An edge computing-oriented low-power reconfigurable electronic component chip scheduling system

Abstract

The application belongs to the technical field of edge computing and integrated circuit, and particularly relates to a low-power reconfigurable electronic component chip scheduling system for edge computing, which comprises a task semantic perception module, a resource state monitoring module, a power consumption trajectory prediction and voltage compensation module, a reconfiguration granularity control module and an integrated scheduling core module. The system extracts a task feature vector, combines a real-time monitored logic array state, predicts power consumption by using a configuration jump Hamming distance, and implements voltage pre-compensation. Local fine-grained reconfiguration is realized by cooperating with bit stream compression and shielding mask. The scheduling core performs resource mapping and strategy generation based on multi-objective optimization logic, realizes efficient and accurate scheduling of reconfigurable resources, and significantly improves the system energy efficiency ratio and operation stability.

Description

Technical Field

[0001] This invention belongs to the field of edge computing and integrated circuit technology, specifically a low-power reconfigurable electronic component chip scheduling system for edge computing. Background Technology

[0002] With the deep evolution of the Internet of Everything and the Industrial Internet, edge computing has become a key foundation for supporting real-time and intelligent applications. Embedded systems and application-specific integrated circuit chips, as the core hardware of edge nodes, determine the robustness of edge sensing networks through their processing efficiency, real-time performance, and power-to-computing ratio. Reconfigurable computing technology dynamically adjusts the hardware interconnection structure through programmable logic arrays, balancing flexibility and high performance, and providing customized computing power support for the edge side.

[0003] Existing chip scheduling solutions for edge computing rely on the operating system kernel scheduler or static resource allocation strategies, and use priority preemptive scheduling combined with DVFS technology to reduce power consumption. In reconfigurable scenarios, reconfigurable resources are abstracted into hardware acceleration operator libraries, and logical states are switched by calling configuration files. In scenarios with fixed computing modes and small load fluctuations, processing efficiency can be improved and power consumption overflow problems can be alleviated.

[0004] As edge scenarios evolve towards ultra-low power consumption, high dynamism, and heterogeneous algorithms, existing scheduling frameworks suffer from deep-seated contradictions. They neglect the time and energy overhead of hardware reconfiguration, and the disconnect between software and hardware leads to an inability to balance resource load. They also struggle to resolve the conflict between reconfiguration power consumption spikes and the steady-state power supply of edge devices. Furthermore, they have limitations such as single perception and insufficient resource orchestration granularity. The core challenge is to achieve elastic and efficient scheduling of reconfigurable resources while balancing extremely low static power consumption with system determinism and long standby capability.

[0005] Therefore, the present invention provides a low-power reconfigurable electronic component chip scheduling system for edge computing. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: A low-power reconfigurable electronic component chip scheduling system for edge computing, comprising a task semantic perception and feature extraction module, a reconfigurable resource status real-time monitoring module, a power consumption trajectory prediction and voltage steady-state compensation module, a dynamic reconfiguration granularity control execution module, and a multi-dimensional constraint integrated scheduling core module. The system is connected to the underlying reconfigurable electronic component logic array via a physical bus and logic control link, realizing dynamic reorganization of hardware logic functions and task mapping.

[0008] The task semantic awareness and feature extraction module is located at the input front end of the system and consists of a task message parser, a computational intensity evaluator, a data dependency analysis unit, and a real-time level identifier. The task message parser is responsible for receiving heterogeneous task streams from the edge computing network layer and extracting the instruction set features, expected operand size, and target algorithm type of the task by deeply decomposing the metadata in the task packet header. The computational intensity evaluator quantifies and evaluates the number of multiply-accumulate operations, logic gate toggling frequency, and memory access bandwidth required for task execution based on the parsed algorithm type. The data dependency analysis unit identifies the parallelization potential and critical paths that must be executed serially by constructing an operand flow graph within the task. The real-time level identifier classifies the task into three levels—hard real-time, soft real-time, and non-real-time—based on the task's deadline requirements and generates a task semantic feature vector accordingly, which serves as the input parameter for subsequent scheduling strategies.

