Generated banner for Matlab Model Serdes Systems
Generated logo for Matlab Model Serdes Systems

Matlab Model Serdes Systems

MathWorksOfficialCreated by MathWorks

  • Skill
  • Tool integration
  • Science and research
  • BSD-2-Clause
  • No issues found

From the author

Model, simulate, and optimize Serializer/Deserializer (SerDes) systems — serial and parallel links — using MATLAB SerDes Toolbox. Design NRZ and PAM-N links (PAM3 through PAM16) — explore equalization architectures (FFE, CTLE, DFE), sweep or optimize parameters with genetic algorithms, and characterize channels from loss models, S-parameter files, or crosstalk scenarios. Process captured waveforms through equalization chains, build eye diagrams, and decompose jitter. Deliver IBIS-AMI models for Tx, Rx, Redriver, or Retimer by exporting to Simulink and compiling .ami/.ibs/.dll/.so files. Covers the full arc from initial design exploration and parameter optimization to compliance testing and compiled model validation, including custom datapath blocks for nonstandard equalization.
— the description by MathWorks

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

This skill provides a comprehensive framework for designing and validating Serializer/Deserializer (SerDes) systems. It covers the full development lifecycle, from initial architecture exploration and channel characterization to parameter optimization and the generation of industry-standard IBIS-AMI models. Users can simulate NRZ and PAM-N links, evaluate equalization architectures like FFE, CTLE, and DFE, and perform compliance testing using statistical and time-domain analysis.

What it does

  • Models and simulates high-speed serial and parallel links including NRZ and PAM-N signaling.
  • Characterizes channels using loss models, S-parameter files, and crosstalk scenarios.
  • Optimizes equalization architectures such as FFE, CTLE, and DFE using genetic algorithms.
  • Performs statistical and time-domain analysis including eye diagrams and jitter decomposition.
  • Exports designs to Simulink for IBIS-AMI model generation and compilation.
  • Validates compiled models against behavioral baselines for compliance and performance.

When to use it

  • When designing SerDes links for specific data rates and channel loss profiles.
  • When evaluating and tuning equalization architectures to meet signal integrity targets.
  • When you need to generate IBIS-AMI models for Tx, Rx, Redriver, or Retimer components.
  • When performing compliance checks using eye masks and jitter budgets.
  • When analyzing captured waveforms from oscilloscopes or simulations.

Advantages

  • Full-arc workflow from design exploration to compiled model delivery.
  • Supports advanced PAM-N modulation schemes up to PAM16.
  • Enables programmatic parameter sweeps and optimization for design space exploration.
  • Provides robust validation tools to ensure compiled models match behavioral references.
  • Integrates seamlessly with Simulink for complex time-domain system simulations.

Usage examples

“Create a SerdesSystem object for a PAM4 link and run a statistical analysis.”
The agent initializes a SerdesSystem with specified Tx/Rx blocks, configures the channel, and executes the analysis() method to generate eye diagrams and performance metrics.
“Export my current SerDes design to a Simulink model for AMI generation.”
The agent uses exportToSimulink(sys) to generate the Simulink representation and prepares the IbisAmiManager settings for model compilation.

Install

  1. Download the skill as a ZIP:

    https://autogineer.com/agent-skills/matlab/matlab-model-serdes-systems/download
  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip matlab-model-serdes-systems.zip -d .agents/skills

Documentation by MathWorks

Modeling and Simulating SerDes Systems

Design, analyze, and deliver high-speed serial link models using SerDes Toolbox. From architecture exploration through IBIS-AMI model generation, covers the full workflow for NRZ and PAM-N links (PAM3 through PAM16).

When to Use

System design and architecture exploration

  • Designing SerDes links for a target data rate, signaling scheme (NRZ, PAM4, PAM-N), and channel loss

  • Evaluating equalization architectures (FFE, CTLE, DFE) and optimizing tap settings

  • Sweeping design parameters to find optimal configurations

  • Using industry reference designs (PCIe, USB4, DDR5, CEI, UCIe) as starting points

  • Characterizing Tx/Rx analog effects (parasitic capacitance, rise time, termination impedance)

  • Building custom datapath blocks for nonstandard equalization

Channel modeling and characterization

  • Loading S-parameter Touchstone files into SerdesSystem

  • Modeling channels with loss profiles, crosstalk (FEXT/NEXT), and aggressors

  • Fitting CTLE transfer functions from measured data via ctlefit

IBIS-AMI model generation

  • Building IBIS-AMI models for Tx, Rx, Redriver, or Retimer configurations

  • Exporting to Simulink, configuring AMI parameters, and compiling .ami/.ibs/.dll/.so

  • Scripting Simulink simulations and parameter sweeps with sim/parsim

Analysis and validation

  • Running statistical and time-domain simulations

  • Processing eye diagrams (eye height, eye width, COM, VEC, bathtub curves)

  • Decomposing jitter (TJ, RJ, DJ, DDJ, DCD, ISI)

  • Validating compiled AMI models against behavioral baselines

  • Running compliance checks with eye masks and jitter budgets

When NOT to Use

  • RF/microwave circuit design, antenna modeling, or baseband DSP filter design

  • General Simulink model scripting unrelated to SerDes

Must-Follow Rules

System Setup
  • SymbolTime / SampleInterval must yield an integer SamplesPerSymbol — fractional ratios cause silent errors

  • TxModel/RxModel are Transmitter/Receiver objects — not cell arrays. Construct with Transmitter('Blocks', {block1, block2}). Transmitter requires single-quoted property names — Transmitter("Blocks", ...) throws an ismember error. Receiver and all other classes accept double quotes

  • Signal conversion functions require column vectors — impulse2pulse, pulse2stateye, etc. error on row vectors

  • Include AnalogModel and JitterAndNoise for realistic results — bare Transmitter/Receiver without analog models produce optimistic COM (1-2 dB higher). See reference/equalization-tuning.md for parameter guidance (rise time, parasitic C, termination R)

