# US 12,001,937 B2 · Optical neural network, data processing method and apparatus based on same, and storage medium

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

- Number: US 12,001,937 B2
- Document ID: US12001937B2
- Status: Granted
- Kind: B2
- Application number: 18/268,665
- Filed: 2021-10-29
- Published: 2024-02-22
- Granted: 2024-06-04
- Earliest claimed: 2021-07-15
- Assignees: Inspur Suzhou Intelligent Technology Co., Ltd.
- Applicants: Inspur Suzhou Intelligent Technology Co., Ltd.
- Inventors: Ping Huang; Ruizhen Wu; Jingjing Chen; Lin Wang
- Web page: https://patentgrep.com/patent/US12001937B2

## CPC

- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/067: using optical means
- Y02D 10/00: Energy efficient computing, e.g. low power processors, power management or thermal management
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- Y02D: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE

## Abstract

Disclosed are a data processing method and apparatus based on an optical neural network, a computer-readable storage medium, and an optical neural network. The method includes: acquiring initial optical information and final output optical information as well as intermediate input optical information and intermediate output optical information at input/output ports of the phase shifter of an input optical signal in a case that beam-splitting ratios of the beam splitters of the two interference optical path structure satisfy a beam-splitting compensation condition; calculating parameters of the internal phase shifters of the two interference optical path structures in a case that the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfy a preset beam-splitting condition of the optical neural network; and performing data processing by using the optical neural network based on the parameters.

## Claims

### Claim 1

A data processing method based on an optical neural network, wherein an optical interference unit of the optical neural network comprises a first interference optical path structure, a phase shifter, and a second interference optical path structure, the first interference optical path structure and the second interference optical path structure each comprise an internal phase shifter and an optical shifter, and the method comprises:
respectively acquiring beam-splitting ratios of beam splitters of the first interference optical path structure and the second interference optical path structure;
in response to the beam-splitting ratios of the beam splitters satisfying a beam-splitting compensation condition, respectively acquiring initial optical information, intermediate input optical information at an input port of the phase shifter, intermediate output optical information at an output port of the phase shifter, and final output optical information of an input optical signal; and
in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculating parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure, so as to perform data processing by using the optical neural network based on the parameters.

### Claim 2 (depends on 1)

The data processing method based on an optical neural network according to claim 1, wherein the input optical signal comprises a first input optical signal and a second input optical signal, and the in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculating parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure comprises:
the first interference optical path structure comprising a first beam splitter, a first internal phase shifter, and a second beam splitter, inputting, by the first beam splitter, the input optical signal of the optical neural network into the first internal phase shifter, and inputting, by the second beam splitter, an optical signal outputted by the first internal phase shifter into the phase shifter; and
invoking a first interference optical path optical parameter calculation relational expression to determine a parameter of the first internal phase shifter, the first interference optical path optical parameter calculation relational expression being as follows:
{
L
1
′
=
(
r
2
⁢
e
i
⁢
θ
L
⁢
r
1
-
t
2
⁢
t
1
)
⁢
L
1
+
(
r
2
⁢
e
i
⁢
θ
L
⁢
i
⁢
t
1
+
i
⁢
t
2
⁢
r
1
)
⁢
L
2
L
2
′
=
(
i
⁢
t
2
⁢
e
i
⁢
θ
L
⁢
r
1
+
i
⁢
r
2
⁢
t
1
)
⁢
L
1
+
(
r
2
⁢
r
1
-
t
2
⁢
t
1
⁢
e
i
⁢
θ
L
)
⁢
L
2
,
wherein, L1′ is the intermediate input optical information of the first input optical signal of the optical neural network at the input port of the phase shifter, L2′ is the intermediate input optical information of the second input optical signal of the optical neural network at the input port of the phase shifter, L1 is the initial optical information of the first input optical signal, L2 is the initial optical information of the second input optical signal, r1 is a reflectance of the first beam splitter, r2 is a reflectance of the second beam splitter, t1 is an intermediate parameter and t1=√{square root over (1−r12)}, t2 is a second intermediate parameter and t2=√{square root over (1−r22)}, i represents an imaginary number, e represents an exponent, and OL is the parameter of the first internal phase shifter.

