A method for solving a quadratic unconstrained binary optimization problem based on an analog computing circuit

By using a programmable resistor memory array and current source module in the analog computing circuit, combined with the feedback loop of the voltage comparator, a single-step exact solution to the QUBO problem is achieved, solving the problem of excessively long solution time in traditional digital computing, and possessing the advantages of speed and low power consumption.

CN122263776APending Publication Date: 2026-06-23PEKING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610362414.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-24
Publication Date
2026-06-23

Smart Images

  • Figure CN122263776A_ABST
    Figure CN122263776A_ABST
Patent Text Reader

Abstract

The application is a QUBO problem solving method based on analog computing circuit, belonging to the field of analog computing and integrated circuit.The method combines the mathematical mapping of QUBO problem with the physical evolution process of analog computing circuit, maps the quadratic term matrix in QUBO problem into the product of resistance value in programmable resistance memory array and output current of current source module, maps the linear coefficient vector into threshold voltage of each row comparator, and takes the output voltage of each row comparator on feedback loop as the output of corresponding logic of state variable to be solved in QUBO problem, and relies on the natural evolution process of voltage in the circuit to realize the solution of QUBO problem in one step.The application does not depend on external digital circuit to provide clock, but updates in continuous time, has the advantages of fast speed, less peripheral circuit and low power consumption, and is suitable for the application scene of QUBO problem which is sensitive to power consumption and requires high solving speed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the fields of semiconductors, analog computing, and integrated circuits, and relates to a simulation computing method for solving the quadratic unconstrained binary optimization (QUBO) problem in one step. Background Technology

[0002] Combinatorial optimization is a core topic in operational science, focusing on finding the optimal solution from a limited set of candidate solutions. This research generally falls into two main directions: one deals with continuous parameters, aiming to obtain a set of real-valued or functional results; the other targets discrete parameters, selecting the optimal term from a finite or countably infinite set of discrete solutions. Discrete combinatorial optimization, due to the separation of variable values, differs fundamentally from continuous optimization in the composition structure of feasible solutions and exploration strategies, thus leading to the development of corresponding solutions. The QUBO problem, as a representative form of discrete combinatorial optimization, has been classified as an NP-hard problem. This means that under traditional computational architectures based on numerical operations, the time required to solve it increases exponentially with the problem size, making it often impossible to obtain an accurate answer within a reasonable timeframe for large-scale real-world examples.

[0003] In recent years, analog computing circuits based on programmable resistive memory cells have been widely used in various matrix operation acceleration scenarios, significantly reducing the execution cost of matrix operations. For example... Figure 1 As shown, analog computing circuits based on programmable resistor storage matrices can handle QUBO problems under specific conditions. However, they suffer from computational approximations. If an analog computing circuit and method can be designed to solve the general QUBO problem accurately in a single step, it is expected to break through the existing constraints of traditional digital computing architectures in terms of computational efficiency and energy consumption. This would accelerate computation in combinatorial optimization problems such as Hopfield networks and Ising models, which focus on energy minimization, and would have significant value in both theoretical development and engineering implementation. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes a QUBO problem-solving method based on analog computing circuits. This method is based on an analog computing circuit consisting of a programmable resistor memory array, a current source module, and voltage comparators connected to form a feedback loop. By mapping the quadratic term matrix in the QUBO problem to the product of the resistance values ​​in the programmable resistor memory array and the output current of the current source module, and mapping the linear coefficient vector to the threshold voltage of each row comparator, the output voltage of each row comparator in the feedback loop is used as the output of the state variable to be solved in the QUBO problem. The QUBO problem is solved in one step by relying on the natural evolution process of the voltage in the circuit.

[0005] The principle underlying this invention is as follows:

[0006] For a given n-dimensional quadratic term matrix and the linear coefficient vector λ=(λ1,λ2,…,λ n ) T The goal is to achieve the following for all state variables X=(x1,x2,…,x…) n ) T ∈{0,1} n In a binary vector, find one or more states that minimize the following energy function:

[0007] ,

[0008] for In updating the rules

[0009]

[0010] Under these conditions, each state update ensures that the energy function decreases or remains unchanged, eventually tending towards stability. Specifically, let the current state be X, and if the state is changed... Let the new state be X', then we have:

[0011] ,

[0012] Since only one component x is changed i Therefore, the difference is entirely determined by the quadratic term in the i-th row and this component. It can be derived that:

[0013] ,

[0014] By enumeration and The state, note the circuit update rule guarantees Therefore, under this update rule, it is guaranteed that the final state converges to a local optimum. With the help of the inherent noise disturbance of the circuit, it is possible to escape the local optimum and eventually gradually approach the global optimum.