Equalization
  • Set WaveType explicitly when using datapath blocks directly in MATLAB — Simulink sets this automatically, but MATLAB defaults to "Sample"

  • Adapted DFE/CTLE parameters are in results.outparams — NOT on the block object. After analysis(), sys.RxModel.Blocks{k}.TapWeights still holds initial values. In system objects chains, adapted taps are the second output: [y, taps] = dfecdr(x)

  • DFECDR Mode=0 is passthrough in Sample mode — DFE only applies with Mode≥1. Pre-load adapted taps from outparams with Mode=1 for instant convergence, or use Mode=2 with 10x EqualizationGain (9.6e-04) for self-converging chains

  • Set Modulation on DFECDR for PAM-N in system objects chains — Simulink inherits it from the model workspace, but MATLAB defaults to 2 (NRZ). Without this, PAM4 DFE adaptation fails silently

Metrics and Waveforms
  • Metrics.summary.EW is in picoseconds (already scaled) — do NOT multiply by 1e12. EH is in volts. PAM-N returns N-1 values per metric (e.g., PAM4 → 3 eyes, PAM8 → 7 eyes)

  • Channel impulse from analysis() is in V/s — when using filter() for time-domain convolution, multiply by dt: filter(impulse * dt, 1, wave). Without scaling, amplitudes blow up by ~10^11

  • pulse2wave operates on the stimulus provided — the output modulation depends on the input pattern (NRZ or PAM-N)

  • Init-Only models cannot adapt — if DFE taps or CDR converge at runtime, you need a Dual model (both Init and GetWave)

  • AMI validation requires Signal Integrity Toolbox — serdes.AMI runner and the AMI Simulink block need both SerDes Toolbox and Signal Integrity Toolbox

  • AMI GetWave: call in a chunked loop — serdes.AMI passes BlockSize (default 1024) to AMI_GetWave, so only BlockSize samples are processed per call. You must call the object in a for loop with BlockSize-length chunks. State is preserved between calls via the DLL memory handle

  • AMI GetWave: set SkipFirstBlock = false when calling from MATLAB — the default (true) is for Simulink's internal signal buffering and causes the first block to pass through unprocessed

  • AMI Init: RowSize must match impulse length — serdes.AMI crashes MATLAB (process termination, no error) if RowSize doesn't match numel(impulse)

  • AMI generation requires Simulink model — use IbisAmiManager GUI or serdes.AMIExport with export() programmatically (see Programmatic AMI Generation)

Workflow

Design Exploration

Most projects start here. The goal is to find the right equalization architecture and settings for your channel.

  1. Design — Create a SerdesSystem with Tx/Rx blocks and channel (loss model or S-parameters)

  2. Analyze — Run analysis for statistical results, plotStatEye for eye diagrams, analysisReport for metrics

  3. Sweep — Vary channel loss, FFE taps, CTLE gain, DFE taps, or jitter to map the design space

  4. Compare — Evaluate architectures (FFE-only vs FFE+CTLE vs FFE+CTLE+DFE) using COM, eye height, eye width

  5. Select — Choose the configuration that meets margin targets, then freeze equalization settings

Use SerdesSystem for programmatic exploration; serdesDesigner for interactive GUI work.

Waveform Processing

When you have a captured or imported waveform (e.g., from an oscilloscope or simulation) and want to equalize and analyze it directly:

  1. Load — Import the waveform and define timing (SampleInterval, SymbolTime)

  2. Equalize — Stream through datapath blocks (FFE, CTLE, DFECDR) with WaveType = "Sample"

  3. Analyze — Build an eye diagram with eyeDiagramSI, extract metrics (eye height, COM, VEC)

  4. Decompose jitter — Run jitter() on the equalized waveform for TJ, RJ, DJ, DDJ, ISI breakdown

DFECDR and DFE require a sample-by-sample for loop in Sample mode; FFE and CTLE accept full vectors. See reference/waveform-processing.md for the Direct Equalization pattern.

IBIS-AMI Model Delivery

When you need compiled models (.ami/.ibs/.dll/.so) for EDA tools or IP delivery:

  1. Export — Call exportToSimulink(sys) to generate a Simulink model from the frozen design

  2. Configure — Set AMI parameters, IBIS component/pin data, and model type via IbisAmiManager or serdes.AMIExport

  3. Generate — Build .ami/.ibs and compile .dll/.so via serdes.AMIExport with export()

  4. Validate — Load compiled DLL/SO with serdes.AMI, compare against behavioral reference (Init for impulse, GetWave for waveform)

  5. Cross-check — The Simulink path (sim with Rx WaveOut) is the preferred time-domain reference. Compare against: statistical analysis(), system objects direct chain, and compiled AMI DLLs. See reference/simulink-serdes-simulation.md

  6. Iterate — Fix discrepancies, re-export, re-validate until all paths agree

AMI Model Types

Choose the model type based on which equalization blocks need to adapt at runtime:

TypeInit_Returns_ImpulseGetWave_ExistsUse For
Init-OnlytruefalseLTI equalization (fixed FFE, CTLE). Supports statistical analysis
GetWave-OnlyfalsetrueTime-domain only. No statistical analysis
DualtruetrueAdaptive equalization (DFE, CDR). All analysis types

Every IBIS-AMI model implements AMI_Init (required), AMI_GetWave (optional), and AMI_Close (required) per the IBIS standard.