[0009] The real-time monitoring module for reconfigurable resource status is located at the boundary between the physical layer and the abstraction layer of the chip. It includes a configuration bitstream status register group, a logic unit hotspot monitoring matrix, an on-chip network congestion sensor, and a power management unit feedback interface. The configuration bitstream status register group records in real time the distribution of the loaded logic function modules in the current reconfigurable electronic component logic array, including the occupancy status of each logic partition, the currently executed logic function ID, and the remaining space of the configuration memory. The logic unit hotspot monitoring matrix consists of multiple miniature temperature sensors distributed inside the logic array, used to capture local temperature rise data caused by high-frequency reconfiguration or high-load operation in real time. The on-chip network congestion sensor monitors the occupancy rate of interconnect resources inside the logic array to ensure that bitstream transmission during the reconfiguration process does not seriously conflict with the task data stream. The power management unit feedback interface obtains the current core voltage, current intensity, and inductor storage energy status in real time.

[0010] The power consumption trajectory prediction and voltage steady-state compensation module is the core component for achieving low-power scheduling. It consists of a configuration transition Hamming distance calculator, an instantaneous current model library, a voltage drop compensation controller, and a clock frequency pre-regulator. The configuration transition Hamming distance calculator calculates the total number of logic switch switching, i.e., the Hamming distance, between the current logic array's configuration bitstream and the target reconfiguration scheme's configuration bitstream, serving as the basis for dynamic power consumption prediction during reconfiguration. The instantaneous current model library stores typical current curves of different logic function modules during startup, operation, and reconfiguration switching. Based on the predicted current spikes, the voltage drop compensation controller pre-compensates for voltage fluctuations within a preset clock cycle before the reconfiguration action by adjusting the output gain of the on-chip low-dropout linear regulator or by prematurely activating the decoupling capacitor array. The clock frequency pre-regulator appropriately reduces the operating frequency of the affected area during reconfiguration to reduce dynamic switching power consumption and ensure system power supply stability.

[0011] The dynamic reconfiguration granularity control execution module is responsible for specific hardware configuration changes. It includes a configuration bitstream compression and decompression engine, a multi-channel direct memory access controller, a reconfiguration mask generator, and a logic isolation gate control unit. The configuration bitstream compression and decompression engine employs a hybrid compression technique based on run-length encoding and Huffman coding to reduce the amount of data in the bitstream during transmission and lower transmission power consumption. The multi-channel direct memory access controller supports parallel loading of multiple configuration bitstreams, significantly shortening the reconfiguration latency. The reconfiguration mask generator allows the system to update the configuration of only the regions in the logic array where functional changes have occurred, while keeping the states of other regions unchanged, thereby achieving local fine-grained reconfiguration. The logic isolation gate control unit physically isolates the target region from the global logic interconnection network during the reconfiguration process to prevent transient random levels during reconfiguration from interfering with the normal operation of the logic functions.

[0012] The multi-dimensional constraint integrated scheduling core module serves as the system's decision-making brain, coordinating the work of the aforementioned modules based on multi-objective optimization logic. This module integrates a resource topology mapper, a power consumption-latency cost evaluation engine, and an adaptive scheduling strategy generator. The resource topology mapper searches for the optimal physical placement location in the logic array based on the task semantic feature vector to minimize the interconnection path length between logic units. The power consumption-latency cost evaluation engine comprehensively considers reconfiguration time overhead, reconfiguration energy loss, task execution time, and static leakage current power consumption to calculate the comprehensive energy efficiency score for each candidate scheduling scheme. The adaptive scheduling strategy generator dynamically selects scheduling decisions based on the current power state, temperature rise limits, and task queue priorities: in-situ waiting, logic reorganization execution, or migration to the general processing core for execution.

[0013] Preferably, the reconfigurable electronic component logic array adopts a configuration storage architecture composed of a hybrid of non-volatile memory cells and SRAM cells; Among them, the non-volatile memory unit is used to store the core logic configuration operators used frequently to achieve non-loss of data and zero-power standby; the SRAM unit is used to store temporary reconfiguration bit streams for specific burst tasks to ensure high-speed configuration switching.

[0014] Preferably, the task semantic perception and feature extraction module employs a hardware-assisted fast feature extraction mechanism when processing task flows. This mechanism sets up a dedicated set of feature extraction hardware logic at the chip's enqueue interface to perform parallel pattern matching on each data packet entering the system. This hardware logic includes a set of programmable mask registers and comparator arrays, which can realize real-time identification of the operator types (such as convolution operations, fast Fourier transforms, encryption and decryption algorithms, etc.) inside the data packets without occupying general-purpose processor cycles.