[0015] The technical solution of this invention is as follows:

[0016] A method for solving the QUBO problem based on an analog computing circuit is used to solve the QUBO problem of an n×n positive quadratic term matrix. The method is characterized by utilizing Ohm's law and Kirchhoff's laws to naturally realize the multiplication and accumulation of the positive quadratic term matrix and the state vector within the circuit, and updating the state through comparators. Finally, the solution result is obtained at the output of each row's comparator. The method is based on an analog computing circuit including an n×n programmable resistive memory array, n current source modules, and n comparators. In this array, transistors and programmable resistive memories are connected in parallel as computing units. The two ends of the programmable resistive memory in each unit are connected to the source and drain of the transistor, respectively. All transistor gates in the same column are connected together. The source and drain of adjacent units in the same row are connected end-to-end. The source of the first transistor in the row is grounded, and the drain of the last transistor in the row is connected to the input of the comparator in that row, with an additional path leading to the current source module in that row. The output of each comparator in each row is connected to the gate of the transistor in the corresponding column of the array, forming n feedback loops. The specific steps of the method include:

[0017] (1) Construct the analog computing circuit;

[0018] (2) Set the output current of each row current source module According to the positive quadratic term matrix Information, for programmable

[0019] Write verification was performed on the resistive memory array to ensure compliance. ,in Let be the memory resistance value in the i-th row and j-th column of the array;

[0020] (3) According to the linear coefficient vector λ=(λ1,λ2,…,λ n ) T Information, setting the threshold voltage of each row comparator to meet the requirements.

[0021] ,in V is the threshold voltage of the comparator in the i-th row; V is the input voltage of the comparator in the i-th row. in Output voltage

[0022] V out When V in ≥λ i At that time, V out =V high At this point, it is considered as x i =0; when V in <λ i At that time, V out =V lowAt this point, it is considered x i =1, where V high and V low These are the high and low power supply voltages of the comparator; x i Let i be the i-th variable in the state vector;

[0023] (4) Turn on the current source modules in each row. The output current of the current source module in the i-th row is: The analog computing circuit begins

[0024] By utilizing Ohm's law and Kirchhoff's laws, the multiplication and accumulation of the positive quadratic term matrix and the state vector can be naturally achieved in the circuit. This allows the voltage V at the end of the i-th row of the programmable resistive memory array to be connected to the input of the comparator for that row. in for ;

[0025] (5) After the analog calculation circuit stabilizes, monitor the output voltage V of each row comparator. out If the voltage V in the i-th row out

[0026] If it is high level, then ,otherwise The state vector obtained on the n feedback loops is X=(x1,x2,…,x…). n ) T This is a solution to the QUBO problem.

[0027] Furthermore, the current source module includes an operational amplifier, resistors R1, R2, and R3, and an n-type transistor circuit. The circuit structure is as follows: the positive input terminal of the operational amplifier is connected to the drain of the n-type transistor, and another path is led out through resistor R3 and connected to the voltage. The inverting input of the operational amplifier is connected to the voltage through resistor R1. The operational amplifier output is connected to the gate of the n-type transistor via resistor R2. The source of the n-type transistor serves as the output interface of the current source module, connected to the input of the comparator in the corresponding row of the analog computing circuit. Based on the virtual short and virtual open characteristics of the operational amplifier, the output current I of the current source module satisfies... .