Key Classes

ClassPurpose
SerdesSystemTop-level system. Methods: analysis, plotStatEye, plotImpulse, plotPulse, plotAlignedPulse, plotWavePattern, analysisReport, exportToSimulink
TransmitterTx container. Construct: Transmitter('Blocks', {serdes.FFE(...)})
ReceiverRx container. Construct: Receiver('Blocks', {serdes.CTLE(...), serdes.DFECDR(...)})
ChannelDataChannel spec. Props: ChannelLossdB, ChannelLossFreq (default 5 GHz — must override to Nyquist), ChannelDifferentialImpedance, or Impulse/dt
JitterAndNoiseIBIS 7.0 jitter/noise. 4 groups: Tx jitter (Rj/Dj/DCD/Sj), Rx jitter, Rx clock recovery (5 params, active with RxClockMode='clocked', 'convolved', or 'normal'), Rx noise. Values in seconds (default) or UI. See reference/serdes-api-reference.md
serdes.AMIRun compiled AMI DLLs/SOs. Call: [waveOut, impulseOut] = ami(waveIn, impulseIn, clockIn)
serdes.AMIExportProgrammatic AMI export (R2026a+). Methods: export, getExportSettings. Props: ModelTypeTx, DLLFiles, LinuxCrossCompile
SParameterChannelS-parameter to impulse response. Handles .s4p through .s16p (multi-port returns Nx(K) matrix: col 1=thru, cols 2+=aggressors). Props: FileName, SampleInterval, StopTime, PortOrder
eyeDiagramSIWaveform eye diagram (R2024a+). Step: eyeObj(wave) — no output. Metrics: eyeHeight, eyeWidth, com, vec, margin
ctlefitCTLE pole/zero fitter. Import: ctlefit.readcsv. Output: GPZ matrix for serdes.CTLE("Specification", "GPZ Matrix")
Datapath Blocks
BlockRoleMode ValuesKey Properties
serdes.FFEFeed-forward equalizer0, 1TapWeights, TapSpacing, Normalize
serdes.CTLEContinuous-time linear EQ0, 1, 2Specification, DCGain, ACGain, PeakingGain, GPZ
serdes.DFECDRDFE + clock recovery0, 1, 2TapWeights, CDRMode, PhaseDetector, Count
serdes.DFEStandalone DFE0, 1, 2TapWeights, EqualizationGain, EqualizationStep
serdes.CDRStandalone CDR0, 1CDRMode, Count, Step, Sensitivity. Mode is deprecated
serdes.AGCAuto gain control0, 1TargetRMSVoltage, MaxGain, AveragingLength
serdes.VGAVariable gain amplifier0, 1Gain
serdes.SaturatingAmplifierLimiting amplifier0, 1Limit, LinearGain, Specification
serdes.PassThroughNo-op placeholder——

Mode values: 0 = Fixed (not exported), 1 = Fixed (exported as AMI parameter), 2 = Adaptive (GetWave). Only CTLE, DFECDR, and DFE support Mode=2.

Set CTLE Specification before setting gain properties — using ACGain with the default spec triggers a warning. GPZ Matrix requires Mode=1 for ConfigSelect to switch rows. See reference/equalization-tuning.md for the 4 specification options.

Patterns

Each reference file includes executable code patterns for its topic. Load the relevant reference before writing code — it contains correct calling conventions, constructor arguments, and property names. See the References table in Conventions below for the full routing map.

Common Mistakes

See reference/common-mistakes.md for 37 documented anti-patterns with wrong/correct code pairs covering API misuse, wrong property names, silent failures, and performance pitfalls.

Conventions

Design
  • Use SerdesSystem for programmatic design; serdesDesigner for interactive exploration. Always run analysis/plotStatEye before exporting

  • CRITICAL: ChannelLossFreq defaults to 5 GHz — always override to Nyquist (1/(2*SymbolTime)). Without this, 15 dB at 5 GHz becomes 40+ dB at Nyquist and the eye appears closed. Use SParameterChannel for real channels

  • analysis() does not auto-optimize FFE taps — use canonical de-emphasis with sum(abs(taps)) ≈ 1.0

  • DFE taps initialized to zero are optimized by the statistical solver automatically

  • Industry reference designs: serdesDesigner or openExample('serdes/PCIe5TransmitterReceiverIBISAMIModelExample'). See reference/equalization-tuning.md for starting-point configurations

Visualization and Metrics
  • All SerdesSystem plot methods (plotStatEye, plotImpulse, plotPulse, plotAlignedPulse, plotWavePattern) render into the current axes — use tiledlayout/nexttile or subplot for multi-panel layouts

  • eyeDiagramSI (R2024a+) for waveform eye diagrams; jitter (R2024b+) for jitter decomposition

  • DFECDR/DFE in "Sample" WaveType require scalar (sample-by-sample) input; FFE/CTLE accept full vectors

  • When generating PRBS waveforms via pulse2wave, ask the user for PRBS order and symbol count — default PRBS-10 (1023 symbols) if unspecified

  • Use serdes.utilities.SignalIntegrityColorMap (not hot/parula) for pulse2stateye plots. For compliance, use margin(eyeObj, eyeMask)

  • See reference/visualization-and-metrics.md for all plot methods, metrics fields/units, eyeDiagramSI, and jitter

AMI Export and Validation
  • Use serdes.AMIExport with export() for programmatic AMI generation — not slbuild (legacy, R2025b and earlier only)

  • Validate compiled DLLs/SOs with serdes.AMI before sending to EDA tools — both Init (impulse correlation) and GetWave (waveform chain)

  • Simulink is the preferred time-domain path — sim() with Rx WaveOut is the ground truth for adaptive systems. Compare against system objects chain and compiled AMI DLLs as cross-checks. Use the same PRBS stimulus across all time-domain paths

  • AMI export parameters on Simulink blocks use *AMI suffix ('ModeAMI', 'TapWeightsAMI'); model-level settings go through IbisAmiManager or serdes.AMIExport

  • No compiled DLLs/SOs ship with the toolbox. Cross-compile for Linux .so via serdes.utilities.createCrossCompiler

Script-First Workflow

For design, analysis, or sweep tasks — write code to .m files on disk, not inline MCP snippets. For quick one-off checks, inline evaluate_matlab_code is fine.