[0015] Preferably, when executing the prediction logic, the power consumption trajectory prediction and voltage steady-state compensation module establishes a dynamic model based on logic flip-flop rate prediction. This model not only considers the static differences in the configuration bit stream, but also predicts the effective flip-flop coefficient of the logic gate during operation by analyzing the data activity of the task to be executed. The voltage drop compensation controller includes a multi-stage charge pump structure. When the instantaneous current is predicted to exceed a preset threshold, the charge pump can inject compensation charge into the power supply rail inside the chip within nanoseconds, thereby controlling the voltage fluctuation range within five percent of the rated voltage.

[0016] Preferably, the dynamic reconfiguration granularity control execution module supports a logical management technology called virtual partitioning; this technology logically divides a physically continuous reconfigurable logical array into multiple virtual partitions of adjustable size and offset position; the reconfiguration mask generator automatically generates configuration bitstream write enable signals based on the boundary information of the virtual partitions, so that the reconfiguration action only acts within the selected virtual partitions; this technology effectively solves the resource fragmentation problem caused by traditional fixed partitioning schemes and improves hardware utilization.

[0017] Preferably, the multi-dimensional constraint integrated scheduling core module introduces a pre-reconfiguration mechanism based on load prediction when generating scheduling strategies. When the adaptive scheduling strategy generator finds that the probability of a certain algorithm task is increasing by analyzing historical task sequences, the system will utilize the current computing idle period or low-power operation period to preload the corresponding logical configuration bit stream into the spare configuration buffer or inactive logical partition. When the actual task arrives, the system only needs to activate the function through clock strobing or a very small number of interconnection switches, thereby minimizing the task reconstruction waiting latency.

[0018] Preferably, the logic unit hotspot monitoring matrix in the real-time monitoring module for reconfigurable resource status adopts a differential temperature sensing architecture; each monitoring point consists of a temperature-sensing diode and a reference diode located in the chip's reference low-temperature region; by measuring the voltage difference between the two, the system can accurately eliminate common-mode interference caused by ambient temperature fluctuations, thereby truly reflecting the microscopic thermal effects inside the chip caused by local high-intensity reconfiguration, and providing data support for thermal safety constraints in scheduling strategies.

[0019] Preferably, the system further includes a global clock management unit that supports independent frequency scaling for different logical partitions. During scheduling, the multi-dimensional constraint integrated scheduling core module applies different operating frequencies to each partition through the global clock management unit based on the real-time urgency of the task. For partitions undergoing reconstruction, the system automatically switches their clock frequency to a low-frequency mode to suppress peak current. For partitions executing critical path tasks, their frequency is appropriately increased to ensure time limits are met.

[0020] The low-power reconfigurable electronic component chip scheduling system for edge computing described in this invention involves complex hardware and software co-design in its engineering implementation process. At the hardware level, the system uses advanced FinFET or FD-SOI technology to build the underlying logic array to take advantage of its excellent low power consumption characteristics. In terms of on-chip interconnection, an asynchronous cross-clock domain transmission protocol based on asynchronous FIFO is adopted to ensure the reliability of data exchange between logic partitions of different frequencies. In terms of configuration links, the system adopts a segmented configuration bus structure to avoid the bandwidth bottleneck of a single bus architecture when multiple partitions are reconstructed at the same time.

[0021] At the software level, this invention is equipped with a set of efficient component configuration driver library and resource scheduling API interface. The driver library is responsible for mapping the algorithm logic described in the high-level programming language to the binary configuration bit stream under the specific hardware architecture. The resource scheduling API interface provides a standardized calling specification for the upper-layer application, so that developers can achieve efficient utilization of reconfigurable hardware resources without having to pay attention to the underlying reconstruction details.

[0022] Preferably, the power supply of each logical partition in the system is controlled by an independent power-gated transistor; when the multi-dimensional constraint integrated scheduling core module detects that a partition has no task processing for a long time and no pre-reconfiguration requirement, it will trigger a shutdown signal to cut off the leakage current path of the partition, thereby significantly improving the battery life of edge devices under extremely low load conditions; in order to shorten the delay of waking up from deep sleep, each partition retains a set of state holding registers to store key configuration state information.

