Opto-electric hybrid analog iterative computing device

Abstract

The application provides an optoelectronic hybrid analog iterative computing device. The optoelectronic hybrid analog iterative computing device comprises a laser generating module, a first beam distribution module, an optoelectronic modulation module, a second beam distribution module, a photonic multiplication and addition module and an iteration module. Through making the proportional operation and the differential operation in the iteration module completed by optical signals, the electric computing processing and the optoelectronic conversion processing in the whole device can be greatly reduced, so that the energy consumption and the computing time delay can be significantly reduced, the bandwidth of the system can be effectively improved, and the precision of the iterative computation can be improved.

Description

Technical Field

[0001] This invention relates to the field of photonic simulation computing, and in particular to an optoelectronic hybrid simulation iterative computing device for solving optimization problems. Background Technology

[0002] Solving optimization problems has numerous applications in real life, such as industrial manufacturing, transportation, investment analysis, energy management, and artificial intelligence. Rapidly solving optimization problems can effectively improve production efficiency, reduce costs, and thus enhance a company's competitiveness. The wide range of applications and significant importance of optimization problems make them an indispensable tool in modern science and technology and social production and life.

[0003] However, solving optimization problems involves numerous matrix-vector multiplications and complex integration and differentiation operations, posing a challenge to existing computer systems and even future quantum computers. To solve optimization problems quickly, existing computer systems can employ methods such as linear programming, nonlinear programming, and integer programming. However, for optimization problems with a large number of iterations and significant scale, these methods still fall short of satisfactory solution time and latency.

[0004] Therefore, solutions to optimization problems using optical computing have been proposed in the past. Optical computing is a novel analog computing method that utilizes the carrier waves and various dimensions of light to perform calculations based on the electro-optic effect. Optical computing features low energy consumption, high clock speed, and low latency, making it highly suitable for large-scale, multi-iterative computations.

[0005] The Analog Iterative Machine (AIM) is a typical optoelectronic hybrid computing architecture that achieves a solution speed and capability far exceeding that of traditional computers by combining optical and electrical computing for combinatorial optimization problems. Figure 4 This diagram illustrates a traditional AIM implementation, using dashed arrows to represent electrical signals and solid arrows to represent optical signals. Figure 4 As shown, the entire system includes a laser (LD), multiple optoelectronic modulators (EOMs), photonic multiply-accumulate units (oMACs), multiple photodiodes (PDs), multiple transimpedance amplifiers (TIAs), and multiple iterative modules. The multiple optoelectronic modulators process the electrical signals x1(t) to x2(t) from the iterative modules, corresponding to the previous calculation result. N (t) These signals are modulated into the laser signals emitted by the laser and provided to the photonic matrix computing unit. The photonic matrix computing unit, in conjunction with multiple photodiodes (PDs) configured at its rear end, performs photonic multiplication and addition calculations on the optical signals from the photoelectric modulator based on the solution matrix W, followed by photoelectric conversion. After passing through a transimpedance amplifier at the rear end, the final product is obtained. electrical analog signal The obtained analog electrical signals are provided to each iterative module. Each iterative module includes a proportional unit consisting of an amplifier and a proportional term (β(t)), and a differentiating unit consisting of a differentiator. The iterative module sums the proportional term from the proportional unit, the differentiating term from the differentiating unit, and the previous calculation result from the iterative module using two summing circuits (add), and then performs nonlinearization using a limiting amplifier (Nonlinear) to obtain the final calculation result. Summary of the Invention

[0006] The technical problem to be solved by the present invention

[0007] While the iteration latency in the existing AIM implementations described above has been significantly reduced compared to traditional computers, only the matrix-vector multiplication part is implemented using optical computing. The differential, proportional, and nonlinear parts are still processed using electrical computing. Due to the limited bandwidth of electrical devices and the high power consumption of analog devices, this AIM implementation cannot effectively leverage the advantages of optical computing, such as high bandwidth, low latency, and low power consumption.

[0008] This invention was made to solve the above-mentioned technical problems. Its purpose is to provide a photoelectric hybrid analog iterative computing device that can significantly reduce the electrical computing and photoelectric conversion processing in the entire device, thereby significantly reducing energy consumption and computing latency, effectively increasing the system bandwidth, and improving the accuracy of iterative computing.

[0009] Technical solutions to solve technical problems

[0010] The first aspect of the present invention relates to a photoelectric hybrid simulation iterative computing device, which performs iterative calculations on a variable vector based on a solution matrix and according to an iterative formula, comprising:

[0011] A laser generating module that outputs a first optical signal;

[0012] A first beam splitting module divides the first optical signal into a second optical signal, a third optical signal, and a fourth optical signal with the same power.

[0013] A photoelectric modulation module modulates the power of the second optical signal according to a first modulation voltage corresponding to the original value of the variable vector, thereby generating a fifth optical signal;

[0014] The second beam splitting module divides the fifth optical signal into a sixth, seventh, and eighth optical signal with the same power.

