Computer memory data processing system

Abstract

The invention relates to the technical field of computers, and discloses a computer memory data processing system comprising a calculation task compiling module used for determining physical setting parameters according to a calculation task and generating a preset target evolution path; the physical calculation execution module is used for performing state evolution according to the physical setting parameters; the dynamic feedback control module is used for monitoring the actual evolution state and generating a calibration signal according to the deviation between the actual evolution state and a preset target evolution path so as to dynamically adjust driving; and the result decoding module is used for converting the steady-state physical quantity into a calculation result after the system reaches a stable equilibrium state. By constructing closed-loop feedback control, evolution deviation generated by non-ideal factors of a physical system is corrected in real time, it can be effectively guaranteed that actual state evolution of physical calculation accurately tends to a preset ideal path, and therefore the accuracy and reliability of data processing through the physical process are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a computer memory data processing system. Background Technology

[0002] Currently, most mainstream computing systems adopt the von Neumann architecture. A key feature of this architecture is the physical separation of its processing and storage units. When processing complex computational tasks, data needs to be transferred frequently between the processing and storage units. This process generates latency and power consumption, constituting a limiting factor in system performance and consequently affecting the overall energy efficiency of the computing system.

[0003] To address the performance limitations of the aforementioned data transmission, one technical solution is to design application-specific integrated circuits (ASICs). This approach integrates the functionality of specific algorithms into hardware circuits, improving processing speed and energy efficiency for specific tasks. However, this solution has drawbacks: its functionality is fixed once designed, lacking programmability and making it difficult to adapt to algorithm updates or diverse task requirements. Furthermore, its development costs and design cycles limit the applicability of this type of dedicated hardware.

[0004] Another type of technical solution utilizes the evolutionary process of a physical system for computation, i.e., physical computation. This approach aims to solve specific mathematical problems by leveraging the parallel nature of physical laws. However, this type of solution faces challenges in terms of accuracy and reliability. Physical systems contain non-ideal factors, such as inhomogeneities in material properties, deviations in manufacturing processes, and variations in ambient temperature. These factors can all affect the evolutionary process of the physical system, leading to deviations in the computational results and making it difficult to meet the accuracy requirements of some applications. Therefore, the reliability of existing physical computation systems is limited, which also affects their practical application. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a computer memory data processing system that solves the energy efficiency bottleneck caused by the separation of storage and computation in traditional computing architectures, as well as the problems of low accuracy and poor reliability of calculation results caused by the inability of existing physical computing schemes to overcome physical non-idealities.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a computer memory data processing system, comprising...

[0007] Preferably, the computing task compilation module is used to determine a set of physical setting parameters to characterize the received computing task.

[0008] The physical calculation execution module is used to perform state evolution based on the physical setting parameters until a stable equilibrium state is reached;

[0009] The dynamic feedback control module is used to monitor the actual evolution state of the physical calculation execution module in real time when the physical calculation execution module is evolving, and generate a calibration signal for correction based on the deviation between the actual evolution state and the preset target evolution path. By dynamically adjusting the drive of the physical calculation execution module, it is used to ensure that the state evolution tends to the preset target evolution path.

[0010] The result decoding module is used to obtain the steady-state physical quantity under the stable equilibrium state and convert the steady-state physical quantity into the original digital value, which is used as the calculation result of the calculation task.

[0011] Preferably, the computation task compilation module includes:

[0012] The interface and physical mapping unit is responsible for receiving and parsing the computing task, translating the mathematical operations contained in the task into a set of driving conditions applied to the boundary of the physical computing execution module according to the preset mapping rules, and generating a corresponding reverse parsing algorithm based on the mapping rules as the preset decoding function.

[0013] An ideal evolution path generation unit is used to perform fast pure digital simulation during the compilation phase based on the driving conditions and initial states determined by the physical mapping unit.

[0014] Preferably, the physical calculation execution module includes:

[0015] Multidimensional computational lattice units are used to utilize their multi-physics coupling characteristics to transform a complex mathematical computation problem into a natural evolution problem of a physical system that occurs within the multidimensional computational lattice and can interact through external driving and internal sensing.