[0023] Preferably, the configuration bitstream compression engine in the dynamic reconfiguration granularity control execution module adopts a compression algorithm based on coordinate differential mapping, taking into account the physical structural characteristics of the reconfigurable components. This algorithm greatly reduces the communication load when processing local fine-tuning tasks by only recording the coordinates of the configuration storage unit whose state has changed and its new state value, rather than transmitting the entire configuration frame. On the logic array side, there is a corresponding hardware decompression unit responsible for restoring the received coordinate differential data into a complete physical configuration signal.

[0024] Preferably, when handling multi-task concurrency scenarios, the system adopts a dynamic allocation strategy based on a resource borrowing mechanism. When the logical resources required by a high-priority task exceed the size of the currently available partition, the scheduling core module will assess the urgency of tasks in adjacent partitions. If the adjacent partition is executing a low-priority or non-real-time task, the system will temporarily preempt the logical resources of the adjacent partition and integrate them into a larger super-large logical cluster through rapid reconstruction to execute the high-priority task. After the task is completed, the system will automatically release the borrowed resources and restore the original logical configuration.

[0025] Preferably, the power consumption trajectory prediction and voltage steady-state compensation module also has online learning and self-calibration functions; by comparing the actual monitored power consumption waveform with the predicted model trajectory, the system will automatically correct the parameter deviation in the internal instantaneous current model library; this closed-loop optimization mechanism enables the system to adapt to the electrical characteristic drift caused by semiconductor aging and extreme working environment, ensuring that the scheduling system can maintain high-precision power consumption control throughout the entire life cycle of the chip.

[0026] Preferably, the multi-dimensional constraint integrated scheduling core module also includes a security verification unit; before performing any reconfiguration action, the security verification unit performs digital signature verification and logical conflict check on the target configuration bit stream to ensure that the reconfigured circuit structure will not produce physical logic errors such as power supply to ground short circuit or signal race hazards, thus ensuring the reliability and security of the scheduling behavior from the hardware level.

[0027] In terms of physical implementation, the system described in this invention allows each functional module to communicate with each other through a hierarchically designed system bus; the task semantic perception module is located within the I / O subsystem and interfaces with external communication modules through a high-speed serial interface; the multi-dimensional constraint integrated scheduling core module serves as the master control node, sending configuration commands to downstream monitoring, prediction, and execution modules through a dedicated control bus; all modules are integrated into a single system-on-a-chip, and through optimized layout and routing, analog compensation circuits and digital logic circuits are reasonably isolated to reduce the impact of electromagnetic interference on prediction accuracy.

[0028] On the data flow path, the raw data stream collected by the edge side first enters the task semantic perception module for splitting and feature labeling; the labeled tasks are pushed into the task priority queue managed by the scheduling core module; the scheduling core module calls the cost evaluation engine to select the optimal path based on the hardware heat, voltage margin and configuration status fed back by the real-time monitoring module; if the selected path involves hardware reconstruction, the power consumption prediction module is notified to start voltage steady-state compensation, and then the instruction reconstruction execution module extracts the bit stream from the configuration storage array for loading; after the logical function is activated, the task data stream is directly imported into the reconstructed logical array through the DMA controller for accelerated processing, and the processing result is fed back to the output buffer through the on-chip network.

[0029] Through the aforementioned system concept and component collaboration, this invention realizes a highly intelligent and deterministic reconfigurable chip scheduling paradigm. It solves the physical stability problem during hardware reconfiguration through predictive power management, improves the flexibility of hardware resource utilization through fine-grained reconfiguration control, and achieves the optimal balance between task execution performance and system power consumption in resource-constrained edge computing environments through integrated scheduling with multi-dimensional constraints. This system architecture is not only applicable to current edge-side industrial IoT gateways and intelligent monitoring terminals, but also provides a scalable and portable scheduling management framework for the future development of heterogeneous integrated circuits.

[0030] The superiority of the system described in this invention lies in its ability to overcome the traditional cognitive limitations of reconstruction as energy consumption and latency, and to incorporate the dynamic process of hardware reconstruction into a predictable and controllable closed-loop management system. It not only achieves task-level logical matching, but also physical-level energy efficiency matching, which greatly improves the robustness of edge computing devices in dealing with complex and ever-changing algorithm tasks.