[0015] A photonic multiply-add module, which uses a second modulation voltage corresponding to the solution matrix to perform photonic multiply-add calculations on the sixth optical signal to generate a ninth optical signal; and

[0016] An iterative module uses the seventh optical signal to generate a tenth optical signal corresponding to the proportional term in the iterative formula, uses the eighth optical signal to generate an eleventh optical signal corresponding to the differential term in the iterative formula, uses the fourth optical signal and the first modulation voltage to generate a twelfth optical signal corresponding to the original value term in the iterative formula, uses the third optical signal to perform photoelectric conversion on the optical domain accumulation value of the ninth, tenth, eleventh, and twelfth optical signals to generate an output current, and uses the output current to generate an output voltage corresponding to the new value of the variable vector.

[0017] Optionally, in the above-mentioned optoelectronic hybrid simulation iterative computing device, the iterative module includes a differential unit, which uses the eighth optical signal, the adjustment voltage, and the third modulation voltage corresponding to the momentum factor in the differential term to generate the eleventh optical signal.

[0018] Optionally, in the above-described optoelectronic hybrid simulation iterative computing device, the differentiating unit includes:

[0019] A first photoelectric modulator modulates the power of the eighth optical signal according to the third modulation voltage to generate a thirteenth optical signal;

[0020] A first beam splitter divides the thirteenth optical signal into two equally powerful optical signals: a fourteenth optical signal and a fifteenth optical signal.

[0021] A phase modulator, which adjusts the time delay and phase of the fourteenth optical signal according to the adjustment voltage to generate a sixteenth optical signal; and

[0022] A beam combiner that combines the fifteenth optical signal and the sixteenth optical signal into the eleventh optical signal.

[0023] Optionally, in the above-described optoelectronic hybrid simulation iterative computing device, the iterative module includes:

[0024] A multiplexing unit that performs optical domain summation on the ninth, tenth, eleventh, and twelfth optical signals to generate a seventeenth optical signal; and

[0025] The photoelectric conversion unit couples the third optical signal and the seventeenth optical signal through multimode interference and performs photoelectric conversion to generate the output current.

[0026] Optionally, in the above-described optoelectronic hybrid simulation iterative computing device, the iterative module includes a scaling unit with a second optoelectronic modulator, which modulates the power of the seventh optical signal according to a fourth modulation voltage corresponding to the annealing factor in the scaling term to generate the tenth optical signal.

[0027] Optionally, in the above-mentioned optoelectronic hybrid analog iterative computing device, the iterative module includes a third optoelectronic modulator, which modulates the power of the fourth optical signal according to the first modulation voltage to generate the twelfth optical signal.

[0028] Optionally, in the above-mentioned optoelectronic hybrid analog iterative computing device, the iterative module includes a transimpedance amplifier unit, which converts the output current into a voltage and amplifies it to obtain the output voltage.

[0029] Optionally, in the above-mentioned optoelectronic hybrid analog iterative calculation device, the voltage amplification factor of the transimpedance amplifier unit is adjusted so that the new value of the variable vector is linearly related to the original value of the variable vector, thereby calibrating the iterative formula.

[0030] Optionally, in the above-mentioned optoelectronic hybrid simulation iterative computing device, the electrical domain portion of the optoelectronic hybrid simulation iterative computing device is electrically connected through an electrically adjustable time delay line, which ensures that the time delay of the electrical domain portion remains consistent.

[0031] Optionally, the above-described optoelectronic hybrid simulation iterative computing device includes N iteration modules, where N is a positive integer greater than 1 corresponding to the dimension of the variable vector.

[0032] The first beam distribution module divides the first optical signal into N second optical signals, N third optical signals, and N fourth optical signals, and provides the N third optical signals and N fourth optical signals to the corresponding iteration modules.

[0033] Optionally, in the above-mentioned optoelectronic hybrid analog iterative computing device, the optoelectronic modulation module includes N fourth optoelectronic modulators, each of which modulates the power of a corresponding second optical signal among the N second optical signals according to the first modulation voltage from the corresponding iterative module among the N iterative modules, thereby generating N fifth optical signals.

[0034] Optionally, in the above-mentioned optoelectronic hybrid simulation iterative computing device, the second beam distribution module includes N second beam distributors, each of which divides each of the fifth optical signals into the sixth optical signal, the seventh optical signal, and the eighth optical signal, and provides each of the seventh optical signal and each of the eighth optical signals to the corresponding iterative module.

[0035] Optionally, in the above-described optoelectronic hybrid simulation iterative computing device, the solution matrix is ​​an N×N matrix.

[0036] The photon multiply-add module is composed of an array of N×N photoelectric conversion devices. The light transmittance of the photoelectric conversion devices corresponds to the scaling factor in the iterative formula. The photoelectric conversion devices modulate the power of the corresponding sixth optical signal according to the second modulation voltage, and accumulate the modulated optical signals in the same column in the optical domain to generate N ninth optical signals, which are then provided to the corresponding iterative modules.

