cause.
210 — Link-Speed Downgrade
A system that trains successfully at a lower generation but not the highest generation is valuable evidence.
It may indicate:
Marginal High-Frequency Channel Performance.
But the exact cause still requires analysis.
211 — Lane-Specific Failure
If only one lane fails repeatedly, investigate lane-specific:
routing
via
connector
assembly
differences.
Multi-Lane Links Provide Built-In Comparative Experiments.
212 — Temperature-Dependent Failure
If failure appears only hot/cold, possible contributors include:
silicon
material
connector
clock
power. Environmental Sensitivity Is a Clue to the Physical Mechanism.
213 — Board-to-Board Variation
If some units pass and others fail: Manufacturing Distribution Has Entered the Electrical Margin.
This is exactly where 365PCB's manufacturing position becomes strategically important.
214 — Failure Correlation With TDR
A failed channel can be compared against a passing channel.
A new impedance discontinuity may reveal:
difference.
Compare Good vs Bad — Not Bad vs Imagination.
215 — Failure Correlation With S-Parameters
Frequency-domain comparison can reveal:
reflection. The Channel Can Tell You How It Changed.
216 — X-Ray / Cross-Section Correlation
If electrical data suggests a via/connector geometry problem, physical analysis can examine the actual structure.
SI Failure Analysis Can Move From Waveform to Metallography.
217 — Microsection Correlation
PCB cross-section can reveal:
plating
via geometry
dielectric thickness
registration.
Physical Geometry Can Explain Electrical Geometry.
218 — Failure Analysis Loop
A mature process is:
Failing Link
↓
Electrical Measurement
↓
Locate Suspected Structure
↓
Physical Inspection
↓
Root Cause
↓
Process Correction
↓
Electrical Re-Verification
Signal Integrity Can Become Manufacturing Root-Cause Engineering.
219 — AI Servers
AI compute platforms combine:
GPUs / accelerators
PCIe / CXL
high-speed networking
high-current PDNs
on extremely dense boards. AI Hardware Is a Signal-Integrity and Power-Integrity Co-Design Problem.
220 — HPC
HPC platforms push:
bandwidth
memory capacity
processor density
simultaneously.
This increases:
channel count
thermal density
routing density. Interconnect Quality Can Limit Compute Scalability.
221 — Networking
800G/1.6T-class networking architectures place enormous pressure on:
PCB materials.
OIF's 224G and emerging 448G work shows this direction clearly.
Networking PCB Design Is Increasingly Channel-Budget Engineering.
222 — Test & Measurement
High-speed instrumentation requires:
often over broad bandwidth.
Measurement Electronics Must Have Better Signal Integrity Than the Signals They Are Trying to Measure.
223 — FPGA Platforms
Modern FPGAs can expose many multi-gigabit transceivers.
The PCB must coordinate:
transceiver banks
reference clocks
power
connectors
channel routing.
FPGA Logic Is Flexible.
PCB Channel Geometry Is Not.
224 — High-Speed Data Acquisition
Fast ADC/DAC systems can involve:
JESD204-class interfaces
extremely clean clocks
analog + digital interaction. Data Converter Performance Can Be Lost After Conversion if the Digital Interconnect Is Poor.
225 — Chiplet / D2D System Context
As package-level D2D becomes faster, the board must connect increasingly high-performance package I/O to:
networking.
The Electrical Margin Saved Inside the Package Cannot Be Wasted on the PCB. This is one of the core ideas behind our 365PCB Chiplet/D2D technical content.
226 — Manufacturing Becomes Electrical Design
At lower speeds, a few microns of trace variation may have little system consequence.
At extreme speeds, the same variation can affect:
loss.
Manufacturing Tolerance Becomes Electrical Performance.
227 — CAD Geometry vs Fabricated Geometry
Simulation frequently uses: nominal geometry.
The customer receives: fabricated geometry.
The Difference Between the Two Is Manufacturing Risk.