[0031] Preferably, the system of the present invention optimizes the internal structure of the reconfigurable logic units in terms of hardware implementation details. Each logic unit includes a four-input lookup table (LUT4), a D flip-flop with synchronous set / reset function, and a dedicated adder logic for fast carry chain processing. This balanced design ensures that the logic units can efficiently implement complex combinational logic while also meeting the clock stability requirements of sequential logic. A three-layer interconnect architecture is adopted for the interconnection structure between logic units: the first layer is a local fast interconnect responsible for communication between units within the same cluster; the second layer is a global long-line interconnect in the horizontal and vertical directions responsible for signal transmission between long-distance partitions; and the third layer is a dedicated configuration bus used to carry the reconfigurable bit stream. This hierarchical interconnect design ensures reconfiguration flexibility while minimizing parasitic capacitance generated by interconnects, thereby reducing signal switching power consumption.

[0032] The system's configuration memory employs a double-buffered synchronization mechanism. While the logic array is currently executing a task, the configuration management unit can asynchronously load the bit stream required for the next stage into the shadow register. When the scheduling core issues a switching instruction, a global switching pulse allows all logic units to update their configuration status from the current value to the shadow register value within a single clock cycle. This background loading and instantaneous switching technology reduces the reconstruction pauses of several milliseconds or even hundreds of milliseconds in traditional solutions to the nanosecond level, fundamentally solving the technical bottleneck of reconstruction overhead interfering with real-time scheduling.

[0033] When dealing with sudden bursts of high-concurrency task flows common in edge computing, the system described in this invention exhibits excellent resilience. When a surge in task traffic causes system power consumption to reach a preset threshold, the multi-dimensional constraint integrated scheduling core module automatically activates an energy efficiency priority mode. In this mode, the system proactively reduces the frequency of tasks on non-critical paths and prioritizes the reuse of ready logical partitions, reducing unnecessary refactoring actions. The power consumption trajectory prediction module dynamically increases the pre-charge voltage of the compensation capacitor to cope with potentially more severe current fluctuations. This adaptive scheduling logic enables edge nodes to maintain logical correctness and stable operation even under high load pressure.

[0034] The system bus of the system described in this invention adopts a variant based on the AXI4 protocol, which has been simplified for low power consumption requirements. It removes unnecessary burst transmission handshake signals and adds dedicated power control pins, allowing the scheduling core to directly send hardware signals to each slave device to enter low power mode via the bus protocol. The bus integrates a data integrity verification unit, which performs CRC verification on each set of configuration bit streams transmitted to prevent logical failures caused by bit stream errors due to electromagnetic pulses in edge environments.

[0035] Preferably, as an extension of the system, the scheduling system also reserves a heterogeneous processor interface for handling some extremely fragmented control flow tasks with overly complex logical branches that are unsuitable for running on the reconfigurable array. The multi-dimensional constraint integrated scheduling core module can automatically determine and execute a seamless switch between hardware acceleration and software processing based on the semantic characteristics of the task. When a task flow is determined to be control-intensive rather than data-intensive, the scheduling core will close the relevant reconfigurable partitions and route the task to the integrated processor core for execution, thereby leveraging the processor's advantages in complex branch prediction and avoiding the efficiency loss caused by frequent logical reorganization of the reconfigurable array.

[0036] The system also includes a complete rollback and recovery mechanism for anomalies. During the reconstruction process, if the real-time monitoring module detects abnormal current fluctuations or voltage drops exceeding the safety limit, the system will immediately terminate the current reconstruction action and use the previous stable configuration snapshot stored in the state holding register to quickly roll back the hardware logic to a safe state, while reporting the reconstruction failure code to the system. This fault-tolerant mechanism ensures that edge devices still have basic self-survival and task assurance capabilities under extremely unstable environmental conditions.

[0037] The present invention discloses a low-power reconfigurable electronic component chip scheduling system for edge computing. Through an integrated and innovative design from the physical layer, logic layer to the system layer, it achieves efficient management of core hardware resources for edge computing. It not only significantly reduces the total static and dynamic power consumption of the chip, but also exhibits near-deterministic low latency performance when handling complex algorithm switching, laying a solid technical foundation for building a high-performance, highly reliable, and long-lasting edge intelligent sensing network.

[0038] In the actual chip layout design, the system has performed specific integration optimization on the inductors and capacitors of the analog compensation module; it adopts on-chip deep trench capacitor technology to achieve high-density energy storage in a very small silicon area, thereby meeting the voltage steady-state compensation module's requirement for instantaneous high current; in order to reduce the substrate coupling noise of digital logic switching on the analog sensing circuit, the system adopts a dual ring protection structure for physical isolation.

