# US 2025/0155484 A1 · Method and apparatus with frequency characteristics analysis of printed circuit board

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0155484 A1
- Document ID: US20250155484A1
- Status: Published application
- Kind: A1
- Application number: 18/945,874
- Filed: 2024-11-13
- Published: 2025-05-15
- Earliest claimed: 2023-11-13
- Assignees: Samsung Electronics Co., Ltd.
- Applicants: Samsung Electronics Co., Ltd.
- Inventors: Doyun KIM; Jaemin PARK; Youngmin OH; Bosun HWANG
- Web page: https://patentgrep.com/patent/US20250155484A1

## CPC

- G01R 27/2611: {Measuring inductance}
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)

## Abstract

An apparatus and method for analyzing a frequency characteristics of a substrate are provided, through the steps of tokenizing a trace of a substrate into a graph, generating a representation vector from the tokens of the graph, and analyzing a frequency characteristics of the substrate based on the representation vector.

## Claims

### Claim 1

An apparatus for determining frequency characteristics of a substrate, the apparatus comprising:
one or more processors and a memory, wherein the memory stores instructions causing the one or more processors to perform a process comprising:
transforming a trace of the substrate to a graph and tokenizing the graph into tokens;
generating representation vectors from the tokens; and
determining the frequency characteristics based on the representation vectors.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the transforming the trace of the substrate to the graph and tokenizing the graph into tokens comprises:
generating the graph to represent a 4-port network included in the substrate; and
tokenizing the graph into a sequence of the tokens.

### Claim 3 (depends on 2)

The apparatus of claim 2, wherein:
the 4-port network includes two components and the trace connecting the two components, and
the generating the graph representing the 4-port network included in the substrate comprises
generating edges respectively corresponding to segments included in the trace and connecting ends of the edges at corresponding nodes.

### Claim 4 (depends on 3)

The apparatus of claim 3, wherein the tokenizing the graph into the sequence of tokens comprises:
extracting features of the segments from trace data of the substrate;
generating positional information by performing positional encoding on the edges; and
generating the tokens respectively corresponding to the edges based on the features and positional information of the corresponding segments.

### Claim 5 (depends on 4)

The apparatus of claim 4, wherein the generating the tokens respectively corresponding to the edges based on the features of the segment and the positional information comprises:
performing linear projection on the features and positional information of the segments using a linear layer of a neural network.

### Claim 6 (depends on 4)

The apparatus of claim 4, wherein each segment comprises corresponding coordinate information thereof the segment and length information thereof.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein the determining the frequency characteristics based on the representation vectors comprises:
predicting scattering parameters of a 4-port network included in the substrate as the frequency characteristics of the substrate.

### Claim 8 (depends on 7)

The apparatus of claim 7, wherein the predicting the scattering parameters of the 4-port network included in the substrate as the frequency characteristics of the substrate comprises:
predicting the scattering parameters by using representation vectors respectively corresponding to segments connected to each component of the 4-port network among segments included in the trace as an input.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein the predicting the scattering parameters by using a representation vector corresponding to a segment connected to each component of the 4-port network among a plurality of segments included in the trace as an input comprises:
outputting a first graph image corresponding to a reflection parameter of the scattering parameters by using a first representation vector corresponding to a first segment connected to a first port of a first component of the 4-port network as a first input;
outputting a second graph image corresponding to a transmission parameter of the scattering parameters by using a second representation vector corresponding to a second segment connected to a second port of a second component of the 4-port network as a second input;
outputting a third graph image corresponding to a near-end crosstalk parameter of the scattering parameters by using a third representation vector corresponding to a third segment connected to a third port of the first component of the 4-port network as a third input; or
outputting a fourth graph image corresponding to a far-end crosstalk parameter of the scattering parameters by using a fourth representation vector corresponding to a fourth segment connected to a fourth port of the second component of the 4-port network as a fourth input.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the process further comprises:
transferring one or more of the graph images to a simulator for signal integrity analysis of the substrate.

### Claim 11

A method for predicting frequency characteristics of a substrate performed by one or more processors, the method comprising:
tokenizing a graph representing a trace of the substrate into a sequence of tokens;
generating a set of representation vectors by encoding the sequence of tokens; and
predicting the frequency characteristics based on the set of representation vectors.

### Claim 12 (depends on 11)

The method of claim 11, wherein the tokenizing the graph representing the trace of the substrate into the sequence of tokens comprises:
generating geometrical information of segments included in the trace from trace data of the substrate;
generating topology information of the segments by performing positional encoding on edges in the graph; and
generating tokens corresponding to the edges based on the geometrical information and the topology information.

### Claim 13 (depends on 11)

The method of claim 11, wherein the generating the set of representation vectors by encoding the sequence of tokens comprises:
encoding the sequence of tokens into the set of representation vectors through encoder layers of a neural network.

### Claim 14 (depends on 13)

The method of claim 13, wherein:
the representation vectors in the set of representation vectors are embeddings of respective segments included in the trace, and
tokens included in the sequence of tokens respectively corresponds to the segments.

### Claim 15 (depends on 11)

The method of claim 11, wherein
the predicting the frequency characteristics based on the set of representation vectors comprises
predicting scattering parameters of a 4-port network on or in the substrate as the frequency characteristics using a trained prediction network.

### Claim 16 (depends on 15)

The method of claim 15, wherein the predicting scattering parameters of the 4-port network on or in the substrate as the frequency characteristics using the trained prediction network comprises:
inputting representation vectors corresponding to segments connected to each component of the 4-port network to the trained prediction network; and
predicting graph images respectively corresponding to each of a reflection parameter, a transmission parameter, a near-end crosstalk parameter, and a far-end crosstalk parameter of the scattering parameters using the trained prediction network.

### Claim 17 (depends on 16)

The method of claim 16, further comprising:
transmitting predicted graph images respectively corresponding to the reflection parameter, the transmission parameter, the near-end crosstalk parameter; and the far-end crosstalk parameter to a simulator for signal integrity analysis of the substrate.

### Claim 18

An apparatus for determining frequency characteristics of a substrate, the apparatus comprises:
one or more processors and a memory, wherein the memory stores instructions configured to cause the one or more processors to perform a process comprising:
transforming a trace of the substrate into a graph;
tokenizing the graph into a sequence of tokens using a linear layer of neural network;
generating a set of representation vectors from the sequence of tokens using an encoder of the neural network; and
using a prediction network to determine the frequency characteristics from the set of representation vectors by using a prediction network.

### Claim 19 (depends on 18)

The apparatus of claim 18, wherein the process further comprises:
training the linear layer, the encoder, and the prediction network using trace data of a labeled substrate and a labeled graph image matched to the trace data,
wherein the labeled substrate is a substrate of which training frequency characteristics have been determined in advance as a ground truth.

### Claim 20 (depends on 19)

The apparatus of claim 19, wherein the training the linear layer, the encoder, and the prediction network using trace data of the labeled substrate and graph images matched to the trace data comprises:
tokenizing a graph representing traces of the labeled substrate using the linear layer;
generating an embedding from tokenized graphs using the encoder;
predicting a training graph image of scattering parameters of the labeled substrate from the embedding using the prediction network; and
updating the linear layer, the encoder, and the prediction network based on calculation of a loss function using the predicted training graph image and the labeled graph image.