228 — Process-Aware SI
The strongest future workflow is therefore:
Design
↓
Simulate
↓
Apply Manufacturing Tolerance
↓
Fabricate
↓
Measure
↓
Correlate
↓
Improve
Signal Integrity Should Close the Loop With Manufacturing.
229 — High-Speed DFM
DFM for a high-speed board cannot ask only:
Are trace/space manufacturable?
It should also ask:
Can manufacturing variation preserve the channel behavior?
High-Speed DFM Is Electrical DFM.
230 — Material Availability
A channel optimized around one exact laminate may create supply risk.
Where product constraints allow, engineering can evaluate qualified material alternatives.
Material Supply Continuity Can Become SI Risk.
231 — Material Substitution
A laminate substitute should not be approved only because: "Dk looks similar."
Also evaluate:
Df
construction
copper
thickness
frequency-dependent behavior.
Material Substitution Is Channel Requalification.
232 — Fabricator-to-Fabricator Transfer
Changing PCB supplier can change:
stack-up
material construction
etch process
impedance geometry.
Same Gerber Does Not Guarantee Same High-Speed Channel.
This is strategically important for365PCB.
233 — Production Transfer Documentation
A robust release should define:
material
stack-up
impedance
critical structures
backdrill
electrical validation.
Transfer the Electrical Intent — Not Only Artwork.
234 — SI Capability Matrix
365PCB should eventually build an evidence-backed public matrix containing verified capabilities such as:
but only where real equipment/process/data support the claim. Capability Should Be Documented — Not Implied.
235 — Avoid "Maximum Gbps" Marketing
I strongly recommend 365PCB avoid writing:
We support 224 Gbps PCB design.
as a standalone claim.
Because:
224G what?
XSR?
VSR?
MR?
Package?
PCB length?
Connector count?
Material?
BER?
Data Rate Without Channel Definition Is Not an Engineering Specification.
236 — Better Capability Language
Use: We evaluate high-speed interconnects according to the complete channel architecture, including data rate, signaling method, reach, materials, package, PCB geometry, vias, connectors, equalization and validation requirements.
This is much stronger.
237 — World-Class SI Engineering
At the highest level:
Protocol Requirement
↓
SerDes Architecture
↓
Channel Budget
↓
Package Model
↓
Stack-Up
↓
Material
↓
Topology
↓
Trace Geometry
↓
Via Architecture
↓
Connector
↓
Return Path
↓
Crosstalk
↓
Loss
↓
Reflection
↓
Mode Conversion
↓
Pre-Layout SI
↓
PCB Layout
↓
3D EM Extraction
↓
Package + PCB + Connector Channel Model
↓
S-Parameters
↓
IBIS / IBIS-AMI
↓
TX / RX Equalization
↓
Jitter
↓
Eye / BER / COM
↓
Process Corners
↓
Manufacturing Tolerance
↓
Post-Layout Sign-Off
↓
Fabrication
↓
TDR / VNA Measurement
↓
Simulation-to-Measurement Correlation
↓
EVT
↓
DVT
↓
PVT
↓
Production Electrical Data
↓
Reliable High-Speed Communication
That is the difference between: Routing a Fast Signal and Engineering a High-Speed Channel.