[0037] Invention Effects

[0038] According to the optoelectronic hybrid analog iterative computing device of the present invention, the electrical computing and optoelectronic conversion processes in the entire device can be significantly reduced, thereby significantly reducing energy consumption and computing latency, effectively increasing the system bandwidth, and improving the accuracy of iterative computing. Attached Figure Description

[0039] Figure 1 This is a schematic diagram showing the structure of the optoelectronic hybrid simulation iterative computing device according to Embodiment 1 of the present invention.

[0040] Figure 2 It is shown Figure 1 A schematic diagram of the structure of a differential unit in the diagram.

[0041] Figure 3 This is a schematic diagram showing the structure of the optoelectronic hybrid simulation iterative computing device according to Embodiment 2 of the present invention.

[0042] Figure 4 This is a schematic diagram illustrating a traditional structure for implementing AIM. Detailed Implementation

[0043] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.

[0044] Furthermore, considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0045] First, the basic principles of simulated iterative calculation will be explained.

[0046] The method of simulating iterative calculation can be implemented based on the iterative formula of the following equation (1).

[0047] [Mathematical Expression 1]

[0048]

[0049] in, Let N be a variable vector with dimensions [N, 1], representing continuous real-valued state variables at time iteration t, where N is the variable vector. Represents the original value of the variable vector. Represents the new value of the variable vector. represents the value of the variable vector in the previous iteration cycle; W is the solution matrix with dimensions [N, N]; α, β, and γ are constants, representing the scaling factor, annealing factor, and momentum factor in the iterative formula, respectively, and are usually less than 1. Additionally, in equation (1)... For the original value term, For matrix operations, For the proportion term, This is the differential term.

[0050] The above iterative formula shows the new values ​​of the variable vector. Compared with the original value The relationship between these relationships applies to binary optimization problems, such as Quadratic Unconstrained Binary Optimization (QUBO) and the Traveling Salesman Problem (TSP). These problems can be effectively solved using the Analog Iterative Machine (AIM) scheme shown in the iterative formulas above. Furthermore, the AIM using the aforementioned iterative formulas can also solve more continuous variable optimization problems, such as financial portfolio optimization problems.

[0051] In the optoelectronic hybrid analog iterative computing device of the present invention, by enabling the proportional, differential and other operations in the iterative module to be completed through optical signals, the electrical computing processing and optoelectronic conversion processing in the entire device can be greatly reduced, thereby significantly reducing energy consumption and computing latency, effectively increasing the system bandwidth and improving the accuracy of iterative calculation.

[0052] Hereinafter, with reference to the accompanying drawings, a method for implementing iterative calculations corresponding to the above-described iterative formulas using the optoelectronic hybrid simulation iterative calculation device according to a preferred embodiment of the present invention will be described in detail.

[0053] Implementation Method 1

[0054] Figure 1 This is a schematic diagram showing the structure of the optoelectronic hybrid simulation iterative computing device according to Embodiment 1. The solid lines in the diagram represent optical signals, and the dashed lines represent electrical signals. The optoelectronic hybrid simulation iterative computing device of this embodiment, based on the solution matrix W, applies the iterative formula described later to the variable vector... Iterative calculations are performed. Furthermore, in this embodiment, a variable vector is defined. The dimension of the variable is [1, 1], which is a one-dimensional vector relative to time. Therefore, the variable vector can also be represented as x(t).

[0055] like Figure 1 As shown, the optoelectronic hybrid simulation iterative computing device includes a laser generation module 1, a first beam distribution module 2, an optoelectronic modulation module 3, a second beam distribution module 4, a photon multiply-accumulate module 5, and an iteration module 6.

[0056] The laser generating module 1 is composed of a device capable of emitting laser light, such as a laser diode (LD), and is used to generate and output a first optical signal, which is, for example, a laser signal with a certain power and phase. In this embodiment, it is assumed that the first optical signal output by the laser generating module 1 can be represented by the following formula (2).

[0057] [Mathematical Expression 2]

[0058]

[0059] Where A0 is the electric field intensity of the optical power, ω0 is the angular frequency of the light, and φ0 is the initial phase of the light.

[0060] The first beam splitting module 2, for example, is composed of a beam splitter (BS), used to split the first optical signal... The optical signals are divided into three optical signals with the same power: a second optical signal, a third optical signal, and a fourth optical signal. In this embodiment, each optical signal in the second to fourth optical signals after being divided by the first beam distribution module 2 can be represented by the following formula (3).

[0061] [Mathematical Expression 3]

[0062]

[0063] The optoelectronic modulation module 3 is, for example, composed of an electro-optic modulator (EOM). This modulator can employ electro-optic conversion devices such as a Mach-Zehnder modulator (MZM) or a voltage-controlled optical attenuator (VOA) combined with linearization modulation technology to achieve linear modulation of the amplitude of the analog electrical signal into an optical signal. The optoelectronic modulation module 3 obtains the second optical signal from the first beam distribution module 2. Based on the first modulation voltage x(t) corresponding to the original value of the variable vector, the acquired second optical signal is... The power is modulated to generate a fifth optical signal. In this embodiment, the fifth optical signal output by the photoelectric modulation module 3 at time t can be represented by the following equation (4).