[0016] The drive and sensing interface unit is used to receive and apply physical stimuli from the outside, and to monitor the physical state of key nodes inside the multidimensional computing lattice unit in real time.

[0017] Preferably, the dynamic feedback control module includes:

[0018] The deviation calculation and control decision unit is used to continuously receive the actual evolution state from the physical calculation execution module and the preset target evolution path from the calculation task compilation module;

[0019] The convergence judgment and drive update unit is used to apply the calibration signal to the drive of the physical calculation execution module in real time.

[0020] Preferably, the result decoding module includes:

[0021] The steady-state physical quantity reading and conversion unit is used to read the final steady-state physical quantity from a designated area of ​​the physical calculation execution module according to a preset decoding rule;

[0022] The digital calibration and formatting unit is used to receive the original digital value and perform linear scaling and offset compensation calibration operations by the decoding function preset by the computing task compilation module.

[0023] Preferably, the multiphysics coupling characteristic of the multidimensional computational lattice unit is the coupling characteristic between the electric field and the thermal field;

[0024] The initial physical state is the initial conductivity distribution of each computational unit in the multidimensional computational lattice;

[0025] The driving condition is the voltage or current applied to the boundary of the multidimensional computational lattice;

[0026] The actual physical properties of the multidimensional computational lattice, including non-uniform heat dissipation and nonlinear effects of material conductivity with temperature, cause a deviation between the actual steady-state temperature distribution formed by the multidimensional computational lattice unit under only initial driving conditions and the ideal result defined by the preset target evolution path.

[0027] The calibration signal generated by the dynamic feedback control module is an additional, dynamically adjusted correction drive. This additional, dynamically adjusted correction drive is superimposed on the driving conditions to actively compensate for the deviation.

[0028] Preferably, the control algorithm built into the deviation calculation and control decision unit is a PID control algorithm, which is used to synthesize the time integral and time derivative of the deviation signal to generate the calibration signal.

[0029] Preferably, the preset mapping rule between the interface and the physical mapping unit is used to map the multiplication and addition operations in the mathematical operations to a comprehensive physical process driven by the driving conditions, which is generated by the Joule heating effect and the heat conduction effect within the multidimensional computing lattice unit due to physical laws.

[0030] Preferably, the computation task compilation module is able to generate different physical setting parameters in response to different computation tasks;

[0031] Preferably, the received computational task's physical setting parameters include an initial state of a physical system and a set of driving conditions applied to the boundary of the physical system, and generate a preset target evolution path of the physical system under ideal conditions.

[0032] This invention provides a computer memory data processing system. It has the following advantages:

[0033] 1. This invention employs a design that integrates data storage and physical computation processes into a single physical unit. Specifically, within a multidimensional computational lattice unit, an initial physical state is set to hold computational data, and the computation is completed through the natural evolution of physical laws. This structure reduces the need for high-frequency data transfer between separate processing and storage units, thereby reducing latency and power consumption caused by data round-trip movement and improving the system's processing parallelism and energy efficiency.

[0034] 2. This invention incorporates a dynamic feedback control module. This module monitors the actual evolution state of the physical calculation execution module in real time and compares it with a preset target evolution path, generating calibration signals to dynamically adjust the driving conditions. This closed-loop control mechanism can proactively compensate for evolution deviations caused by factors such as non-ideal material properties and environmental changes. Compared to open-loop physical calculation systems, this invention improves the accuracy and reliability of the calculation results through this approach.

[0035] 3. This invention achieves system programmability by setting up a computational task compilation module. This module can receive different computational tasks and map the mathematical entities in these tasks to different physical setting parameters. This allows the same set of physical computation execution module hardware to be used to solve different types of computational problems without modifying the physical structure. Compared with the application-specific integrated circuit (ASIC) solution that embeds the algorithm into the hardware, this invention solves the problems of limited hardware functionality and narrow applicability, and enhances the system's task adaptability. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the structure of a computer memory data processing system according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram illustrating the working principle of a computer memory data processing system according to an embodiment of the present invention;

[0038] Figure 3 This is a flowchart illustrating a computer memory data processing method according to an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Please see the appendix Figure 1 - Appendix Figure 3 This invention provides a computer memory data processing system, including: a computing task compilation module, a physical computing execution module, a dynamic feedback control module, and a result decoding module.

