# US 2022/0277189 A1 · Method for setting of semiconductor manufacturing parameter and computing device for executing the method

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

- Number: US 2022/0277189 A1
- Document ID: US20220277189A1
- Status: Published application
- Kind: A1
- Application number: 17/551,450
- Filed: 2021-12-15
- Published: 2022-09-01
- Earliest claimed: 2020-10-15
- Assignees: Research Cooperation Foundation of Yeungnam University; Postech Research and Business Development Foundation
- Applicants: Research Cooperation Foundation of Yeungnam University; POSTECH Research and Business Development Foundation
- Inventors: Hyun Chul CHOI; Rock Hyun BAEK; Jun Sik YOON; Hyeok YUN
- Web page: https://patentgrep.com/patent/US20220277189A1

## CPC

- G03F 7/705: {Modelling or simulating from physical phenomena up to complete wafer processes or whole workflow in wafer productions}
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/09: Supervised learning
- G03F: PHOTOMECHANICAL PRODUCTION OF TEXTURED OR PATTERNED SURFACES, e.g. FOR PRINTING, FOR PROCESSING OF SEMICONDUCTOR DEVICES; MATERIALS THEREFOR; ORIGINALS THEREFOR; APPARATUS SPECIALLY ADAPTED THEREFOR; (phototypographic composing devices B41B; photosensitive materials or processes for photographic purposes G03C; electrophotography, sensitive layers or processes therefor G03G)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method for setting of a semiconductor manufacturing parameter according to an embodiment is a method performed in a computing device including one or more processors, and a memory for storing one or more programs executed by the one or more processors, the method including an operation of inputting manufacturing parameters for manufacturing a semiconductor to a neural network model and an operation of training the neural network model to predict at least one of power and delay of the semiconductor based on the input manufacturing parameters.

## Claims

### Claim 1

A method for setting of a semiconductor manufacturing parameter performed in a computing device comprising one or more processors and a memory for storing one or more programs executed by the one or more processors, the method comprising:
an operation of inputting manufacturing parameters for manufacturing a semiconductor to a neural network model; and
an operation of training the neural network model to predict at least one of power and delay of the semiconductor based on the input manufacturing parameters.

### Claim 2 (depends on 1)

The method of claim 1, wherein the operation of inputting into the neural network model inputs a value obtained by taking a log of the manufacturing parameters to the neural network model.

### Claim 3 (depends on 1)

The method of claim 1, wherein the operation of training the neural network model trains the neural network model to predict at least one of the power and delay of the semiconductor within a range of a preset minimum value to maximum value for each of the manufacturing parameters.

### Claim 4 (depends on 3)

The method of claim 3, wherein the neural network model comprises:
a first neural network model configured to receive each manufacturing parameter and is trained to predict the power of the semiconductor from each received manufacturing parameter; and
a second neural network model configured to receive each manufacturing parameter and is trained to predict the delay of the semiconductor from each received manufacturing parameter.

### Claim 5 (depends on 4)

The method of claim 4, wherein the operation of inputting into the neural network model inputs a value obtained by taking a log of the manufacturing parameters to the neural network model; and
loss functions of the first neural network model and the second neural network model is represented by Equation 1 below:
L
y
=
1
N
⁢
∑
i
N
⁢

log
⁡
(
O
y
i
)
-
log
⁡
(
D
y
i
)

2
,
y
=
power
⁢
⁢
or
⁢
⁢
delay
[
Equation
⁢
⁢
1
]
where Ly: loss function of the first neural network model when y=power, loss function of the second neural network model when y=delay;
N: the number of training data of the first neural network model and the second neural network model;
i: i-th training data;
log(Oyi) : predicted value of the first neural network model (when y=power) and the second neural network model (when y=delay) for the i-th training data; and
log(Dyi) : target value of the first neural network model (when y=power) and the second neural network model (when y=delay) for the i-th training data.

### Claim 6 (depends on 3)

The method of claim 3, wherein the neural network model is trained to predict each of the power and the delay of the semiconductor from each inputted manufacturing parameter; and
the method for setting of the semiconductor manufacturing parameter further comprises an operation of calculating power delay products (PDP) based on the power and the delay predicted by the neural network model.

### Claim 7 (depends on 6)

The method of claim 6, wherein the method for setting of the semiconductor manufacturing parameter further comprises an operation of extracting values of manufacturing parameters that minimize the PDP by using the trained neural network model.

### Claim 8 (depends on 7)

The method of claim 7, wherein
the operation of extracting the values of the manufacturing parameters that minimize the PDP extracts values of the manufacturing parameters that minimize the PDP by applying a gradient descent method to the trained neural network model.

### Claim 9 (depends on 8)

The method of claim 8, wherein the operation of extracting the values of the manufacturing parameters that minimize the PDP by applying the gradient descent method comprises:
an operation of setting an initial value of each manufacturing parameter and a learning rate of the neural network model;
an operation of calculating a slope of the PDP with respect to the initial value of the manufacturing parameter and updating the manufacturing parameter for moving the slope of the PDP in a negative direction using the learning rate; and
an operation of repeating the operation of updating until a preset interruption condition is satisfied.

### Claim 10 (depends on 9)

The method of claim 9, wherein the operation of inputting to the neural network model inputs the value obtained by taking a log of the manufacturing parameters to the neural network model; and
a slope G(x) of the PDP is calculated through Equation 2 below, and the update of the manufacturing parameter is performed through Equation 3 below:
G
⁡
(
x
)
=
d
⁡
(
log
⁡
(
P
⁢
D
⁢
P
)
)
d
⁢
x
[
Equation
⁢
⁢
2
]
x
t
+
1
=
x
t
-
γ
·
G
⁡
(
x
t
)
,
x
t
=
log
⁡
(
I
p
t
)
[
Equation
⁢
⁢
3
]
where γ: learning rate of neural network model;
Ipt: manufacturing parameter value input to neural network model at time t; and
xt: log-scaled manufacturing parameter value input to neural network model at time.