  1. Create a .m file (e.g., serdes_pam4_design.m), run via run_matlab_file, iterate by editing and re-running

  2. Deliver a polished script — %% Parameters block at top, %% Section headers, summary fprintf at end

References
Load when...Reference
Building or modifying a SerdesSystemreference/serdes-api-reference.md
Debugging wrong API names or silent failuresreference/common-mistakes.md
Working with S-parameter files or crosstalkreference/channel-modeling.md
Tuning CTLE, FFE, DFE, or fitting transfer functionsreference/equalization-tuning.md
Sweeping parameters or running GA optimizationreference/optimization.md
Plotting eyes, reading metrics, or decomposing jitterreference/visualization-and-metrics.md
Equalizing captured or imported waveforms directlyreference/waveform-processing.md
Exporting to Simulink, compiling AMI DLLs, or validating AMI modelsreference/programmatic-ami-generation.md
Scripting Simulink simulations or parsim sweepsreference/simulink-serdes-simulation.md
Building custom datapath blocksreference/custom-datapath-blocks.md
Looking up utility functions (SNR, ICN, resample)reference/serdes-utilities.md

Files

reference/programmatic-ami-generation.md

Programmatic AMI Generation

Build .ami, .ibs, and compiled .dll (Windows) or .so (Linux) files without the IbisAmiManager GUI.

New in R2026a, serdes.AMIExport provides full programmatic control over AMI export. It mirrors the IbisAmiManager GUI — changes in either are synchronized in real-time.

Construction
AMIExport = serdes.AMIExport              % from currently selected model
AMIExport = serdes.AMIExport("modelName") % from a specific model
Properties
PropertyDefaultDescription
ModelName(read-only)Simulink model name
ModelConfiguration"TxandRx""TxandRx", "IO", "Redriver", "Retimer"
ModelNameTxstringTransmitter model name
ModelNameRxstringReceiver model name
ModelTypeTx"Dual model""Dual model", "GetWave only", "Init only"
ModelTypeRx"Dual model""Dual model", "GetWave only", "Init only"
ModelsToExport"Both Tx and Rx""Both Tx and Rx", "Tx only", "Rx only"
IBISFiletrueExport .ibs file
AMIFilestrueExport .ami files
DLLFilestrueExport compiled DLL/SO
LinuxCrossCompilefalseBuild Linux .so from Windows
LinuxSourceCodefalseExport Linux C source code
ObfuscatefalseObfuscate executables (IP protection)
TargetDirstringOutput directory
IBISFileNamecharIBIS file name
CornerFactorscalarCorner factor (0-50)
BitsToIgnoreTxintegerBits to skip in Tx time-domain sim
BitsToIgnoreRxintegerBits to skip in Rx time-domain sim
CheckAMIValuetrueValidate input string against IBIS standard
QuietStatusfalseSuppress verbose output
Methods
MethodDescription
export(AMIExport)Export with current settings
getExportSettings(AMIExport)Display current settings
Full Workflow Example
% 1. Design and export to Simulink
sys = SerdesSystem(...
    'TxModel', Transmitter('Blocks', {serdes.FFE('TapWeights', [-0.1 0.8 -0.1])}), ...
    'RxModel', Receiver('Blocks', {serdes.CTLE, serdes.DFECDR}), ...
    'SymbolTime', 100e-12, ...
    'SamplesPerSymbol', 16);
exportToSimulink(sys);

% 2. Configure and export programmatically
AMIExport = serdes.AMIExport("untitled");
AMIExport.ModelsToExport = "Both Tx and Rx";
AMIExport.ModelTypeTx = "Init only";
AMIExport.ModelTypeRx = "Dual model";
AMIExport.DLLFiles = true;
AMIExport.LinuxCrossCompile = true;
AMIExport.TargetDir = fullfile(pwd, "export");
AMIExport.Obfuscate = false;

% 3. Review and export
getExportSettings(AMIExport);
export(AMIExport);
Repeater Configurations (Redriver and Retimer)

ModelConfiguration controls the IBIS model type:

ConfigUse CaseTypical BlocksCDR?
"TxAndRx"Standard endpointFFE + CTLE + DFECDRYes
"Redriver"Linear repeaterFFE + CTLE + VGA (no CDR)No
"Retimer"Regenerating repeaterFFE + CTLE + DFECDRYes
"IO"I/O buffer modelVariesVaries

Key behavior: For "Redriver" and "Retimer", the export generates separate Tx and Rx DLLs and .ami files, but AMIExport.export() builds only the Tx DLL in a single pass. To get the Rx DLL, either:

  1. Call export() twice (once with "Tx only", once re-generating Rx via slbuild), or

  2. Use slbuild(mdl + "/Rx") directly to generate Rx code, then compile.

Redriver Design Pattern

A Redriver is a linear repeater — no CDR, no data recovery. It amplifies and equalizes the signal without retiming. Use CTLE + VGA (Variable Gain Amplifier) on the Rx side, with no DFECDR block.

% Redriver: linear amplifier (CTLE + VGA, no CDR)
% Rx side: CTLE for frequency-dependent EQ, VGA for flat gain
sysRD = SerdesSystem(...
    'TxModel', Transmitter('Blocks', {serdes.FFE('TapWeights', [-0.1 0.8 -0.1])}), ...
    'RxModel', Receiver('Blocks', ...
        {serdes.CTLE('Specification', "DC Gain and AC Gain", 'DCGain', 0, 'ACGain', 10), ...
         serdes.VGA('Gain', 1.5)}), ...
    'SymbolTime', 100e-12, 'SamplesPerSymbol', 16, 'Modulation', 2);
exportToSimulink(sysRD);

ae = serdes.AMIExport("untitled");
ae.ModelConfiguration = "Redriver";
ae.TargetDir = fullfile(pwd, "export_redriver");
export(ae);  % builds AMI/IBIS/DLL for Tx; Rx code requires slbuild

Design considerations for Redrivers:

  • VGA gain compensates for insertion loss without frequency shaping

  • CTLE provides frequency-dependent equalization (peaking at Nyquist)

  • No DFE or CDR — jitter passes through (jitter amplification is a concern)

  • Typical use: PCIe/USB mid-board repeaters for long backplane traces

Retimer Design Pattern

A Retimer recovers the clock and data, then retransmits — it includes CDR and DFE for full signal regeneration. Jitter is cleaned (not passed through).