[0041] The computation task compilation module is used to determine a set of physical setting parameters to characterize the received computation task, and to generate a preset target evolution path of the physical system under ideal conditions. In one embodiment, the function of this module is accomplished by its internal interface and physical mapping unit and ideal evolution path generation unit.

[0042] The interface and physical mapping unit is responsible for receiving and parsing computational tasks. Taking the task of solving a system of linear equations as an example, this task can be mathematically represented as:

[0043] Ax=b (1)

[0044] In the formula: Given a known coefficient matrix; Given a known constant vector; The unknown vector to be solved. This unit, according to a set of preset mapping rules, transforms the mathematical components of the task into physical parameters of the physical system. These parameters include the initial state of the physical system and a set of driving conditions applied to the boundary of the physical system. The initial state of the physical system is set as the initial conductivity distribution of each computational cell in the multidimensional computational lattice.

[0045] This unit uses a pre-defined linear mapping function to map each element a of the coefficient matrix A. ij Converted to the initial conductivity value σ in the corresponding physical space coordinates (i,j) ij (0).

[0046] σ ij (0)=c1·a ij +c2 (2)

[0047] In the formula: σ ij (0) represents the conductivity of the computational unit at coordinates (i,j) at the initial time t=0, in Siemens per meter (S / m); c1 is a calibrated scaling factor used to convert the dimensionless mathematical value a ij Converted to conductivity values; c2 is a calibrated offset constant used to ensure that the mapped conductivity values ​​are within the positive operating range allowed by the physical material. The driving conditions applied to the boundary of this physical system are set as a set of voltages applied to the boundary electrodes of the multidimensional computational lattice unit. This unit, through another preset linear mapping function, converts each element b of the constant vector b... i Converted to the driving voltage value V applied to the i-th boundaryi。

[0048] V i =c3·b i +c4 (3)

[0049] In the formula: V i c3 is the initial driving voltage applied to the i-th boundary, in volts (V); c4 is the scaling factor determined according to the gain of the driving circuit; c5 is the offset voltage determined according to the circuit reference ground potential. Simultaneously, this interface and the physical mapping unit generate a corresponding reverse analytical algorithm based on the adopted mapping functions (2) and (3) and the system's physical characteristics, which serves as the preset decoding function and is then transmitted to the result decoding module. The ideal evolution path generation unit is used to perform digital simulation based on the physical setting parameters during the compilation phase to generate a preset target evolution path. This unit establishes an idealized mathematical model of the physical system, which ignores non-ideal effects and assumes that the electrical and thermal conductivity of the material are constants that do not change with temperature.

[0050] In this ideal model, mathematical operations are mapped to a combined physical process driven by conditions, involving both Joule heating and heat conduction effects generated within a multidimensional computational lattice. The transient behavior of this process is described by the following ideal heat conduction equation:

[0051]

[0052] In the formula: T(t) is the temperature field that varies with time t, and the unit is Kelvin (K); ρ is the material density, and the unit is kilograms per cubic meter (kg / m³). 3 );c p K is the specific heat capacity of the material, expressed in joules per kilogram of Kelvin (J / (kg·K)); ideal The idealized, uniform thermal conductivity of the material is expressed in watts per meter Kelvin (W / (m·K)); σ ideal For the initial state {σ ij (0)} constitutes an ideal conductivity field; φ is the electric conductivity field determined by the driving condition {V i} and conductivity field σ ideal The determined, steady-state current continuity equation The electric potential field.

[0053] Subsequently, the unit performs transient simulation of formula (4) using numerical calculation methods, and outputs a time series data that records the temperature evolution trajectory of each key sensing node position in the lattice under ideal conditions. This trajectory is the preset target evolution path T. ref (t). This path is transmitted to the dynamic feedback control module.