[0039] In terms of the implementation of the software protocol stack, the system described in this invention supports task preemption and context switching. When a higher-priority hard real-time task arrives, the scheduling core instructs the execution module to quickly save the current partition state and perform priority preemption reconstruction. The intermediate computation data of the preempted task will be temporarily stored in a dedicated context SRAM block. After the high-priority task is completed, the system restores the hardware state through recovery reconstruction and continues to execute the original task. This context switching mechanism based on hardware reconstruction enables the reconfigurable chip on the edge side to have the multi-task concurrent processing capability of a general-purpose processor.

[0040] The beneficial effects of this invention are as follows: 1. The low-power reconfigurable electronic component chip scheduling system for edge computing described in this invention, through the setting of a power consumption trajectory prediction and voltage steady-state compensation module, can accurately predict power consumption spikes before the reconfiguration action occurs, based on the Hamming distance of the configured bit current and the instantaneous current model library; in conjunction with a voltage drop compensation controller and a clock frequency pre-regulator, it can inject compensation charge or dynamically reduce the frequency to the power supply rail within nanoseconds, controlling the voltage fluctuation range within 5% of the rated voltage. This ensures that the power rail of the edge device remains stable during frequent dynamic reconfiguration, avoiding system crashes or logic errors caused by sudden voltage drops.

[0041] 2. The low-power reconfigurable electronic component chip scheduling system for edge computing described in this invention introduces a compression algorithm based on coordinate differential mapping and a reconfiguration mask generator through a dynamically configured reconfiguration granularity control execution module. This allows the system to precisely update only local areas (virtual partitions) in the logic array where functional changes occur, rather than configuring the entire system. Combined with a double-buffered synchronization mechanism, it achieves background loading and instant switching, reducing the reconfiguration pause of several milliseconds in traditional solutions to the nanosecond level. This not only significantly improves the utilization efficiency of hardware resources but also enables reconfigurable chips to have real-time response capabilities to cope with the highly dynamic task flow of edge computing.

[0042] 3. The low-power reconfigurable electronic component chip scheduling system for edge computing described in this invention extracts computational intensity, data dependency, and real-time level from the task flow entry point through a task semantic awareness module, generating refined feature vectors. The multi-dimensional constraint integrated scheduling core module acts as the decision-making brain, comprehensively considering real-time hardware heat, voltage margin, and reconfiguration costs, and introducing a pre-reconfiguration mechanism and resource borrowing mechanism based on load prediction. This closed-loop management can dynamically select the optimal execution path (such as waiting in place, logic reorganization, or migration to a general-purpose core). While meeting task time limits, it minimizes static leakage current and dynamic power consumption through power gating and independent frequency scaling, significantly improving the energy efficiency ratio and long standby capability of edge devices.

[0043] 4. The low-power reconfigurable electronic component chip scheduling system for edge computing described in this invention physically isolates the target area during reconfiguration by setting a logic-isolated gate control unit to prevent transient interference; the abnormal rollback and recovery mechanism can quickly restore to a safe configuration snapshot when an abnormal current is detected, ensuring system survivability; the power trajectory prediction module has online learning and self-calibration functions, and can correct model parameters according to the actual monitored waveform to compensate for characteristic changes caused by semiconductor aging and environmental drift, ensuring that the scheduling system maintains high-precision power control throughout its entire life cycle. Attached Figure Description

[0044] The invention will now be further described with reference to the accompanying drawings.

[0045] Figure 1 This is a structural block diagram of a low-power reconfigurable electronic component chip scheduling system for edge computing in this invention. Detailed Implementation