% Retimer: regenerating repeater (full CDR + DFE)
% Rx side: CTLE + DFECDR with adaptation for clock/data recovery
sysRT = SerdesSystem(...
    'TxModel', Transmitter('Blocks', ...
        {serdes.FFE('TapWeights', [-0.1 0.7 -0.2])}), ...
    'RxModel', Receiver('Blocks', ...
        {serdes.CTLE('Specification', "DC Gain and AC Gain", ...
            'DCGain', 0, 'ACGain', 12, 'PeakingFrequency', 14e9), ...
         serdes.DFECDR('Mode', 2, 'TapWeights', zeros(1, 5), ...
            'SymbolTime', 1/28e9, 'Modulation', 4)}), ...
    'SymbolTime', 1/28e9, 'SamplesPerSymbol', 16, 'Modulation', 4);
exportToSimulink(sysRT);

ae = serdes.AMIExport("untitled");
ae.ModelConfiguration = "Retimer";
ae.ModelTypeTx = "Dual model";
ae.ModelTypeRx = "Dual model";
ae.TargetDir = fullfile(pwd, "export_retimer");
export(ae);

Design considerations for Retimers:

  • Full signal regeneration — jitter is cleaned, not amplified

  • CDR loop bandwidth determines jitter transfer characteristics

  • DFE in adaptive mode (Mode=2) handles varying channel conditions

  • Typical use: PCIe 5.0/6.0 retimers, 400G Ethernet, USB4

Redriver vs Retimer Selection Guide
CriterionRedriverRetimer
LatencyLow (~1 UI)Higher (~100s of UI for CDR lock)
JitterPasses through (amplifies)Cleans (CDR filtering)
PowerLowerHigher
ComplexitySimpler (no CDR)More complex
Channel loss budgetModerate (< 20 dB)High (> 20 dB)
Standard complianceSimpler AMI modelFull AMI with Init + GetWave
Batch Export Script
models = {"pcie5_link", "cei56g_link", "usb4_link"};
configs = {"Dual model", "Dual model", "Init only"};

for i = 1:numel(models)
    load_system(models{i});
    ae = serdes.AMIExport(models{i});
    ae.ModelsToExport = "Both Tx and Rx";
    ae.ModelTypeRx = configs{i};
    ae.DLLFiles = true;
    ae.LinuxCrossCompile = true;
    ae.TargetDir = fullfile(pwd, "export", models{i});
    export(ae);
    close_system(models{i}, 0);
end

slbuild (R2025b and earlier)

For releases before R2026a, use slbuild with Simulink Coder and Embedded Coder.

Prerequisites
RequirementWhy
Simulink CoderCode generation from Simulink model
Embedded CoderStandalone executable / shared library build
C/C++ compilerMSVC (Windows), GCC (Linux), or MinGW
Configured Simulink modelMust have AMI parameter tree set up

Workflow

Step 1: Design and Export
% Design the SerDes system
sys = SerdesSystem(...
    'TxModel', Transmitter('Blocks', {serdes.FFE('TapWeights', [-0.1 0.8 -0.1])}), ...
    'RxModel', Receiver('Blocks', {serdes.CTLE, serdes.DFECDR}), ...
    'SymbolTime', 100e-12, ...
    'SamplesPerSymbol', 16, ...
    'Modulation', 2);

% Export to Simulink — creates a model with Tx, Rx, and Configuration blocks
exportToSimulink(sys);
mdl = 'untitled';  % default model name
Step 2: Configure AMI Parameters

Before building, the AMI parameter tree must be configured. This is typically done once via the IbisAmiManager GUI, then saved in the Simulink model. The configuration persists in the model file (.slx), so subsequent builds via slbuild reuse it automatically.

For first-time configuration:

  1. Open the exported Simulink model

  2. Double-click the Configuration block

  3. Click "Open SerDes IBIS-AMI Manager"

  4. Configure: AMI parameter tree, IBIS component/model/pin data, output paths

  5. Click "Apply" to save configuration to the model (do not need to "Generate" from GUI)

  6. Save the model

Step 3: Build Programmatically
% Build Tx and Rx AMI models
% Note: slbuild produces .ami/.ibs/.dll only when IbisAmiManager has been
% configured and saved in the model. Without configuration, slbuild
% generates only the .dll/.so (code generation without AMI metadata).
slbuild([mdl '/Tx']);
slbuild([mdl '/Rx']);

% Output directory structure (with IbisAmiManager configured):
%   <modelName>_build/
%   ├── Tx/
%   │   ├── <TxModelName>.ami
%   │   ├── <TxModelName>.ibs
%   │   └── <TxModelName>.dll  (or .so on Linux)
%   └── Rx/
%       ├── <RxModelName>.ami
%       ├── <RxModelName>.ibs
%       └── <RxModelName>.dll  (or .so on Linux)
%
% Output directory structure (without IbisAmiManager configured):
%   <modelName>_build/
%   ├── Tx/
%   │   └── <TxModelName>.dll  (or .so on Linux)  ← no .ami or .ibs
%   └── Rx/
%       └── <RxModelName>.dll  (or .so on Linux)
Step 4: Validate

Chain the compiled Tx and Rx DLLs and compare against the Simulink behavioral reference from analysis().

DLL naming: export() produces serdes_tx_win64.dll and serdes_rx_win64.dll. The LibraryName is the filename without extension ("serdes_tx_win64"). The InputString root name comes from the .ami file ("(serdes_tx)").