[0064] [Mathematical Expression 4]

[0065]

[0066] The second beam splitting module 4, for example, is composed of a beam splitter, used to split the fifth optical signal. The light is divided into three optical signals with the same power: a sixth optical signal, a seventh optical signal, and an eighth optical signal. In this embodiment, the sixth optical signal after being divided by the second beam splitting module 4 can be represented by the following formula (5). Seventh optical signal and the eighth optical signal

[0067] [Mathematical Expression 5]

[0068]

[0069] The photonic multiply-accumulate module 5, for example, is composed of an optical multiply-accumulate unit (oMAC), which obtains the second modulation voltage corresponding to the solution matrix W from an external source and obtains the sixth optical signal from the second beam distribution module 4. Using the second modulation voltage, the sixth optical signal Perform photon multiplication and addition calculations to generate the ninth optical signal.

[0070] Specifically, in this embodiment, the solution matrix W is, for example, an N×1 matrix, where N is a positive integer. In this case, the photonic multiply-add module 5 can be composed of an array of photoelectric conversion devices such as N×1 micro-ring modulators, Mach-Zehnder modulators, and electro-absorption modulators, with the light transmittance of each photoelectric conversion device corresponding to the scaling factor α in the iterative formula. Each photoelectric conversion device modulates the power of the sixth optical signal according to the second modulation voltage of each row. The photonic multiply-add module 5 performs optical domain accumulation on the modulated optical signal to generate the ninth optical signal.

[0071] In this embodiment, the ninth optical signal after photon multiplication and addition calculation by photon multiplication and addition module 5 can be represented by the following formula (6).

[0072] [Mathematical Expression 6]

[0073]

[0074] Iteration module 6 obtains the third optical signal from the first beam allocation module 2. and the fourth optical signal The seventh optical signal is obtained from the second beam distribution module 4. and the eighth optical signal The ninth optical signal is obtained from photon multiplying module 5. Using the seventh optical signal To generate the proportional term in the iterative formula The corresponding tenth optical signal Using the eighth optical signal To generate differential terms in the iterative formula The corresponding eleventh optical signal Using the fourth optical signal The first modulation voltage x(t) is used to generate the original value term in the iterative formula. The corresponding twelfth optical signal Using third optical signals For the ninth optical signal Tenth optical signal Eleventh optical signal and the twelfth optical signal The optical domain accumulation value is used for photoelectric conversion to generate the output current I. out Using the output current I out To generate new values ​​for the variable vector The corresponding output voltage.

[0075] Specifically, such as Figure 1As shown, in this embodiment, the iteration module 6 includes a differentiating unit 11, a proportional unit 12, a photoelectric modulator 13 (corresponding to the third photoelectric modulator), a multiplexing unit 14, a photoelectric conversion unit 15, and a transimpedance amplifier unit 16.

[0076] Differential unit 11 obtains the eighth optical signal from the second beam distribution module 4. Using the eighth optical signal Adjusting voltage V diff Sum and Differential Terms The momentum factor γ in the signal corresponds to the third modulation voltage, which is used to generate the eleventh optical signal.

[0077] Below, using Figure 2 The specific structural example of differential unit 11 is explained.

[0078] Figure 2 It is shown Figure 1 A schematic diagram of the structure of the differential unit 11 in the diagram is shown. Figure 2 As shown, the differential unit 11 may include an optoelectronic modulator 101 (corresponding to the first optoelectronic modulator), a first beam splitter 102, a phase modulator 103, and a beam combiner 104.

[0079] The optoelectronic modulator 101 can be constructed using an existing optoelectronic modulator (EOM), and modulates the eighth optical signal from the second beam distribution module 4 according to the third modulation voltage corresponding to the momentum factor γ obtained from the outside. The power is modulated to generate the thirteenth optical signal.

[0080] The first beam splitter 102 can be constructed using an existing beam splitter (PS: Power Separator) to split the thirteenth optical signal from the optoelectronic modulator 101 into two optical signals with the same power, namely the fourteenth optical signal and the fifteenth optical signal. The fifteenth optical signal directly enters the beam combiner 104, which will be described later, while the fourteenth optical signal enters the phase modulator 103.

[0081] The phase modulator 103 can be constructed using an existing phase modulator PM (PM: Phase Modulator), and the variable adjustment voltage V for adjusting the time delay of the optical signal is obtained externally. diff Adjust the voltage V diff For V π (1+ω0τ / π), where V πω0 is the voltage required to rotate the phase modulator by π, ω0 is the angular frequency of the light, and τ is the time delay difference (the time it takes for the entire optical signal to travel from the photoelectric modulation module 3 through the second beam distribution module 4, the photon multiply-accumulate module 5, the multiplexing unit 14, the photoelectric conversion unit 15, the transimpedance amplifier unit 16, and finally back to the photoelectric modulation module 3).

[0082] Phase modulator 103 adjusts the voltage V according to the voltage V diff The time delay and phase of the fourteenth optical signal from the first beam splitter 102 are adjusted to generate the sixteenth optical signal.