[0046] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0047] like Figure 1 As shown in the embodiment of the present invention, a low-power reconfigurable electronic component chip scheduling system for edge computing includes a task semantic perception and feature extraction module that undertakes the responsibility of deep analysis and preprocessing of heterogeneous task flows. Specifically, the task message parser in this module uses a set of hardware state machines with high-speed sampling capabilities to disassemble AXI4-Stream bus or high-speed serial interface data packets entering the system in real time. The task message parser integrates dedicated message header recognition logic, which can automatically locate key fields in the task descriptor, including but not limited to instruction set identifier, expected operator type, and starting address of data block. The computational intensity evaluator connected to it uses a pre-built hardware feature library to quantize and map the parsed algorithm type; When a task is identified as a convolution operation, the computational intensity estimator estimates the total amount of multiply-accumulate operations required for the task and the resulting on-chip network bandwidth requirements based on the size of the convolution kernel, stride, and number of input channels by flipping the probability lookup table of the internal logic gates. The data dependency analysis unit maintains a dynamic task flow tracking array to identify the precedence and succession relationships between subtasks, thereby determining which operations can be mapped to different physical partitions of the reconfigurable logical array for parallel execution and which must follow a strict timing sequence. The real-time level identifier then attaches a time constraint label to the task based on the priority field in the task message, ensuring that the subsequent scheduling core can prioritize hard real-time tasks with strict deadline requirements.

[0048] To achieve precise control over the underlying hardware status, a real-time monitoring module for reconfigurable resource status is deployed on the physical critical path of the logic array. The configuration bitstream status register group adopts a bit-mapped storage structure, where each bit corresponds to the occupancy status of a minimum reconfigurable region in the logic array, enabling the scheduling system to know the current resource distribution map within a single clock cycle. The logic unit hotspot monitoring matrix consists of miniature temperature-sensing diodes distributed inside the logic array and arranged in a cross-point layout; These diodes are configured in a differential sensing architecture, with each monitoring point containing a temperature-sensing tube exposed in the computing core area and a reference tube located in the low-temperature region at the edge of the chip. Through a high-precision successive approximation analog-to-digital converter, the system can acquire real-time microscopic temperature gradient data within the chip, thereby effectively identifying localized hotspots generated by prolonged high-load operation. The on-chip network congestion sensor monitors the real-time traffic of interconnect resources between logical partitions. When the buffer occupancy rate of an interconnect node exceeds 80%, the sensor will trigger a congestion signal, prompting the scheduling core to adjust the subsequent resource mapping scheme to avoid the congested path. The power management unit feedback interface obtains the current core voltage fluctuations and current intensity at a nanosecond-level sampling frequency, providing the most accurate physical layer data for subsequent power consumption compensation.

[0049] The power consumption trajectory prediction and voltage steady-state compensation module is the key support for achieving low-power stable operation in this invention. After the reconstruction command is issued, the configuration transition Hamming distance calculator compares the currently running configuration bitstream data and the target reconstruction bitstream data bit by bit through hardware logic, and calculates the total number of bits where the logic level has flipped between the two. Since logic gate switching is the primary source of dynamic power consumption, this Hamming distance data was directly input into the instantaneous current model library. This model library can predict the slope of the current change during the instantaneous reconfiguration switching by storing the current characteristic curves of different logic function blocks at typical operating frequencies. Based on the predicted current spikes, the voltage sag compensation controller injects compensation charge into the power supply rails of the affected area through a voltage injection unit consisting of a multi-stage charge pump ten clock cycles before the actual reconfiguration operation occurs. This advanced compensation mechanism can effectively suppress the inductive voltage drop caused by sudden current changes, ensuring that the core voltage remains stable at more than 95% of the rated voltage. The clock frequency pre-regulator dynamically reduces the clock tree drive strength of the target reconstruction region based on the predicted power consumption pressure, thereby reducing transient power consumption overhead by lowering the frequency.

[0050] At the specific execution level of the refactoring, the dynamic refactoring granularity control execution module introduces a fine-grained management mechanism; The configuration bitstream compression and decompression engine employs a hybrid compression algorithm based on coordinate differential mapping. In practice, the system does not transmit the complete configuration frame, but only extracts the row and column coordinates and the new configuration bit values ​​of the configuration storage unit whose state has changed. This compression method, designed for localized fine-tuning, significantly reduces the amount of configuration data transmitted. The multi-channel direct memory access controller supports parallel transmission of four or more configuration streams, ensuring rapid loading of large-capacity logic modules. The reconstruction mask generator is the core of local reconstruction. Based on the reconstruction region boundary determined by the scheduling core, it automatically generates a set of write enable mask signals, so that the reconstruction bit stream can only be written to the selected physical partition, while the adjacent running task regions are logically masked and are not affected by the configuration bus flip. During the reconfiguration process, the logic isolation gate control unit temporarily pulls the output of the target partition low or keeps it in a high impedance state to prevent random glitches during the reconfiguration process from penetrating into the global interconnect network and causing logic errors.