[0028] On the other hand, this invention also proposes a method for solving the QUBO problem of an n×n real quadratic term matrix. This method is based on an analog computing circuit for solving the QUBO problem of a real quadratic term matrix. This circuit, based on the aforementioned analog computing circuit for solving the QUBO problem of a positive quadratic term matrix, adds an n×n programmable resistive memory array, n current source modules, and n voltage subtraction modules of the same structure, thus forming a circuit for solving the problem of a matrix with n state variables X=(x1,x2,…,x…).n ) T A simulation circuit for the QUBO problem of a real quadratic term matrix; the drains of the row-end transistors of the newly added programmable resistive memory array and the original programmable resistive memory array are respectively connected to the non-inverting and inverting input terminals of the newly added voltage subtraction module in that row, and the drains of the row-end transistors are respectively connected to the current source module of the corresponding row. The gates of all transistors in the corresponding column are connected. The output terminal of the voltage subtraction module in each row is connected to the input terminal of the comparator in that row. The output terminal of the comparator in each row is connected to the gate terminals of the corresponding column transistors in the two arrays, forming n closed-loop feedback loops; the specific steps of this method include:

[0029] (1) Construct a simulation circuit for solving the QUBO problem of the real quadratic term matrix;

[0030] (2) Set the output current of each row current source module and ;

[0031] (3) Based on the real quadratic term matrix Information, for Perform the difference as follows and ,

[0032] satisfy Then, the programmable resistive memory array is written to verify that it meets the requirements. and ,in and Let be the memory resistance values ​​in the i-th row and j-th column of the two arrays, respectively; and satisfy . ;

[0033] (4) Output voltage of the voltage subtraction module ,in , and They are respectively voltage reduction

[0034] The output voltage, positive input voltage, and inverting input voltage of the module; among which,

[0035] , , ;

[0036] (5) According to the linear coefficient vector λ=(λ1,λ2,…,λ n ) T Information, setting the threshold voltage of each row comparator to meet the requirements. ,

[0037] in V is the threshold voltage of the comparator in the i-th row; V is the input voltage of the comparator in the i-th row. in Output voltage V out When V in ≥λ i At that time, V out =V high At this point, it is considered as x i =0; when V in <λ i At that time, V out =V low At this point, it is considered x i =1, where V high and V low These are the high and low power supply voltages of the comparator; x i Let i be the i-th variable in the state vector;

[0038] (6) Turn on the current source module. The output current magnitudes of the current source modules in the i-th row of the two arrays are respectively and ,electricity

[0039] The circuit starts working. After the circuit stabilizes, the voltage is monitored at the output of each row comparator. If the voltage of the i-th row is high, then... ,otherwise The state vector obtained on the n feedback loops is X=(x1,x2,…,x…). n ) T This is a solution to the QUBO problem.

[0040] Furthermore, the voltage subtraction module includes two buffers and one voltage subtractor. The two buffers are source follower structures, and are directly connected to the inverting input and output of an operational amplifier. Their non-inverting inputs are respectively connected to voltage V. pos and V neg The voltage subtractor consists of an operational amplifier and four resistors of the same value.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention proposes a QUBO problem-solving method based on analog computing circuits. Utilizing a programmable resistive memory array, a current source module, a voltage comparator, and a voltage subtraction module, this method achieves a one-step solution to the QUBO problem and possesses good versatility. The calculation results of each row in the array are directly fed into the comparator input in voltage form, eliminating the need for current-to-voltage conversion via a transimpedance amplifier. Furthermore, the comparator's output signal is fed back to the gates of the interconnected transistors in each column of the programmable resistive memory array. The high impedance of these gates provides excellent signal isolation and insensitivity to source-drain voltage changes, thus significantly reducing signal disturbances caused by array read / write operations. This invention combines the mathematical mapping of the QUBO problem with the physical evolution of the analog computing circuit. By mapping the quadratic term matrix and the linear coefficient vector to the resistance-current product and the comparator threshold voltage in the circuit, a physical system directly corresponding to the energy function of the optimization problem is constructed, simulating the iterative descent process in discrete optimization. This method does not rely on external digital circuits to provide a clock, but updates continuously, and has advantages such as high speed, fewer external circuits, and low power consumption. It is suitable for QUBO problem application scenarios that are sensitive to power consumption and have high solution speed requirements. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of an existing analog computing circuit for solving the QUBO problem;

[0044] Figure 2 This is a schematic diagram of the structure of a simulation computing circuit for solving a positive quadratic term matrix in this invention;

[0045] Figure 3 This is a circuit diagram illustrating one specific implementation of the current source module in this invention.

[0046] Figure 4 This is a schematic diagram of the structure of a simulation computing circuit for solving the QUBO problem of a real quadratic term matrix in this invention;

[0047] Figure 5 This is a circuit diagram illustrating one specific implementation of the voltage subtraction module in this invention.