[0083] Beam combiner 104 can be constructed using an existing beam combiner (BC) to combine the fifteenth optical signal from the first beam splitter 102 with the sixteenth optical signal from the phase modulator 103 into an eleventh optical signal. Since the phase modulator 103 uses an adjustment voltage to adjust the time delay and phase of the fourteenth optical signal, which has the same power and phase as the fifteenth optical signal, it enables the combined eleventh optical signal to achieve the desired effect. To achieve signal subtraction in the optical domain, i.e. Therefore, the eleventh optical signal It can be calculated using the following formula (7).

[0084] [Mathematical Expression 7]

[0085]

[0086] The scaling unit 12, for example, has a photoelectric modulator (corresponding to the second photoelectric modulator). This photoelectric modulator can be constructed using an existing photoelectric modulator (EOM), depending on the scaling term. The annealing factor β corresponds to the fourth modulation voltage β(t) for the seventh optical signal. The power is modulated to generate the tenth optical signal. In this embodiment, the scaling unit 12 can calculate the tenth optical signal using the following formula (8).

[0087] [Mathematical Expression 8]

[0088]

[0089] The photoelectric modulator 13 modulates the fourth optical signal according to the first modulation voltage x(t). The power is modulated to generate the twelfth optical signal. In this embodiment, the photoelectric modulator 13 can calculate the twelfth optical signal using the following formula (9).

[0090] [Mathematical Expression 9]

[0091]

[0092] The multiplexing unit 14, for example, is composed of a multiplexer (MUX) to multiplex the ninth optical signal. Tenth optical signal Eleventh optical signal and the twelfth optical signal Perform optical domain accumulation to generate the seventeenth optical signal. In this embodiment, the multiplexing unit 14 can calculate the seventeenth optical signal using the following formula (10).

[0093] [Mathematical Expression 10]

[0094]

[0095] The photoelectric conversion unit 15, for example, is composed of a balanced photodiode (BPD) and may include a multimode interferometer and two photodiodes. The photoelectric conversion unit 15 acquires the third optical signal from the first beam distribution module 2. The seventeenth optical signal is obtained from multiplexing unit 14. Using a multimode interferometer to transmit the third optical signal and the seventeenth optical signal Coupled together, and using two photodiodes for photoelectric conversion, an output current I is generated. out In this embodiment, the photoelectric conversion unit 15 can convert the third optical signal using the following formula (11). and the seventeenth optical signal Multimode interference is performed to achieve coupling.

[0096] [Mathematical Expression 11]

[0097]

[0098] Furthermore, the photoelectric conversion unit 15 can use two photodiodes to convert the two optical signals in the above formula (10) into two optical signals. and The signal is converted into an electrical signal, and the output current I is calculated using the following equation (12). out .

[0099] [Mathematical Expression 12]

[0100]

[0101] The transimpedance amplifier unit 16, for example, is composed of a transimpedance amplifier (TIA), which will output current I... out It is converted into voltage and amplified to obtain a new value relative to the variable vector. The corresponding output voltage.

[0102] In this embodiment, the transimpedance amplifier unit 16 can calculate the new value of the variable vector using the following equation (13). The corresponding output voltage.

[0103] [Mathematical Expression 13]

[0104]

[0105] Where, μ TIA The voltage amplification factor of the transimpedance amplifier unit 16 is determined by μ. TIA Adjustments can make That is, to make the new value of the variable vector Compared with the original value The relationship is linear, which allows the iteratively obtained value to be scaled back to the original range, effectively calibrating the iterative formula. α″, β″(t), and γ″ are the scaling factor, annealing factor, and momentum factor in the simplified iterative formula. Through parameter matching, adjustment, and calibration, the matching between α″, β″(t), and γ″ and α, β(t), and γ can be achieved. Let be the noise vector of the system, representing the computational error of the system. Through system optimization, it can be... Adjust to zero.

[0106] As described above, in the optoelectronic hybrid analog iterative computing device according to this embodiment, except for the matrix-vector multiplication in the photon multiply-add module 5, the scaling, differentiation, and other operations in the iterative module 6 are also performed using optical signals. Only the photoelectric conversion unit 15 (balanced detector BPD) in the iterative module 6 needs to be converted into an electrical signal. Therefore, by replacing analog electrical computation with optical computation, the system bandwidth can be effectively improved, the computational accuracy increased, and the overall system processing latency significantly lower than that of traditional AIM.

[0107] Furthermore, the optoelectronic hybrid analog iterative computing device according to this embodiment eliminates numerous electrical operational amplifiers, amplifiers, and electrical processing units, thus the energy consumption of the entire system is significantly lower than that of traditional AIM.

[0108] Furthermore, the applicant discovered that existing optoelectronic modulators (EOMs) can also function as limiting amplifiers. Therefore, in the structure of the optoelectronic hybrid analog iterative computing device involved in this invention, there is no need to set up a separate limiting amplifier to achieve nonlinearity.