% Get reference from behavioral analysis
results = analysis(sys);
channelImpulse = results.impulse(:, 1);  % unequalized channel
simReference   = results.impulse(:, 2);  % equalized (behavioral)
N = numel(channelImpulse);
exportDir = fullfile(pwd, "export");

% Tx Init
txAmi = serdes.AMI;
txAmi.LibraryName = "serdes_tx_win64";
txAmi.LibraryPath = exportDir;
txAmi.SymbolTime = sys.SymbolTime;
txAmi.SampleInterval = sys.dt;
txAmi.InitOnly = true;
txAmi.RowSize = N;
txAmi.InputString = "(serdes_tx)";  % root name from .ami file
[~, txOut] = txAmi(zeros(N,1), channelImpulse, -1);

% Rx Init
rxAmi = serdes.AMI;
rxAmi.LibraryName = "serdes_rx_win64";
rxAmi.LibraryPath = exportDir;
rxAmi.SymbolTime = sys.SymbolTime;
rxAmi.SampleInterval = sys.dt;
rxAmi.InitOnly = true;
rxAmi.RowSize = N;
rxAmi.InputString = "(serdes_rx)";
[~, rxOut] = rxAmi(zeros(N,1), txOut, -1);

% Compare against behavioral reference — expect correlation ≈ 1.0
corrCoef = corrcoef(simReference, rxOut);
fprintf("Correlation: %.6f\n", corrCoef(1,2));

GetWave Validation (Waveform Chain)

Init-only validation (above) checks the impulse path. For full Dual-model validation, run the compiled DLLs in GetWave mode and compare against Simulink Rx WaveOut (the preferred ground truth) or the system objects direct chain.

Critical: serdes.AMI passes BlockSize (default 1024) to AMI_GetWave — only that many samples are processed per call. You must call the object in a for loop with BlockSize-length chunks. The DLL state (memory handle) is preserved between calls, just as Simulink processes the waveform block by block. Also set SkipFirstBlock = false when calling from MATLAB (the default true is for Simulink's internal signal buffering).

The strongest validation is sample-for-sample comparison against Simulink: log the channel output (chOut) from Simulink and feed the exact same waveform to the DLL. This should achieve perfect correlation (max diff at machine epsilon).

Note: The Rx output (rxOut) is logged by default after exportToSimulink. The channel output (chOut) is not — you must manually enable signal logging on the Analog Channel output port and name it chOut.

% 1. Run Simulink to get ground-truth channel output and Rx output
mdl = "mySerDesModel";
load_system(mdl);
% NOTE: 'rxOut' is logged by default. For 'chOut', manually enable signal
% logging on the Analog Channel output port in the Simulink model.
set_param(mdl, 'SaveOutput', 'on', 'SaveFormat', 'Dataset');
simOut = sim(mdl);
chSimWave = simOut.logsout.get('chOut').Values.Data(:);
rxSimWave = simOut.logsout.get('rxOut').Values.Data(:);

% 2. Set up Rx DLL
results = analysis(sys);
chImpulse = results.impulse(:, 1);
N = numel(chImpulse);
exportDir = fullfile(pwd, "export");
blockSize = 1024;

rxAmi = serdes.AMI;
rxAmi.LibraryName    = "serdes_rx_win64";
rxAmi.LibraryPath    = exportDir;
rxAmi.SymbolTime     = sys.SymbolTime;
rxAmi.SampleInterval = sys.dt;
rxAmi.InitOnly       = false;       % Init + GetWave
rxAmi.SkipFirstBlock = false;       % required when calling from MATLAB
rxAmi.RowSize        = N;           % must match numel(chImpulse)
rxAmi.BlockSize      = blockSize;
rxAmi.InputString    = "(serdes_rx)";

% 3. Run DLL in chunked loop — one call per BlockSize chunk
Nsim = numel(chSimWave);
nPad = ceil(Nsim / blockSize) * blockSize;
wavePad = [chSimWave; zeros(nPad - Nsim, 1)];
waveOut = zeros(nPad, 1);
clockIn = -ones(blockSize, 1);

for k = 1:nPad/blockSize
    idx = (k-1)*blockSize + (1:blockSize);
    [waveOut(idx), ~] = rxAmi(wavePad(idx), chImpulse, clockIn);
end
dllWave = waveOut(1:Nsim);

% 4. Verify sample-for-sample correlation
R = corrcoef(rxSimWave, dllWave);
fprintf("GetWave vs Simulink correlation: %.15f\n", R(1,2));  % expect 1.0
fprintf("Max abs diff: %.2e\n", max(abs(rxSimWave - dllWave)));

% 5. Analyze with eyeDiagramSI
eyeObj = eyeDiagramSI;
eyeObj.SampleInterval = sys.dt;
eyeObj.SymbolTime = sys.SymbolTime;
eyeObj.Modulation = sys.Modulation;
eyeObj(dllWave);
fprintf("AMI GetWave — EH: %.1f mV, EW: %.1f ps, COM: %.2f dB\n", ...
    eyeHeight(eyeObj)*1e3, eyeWidth(eyeObj)*1e12, com(eyeObj));
System Objects Direct Chain

The same validation can be done using the system objects from sys directly, without compiled DLLs. This exercises the behavioral MATLAB code path.

% Extract blocks and set WaveType = "Sample"
txBlocks = sys.TxModel.Blocks;
rxBlocks = sys.RxModel.Blocks;
for i = 1:numel(txBlocks)
    txBlocks{i}.SymbolTime = sys.SymbolTime;
    txBlocks{i}.SampleInterval = sys.dt;
    txBlocks{i}.WaveType = "Sample";
end
for i = 1:numel(rxBlocks)
    rxBlocks{i}.SymbolTime = sys.SymbolTime;
    rxBlocks{i}.SampleInterval = sys.dt;
    rxBlocks{i}.WaveType = "Sample";
end

% Stream through Tx chain (FFE accepts full vector)
w = prbsWave;
for i = 1:numel(txBlocks)
    w = txBlocks{i}(w);
end

% Stream through Rx chain — DFECDR/DFE require sample-by-sample
for i = 1:numel(rxBlocks)
    block = rxBlocks{i};
    if isa(block, 'serdes.DFECDR') || isa(block, 'serdes.DFE')
        wOut = zeros(size(w));
        for k = 1:numel(w)
            wOut(k) = block(w(k));
        end
        w = wOut;
    else
        w = block(w);
    end
end

