# US 2025/0155702 A1 · Method and device for optical system online designing based on intelligent light computing, 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 2025/0155702 A1
- Document ID: US20250155702A1
- Status: Published application
- Kind: A1
- Application number: 18/592,311
- Filed: 2024-02-29
- Published: 2025-05-15
- Earliest claimed: 2023-11-15
- Assignees: Tsinghua University
- Applicants: Tsinghua University
- Inventors: Lu FANG; Zhiwei XUE; Tiankuang ZHOU
- Web page: https://patentgrep.com/patent/US20250155702A1

## CPC

- G02B 27/0012: {Optical design, e.g. procedures, algorithms, optimisation routines}
- G06N 3/0499: Feedforward networks
- G02B 27/12: operating by refraction only
- G06N 3/0675: {using electro-optical, acousto-optical or opto-electronic means}
- G06N 3/09: Supervised learning
- G02B: OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method and a device for optical system online designing based on intelligent optical computing are disclosed. The method includes: constructing an optical system based on a differentiable online neural network and mapping light propagation input data to the differentiable online neural network to obtain a network mapping result; training the online neural network based on the network mapping result and a fully-forward-mode to compute first light propagation output data of the optical system; computing an error field according to the first light propagation output data and application target data, and inputting the error field to the optical system in the fully-forward-mode to output second light propagation output data; and computing gradient information of an online neural network parameter, and updating the online neural network parameter to obtain a trained optical system parameter.

## Claims

### Claim 1

A method for optical system online designing based on intelligent optical computing, comprising:
constructing an optical system based on a differentiable online neural network and mapping light propagation input data to the differentiable online neural network to obtain a network mapping result;
training the differentiable online neural network based on the network mapping result and a fully-forward-mode, to compute first light propagation output data of the optical system;
computing an error field according to the first light propagation output data and application target data, and inputting the error field to the optical system in the fully-forward-mode to output second light propagation output data; and
computing gradient information of an online neural network parameter based on the second light propagation output data, and updating the online neural network parameter according to the gradient information to obtain a trained optical system parameter, such that the trained optical system outputs data corresponding to the application target data to enable self-designing of the optical system.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the optical system is constructed based on a free-space optical system and an integrated photonic system, wherein the free-space optical system and the integrated photonic system each comprises a corresponding modulation region and propagation region.

### Claim 3 (depends on 2)

The method according to claim 2, wherein the application target data is obtained by:
obtaining first target data by focusing or imaging in an application target by constructing a wavefront by the free-space optical system;
obtaining second target data through neuromorphic computation in the application target by the integrated photonic system; and
obtaining the application target data according to the first target data and the second target data.

### Claim 4 (depends on 3)

The method according to claim 3, wherein a complex refractive index of the optical system is:
(
RI
)
⁢
n
=
n
R
+
in
I
where the RI is the complex refractive index of the optical system, np is a real part value of the complex refractive index of the optical system, nI is an imaginary part value of the complex refractive index of the optical system, wherein the RI is kept fixed in the propagation region and the RI is electro-optically or all-optically reconstructed in the modulation region.

### Claim 5 (depends on 3)

The method according to claim 3, wherein mapping the light propagation input data to the differentiable online neural network to obtain the network mapping result, comprises:
mapping a reconfigurable modulation portion of the light propagation input data to a differentiable and modulatable weight unit in the differentiable online neural network to obtain a first mapping result;
mapping a light propagation portion of the light propagation input data to neuron connections between different layers in the differentiable online neural network to obtain a second mapping result; and
obtaining the network mapping result of the differentiable online neural network based on the first mapping result and the second mapping result.

### Claim 6 (depends on 3)

The method according to claim 3, wherein mapping the light propagation input data to the differentiable online neural network to obtain the network mapping result, further comprises:
re-parameterizing an optical system controlled based on Maxwell's equation ∇×∇×−μ0∈0ω02∈r)E=−jμ0μrω0J, where E is an electric field, J is a current density, ω0 is an angular frequency, μ0 is a magnetic permeability of vacuum, μr is a magnetic permeability of a medium, ∈0 is a dielectric constant in vacuum, ∈r is a dielectric constant of the medium, ∇× is a curl operator, through a real part and an imaginary part of a complex refractive index as a differentiable embedded photonic online neural network, the expression being:
y
=
W
⁢
(
n
R
(
0
)
,
n
I
(
0
)
)
⁢
(
M
⁢
(
n
R
(
t
)
,
n
I
(
t
)
)
·
x
)
where x and y are an input electric field and an output electric field, respectively, W is a propagation characterized by a fixed refractive index nR(0)+inI(0), and M is a modulation region with a modulatable refractive index nR(t)+inI(t).

### Claim 7

An electronic device, comprising:
a processor; and
a memory storing instructions executable by the processor,
wherein the processor is configured to:
construct an optical system based on a differentiable online neural network and map light propagation input data to the differentiable online neural network to obtain a network mapping result;
train the differentiable online neural network based on the network mapping result and a fully-forward-mode, to compute first light propagation output data of the optical system;
compute an error field according to the first light propagation output data and application target data, and input the error field to the optical system in the fully-forward-mode to output second light propagation output data; and
compute gradient information of an online neural network parameter based on the second light propagation output data, and update the online neural network parameter according to the gradient information to obtain a trained optical system parameter, such that the trained optical system outputs data corresponding to the application target data to enable self-designing of the optical system.

### Claim 8 (depends on 7)

The electronic device according to claim 7, wherein the optical system is constructed based on a free-space optical system and an integrated photonic system, wherein the free-space optical system and the integrated photonic system each comprises a corresponding modulation region and propagation region.