[0109] Furthermore, in this invention, the electric domain portion of the optoelectronic hybrid analog iterative computing device, i.e. Figure 1 The circuit section shown by the dashed line can also be electrically connected using an adjustable delay line. By adjusting the length of this adjustable delay line, the delay of each electrical domain can be kept consistent, thereby ensuring the consistency of the timing of the calculation results for each electrical domain and further ensuring the accuracy of the iterative calculation.

[0110] Implementation Method 2

[0111] Figure 3 This is a schematic diagram showing the structure of the optoelectronic hybrid simulation iterative computing device according to Embodiment 2. The solid lines in the diagram represent optical signals, and the dashed lines represent electrical signals. The difference between the optoelectronic hybrid simulation iterative computing device according to Embodiment 2 and Embodiment 1 is that the optoelectronic hybrid simulation iterative computing device in this embodiment uses a solution matrix W of dimension [N, N] to process a variable vector of dimension [N, 1]. Iterative calculations are performed, where N is a positive integer greater than 1. Therefore, based on Embodiment 1 described above, this embodiment modifies the structure of each part of the optoelectronic hybrid simulation iterative calculation device except for the laser generation module 1. The optoelectronic hybrid simulation iterative calculation device involved in this embodiment will be described below, focusing on the differences between this embodiment and Embodiment 1.

[0112] like Figure 3 As shown, the optoelectronic hybrid simulation iterative computing device includes a laser generation module 1, a first beam distribution module 2, an optoelectronic modulation module 3, a second beam distribution module 4, a photon multiply-accumulate module 5, and N iteration modules 6 (X1(t) to X... in the figure). N The iterative module of (t), where N is the variable vector. (The dimension corresponds to a positive integer greater than 1).

[0113] The structure of the laser generating module 1 is the same as that of Embodiment 1 described above, and the description is omitted here.

[0114] The first beam distribution module 2 will distribute the first optical signal The optical signals are divided into N groups of N×3 optical signals, consisting of N second optical signals, N third optical signals, and N fourth optical signals with the same power, and each optical signal is provided to the corresponding iteration module. Specifically, the first group of three optical signals (the second, third, and fourth optical signals) is provided to the x1(t) iteration module 6, the second group of three optical signals is provided to the x2(t) iteration module 6, and so on. In this embodiment, the second to fourth optical signals in each group of optical signals after being divided by the first beam distribution module 2 can be represented by the following formula (14).

[0115] [Mathematical Expression 14]

[0116]

[0117] The photoelectric modulation module 3 includes N photoelectric modulators 18 (corresponding to the fourth photoelectric modulator), each photoelectric modulator 18 modulates according to the first modulation voltage (x1(t), x2(t), ... x) from the corresponding iteration module 6 among the N iteration modules 6. N (t)), for N second optical signals The power of the corresponding second optical signal is modulated to generate N fifth optical signals. In this embodiment, the fifth optical signal output by the i-th photoelectric modulator 18 in the photoelectric modulation module 3 at time t can be represented by the following formula (15).

[0118] [Mathematical Expression 15]

[0119]

[0120] The second beam splitting module 4 includes N second beam splitters 19, each of which splits the fifth optical signal into multiple beams. The light signal is divided into a sixth, a seventh, and an eighth optical signal, and each of the seventh and eighth optical signals is provided to the corresponding iteration module 6. In this embodiment, the sixth optical signal after being divided by the second beam splitter 19 in the second beam splitting module 4 can be represented by the following formula (16). Seventh optical signal and the eighth optical signal

[0121] [Mathematical Expression 16]

[0122]

[0123] The photon multiply-add module 5 consists of an N×N array of photoelectric conversion devices. The light transmittance of each photoelectric conversion device corresponds to the scaling factor α in the iterative formula. Each photoelectric conversion device modulates the corresponding sixth optical signal according to the second modulation voltage corresponding to each element in the solution matrix W. The power is modulated, and the modulated optical signals in the same column are accumulated in the optical domain to generate N ninth optical signals, which are then provided to the corresponding iteration module 6. In this embodiment, the ninth optical signal of the i-th path among the N ninth optical signals after photon multiplication and addition calculation by the photon multiplication and addition module 5 can be represented by the following formula (17).

[0124] [Mathematical Expression 17]

[0125]

[0126] Where W(i, n) represents the element in the i-th row and n-th column of matrix W, and W(i, ∶) represents all elements in the i-th row of matrix W.

[0127] In this embodiment, the structures of each of the N iteration modules 6 are identical. The following explanation will use the x1(t) iteration module 6 as an example.

[0128] like Figure 3 As shown, the iterative module 6 includes a differentiating unit 11, a proportional unit 12, a photoelectric modulator 13, a multiplexing unit 14, a photoelectric conversion unit 15, and a transimpedance amplifier unit 16.

[0129] The structure of the differentiating unit 11 is the same as that in Embodiment 1 described above, therefore, repeated descriptions are omitted here. The i-th differential unit 11 obtains the eighth optical signal from the second beam distribution module 4. Using the eighth optical signal Adjusting voltage V diff Sum and Differential Terms The momentum factor γ in the third modulation voltage is used to generate the eleventh optical signal expressed by the following equation (18).