### Claim 3 (depends on 2)

The data processing method based on an optical neural network according to claim 2, wherein the in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculating parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure comprises:
the preset beam-splitting condition being 50:50, solving the first interference optical path optical parameter calculation relational expression to obtain a first parameter calculation relational expression; and
determining the parameter of the first internal phase shifter according to the first parameter calculation relational expression, the first parameter calculation relational expression being as follows:
cos
⁢
θ
L
=
(
t
2
⁢
t
1
)
2
+
(
r
2
⁢
r
1
)
2
-
1
/
2
2
⁢
r
2
⁢
t
1
⁢
t
2
⁢
r
1
.

### Claim 4 (depends on 1)

The data processing method based on an optical neural network according to claim 1, wherein the input optical signal comprises a first input optical signal and a second input optical signal, and the in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculating parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure comprises:
the second interference optical path structure comprising a third beam splitter, a second internal phase shifter, and a fourth beam splitter, inputting, by the third beam splitter, an input optical signal of the phase shifter into the second internal phase shifter, and splitting and outputting, by the fourth beam splitter, an optical signal outputted by the second internal phase shifter; and
invoking a second interference optical path optical parameter calculation relational expression to determine a parameter of the second internal phase shifter, the second interference optical path optical parameter calculation relational expression being as follows:
{
L
3
′
=
(
r
4
⁢
e
i
⁢
θ
R
⁢
r
3
-
t
4
⁢
t
3
)
⁢
L
3
+
(
r
4
⁢
e
i
⁢
θ
R
⁢
i
⁢
t
3
+
i
⁢
t
4
⁢
r
3
)
⁢
L
4
L
4
′
=
(
i
⁢
t
4
⁢
e
i
⁢
θ
R
⁢
r
3
+
i
⁢
r
4
⁢
t
3
)
⁢
L
3
+
(
r
4
⁢
r
3
-
t
4
⁢
t
3
⁢
e
i
⁢
θ
R
)
⁢
L
4
,
wherein, L3′ is the final output optical information of the first input optical signal, L4′ is the final output optical information of the second input optical signal, L3 is the intermediate output optical information of the first input optical signal at the output port of the phase shifter, L4 is the intermediate output optical information of the second input optical signal at the output port of the phase shifter, r3 is a reflectance of the third beam splitter, r4 is a reflectance of the fourth beam splitter, t3 is an intermediate parameter and t3=√{square root over (1−r32)}, t4 is a second intermediate parameter and t4=√{square root over (1−r42)}, i represents an imaginary number, e represents an exponent, and OR is the parameter of the second internal phase shifter.

### Claim 5 (depends on 4)

The data processing method based on an optical neural network according to claim 4, wherein the in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculating parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure comprises:
the preset beam-splitting condition being 50:50, solving the second interference optical path optical parameter calculation relational expression to obtain a second parameter calculation relational expression; and
determining the parameter of the second internal phase shifter according to the second parameter calculation relational expression, the second parameter calculation relational expression being as follows:
cos
⁢
θ
R
=
(
t
3
⁢
t
4
)
2
+
(
r
3
⁢
r
4
)
2
-
1
/
2
2
⁢
r
3
⁢
r
4
⁢
t
3
⁢
t
4
.

### Claim 6 (depends on 1)

The data processing method based on an optical neural network according to claim 1, wherein the preset beam-splitting condition is 50:50, and the data processing method comprises:
determining whether the beam-splitting ratios of the beam splitters satisfy the beam-splitting compensation condition, wherein the determining whether the beam-splitting ratios of the beam splitters satisfy the beam-splitting compensation condition comprises:
determining whether the beam-splitting ratios of the beam splitters are within a range of 15:85 to 85:15.

### Claim 7 (depends on 1)

The data processing method based on an optical neural network according to claim 1, wherein the initial optical information comprises an optical power.

### Claim 8 (depends on 1)

The data processing method based on an optical neural network according to claim 1, further comprising:
in response to the beam-splitting ratios of any one of the beam splitters not satisfying the beam-splitting compensation condition, not compensating the beam splitters of the optical neural network.

### Claim 9 (depends on 1)

The data processing method based on an optical neural network according to claim 1, wherein the optical interference unit of the optical neural network is integrated in a silicon photonic circuit.