### Claim 11 (depends on 9)

The method of claim 9, wherein the preset interruption condition includes one or more of a first condition in which the calculated PDP is less than or equal to a preset target value, a second condition in which the slope of the PDP is less than or equal to a preset threshold slope, and a third condition in which an updated amount of change of the manufacturing parameter is equal to or less than a preset threshold change amount.

### Claim 12 (depends on 9)

The method of claim 9, wherein the operation of extracting the values of the manufacturing parameters that minimize the PDP by applying the gradient descent method further comprises:
an operation of limiting a range of input values of the manufacturing parameters so that the manufacturing parameters are input to the neural network model within a range of a pre-learned minimum value to maximum value.

### Claim 13 (depends on 12)

The method of claim 12, wherein the operation of limiting the range of input values of the manufacturing parameters further comprises:
an operation of receiving an input of a latent value having an unrestricted arbitrary value; and
an operation of outputting the input latent value to the neural network model by allowing the input latent value to be within the range of the pre-learned minimum value to maximum value by using a limiter function.

### Claim 14 (depends on 13)

The method of claim 13, wherein the operation of inputting into the neural network model inputs a value obtained by taking a log of the manufacturing parameters to the neural network model; and
the operation of allowing the latent value to be within the range of the pre-learned minimum value to maximum value is implemented through Equation 4 below,
g(zp)=sig(zp)·(log(Ipmax)−log(Ipmin))+log(Ipmin) [Equation 4]
where g(zp): limiter function;
zp: latent value;
Ipmin: pre-learned minimum value of manufacturing parameter;
Ipmax: pre-learned maximum value of manufacturing parameter; and
sig: sigmoid function,
sig
⁡
(
z
p
)
=
1
1
+
e
-
z
p
.

### Claim 15 (depends on 14)

The method of claim 14, wherein the slope Ĝ(x) of the PDP is calculated through Equation 5 below, and the update of the manufacturing parameter is performed through Equation 6 below:
G
^
⁡
(
x
)
=
log
⁡
(
PDP
·
g
⁡
(
x
+
τ
2
)
)
-
log
⁡
(
PDP
·
g
⁡
(
x
-
τ
2
)
)
τ
,
τ
≈
0
[
Equation
⁢
⁢
5
]
⁢
x
t
+
1
=
x
t
-
γ
·
G
^
⁡
(
x
t
)
,
x
t
=
z
p
t
[
Equation
⁢
⁢
6
]
where γ: learning rate of neural network model; and
zpt: latent value (manufacturing parameter value) input to the neural network model at time t.

### Claim 16 (depends on 7)

The method of claim 7, wherein the method for setting of the semiconductor manufacturing parameter further comprises an operation of calculating a PDP sensitivity for the PDP for each manufacturing parameter.

### Claim 17 (depends on 16)

The method of claim 16, wherein the operation of calculating the PDP sensitivity calculates the PDP sensitivity of each manufacturing parameter through a difference between a PDP value calculated when the minimum value of each manufacturing parameter is input and a PDP value calculated when the maximum value of each manufacturing parameter is input.

### Claim 18 (depends on 16)

The method of claim 16, wherein the operation of extracting the values of the manufacturing parameters that minimize the PDP includes an operation of selecting manufacturing parameters for which the PDP sensitivity is greater than or equal to a preset reference among the respective manufacturing parameters.

### Claim 19 (depends on 7)

The method of claim 7, wherein the method for setting of the semiconductor manufacturing parameter further comprises an operation of calculating each of a power sensitivity to the power and a delay sensitivity to the delay for the respective manufacturing parameters.

### Claim 20 (depends on 19)

The method of claim 19, wherein the operation of calculating the power sensitivity calculates the power sensitivity of each manufacturing parameter through a difference between a power value calculated when a minimum value of each manufacturing parameter is input and a power value calculated when a maximum value of each manufacturing parameter is input; and
the operation of calculating the delay sensitivity calculates the delay sensitivity of each manufacturing parameter through a difference between a delay value calculated when the minimum value of each manufacturing parameter is input and a delay value calculated when the maximum value of each manufacturing parameter is input.

### Claim 21 (depends on 19)

The method of claim 19, wherein the method for setting of the semiconductor manufacturing parameter further comprises an operation of classifying the respective manufacturing parameters into a plurality of preset groups based on the power sensitivity and the delay sensitivity.

### Claim 22 (depends on 21)

The method of claim 21, wherein the operation of classifying the respective manufacturing parameters into the plurality of preset groups classifies manufacturing parameters insensitive to both the power and the delay into a first group, classifies the manufacturing parameters sensitive only to the delay into a second group, classifies the manufacturing parameters sensitive only to the power into a third group, classifies manufacturing parameters sensitive to both the power and the delay into a fourth group, and classifies manufacturing parameters that are sensitive to both the power and the delay but are sensitive in opposite directions from each other into a fifth group.

### Claim 23

A computing device comprising:
one or more processors;
a memory; and
one or more programs,
wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors; and
the one or more programs comprises:
an instruction for inputting manufacturing parameters for manufacturing a semiconductor to a neural network model; and
an instruction for training the neural network model to predict at least one of power and delay of the semiconductor based on the input manufacturing parameters.

### Claim 24

A computer program stored in a non-transitory computer readable storage medium, wherein the computer program including one or more instructions, the instructions that, when executed by a computing device including one or more processors, causes the computing device to:
input manufacturing parameters for manufacturing a semiconductor to a neural network model; and
train the neural network model to predict at least one of power and delay of the semiconductor based on the input manufacturing parameters.