Typical High-Speed PCB Design & Signal Integrity Deliverables
Depending on project scope, a 365PCB ODM high-speed program may include:
High-Speed Interface Requirements
Channel Architecture Definition
Protocol Review
SerDes Channel Budget
NRZ / PAM4 Architecture Inputs
Pre-Layout SI Analysis
Stack-Up Optimization
Controlled-Impedance Design
Differential-Pair Geometry
Single-Ended Transmission-Line Design
Propagation-Delay Analysis
Material Selection Inputs
Dk / Df Evaluation
Copper-Roughness Inputs
Loss-Budget Analysis
Insertion-Loss Analysis
Return-Loss Analysis
Crosstalk Analysis
NEXT / FEXT Analysis
Differential / Common-Mode Analysis
Mode-Conversion Analysis
Via Modeling
Antipad Optimization
Ground / Return-Via Optimization
Via-Stub Analysis
Back-Drill Requirements
Blind / Microvia Trade Study
Coaxial / Shielded Via Inputs where appropriate
BGA Escape SI Review
Connector-Launch Modeling
Cable / Backplane Inputs
Package Model Integration
Package + PCB Co-Simulation
PCIe Channel Engineering
PCIe 6.x / 7.0 Technology Inputs
CXL Channel Engineering
CXL 4.0 Technology Inputs
DDR SI Analysis
DQ / DQS Timing Inputs
Memory Topology Review
FPGA SerDes Channels
Ethernet / Networking SerDes Inputs
JESD204 Channel Inputs
224G-Class Interconnect Evaluation where applicable
IBIS Simulation
IBIS-AMI Analysis where models are available
Equalization Analysis
TX FFE Inputs
RX CTLE / DFE Inputs
Jitter Analysis
Eye-Diagram Analysis
Statistical Eye Analysis
BER Prediction Inputs
COM / Compliance-Metric Inputs where applicable
S-Parameter Analysis
Mixed-Mode S-Parameters
Touchstone Model Handling
2D Field-Solver Analysis
3D EM Analysis
Post-Layout Extraction
SI / PI Co-Design Inputs
PDN-Induced Jitter Review
EMC / Common-Mode Review
Process-Corner Analysis
Manufacturing-Tolerance Analysis
Impedance-Tolerance Analysis
Channel Sensitivity Analysis
Statistical / Monte Carlo Inputs
TDR Test Plan
VNA / S-Parameter Test Plan
Test-Coupon Architecture
De-Embedding Strategy Inputs
Simulation-to-Measurement Correlation
Good-vs-Bad Channel Comparison
High-Speed Failure-Analysis Inputs
EVT Channel Validation
DVT Channel Validation
PVT Manufacturing Inputs
Production Impedance Monitoring
Production Electrical Correlation
High-Speed Manufacturing Control Requirements
Channel Release / Sign-Off Documentation
The exact engineering depth should follow: Data Rate + Modulation + Channel Reach + Protocol + Package + PCB Length + Connector Count + Material + BER Requirement + Production Risk.
High-speed channel capability is architecture-specific. Achievable data rate and channel reach depend on the transmitter and receiver architecture, modulation, equalization, package models, PCB materials, routing length, via structures, connectors, crosstalk, power integrity, manufacturing tolerances and required system margin.
We Don't Claim a Data Rate From Trace Impedance Alone.
We Evaluate the Complete Channel From Transmitter to Receiver.
A 100-Ohm Differential Pair Is Not Automatically a Good High-Speed Channel.
Bring Us the Channel — Not Just the Trace Width
You can begin with:
Protocol
Data Rate
Processor / FPGA
SerDes
Schematic
Stack-Up
PCB Layout
Package Models
Connector Models
IBIS / IBIS-AMI Models
S-Parameters
TDR Data
VNA Data
Existing Link Failure
or simply: Tell Us What the Transmitter Must Deliver to the Receiver — and What Exists Between Them.
365PCB can help translate: Protocol → Channel Budget → PCB Geometry → Simulation → Manufacturing → Measurement → Production.
Don't Just Route the Differential Pair.
Understand the Edge Rate.
Define the Channel Budget.
Control the Impedance.
Protect the Return Path.
Engineer the Via.
Minimize the Stub.
Control the Crosstalk.
Preserve Differential Symmetry.
Understand the Material Loss.
Model the Connector.
Include the Package.
Understand the Equalization.
Analyze the Jitter.
Predict the Eye.
Evaluate BER.
Model Manufacturing Variation.
Measure the Physical Channel.
Correlate Simulation With Reality.
Make High-Speed Performance Repeatable in Production.
365PCB High-Speed PCB Design & Signal Integrity connects: SerDes + Electromagnetics + PCB + Package + Materials + SI + PI + Measurement + Manufacturing
English