### Claim 10

A data processing apparatus based on an optical neural network, comprising a processor configured to execute a computer program stored in a memory, wherein an optical interference unit of the optical neural network comprises a first interference optical path structure, a phase shifter, and a second interference optical path structure, the first interference optical path structure and the second interference optical path structure each comprise an internal phase shifter and an optical shifter, and upon execution of the computer program, the processor is configured to:
respectively acquire beam-splitting ratios of beam splitters of the first interference optical path structure and the second interference optical path structure;
in response to the beam-splitting ratios of the beam splitters satisfying a beam-splitting compensation condition, respectively acquire initial optical information, intermediate input optical information at an input port of the phase shifter, intermediate output optical information at an output port of the phase shifter, and final output optical information of an input optical signal; and
in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculate parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure, so as to perform data processing by using the optical neural network based on the parameters.

### Claim 11 (depends on 10)

The data processing apparatus based on an optical neural network according to claim 10, wherein the input optical signal comprises a first input optical signal and a second input optical signal, and the processor, upon execution of the computer program, is configured to:
the first interference optical path structure comprising a first beam splitter, a first internal phase shifter, and a second beam splitter, input, by the first beam splitter, the input optical signal of the optical neural network into the first internal phase shifter, and input, by the second beam splitter, an optical signal outputted by the first internal phase shifter into the phase shifter; and
invoke a first interference optical path optical parameter calculation relational expression to determine a parameter of the first internal phase shifter, the first interference optical path optical parameter calculation relational expression being as follows:
{
L
1
′
=
(
r
2
⁢
e
i
⁢
θ
L
⁢
r
1
-
t
2
⁢
t
1
)
⁢
L
1
+
(
r
2
⁢
e
i
⁢
θ
L
⁢
i
⁢
t
1
+
i
⁢
t
2
⁢
r
1
)
⁢
L
2
L
2
′
=
(
i
⁢
t
2
⁢
e
i
⁢
θ
L
⁢
r
1
+
i
⁢
r
2
⁢
t
1
)
⁢
L
1
+
(
r
2
⁢
r
1
-
t
2
⁢
t
1
⁢
e
i
⁢
θ
L
)
⁢
L
2
,
wherein, L1′ is the intermediate input optical information of the first input optical signal of the optical neural network at the input port of the phase shifter, L2′ is the intermediate input optical information of the second input optical signal of the optical neural network at the input port of the phase shifter, L1 is the initial optical information of the first input optical signal, L2 is the initial optical information of the second input optical signal, r1 is a reflectance of the first beam splitter, r2 is a reflectance of the second beam splitter, t1 is an intermediate parameter and t1=√{square root over (1−r12)}, t2 is a second intermediate parameter and t2=√{square root over (1−r22)}, i represents an imaginary number, e represents an exponent, and θL is the parameter of the first internal phase shifter.

### Claim 12 (depends on 11)

The data processing apparatus based on an optical neural network according to claim 11, wherein the processor, upon execution of the computer program, is configured to:
the preset beam-splitting condition being 50:50, solve the first interference optical path optical parameter calculation relational expression to obtain a first parameter calculation relational expression; and
determine the parameter of the first internal phase shifter according to the first parameter calculation relational expression, the first parameter calculation relational expression being as follows:
cos
⁢
θ
L
=
(
t
2
⁢
t
1
)
2
+
(
r
2
⁢
r
1
)
2
-
1
/
2
2
⁢
r
2
⁢
t
1
⁢
t
2
⁢
r
1
.