[0054] The physics computation execution module receives the physical setting parameters generated by the computation task compilation module and performs state evolution in a multiphysics coupled system based on these parameters until a stable equilibrium state is reached. This module includes a multidimensional computational lattice unit and a drive and sensing interface unit.

[0055] Multidimensional computational lattice units are the medium for performing physical calculations. Physically, they can be constructed as a two-dimensional or three-dimensional solid array composed of a large number of computational units, each of which contains materials with specific electrical and thermal conductivity properties.

[0056] The drive and sensing interface unit is used to apply drive signals and monitor status information. The drive section consists of a digital-to-analog converter array, a power amplifier, and an electrode array. The sensing section consists of a temperature sensor network embedded within a multidimensional computing lattice unit, signal conditioning circuitry, and an analog-to-digital converter array.

[0057] The dynamic feedback control module is used to construct a real-time closed-loop control system during the state evolution of the physical calculation execution module. This module includes a deviation calculation and control decision unit and a convergence judgment and drive update unit.

[0058] The deviation calculation and control decision unit, in each discrete control cycle, calculates the actual evolution state vector T from the physical calculation execution module. actual (k) and the preset target evolution path vector T from the computation task compilation module ref (k) Perform element-by-element subtraction to obtain a real-time deviation vector e. k To generate the calibration signal for error correction, this unit can employ a PID control algorithm. In its discrete-time digital implementation, the algorithm is calculated as follows:

[0059]

[0060] In the formula: u k e is the calibration signal vector generated at time k; k With e k-1 These are the deviation vectors for the current time step and the previous time step, respectively; T s K is the fixed sampling period of the control system. p ,K i ,K d The proportional, integral, and derivative gain parameters of the PID controller can be a diagonal matrix in the multi-input multi-output system of this embodiment.

[0061] The convergence judgment and drive update unit receives the calibration signal vector u kThis unit performs a vector addition operation between the dynamic calibration signal and the static initial driving condition vector V provided by the computation task compilation module, resulting in a new, dynamically adjusted total driving vector V. total (k):

[0062] V total (k)=V+u k (6)

[0063] This unit transmits the total driving vector in real time to the driving and sensing interface unit of the physical calculation execution module to update the physical excitation applied to the boundary of the multidimensional computational lattice. This process can compensate for evolutionary deviations caused by physical non-ideals.

[0064] This unit transmits the total driving vector in real time to the driving and sensing interface unit of the physical calculation execution module to update the physical excitation applied to the boundary of the multidimensional computational lattice. This process can compensate for evolutionary deviations caused by physical non-ideals.

[0065] While updating the driver, this unit determines whether the physical system has reached a stable equilibrium state based on the magnitude of the deviation vector. In each control cycle, this unit calculates the L2 norm of the deviation vector. When this norm value is less than a preset convergence threshold ∈ [a certain value] within a preset duration window, the system is considered converged.

[0066] ||e k ||2<∈ (7)

[0067] Once this condition is met, the unit sends a trigger signal to the result decoding module. The result decoding module's task is to convert the final physical state into a standard digital format after the physical calculation process is complete. This module includes a steady-state physical quantity reading and conversion unit and a digital calibration and formatting unit.

[0068] Upon receiving a trigger signal, the steady-state physical quantity reading and conversion unit sends a reading command to the drive and sensing interface unit of the physical calculation execution module to obtain the final steady-state physical quantity in the specified region, such as the final steady-state temperature vector T. final .

[0069] The digital calibration and formatting unit receives this steady-state physical quantity vector and processes it using a decoding function pre-generated by the computation task compilation module. This function performs an affine transformation:

[0070] x calc =M scale ·T final +V offset (8)

[0071] In the formula: x calcThis is the final output vector of the decoded computation result. T final M is the input steady-state temperature vector. scale This is a scaling calibration matrix, calculated by the computation task compiler module based on the system gain characteristics and mapping scale, used to convert measurements in physical units back to dimensionless numerical space. V offset This is an offset compensation calibration vector, pre-determined by the compiler module, used to compensate for systematic biases. After numerical calibration is completed, this unit can adjust the calculated result vector x. calc Perform formatting and output it as the result of the computation task.