% Analyze
eyeObj2 = eyeDiagramSI;
eyeObj2.SampleInterval = sys.dt;
eyeObj2.SymbolTime = sys.SymbolTime;
eyeObj2.Modulation = sys.Modulation;
eyeObj2(w);
fprintf("SysObj chain — EH: %.1f mV, EW: %.1f ps, COM: %.2f dB\n", ...
    eyeHeight(eyeObj2)*1e3, eyeWidth(eyeObj2)*1e12, com(eyeObj2));

Toolchain Selection

The compiler used for DLL/SO generation is controlled by MATLAB's MEX configuration. Check and change the active toolchain:

% Check current C compiler
mex -setup C

% List available compilers
compilersC = mex.getCompilerConfigurations('C');
disp({compilersC.Name}')
Supported Toolchains
PlatformToolchainOutputNotes
WindowsMicrosoft Visual C++ (MSVC).dllDefault on Windows with Visual Studio
WindowsMinGW-w64 GCC.dllFree alternative; install via Add-Ons
LinuxGCC.soDefault on Linux
Cross-compileLinux cross-compiler.soBuild Linux .so from Windows (see below)
Cross-Compilation for Linux .so

Build Linux-compatible .so files from a Windows machine using the SerDes cross-compiler utility:

% One-time setup — downloads and configures Linux GCC toolchain
% Takes ~2.5 hours, requires ~12.5 GB disk space
serdes.utilities.createCrossCompiler

% After setup, select the cross-compiler toolchain in IbisAmiManager
% or set it programmatically before slbuild:
%   Model Configuration Parameters > Code Generation > Toolchain
%   Select: "SerDes Linux Cross-Compiler"

AMI Parameter Access (Programmatic)

Each SerDes block in the Simulink model exposes *AMI suffix parameters that control which parameters appear in the exported .ami file. These are accessible via get_param/set_param.

Reserved AMI Parameter Names

The IBIS standard defines reserved parameter names that the framework manages automatically. Do not use these as custom parameter names:

ParameterDescription
Init_Returns_ImpulseTrue if Init modifies the impulse response
GetWave_ExistsTrue if the model implements GetWave
Use_Init_OutputTrue to use Init impulse for statistical analysis
Max_Init_AggressorsMax number of aggressor channels for Init
Ignore_BitsNumber of initial bits to discard in GetWave

These are set via IbisAmiManager or serdes.AMIExport model type settings — not directly in the .ami file.

% Check which FFE parameters will be exported to AMI
get_param('untitled/Tx/FFE', 'ModeAMI')         % 'on' or 'off'
get_param('untitled/Tx/FFE', 'TapWeightsAMI')   % 'on' or 'off'

% CTLE AMI parameters
get_param('untitled/Rx/CTLE', 'ModeAMI')
get_param('untitled/Rx/CTLE', 'ConfigSelectAMI')
get_param('untitled/Rx/CTLE', 'SliceSelectAMI')

% DFECDR AMI parameters
get_param('untitled/Rx/DFECDR', 'ModeAMI')
get_param('untitled/Rx/DFECDR', 'TapWeightsAMI')
get_param('untitled/Rx/DFECDR', 'PhaseOffsetAMI')
get_param('untitled/Rx/DFECDR', 'ReferenceOffsetAMI')

% Disable a parameter from AMI export
set_param('untitled/Tx/FFE', 'ModeAMI', 'off');

With R2026a, serdes.AMIExport handles the full export — no need for internal APIs. The *AMI flags and block parameters persist in the model, so AMIExport.export() picks them up automatically.

Batch Build Script Example

% Batch-build AMI models for multiple configurations
models = {"pcie5_link", "cei56g_link", "usb4_link"};
subsystems = {"Tx", "Rx"};

for i = 1:numel(models)
    mdl = models{i};
    load_system(mdl);
    for j = 1:numel(subsystems)
        fprintf("Building %s/%s...\n", mdl, subsystems{j});
        slbuild([mdl '/' subsystems{j}]);
    end
    close_system(mdl, 0);
end
fprintf("All builds complete.\n");

Troubleshooting

SymptomCauseFix
slbuild error: "No compiler found"No C compiler configuredRun mex -setup C and select a compiler
slbuild error: "License checkout failed"Missing Simulink Coder or Embedded CoderInstall required toolboxes
.so needed but .dll generatedBuilding on Windows without cross-compilerSet up cross-compiler via serdes.utilities.createCrossCompiler
AMI parameters missing in outputConfiguration block not savedOpen IbisAmiManager, click Apply, save model
AMI parameter typo uses default silentlyParameter name mismatch in .ami fileAMI parameter name typos do NOT throw errors — the DLL silently uses defaults. Always validate parameter names against the .ami file and compare Init output against known-good reference
DLL loads in MATLAB but fails in EDA tool32-bit vs 64-bit DLL mismatchEDA tools (ADS, Sigrity, HyperLynx) may require 32-bit or 64-bit DLLs. Check the EDA tool's documentation. Build both via separate MSVC targets or use serdes.AMIExport with appropriate toolchain
Build succeeds but DLL fails validationModel changed after last IbisAmiManager configRe-open IbisAmiManager, verify settings, Apply, rebuild
slbuild generates .dll but no .ami/.ibsIbisAmiManager never configured on this modelOpen IbisAmiManager, configure AMI parameter tree + IBIS data, Apply, save model, rebuild
'Tx.bat' is not recognized after slbuild or export()Windows NoDefaultCurrentDirectoryInExePath=1 prevents .bat files from being found in the current directorySee Build Fails: Tx.bat Not Recognized below
serdes.AMI error: "Unable to resolve 'lib.pointer'"MATLAB Interface to Shared C Libraries component not installedInstall via Add-On Explorer: search "MATLAB Interface to Shared C Libraries"
DLL loads but InputString root name mismatchUsing model name as root instead of default serdes_tx/serdes_rxUse "(serdes_tx)" or "(serdes_rx)" — or extract from .ami file via generateDefaultInputString
GetWave output only processes first 1024 samples, rest is passthroughPassing full waveform in a single call with default BlockSize=1024Call serdes.AMI in a for loop with BlockSize-length chunks. AMI_GetWave processes exactly BlockSize samples per call. State is preserved between calls via the DLL memory handle
GetWave first block is passthroughSkipFirstBlock defaults to true (Simulink buffering)Set SkipFirstBlock = false when calling from MATLAB
MATLAB crash (access violation) in AMI_InitRowSize does not match the impulse array lengthAlways set RowSize = numel(impulse) — mismatched sizes cause buffer overrun in the DLL
Build Fails: Tx.bat Not Recognized