[0130] [Mathematical Expression 18]

[0131]

[0132] Proportional unit 12 is based on the proportional term The annealing factor β corresponds to the third modulation voltage β(t) for the seventh optical signal. The power is modulated to generate the tenth optical signal. In this embodiment, the scaling unit 12 can calculate the tenth optical signal using the following formula (19).

[0133] [Mathematical Expression 19]

[0134]

[0135] The photoelectric modulator 13 modulates the fourth optical signal according to the first modulation voltage x1(t). The power is modulated to generate the twelfth optical signal. In this embodiment, the photoelectric modulator 13 can calculate the twelfth optical signal using the following formula (20).

[0136] [Mathematical Expression 20]

[0137]

[0138] Multiplexing unit 14 will use the ninth optical signal Tenth optical signal Eleventh optical signal and the twelfth optical signal Perform optical domain accumulation to generate the seventeenth optical signal. In this embodiment, the multiplexing unit 14 can calculate the seventeenth optical signal using the following formula (21).

[0139] [Mathematical Expression 21]

[0140]

[0141] The photoelectric conversion unit 15 obtains the third optical signal from the first beam distribution module 2. The seventeenth optical signal is obtained from multiplexing unit 14. Using a multimode interferometer to transmit the third optical signal and the seventeenth optical signal Coupled together, and using two photodiodes for photoelectric conversion, an output current I is generated. out In this embodiment, the second photoelectric conversion unit 15 can convert the third optical signal using the following formula (22). and the seventeenth optical signal Multimode interference is performed to achieve coupling.

[0142] [Mathematical Expression 22]

[0143]

[0144] Furthermore, the photoelectric conversion unit 15 can use two photodiodes to convert the two optical signals in the above equation (22) into two optical signals. and It is converted into an electrical signal, and the output current I is calculated using the following formula (23). out .

[0145] [Mathematical Expression 23]

[0146]

[0147] Transimpedance amplifier unit 16 outputs current I out The signal is converted to voltage and amplified. The N signals are then written as vectors to obtain the new values ​​of the variable vectors expressed by the following equation (24). The corresponding output voltage.

[0148] [Mathematical Expression 24]

[0149]

[0150] Where, μ TIA The voltage amplification factor of the transimpedance amplifier unit 16 is determined by μ. TIA Adjustments can make That is, to make the new value of the variable vector Compared with the original value The relationship is linear, which allows the iteratively obtained value to be scaled back to the original range, effectively calibrating the iterative formula. α′′, β′′(t), and γ″ are the scaling factor, annealing factor, and momentum factor in the simplified iterative formula. Through parameter matching, adjustment, and calibration, the matching between α″", β″(t), and γ can be achieved with α, β(t), and γ. Let be the noise vector of the system, representing the computational error of the system. Through system optimization, it can be... Adjust to zero.

[0151] As described above, the optoelectronic hybrid simulation iterative computing device according to this embodiment can perform high-precision calculations on [N,1]-dimensional variable vectors based on the [N,N]-dimensional solution matrix W with high bandwidth, low power consumption, and low latency. Perform iterative calculations.

[0152] The optoelectronic hybrid simulation iterative computing device of the present invention has been described above. It should be considered that all aspects of the embodiments disclosed herein are merely illustrative and not restrictive. The scope of this disclosure is defined by the claims, not by the above embodiments, and also includes all modifications and variations within the meaning and scope equivalent to the claims.

[0153] Industrial practicality

[0154] As described above, the optoelectronic hybrid analog iterative computing device according to the present invention is useful for solving combinatorial optimization problems and any other optimization problems in scenarios such as logistics and supply chain management, production and manufacturing, financial portfolio optimization, network and communication, and artificial intelligence learning.

[0155] Label Explanation

[0156] 1. Laser Generating Module

[0157] 2 First Beam Assignment Module

[0158] 3. Optoelectronic modulation module

[0159] 4 Second Beam Allocation Module

[0160] 5 Photon Multiply-Accumulate Module

[0161] 6. Iteration Module

[0162] 11 Differential Units

[0163] 12 Scale Units

[0164] 13. Optoelectronic modulator

[0165] 14 Multiplexing Units

[0166] 15 Photoelectric Conversion Units

[0167] 16 Transimpedance Amplifier Units

[0168] 18. Optoelectronic modulator

[0169] 19 Second Beam Splitter

[0170] 101 Optoelectronic Modulator

[0171] 102 First Beam Splitter

[0172] 103 Phase Modulator

[0173] 104 Beam synthesizer.