[0072] Working principle: The system transforms a mathematical computation task into a controlled physical evolution process within a specific physical system. This process completes the computation through a closed-loop workflow of "compilation-execution-control-decoding." The system utilizes real-time digital feedback to compensate for non-ideal factors in the physical system, thereby improving the accuracy of the computational results.

[0073] The physicalization of the computation task is accomplished by the computation task compilation module, which transforms the input mathematical problem into a set of initial parameters for the physical system and a preset evolution reference path.

[0074] The interface and physical mapping unit receives the computational task. This unit converts mathematical entities into physical parameters according to preset mapping rules. Specifically, each element of the coefficient matrix is ​​linearly mapped to the initial conductivity at the corresponding spatial location in the multidimensional computational lattice unit; each element of the constant vector is linearly mapped to the driving voltage applied to the lattice boundary. This process transforms abstract mathematical equations into a set of explicit physical parameters.

[0075] Based on the aforementioned physical parameters, the ideal evolution path generation unit calculates the entire temperature evolution process of the system from its initial state to its theoretical steady state using numerical simulation methods within an idealized digital model. Its output is a time series vector, representing the preset target evolution path, which is used as a reference signal for subsequent physical processes.

[0076] Controlled physical computation execution is performed by the physical computation execution module and the dynamic feedback control module working together to perform computations in the physical system.

[0077] The driving and sensing interface unit of the physical calculation execution module applies the initial driving voltage determined during the compilation phase to the boundary of the multidimensional computational lattice unit. Based on the preset initial conductivity distribution, the internal temperature field of the lattice begins to evolve over time due to the Joule heating effect and thermal conduction effect under electrical drive.

[0078] Due to the non-ideal factors such as nonlinearity and inhomogeneity in real materials, the actual temperature evolution path will deviate from the preset target evolution path. The dynamic feedback control module performs the following operations to reduce this deviation. The deviation calculation and control decision unit acquires the actual temperature state through the sensor network at a certain frequency and compares it with the preset target evolution path to generate a real-time deviation vector. The PID control algorithm built into this unit calculates a calibration signal vector based on the current deviation, the accumulation of historical deviations, and the rate of change of the deviation. The convergence judgment and drive update unit adds this calibration signal vector to the original drive voltage vector to form a dynamically adjusted total drive vector. This total drive vector is applied back to the physical system in real time to adjust its state evolution and make it approach the preset target evolution path.

[0079] The decoding and output of the calculation results are performed by the result decoding module, whose function is to convert the final physical state back to a standard digital format after the calculation converges. After receiving the convergence trigger signal from the control module, the steady-state physical quantity reading and conversion unit reads and records the final steady-state physical quantity of a specified region inside the physical lattice at this moment through the sensor network, and obtains a steady-state temperature vector.

[0080] The digital calibration and formatting unit receives the steady-state temperature vector. This unit applies a decoding function, pre-generated during the compilation phase and the inverse of the physical mapping process, to linearly scale and offset the raw temperature data. This operation converts the measured values, with physical units, into a dimensionless solution vector in the original mathematical problem space, which is then output as the final calculation result.

[0081] This invention constructs a complete and high-precision physical computation method by mapping mathematical problems to physical processes, pre-setting ideal paths for these processes, correcting deviations in real time through closed-loop feedback during execution, and reversing the physical results back to digital solutions after stabilization. It cleverly combines the high parallelism of physical systems with the high precision of digital control systems, enabling efficient and reliable solutions to complex computational tasks.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A computer memory data processing system, characterized in that, include: The computation task compilation module is used to determine a set of physical setting parameters to characterize the received computation task. The physical calculation execution module is used to perform state evolution based on the physical setting parameters until a stable equilibrium state is reached; The dynamic feedback control module is used to monitor the actual evolution state of the physical calculation execution module in real time when the physical calculation execution module is evolving, and generate a calibration signal for correction based on the deviation between the actual evolution state and the preset target evolution path. By dynamically adjusting the drive of the physical calculation execution module, it is used to ensure that the state evolution tends to the preset target evolution path. The result decoding module is used to obtain the steady-state physical quantity under the stable equilibrium state and convert the steady-state physical quantity into the original digital value, which is used as the calculation result of the calculation task.