[0048] Figure 6 This is a schematic diagram of the overall execution flow of the method of the present invention. Detailed Implementation

[0049] To more clearly illustrate the objectives, technical solutions, and advantages of this invention, a further detailed description is provided below in conjunction with the accompanying drawings. The description herein is merely illustrative and not intended to limit the scope of the invention.

[0050] Figure 2This refers to the analog computing circuit used in the method of this invention, which is used to solve the QUBO problem of an n×n positive quadratic term matrix. The circuit comprises an n×n programmable resistive memory array, n current source modules, and n comparators, and is used to solve the QUBO problem. In the circuit structure, the transistor units in each row are connected sequentially, with their sources and drains interconnected to form a chain-like path. The source of the first column of transistors is connected to ground, while the drain of the last column is connected to the signal input of the comparator. Additionally, a branch is led out from this node and connected to the current source module. The comparator's output signal is sent to the gate of the corresponding column of transistors in that row, thus forming a closed feedback control loop.

[0051] positive quadratic term matrix The element is mapped to the product of the analog resistance value of the programmable resistor memory in the programmable resistor memory array and the output current of the current source module. The mapping rule is as follows: , where I i Let V be the output current of the current source module in the i-th row; assuming the dual power supply voltages of the comparator are V... high and V low Its threshold voltage is controlled by a linear coefficient vector λ, satisfying ,in The threshold voltage of the comparator in the i-th row is defined when its input voltage is greater than or equal to the threshold voltage. At that time, the comparator output voltage V high , regarded as x i =0, at which point the corresponding column transistor channel is turned on, which is considered a short circuit; otherwise, the output voltage is V. low , regarded as x i =1, at which point the corresponding column transistor channel is closed, which is considered as the transistor being open-circuited.

[0052] For the input voltage V of the comparator in row i in ,have:

[0053] ,

[0054] Therefore, the circuit satisfies:

[0055]

[0056] The above rule is equivalent to choosing a variable x at each time step. i And based on the current other variables x j (j≠i) state update x i Make the system energy function: Decrease or remain unchanged.

[0057] The current source modules for each row are activated, and the analog calculation circuit begins to operate. After the circuit stabilizes, the output voltage of each row comparator is monitored. The state vector X=(x1,x2,…,x) is obtained from the n feedback loops. n ) T This is a solution to the QUBO problem. The state vector X is updated naturally each time by the feedback loop.

[0058] Figure 3 This is a circuit diagram illustrating a specific implementation of the current source module in this invention, wherein the inverting input terminal of the operational amplifier is connected to the voltage via resistor R1. The connection is made by connecting the non-inverting input terminal to the drain terminal of the NMOS transistor, and then connecting it to the voltage through resistor R3. The output terminal is connected to the NMOS gate via resistor R2, and the NMOS source terminal serves as the output port of the current source module. When the circuit is operating normally, the current I generated at the NMOS source terminal satisfies:

[0059] ,

[0060] By adjusting and The voltage value can be used to adjust the magnitude of the constant current.

[0061] Figure 4 This is a schematic diagram of the circuit structure used in the simulation calculation method of the present invention for solving the QUBO problem of a real quadratic term matrix. The circuit includes two n×n programmable resistive memory arrays, 2×n current source modules, n voltage subtraction modules, and n comparators. In the same row of the two arrays, the source and drain terminals of the transistors are connected sequentially from end to end. The source terminal of the first transistor in the row is grounded, and the drain terminals of the last transistor in the row are connected to the non-inverting and inverting input terminals of the voltage subtraction modules, respectively. Each module also has a corresponding current source module output. The output terminal of the voltage subtraction module is connected to the input terminal of the comparator, and the output terminal of the comparator in the i-th row is connected to the gate terminal of the transistor in the i-th column of the two arrays, forming a closed-loop feedback loop.