### Claim 13 (depends on 10)

The data processing apparatus based on an optical neural network according to claim 10, wherein the input optical signal comprises a first input optical signal and a second input optical signal, and the processor, upon execution of the computer program, is configured to:
the second interference optical path structure comprising a third beam splitter, a second internal phase shifter, and a fourth beam splitter, input, by the third beam splitter, an input optical signal of the phase shifter into the second internal phase shifter, and split and output, by the fourth beam splitter, an optical signal outputted by the second internal phase shifter; and
invoke a second interference optical path optical parameter calculation relational expression to determine a parameter of the second internal phase shifter, the second interference optical path optical parameter calculation relational expression being as follows:
{
L
3
′
=
(
r
4
⁢
e
i
⁢
θ
R
⁢
r
3
-
t
4
⁢
t
3
)
⁢
L
3
+
(
r
4
⁢
e
i
⁢
θ
R
⁢
i
⁢
t
3
+
i
⁢
t
4
⁢
r
3
)
⁢
L
4
L
4
′
=
(
i
⁢
t
4
⁢
e
i
⁢
θ
R
⁢
r
3
+
i
⁢
r
4
⁢
t
3
)
⁢
L
3
+
(
r
4
⁢
r
3
-
t
4
⁢
t
3
⁢
e
i
⁢
θ
R
)
⁢
L
4
,
wherein, L3′ is the final output optical information of the first input optical signal, L4′ is the final output optical information of the second input optical signal, L3 is the intermediate output optical information of the first input optical signal at the output port of the phase shifter, L4 is the intermediate output optical information of the second input optical signal at the output port of the phase shifter, r3 is a reflectance of the third beam splitter, r4 is a reflectance of the fourth beam splitter, t3 is an intermediate parameter and t3=√{square root over (1−r32)}, t4 is a second intermediate parameter and t4=√{square root over (1−r42)}, i represents an imaginary number, e represents an exponent, and θR is the parameter of the second internal phase shifter.

### Claim 14 (depends on 13)

The data processing apparatus based on an optical neural network according to claim 13, wherein the processor, upon execution of the computer program, is configured to:
the preset beam-splitting condition being 50:50, solve the second interference optical path optical parameter calculation relational expression to obtain a second parameter calculation relational expression; and
determine the parameter of the second internal phase shifter according to the second parameter calculation relational expression, the second parameter calculation relational expression being as follows:
cos
⁢
θ
R
=
(
t
3
⁢
t
4
)
2
+
(
r
3
⁢
r
4
)
2
-
1
/
2
2
⁢
r
3
⁢
r
4
⁢
t
3
⁢
t
4
.

### Claim 15 (depends on 10)

The data processing apparatus based on an optical neural network according to claim 10, wherein the preset beam-splitting condition is 50:50, and the processor, upon execution of the computer program, is configured to:
determine whether the beam-splitting ratios of the beam splitters are within a range of 15:85 to 85:15.

### Claim 16

A non-transitory computer-readable storage medium, storing a computer program executable by a processor, wherein upon execution by the processor, the computer program is configured to cause the processor to:
respectively acquire beam-splitting ratios of beam splitters of a first interference optical path structure and a second interference optical path structure; wherein an optical interference unit of an optical neural network comprises the first interference optical path structure, a phase shifter, and the second interference optical path structure, the first interference optical path structure and the second interference optical path structure each comprise an internal phase shifter and an optical shifter;
in response to the beam-splitting ratios of the beam splitters satisfying a beam-splitting compensation condition, respectively acquire initial optical information, intermediate input optical information at an input port of the phase shifter, intermediate output optical information at an output port of the phase shifter, and final output optical information of an input optical signal; and
in response to the initial optical information and the intermediate input optical information as well as the intermediate output optical information and the final output optical information satisfying a preset beam-splitting condition of the optical neural network, calculate parameters of internal phase shifters of the first interference optical path structure and the second interference optical path structure, so as to perform data processing by using the optical neural network based on the parameters.

### Claim 17 (depends on 16)

The non-transitory computer-readable storage medium, according to claim 16, wherein the input optical signal comprises a first input optical signal and a second input optical signal, and the processor, upon execution of the computer program, is configured to:
the first interference optical path structure comprising a first beam splitter, a first internal phase shifter, and a second beam splitter, input, by the first beam splitter, the input optical signal of the optical neural network into the first internal phase shifter, and input, by the second beam splitter, an optical signal outputted by the first internal phase shifter into the phase shifter; and
invoke a first interference optical path optical parameter calculation relational expression to determine a parameter of the first internal phase shifter, the first interference optical path optical parameter calculation relational expression being as follows:
{
L
1
′
=
(
r
2
⁢
e
i
⁢
θ
L
⁢
r
1
-
t
2
⁢
t
1
)
⁢
L
1
+
(
r
2
⁢
e
i
⁢
θ
L
⁢
i
⁢
t
1
+
i
⁢
t
2
⁢
r
1
)
⁢
L
2
L
2
′
=
(
i
⁢
t
2
⁢
e
i
⁢
θ
L
⁢
r
1
+
i
⁢
r
2
⁢
t
1
)
⁢
L
1
+
(
r
2
⁢
r
1
-
t
2
⁢
t
1
⁢
e
i
⁢
θ
L
)
⁢
L
2
,
wherein, L1′ is the intermediate input optical information of the first input optical signal of the optical neural network at the input port of the phase shifter, L2′ is the intermediate input optical information of the second input optical signal of the optical neural network at the input port of the phase shifter, L1 is the initial optical information of the first input optical signal, L2 is the initial optical information of the second input optical signal, r1 is a reflectance of the first beam splitter, r2 is a reflectance of the second beam splitter, t1 is an intermediate parameter and t1=√{square root over (1−r12)}, t2 is a second intermediate parameter and t2=√{square root over (1−r22)}, i represents an imaginary number, e represents an exponent, and θL is the parameter of the first internal phase shifter.