2. The computer memory data processing system according to claim 1, characterized in that, The computation task compilation module includes: The interface and physical mapping unit is responsible for receiving and parsing the computing task, translating the mathematical operations contained in the task into a set of driving conditions applied to the boundary of the physical computing execution module according to the preset mapping rules, and generating a corresponding reverse parsing algorithm based on the mapping rules as the preset decoding function. An ideal evolution path generation unit is used to perform fast pure digital simulation during the compilation phase based on the driving conditions and initial states determined by the physical mapping unit.

3. The computer memory data processing system according to claim 1, characterized in that, The physics calculation execution module includes: Multidimensional computational lattice units are used to utilize their multi-physics coupling characteristics to transform a complex mathematical computation problem into a natural evolution problem of a physical system that occurs within the multidimensional computational lattice and can interact through external driving and internal sensing. The drive and sensing interface unit is used to receive and apply physical stimuli from the outside, and to monitor the physical state of key nodes inside the multidimensional computing lattice unit in real time.

4. A computer memory data processing system according to claim 1, characterized in that, The dynamic feedback control module includes: The deviation calculation and control decision unit is used to continuously receive the actual evolution state from the physical calculation execution module and the preset target evolution path from the calculation task compilation module; The convergence judgment and drive update unit is used to apply the calibration signal to the drive of the physical calculation execution module in real time.

5. A computer memory data processing system according to claim 1, characterized in that, The result decoding module includes: The steady-state physical quantity reading and conversion unit is used to read the final steady-state physical quantity from a designated area of ​​the physical calculation execution module according to a preset decoding rule; The digital calibration and formatting unit is used to receive the original digital value and perform linear scaling and offset compensation calibration operations by the decoding function preset by the computing task compilation module.

6. A computer memory data processing system according to claim 3, characterized in that, The multiphysics coupling characteristics of the multidimensional computational lattice unit are the coupling characteristics between the electric field and the thermal field. The initial physical state is the initial conductivity distribution of each computational unit in the multidimensional computational lattice; The driving condition is the voltage or current applied to the boundary of the multidimensional computational lattice; The actual physical properties of the multidimensional computational lattice, including non-uniform heat dissipation and nonlinear effects of material conductivity with temperature, cause a deviation between the actual steady-state temperature distribution formed by the multidimensional computational lattice unit under only initial driving conditions and the ideal result defined by the preset target evolution path. The calibration signal generated by the dynamic feedback control module is an additional, dynamically adjusted correction drive. This additional, dynamically adjusted correction drive is superimposed on the driving conditions to actively compensate for the deviation.

7. A computer memory data processing system according to claim 3, characterized in that, The control algorithm built into the deviation calculation and control decision unit is a PID control algorithm, which is used to combine the time integral and time derivative of the deviation signal to generate the calibration signal.

8. A computer memory data processing system according to claim 3, characterized in that, The preset mapping rules between the interface and the physical mapping unit are used to map the multiplication and addition operations in the mathematical operations to a comprehensive physical process driven by the driving conditions, which is generated by the Joule heating effect and the heat conduction effect within the multidimensional computing lattice unit due to physical laws.

9. A computer memory data processing system according to claim 1, characterized in that, The computation task compilation module can generate different physical setting parameters in response to different computation tasks.

10. A computer memory data processing system according to claim 1, characterized in that, The received computational task's physical setting parameters include an initial state of a physical system and a set of driving conditions applied to the boundary of the physical system, and generate a preset target evolution path for the physical system under ideal conditions.

Images