[0062] The resistance values ​​of the programmable resistor memory in the i-th row and j-th column of each cell in the two arrays of the circuit are respectively and The constant current of the current source module in the i-th row is set to... and ,satisfy The voltage subtraction module satisfies the output voltage. ,in , and These represent the output voltage, positive input voltage, and inverting input voltage of the voltage subtraction module, respectively. Assume the comparator's dual power supply voltages are V...high and V low The threshold voltage of the comparator in the i-th row is Its threshold voltage is controlled by a linear coefficient vector λ, satisfying When the input voltage is higher than or equal to the threshold voltage, the output voltage is V. high , regarded as x i =0, otherwise the output voltage is V low , regarded as x i =1. The input voltage V of the comparator in the i-th row is... in satisfy:

[0063] ,

[0064] Therefore, the circuit satisfies:

[0065] ,

[0066] Similar to the circuit update rule for solving the QUBO problem with positive quadratic terms, this circuit selects x that makes the energy change non-positive (≤0) for each update. i This ensures that the energy function is non-increasing. Therefore, during the natural evolution of the physical circuit, the energy function E of this system is always monotonically non-increasing and eventually tends to stabilize, thus ensuring convergence to a local optimum. Considering factors such as noise in the circuit, it will eventually gradually approach the global optimum.

[0067] The current source modules for each row are activated, and the analog calculation circuit begins to operate. After the circuit stabilizes, the output voltage of each row comparator is monitored. The state vector X=(x1,x2,…,x) is obtained from the n feedback loops. n ) T This is a solution to the QUBO problem. The state vector X is updated naturally each time by the feedback loop.

[0068] Figure 5 This is a circuit diagram illustrating a specific implementation of the voltage subtraction module in this invention. It includes two buffers and a voltage subtractor. The two buffers are source follower structures, directly connected to the inverting input and output of an operational amplifier, with their non-inverting inputs connected to voltage subtractors. and The voltage subtractor consists of an operational amplifier and four resistors of the same value, with the output voltage set to... Based on the virtual short and virtual open characteristics of the operational amplifier, the following conditions are met: In the circuit, the voltages at the two input terminals of the voltage subtraction module in the i-th row satisfy the following conditions:

[0069] , ,

[0070] The output voltage then satisfies:

[0071] ,

[0072] Figure 6 The diagram shows the overall execution flow of the method of the present invention. Step 1: Construct the analog calculation circuit for QUBO solving, including components such as a programmable resistor memory array, a current source module, and a comparator; Step 2: Based on the given quadratic term matrix... Output current of the current source module This is mapped to the resistance value R in the array, satisfying... or Step 3: Based on the linear coefficient vector Step 4: Set the threshold voltage for each row comparator; Step 5: Turn on the current source module to start the circuit and allow the system to enter a natural evolution state; Step 6: After the circuit stabilizes, monitor the output level of each row comparator and output the state vector based on the state and output voltage transition relationship. This is a solution to the QUBO problem.

[0073] The embodiments described above are not intended to limit the present invention. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention is defined by the scope of the claims.

Claims

1. A method for solving the QUBO problem based on analog computing circuits, used to solve the QUBO problem of an n×n positive quadratic term matrix, characterized in that, Using Ohm's law and Kirchhoff's laws, the multiplication and accumulation of the positive quadratic term matrix and the state vector are naturally realized in the circuit, and the state is updated through comparators. Finally, the solution is obtained at the output of each row comparator. This method is based on an analog computing circuit including an n×n programmable resistive memory array, n current source modules, and n comparators. In this array, transistors and programmable resistive memories are connected in parallel as computing units. The two ends of the programmable resistive memory of each unit are connected to the source and drain of the transistor, respectively. All transistor gates in the same column are connected. The source and drain of adjacent units in the same row are connected end to end. The source of the first transistor in the row is grounded, and the drain of the last transistor in the row is connected to the input of the comparator in that row. An additional path is led out to connect to the current source module in that row. The output of the comparator in each row is connected to the gate of the transistor in the corresponding column of the array, forming n feedback loops. The specific steps of this method include: (1) Construct the analog computing circuit; (2) Set the output current of each row current source module According to the positive quadratic term matrix Information, for programmable Write verification was performed on the resistive memory array to ensure compliance. ,in Let be the memory resistance value in the i-th row and j-th column of the array; (3) According to the linear coefficient vector λ=(λ1,λ2,…,λ n ) T Information, setting the threshold voltage of each row comparator to meet the requirements. ,in V is the threshold voltage of the comparator in the i-th row; V is the input voltage of the comparator in the i-th row. in Output voltage V out When V in ≥λ i At that time, V out =V high At this point, it is considered as x i =0; when V in <λ i At that time, V out =V low At this point, it is considered x i =1, where V high and V low These are the high and low power supply voltages of the comparator; x i Let i be the i-th variable in the state vector; (4) Turn on the current source modules in each row. The output current of the current source module in the i-th row is: The analog computing circuit is open To begin operation, utilizing Ohm's law and Kirchhoff's laws, the multiplication and accumulation of the positive quadratic term matrix and the state vector can be naturally achieved in the circuit. This allows the voltage V at the end of the i-th row of the programmable resistive memory array to be connected to the input of the comparator for that row. in for ; (5) After the analog calculation circuit stabilizes, monitor the output voltage V of each row comparator. out If the voltage V in the i-th row out If it is high level, then ,otherwise The state vector obtained on the n feedback loops is X=(x1,x2,…,x…). n ) T This is a solution to the QUBO problem.