export() and slbuild generate .bat files (e.g., Tx.bat, setup_msvc.bat) in the build directory and call them without path prefixes. When the Windows environment variable NoDefaultCurrentDirectoryInExePath is set to 1, cmd.exe refuses to find .bat files in the current directory — causing the build to fail:

'Tx.bat' is not recognized as an internal or external command

This is common in enterprise environments with security hardening, and also occurs when MATLAB is launched from Git Bash (which sets SHELL to bash).

Root cause: NoDefaultCurrentDirectoryInExePath=1 is a Windows security setting that prevents cmd.exe from searching . for executables. The build system calls Tx.bat which calls setup_msvc.bat — both without .\ prefixes.

Fix: Clear the env var before export() or slbuild:

% Save and clear the security setting
origVal = getenv('NoDefaultCurrentDirectoryInExePath');
setenv('NoDefaultCurrentDirectoryInExePath', '');

% Run export — both Tx and Rx build successfully
ae = serdes.AMIExport(mdl);
ae.ModelsToExport = "Both Tx and Rx";
ae.DLLFiles = true;
ae.TargetDir = fullfile(pwd, "export");
export(ae);

% Restore
setenv('NoDefaultCurrentDirectoryInExePath', origVal);

Note: setenv('SHELL', '') alone is NOT sufficient — the SHELL variable is not the root cause. The .bat resolution is blocked by the Windows security setting regardless of which shell MATLAB uses.

exportToSimulink(sys);
% Then: open model → Configuration block → "Open SerDes IBIS-AMI Manager"
% → configure AMI parameters + IBIS component/pin data → Generate

Pattern: Validate Compiled AMI DLL/SO

results = analysis(sys);
chImpulse = results.impulse(:, 1);  % unequalized channel
N = numel(chImpulse);
exportDir = fullfile(pwd, "export");

% LibraryName = DLL filename without .dll (export() produces serdes_tx_win64.dll)
% InputString root = .ami file root name (serdes_tx)
txAmi = serdes.AMI;
txAmi.LibraryName = "serdes_tx_win64";  txAmi.LibraryPath = exportDir;
txAmi.SymbolTime = sys.SymbolTime;      txAmi.SampleInterval = sys.dt;
txAmi.InitOnly = true;  txAmi.RowSize = N;  txAmi.InputString = "(serdes_tx)";
[~, txImp] = txAmi(zeros(N, 1), chImpulse, -1);

rxAmi = serdes.AMI;
rxAmi.LibraryName = "serdes_rx_win64";  rxAmi.LibraryPath = exportDir;
rxAmi.SymbolTime = sys.SymbolTime;      rxAmi.SampleInterval = sys.dt;
rxAmi.InitOnly = true;  rxAmi.RowSize = N;  rxAmi.InputString = "(serdes_rx)";
[~, rxImp] = rxAmi(zeros(N, 1), txImp, -1);

R = corrcoef(rxImp, results.impulse(:, 2));
fprintf("Init correlation: %.6f\n", R(1, 2));  % expect ≈ 1.0

% GetWave validation — chunked loop (requires manual signal logging on chOut)
% Log Analog Channel output in Simulink as 'chOut', Rx output as 'rxOut'
blockSize = 1024;
rxAmiGW = serdes.AMI;
rxAmiGW.LibraryName = "serdes_rx_win64";  rxAmiGW.LibraryPath = exportDir;
rxAmiGW.SymbolTime = sys.SymbolTime;      rxAmiGW.SampleInterval = sys.dt;
rxAmiGW.InitOnly = false;       rxAmiGW.SkipFirstBlock = false;
rxAmiGW.RowSize = N;            rxAmiGW.BlockSize = blockSize;
rxAmiGW.InputString = "(serdes_rx)";

Nsim = numel(chSimWave);  % chSimWave from Simulink logsout 'chOut'
nPad = ceil(Nsim / blockSize) * blockSize;
wavePad = [chSimWave; zeros(nPad - Nsim, 1)];
waveOut = zeros(nPad, 1);  clockIn = -ones(blockSize, 1);
for k = 1:nPad/blockSize
    idx = (k-1)*blockSize + (1:blockSize);
    [waveOut(idx), ~] = rxAmiGW(wavePad(idx), chImpulse, clockIn);
end
dllWave = waveOut(1:Nsim);
R = corrcoef(rxSimWave, dllWave);  % rxSimWave from Simulink logsout 'rxOut'
fprintf("GetWave vs Simulink: %.15f\n", R(1, 2));  % expect 1.0

Pattern: Programmatic AMI Generation

% R2026a+: Full programmatic control via serdes.AMIExport
AMIExport = serdes.AMIExport("mySerDesModel");
AMIExport.ModelsToExport = "Both Tx and Rx";
AMIExport.DLLFiles = true;
AMIExport.TargetDir = fullfile(pwd, "export");
export(AMIExport);  % generates .ami, .ibs, .dll/.so

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Security

No issues found by our static scan· No issues found

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by
MathWorks
Publisher
MathWorks (Official)
Source
matlab/matlab-agentic-toolkit · skills-catalog/rf-and-mixed-signal/matlab-model-serdes-systems · commit 2a7f215
Licence
BSD 2-Clause License (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE.md)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by MathWorks or MathWorks.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download matlab-model-serdes-systems.zip