### Claim 18 (depends on 17)

The non-transitory computer-readable storage medium, according to claim 17, wherein the processor, upon execution of the computer program, is configured to:
the preset beam-splitting condition being 50:50, solve the first interference optical path optical parameter calculation relational expression to obtain a first parameter calculation relational expression; and
determine the parameter of the first internal phase shifter according to the first parameter calculation relational expression, the first parameter calculation relational expression being as follows:
cos
⁢
θ
L
=
(
t
2
⁢
t
1
)
2
+
(
r
2
⁢
r
1
)
2
-
1
/
2
2
⁢
r
2
⁢
t
1
⁢
t
2
⁢
r
1
.

### Claim 19 (depends on 16)

The non-transitory computer-readable storage medium, according to claim 16, wherein the input optical signal comprises a first input optical signal and a second input optical signal, and the processor, upon execution of the computer program, is configured to:
the second interference optical path structure comprising a third beam splitter, a second internal phase shifter, and a fourth beam splitter, input, by the third beam splitter, an input optical signal of the phase shifter into the second internal phase shifter, and split and output, by the fourth beam splitter, an optical signal outputted by the second internal phase shifter; and
invoke a second interference optical path optical parameter calculation relational expression to determine a parameter of the second internal phase shifter, the second interference optical path optical parameter calculation relational expression being as follows:
{
L
3
′
=
(
r
4
⁢
e
i
⁢
θ
R
⁢
r
3
-
t
4
⁢
t
3
)
⁢
L
3
+
(
r
4
⁢
e
i
⁢
θ
R
⁢
i
⁢
t
3
+
i
⁢
t
4
⁢
r
3
)
⁢
L
4
L
4
′
=
(
i
⁢
t
4
⁢
e
i
⁢
θ
R
⁢
r
3
+
i
⁢
r
4
⁢
t
3
)
⁢
L
3
+
(
r
4
⁢
r
3
-
t
4
⁢
t
3
⁢
e
i
⁢
θ
R
)
⁢
L
4
,
wherein, L3′ is the final output optical information of the first input optical signal, L4′ is the final output optical information of the second input optical signal, L3 is the intermediate output optical information of the first input optical signal at the output port of the phase shifter, L4 is the intermediate output optical information of the second input optical signal at the output port of the phase shifter, r3 is a reflectance of the third beam splitter, r4 is reflectance of the fourth beam splitter, t3 is an intermediate parameter and t3=√{square root over (1−r32)}, t4 is a second intermediate parameter and t4=√{square root over (1−r42)}, i represents an imaginary number, e represents an exponent, and θR is the parameter of the second internal phase shifter.

### Claim 20 (depends on 19)

The non-transitory computer-readable storage medium, according to claim 19, wherein the processor, upon execution of the computer program, is configured to:
the preset beam-splitting condition being 50:50, solve the second interference optical path optical parameter calculation relational expression to obtain a second parameter calculation relational expression; and
determine the parameter of the second internal phase shifter according to the second parameter calculation relational expression, the second parameter calculation relational expression being as follows:
cos
⁢
θ
R
=
(
t
3
⁢
t
4
)
2
+
(
r
3
⁢
r
4
)
2
-
1
/
2
2
⁢
r
3
⁢
r
4
⁢
t
3
⁢
t
4
.