### Claim 9 (depends on 8)

The electronic device according to claim 8, wherein the processor is further configured to:
obtain first target data by focusing or imaging in an application target by constructing a wavefront by the free-space optical system;
obtain second target data through neuromorphic computation in the application target by the integrated photonic system; and
obtain the application target data according to the first target data and the second target data.

### Claim 10 (depends on 9)

The electronic device according to claim 9, wherein a complex refractive index of the optical system is:
(
RI
)
⁢
n
=
n
R
+
in
I
where the RI is the complex refractive index of the optical system, nR is a real part value of the complex refractive index of the optical system, nI is an imaginary part value of the complex refractive index of the optical system, wherein the RI is kept fixed in the propagation region and the RI is electro-optically or all-optically reconstructed in the modulation region.

### Claim 11 (depends on 9)

The electronic device according to claim 9, wherein the processor is further configured to:
map a reconfigurable modulation portion of the light propagation input data to a differentiable and modulatable weight unit in the differentiable online neural network to obtain a first mapping result;
map a light propagation portion of the light propagation input data to neuron connections between different layers in the differentiable online neural network to obtain a second mapping result; and
obtain the network mapping result of the differentiable online neural network based on the first mapping result and the second mapping result.

### Claim 12 (depends on 9)

The electronic device according to claim 9, wherein the processor is further configured to:
re-parameterize an optical system controlled based on Maxwell's equation ∇×∇×−μ0∈0ω02∈r)E=—jμ0μrω0J, where E is an electric field, J is a current density, ω0 is an angular frequency, μ0 is a magnetic permeability of vacuum, μr is a magnetic permeability of a medium, ∈0 is a dielectric constant in vacuum, ∈r is a dielectric constant of the medium, ∇× is a curl operator, through a real part and an imaginary part of a complex refractive index as a differentiable embedded photonic online neural network, the expression being:
y
=
W
⁢
(
n
R
(
0
)
,
n
I
(
0
)
)
⁢
(
M
⁢
(
n
R
(
t
)
,
n
I
(
t
)
)
·
x
)
where x and y are an input electric field and an output electric field, respectively, W is a propagation characterized by a fixed refractive index nR(0)+inI(0), and M is a modulation region with a modulatable refractive index nR(t)+inI(t).

### Claim 13

A non-transitory computer-readable storage medium having stored thereon instructions that, when executed by a processor of an electronic device, cause the electronic device to perform a method for optical system online designing based on intelligent optical computing, the method comprising:
constructing an optical system based on a differentiable online neural network and mapping light propagation input data to the differentiable online neural network to obtain a network mapping result;
training the differentiable online neural network based on the network mapping result and a fully-forward-mode, to compute first light propagation output data of the optical system;
computing an error field according to the first light propagation output data and application target data, and inputting the error field to the optical system in the fully-forward-mode to output second light propagation output data; and
computing gradient information of an online neural network parameter based on the second light propagation output data, and updating the online neural network parameter according to the gradient information to obtain a trained optical system parameter, such that the trained optical system outputs data corresponding to the application target data to enable self-designing of the optical system.

### Claim 14 (depends on 13)

The non-transitory computer-readable storage medium according to claim 13, wherein the optical system is constructed based on a free-space optical system and an integrated photonic system, wherein the free-space optical system and the integrated photonic system each comprises a corresponding modulation region and propagation region.

### Claim 15 (depends on 14)

The non-transitory computer-readable storage medium according to claim 14, wherein the application target data is obtained by:
obtaining first target data by focusing or imaging in an application target by constructing a wavefront by the free-space optical system;
obtaining second target data through neuromorphic computation in the application target by the integrated photonic system; and
obtaining the application target data according to the first target data and the second target data.

### Claim 16 (depends on 15)

The non-transitory computer-readable storage medium according to claim 15, wherein a complex refractive index of the optical system is:
(
RI
)
⁢
n
=
n
R
+
in
I
where the RI is the complex refractive index of the optical system, nR is a real part value of the complex refractive index of the optical system, nI is an imaginary part value of the complex refractive index of the optical system, wherein the RI is kept fixed in the propagation region and the RI is electro-optically or all-optically reconstructed in the modulation region.

### Claim 17 (depends on 15)

The non-transitory computer-readable storage medium according to claim 15, wherein mapping the light propagation input data to the differentiable online neural network to obtain the network mapping result, comprises:
mapping a reconfigurable modulation portion of the light propagation input data to a differentiable and modulatable weight unit in the differentiable online neural network to obtain a first mapping result;
mapping a light propagation portion of the light propagation input data to neuron connections between different layers in the differentiable online neural network to obtain a second mapping result; and
obtaining the network mapping result of the differentiable online neural network based on the first mapping result and the second mapping result.

### Claim 18 (depends on 15)

The non-transitory computer-readable storage medium according to claim 15, wherein mapping the light propagation input data to the differentiable online neural network to obtain the network mapping result, further comprises:
re-parameterizing an optical system controlled based on Maxwell's equation ∇×∇×−μ0∈0ω02∈r)E=−jμ0μrω0J, where E is an electric field, J is a current density, ω0 is an angular frequency, μ0 is a magnetic permeability of vacuum, μr is a magnetic permeability of a medium, ∈0 is a dielectric constant in vacuum, ∈r is a dielectric constant of the medium, ∇× is a curl operator, through a real part and an imaginary part of a complex refractive index as a differentiable embedded photonic online neural network, the expression being:
y
=
W
⁡
(
n
R
(
0
)
,
n
I
(
0
)
)
⁢
(
M
⁡
(
n
R
(
t
)
,
n
I
(
t
)
)
·
x
)
where x and y are an input electric field and an output electric field, respectively, W is a propagation characterized by a fixed refractive index nR(0)+inI(0) and M is a modulation region with a modulatable refractive index nR(t)+inI(t).
