SerDes Modeling With Neural Networks for Faster Accurate Simulation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for modeling serializer/deserializer (SerDes) models struggle to accurately match the performance of actual SerDes across various operating environments, leading to increased simulation times and difficulty in ensuring consistency.

Innovation Solution

A method utilizing machine learning, specifically neural networks, to model and manufacture SerDes models by generating data sets, training a machine learning model with noise simulation and output measurement data, and applying the trained model to the SerDes model to simplify the process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional simulation methods are used to model SerDes, then the model can be constructed, but the simulation time increases and consistency with actual SerDes becomes difficult to ensure

Engineering Contradiction:

Improvemodel consistencyVSAvoidsimulation time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent creates a virtual SerDes model that copies the behavioral characteristics of the actual SerDes device. By training a machine learning model with input-output data pairs from the actual SerDes, the virtual model replicates the device's response across various operating conditions, achieving consistency without time-consuming simulations

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary training of the machine learning model using comprehensive input-output data collected from the actual SerDes across multiple operating environments. This preliminary action creates a pre-trained virtual model that can quickly predict SerDes behavior without requiring subsequent time-consuming simulations

Inventive Principle:

Principle #10Preliminary action

2Reliability

If complex control algorithms and analog front-end models are used, then the SerDes model can be created, but the device complexity increases

Engineering Contradiction:

Improvemodel accuracyVSAvoidmodel complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces complex mechanical/control algorithms and analog front-end models with a machine learning-based virtual model. The ML model learns the input-output relationship directly from data, substituting the need for complex analytical models and control algorithms while maintaining prediction accuracy

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental approach from using complex control parameters and analog model parameters to using trained neural network weights and biases. The virtual model achieves accuracy by learning optimal parameter representations from data rather than relying on complex theoretical models

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentEP4181028B1Method for modeling serializer/deserializer model and method for manufacturing serializer/deserializer

Publication Date: 2025.09.17 SAMSUNG ELECTRONICS CO LTD

Figure 1
Figure 2
Figure 3

AI summary

A method for modeling a serializer/deserializer (SerDes) model includes generating plural data sets including noise simulation data of the SerDes model and output measurement data of an actual SerDes, training a machine learning model based on the plural data sets, and applying the trained machine learning model and an estimation model to a model included in the SerDes model. The estimation model provides the noise simulation data as an input to the trained machine learning model.