2. The method as described in claim 1, characterized in that, The current source module includes an operational amplifier, resistors R1, R2, and R3, and an n-type transistor circuit. The circuit structure is as follows: the positive input terminal of the operational amplifier is connected to the drain of the n-type transistor, and another path is led out through resistor R3 and connected to the voltage. The inverting input of the operational amplifier is connected to the voltage through resistor R1. The operational amplifier output is connected to the gate of the n-type transistor via resistor R2. The source of the n-type transistor serves as the output interface of the current source module, connected to the input of the comparator in the corresponding row of the analog computing circuit. Based on the virtual short and virtual open characteristics of the operational amplifier, the output current I of the current source module satisfies... .

3. A method for solving the QUBO problem of an n×n real quadratic term matrix, characterized in that, This method is based on a simulation circuit for solving the QUBO problem of a real quadratic term matrix. This circuit, based on the simulation circuit for solving the QUBO problem of a positive quadratic term matrix as described in claim 1, adds an n×n programmable resistive memory array, n current source modules, and n voltage subtraction modules of the same structure, thus forming a circuit for solving the problem of a QUBO problem with n state variables X=(x1,x2,…,x…). n ) T A simulation circuit for the QUBO problem of a real quadratic term matrix; the drains of the row-end transistors of the newly added programmable resistive memory array and the original programmable resistive memory array are respectively connected to the non-inverting and inverting input terminals of the newly added voltage subtraction module in that row, and the drains of the row-end transistors are respectively connected to the current source module of the corresponding row. The gates of all transistors in the corresponding column are connected. The output terminal of the voltage subtraction module in each row is connected to the input terminal of the comparator in that row. The output terminal of the comparator in each row is connected to the gate terminals of the corresponding column transistors in the two arrays, forming n closed-loop feedback loops; the specific steps of this method include: (1) Construct a simulation circuit for solving the QUBO problem of the real quadratic term matrix; (2) Set the output current of each row current source module and ; (3) Based on the real quadratic term matrix Information, for Perform the difference as follows and , satisfy Then, the programmable resistive memory array is written to verify that it meets the requirements. and ,in and Let be the memory resistance values ​​in the i-th row and j-th column of the two arrays, respectively; and satisfy . ; (4) Output voltage of the voltage subtraction module ,in , and Voltage The subtraction module's output voltage, positive input voltage, and inverting input voltage; among which, , , ; (5) According to the linear coefficient vector λ=(λ1,λ2,…,λ n ) T Information, setting the threshold voltage of each row comparator to meet the requirements. , in V is the threshold voltage of the comparator in the i-th row; V is the input voltage of the comparator in the i-th row. in Output voltage V out When V in ≥λ i At that time, V out =V high At this point, it is considered as x i =0; when V in <λ i At that time, V out =V low At this point, it is considered x i =1, where V high and V low These are the high and low power supply voltages of the comparator; x i Let i be the i-th variable in the state vector; (6) Turn on the current source module. The output current magnitudes of the current source modules in the i-th row of the two arrays are respectively and , The circuit starts working. After the circuit stabilizes, the voltage at the output of each row comparator is monitored. If the voltage of the i-th row is high, then... ,otherwise The state vector obtained on the n feedback loops is X=(x1,x2,…,x…). n ) T This is a solution to the QUBO problem.

4. The method as described in claim 3, characterized in that, The voltage subtraction module includes two buffers and one voltage subtractor. The two buffers are source follower structures, directly connected to the inverting input and output of an operational amplifier, with their non-inverting inputs connected to voltage V. pos and V neg The voltage subtractor consists of an operational amplifier and four resistors of the same value.