Claims

1. A photoelectric hybrid simulation iterative computing device, which performs iterative calculations on a variable vector based on a solution matrix and according to an iterative formula, characterized in that... include: A laser generating module that outputs a first optical signal; A first beam splitting module divides the first optical signal into a second optical signal, a third optical signal, and a fourth optical signal with the same power. A photoelectric modulation module modulates the power of the second optical signal according to a first modulation voltage corresponding to the original value of the variable vector, thereby generating a fifth optical signal; The second beam splitting module divides the fifth optical signal into a sixth, seventh, and eighth optical signal with the same power. A photonic multiply-add module, which uses a second modulation voltage corresponding to the solution matrix to perform photonic multiply-add calculations on the sixth optical signal to generate a ninth optical signal; as well as An iterative module uses the seventh optical signal to generate a tenth optical signal corresponding to the proportional term in the iterative formula, uses the eighth optical signal to generate an eleventh optical signal corresponding to the differential term in the iterative formula, uses the fourth optical signal and the first modulation voltage to generate a twelfth optical signal corresponding to the original value term in the iterative formula, uses the third optical signal to perform photoelectric conversion on the optical domain accumulation value of the ninth, tenth, eleventh, and twelfth optical signals to generate an output current, and uses the output current to generate an output voltage corresponding to the new value of the variable vector.

2. The optoelectronic hybrid analog iterative computing device as described in claim 1, characterized in that, The iterative module includes a differentiating unit that uses the eighth optical signal, the adjustment voltage, and the third modulation voltage corresponding to the momentum factor in the differential term to generate the eleventh optical signal.

3. The optoelectronic hybrid simulation iterative computing device as described in claim 2, characterized in that, The differential unit includes: A first photoelectric modulator modulates the power of the eighth optical signal according to the third modulation voltage to generate a thirteenth optical signal; A first beam splitter divides the thirteenth optical signal into two equally powerful optical signals: a fourteenth optical signal and a fifteenth optical signal. A phase modulator, which adjusts the time delay and phase of the fourteenth optical signal according to the adjustment voltage to generate a sixteenth optical signal; and A beam combiner that combines the fifteenth optical signal and the sixteenth optical signal into the eleventh optical signal.

4. The optoelectronic hybrid simulation iterative computing device according to any one of claims 1 to 3, characterized in that, The iteration module includes: A multiplexing unit that performs optical domain summation on the ninth, tenth, eleventh, and twelfth optical signals to generate a seventeenth optical signal; and The photoelectric conversion unit couples the third optical signal and the seventeenth optical signal through multimode interference and performs photoelectric conversion to generate the output current.

5. The optoelectronic hybrid simulation iterative computing device according to any one of claims 1 to 3, characterized in that, The iterative module includes a scaling unit with a second photoelectric modulator, which modulates the power of the seventh optical signal according to a fourth modulation voltage corresponding to the annealing factor in the scaling term to generate the tenth optical signal.

6. The optoelectronic hybrid analog iterative computing device according to any one of claims 1 to 3, characterized in that, The iterative module includes a third photoelectric modulator, which modulates the power of the fourth optical signal according to the first modulation voltage to generate the twelfth optical signal.

7. The optoelectronic hybrid simulation iterative computing device according to any one of claims 1 to 3, characterized in that, The iterative module includes a transimpedance amplifier unit that converts the output current into a voltage and amplifies it to obtain the output voltage.

8. The optoelectronic hybrid analog iterative computing device as described in claim 7, characterized in that, The voltage amplification factor of the transimpedance amplifier unit is adjusted so that the new value of the variable vector is linearly related to the original value of the variable vector, thereby calibrating the iterative formula.

9. The optoelectronic hybrid analog iterative computing device according to any one of claims 1 to 3, characterized in that, The electrical domain portion of the optoelectronic hybrid simulation iterative computing device is electrically connected via an electrically adjustable time delay line, which ensures that the time delay of the electrical domain portion remains consistent.

10. The optoelectronic hybrid analog iterative computing device according to any one of claims 1 to 3, characterized in that, It includes N iteration modules, where N is a positive integer greater than 1 corresponding to the dimension of the variable vector. The first beam distribution module divides the first optical signal into N second optical signals, N third optical signals, and N fourth optical signals, and provides the N third optical signals and N fourth optical signals to the corresponding iteration modules.

11. The optoelectronic hybrid analog iterative computing device as described in claim 10, characterized in that, The photoelectric modulation module includes N fourth photoelectric modulators. Each of the fourth photoelectric modulators modulates the power of the corresponding second optical signal among the N second optical signals according to the first modulation voltage from the corresponding iteration module among the N iteration modules, thereby generating N fifth optical signals.

12. The optoelectronic hybrid analog iterative computing device as described in claim 11, characterized in that, The second beam splitting module includes N second beam splitters. Each second beam splitter divides each of the fifth optical signals into the sixth optical signal, the seventh optical signal, and the eighth optical signal, and provides each of the seventh optical signal and each of the eighth optical signals to the corresponding iteration module.

13. The optoelectronic hybrid analog iterative computing device as described in claim 12, characterized in that, The solution matrix is ​​an N×N matrix. The photon multiply-add module is composed of an array of N×N photoelectric conversion devices. The light transmittance of the photoelectric conversion devices corresponds to the scaling factor in the iterative formula. The photoelectric conversion devices modulate the power of the corresponding sixth optical signal according to the second modulation voltage, and accumulate the modulated optical signals in the same column in the optical domain to generate N ninth optical signals, which are then provided to the corresponding iterative modules.

Images