[0051] The multi-dimensional constraint integrated scheduling core module serves as the system's global decision-making center, responsible for finding the optimal scheduling scheme under complex constraints. The resource topology mapper not only considers the number of logical units required, but also finds the free clusters in the physical array that are closest to the input / output ports and have the shortest interconnections by analyzing the data communication overhead between tasks. The power consumption-latency cost assessment engine establishes a multi-objective evaluation system, whose input parameters include the predicted reconfiguration energy loss, the expected task execution time, and the additional latency caused by partition switching. The engine calculates a comprehensive energy efficiency score for each possible scheduling path; The adaptive scheduling strategy generator combines the heat information and task priority feedback from the real-time monitoring module to dynamically decide whether to immediately refactor, reuse the existing logic configuration, or redirect the task to the low-power RISC-V processing core integrated inside the chip.

[0052] The reconfigurable electronic component logic array adopts a heterogeneous integrated memory architecture; Non-volatile memory cells (such as random access memory based on gyromagnetic torque mechanism) are embedded in the configuration storage nodes of the logic cells to store core logic that is used frequently, such as encryption algorithms and basic communication protocols. It allows the system to maintain its core logic state and achieve instant wake-up when entering a deep sleep mode with extremely low power consumption; SRAM cells are used to store temporary acceleration operators for specific application scenarios, leveraging their extremely high read and write speeds to achieve millisecond-level fast function switching.

[0053] Furthermore, the system includes a global clock management unit that supports independent frequency scaling for different logical partitions; Each logical partition has its own independent digital phase-locked loop and clock gating logic; When performing high-throughput tasks such as image processing, the scheduling core instructs the clock management unit to increase the frequency of the corresponding partition; while for low-frequency tasks such as sensor signal monitoring, the frequency is reduced to the megahertz level to achieve optimal energy efficiency. The security verification unit adds a digital signature verification step to the reconstructing link. Only bit streams that pass the verification are allowed to enter the loading stage, preventing malicious configuration code from damaging the chip's physical structure at the hardware source.

[0054] Example 1: This embodiment employs a low-power reconfigurable electronic component chip scheduling system for edge computing as described in this invention; The chip is manufactured using a 14nm FinFET process, and its logic array contains 50k reconfigurable logic cells, divided into 16 virtual partitions. The system integrates a 256-layer instantaneous current model library; In the test, the input stream was a heterogeneous task sequence that combined AES encryption, FFT operation and CNN convolution operator, and the task arrival rate followed a Poisson distribution.

[0055] Comparative Example 1: The traditional static resource pre-allocation scheme is adopted, in which each logical partition is fixedly allocated to a specific algorithm module. Dynamic reconstruction is not supported. When the task type changes, a global reset and reload of all bit streams are required.

[0056] Comparative Example 2: It adopts a conventional dynamic reconfiguration scheduling scheme, which has basic task mapping capabilities, but does not include the power consumption trajectory prediction, voltage steady-state compensation, and Hamming distance-based prediction logic described in this invention. The reconfiguration granularity is a coarse-grained fixed partition.

[0057] For the above embodiments and comparative examples, a continuous 24-hour operation test was conducted under the same task load, and the obtained quantitative comparison data is shown in Table 1.

[0058] Table 1: Comparison of Performance Indicators between Embodiments and Comparative Examples of the Invention Test metrics Comparative Example 1 (Static Allocation) Comparative Example 2 (Conventional Dynamic Scheduling) Example 1 (System of the Invention) Average power consumption during task execution (mW) 485.6 3.4 185.2 Reconstructing instantaneous peak current (mA) 10.5 980.2 315.6 Average reconfiguration handover delay (μs) 5500.0 850.5 42.8 Maximum voltage drop (%) of power supply rail 12.5% 8.2% 1.8% Hardware resource utilization rate (%) 28.4% 62.5% 91.2% Task deadline violation rate (%) 15.6% 5.2% 0.3% Local maximum temperature rise (°C) 45.2 32.5 12.8 As can be clearly observed from the data in Table 1, Example 1 demonstrates significant superiority in multiple core indicators; In terms of power consumption control, the system of the present invention avoids invalid logic flips through accurate task semantic perception and feature extraction, and its average power consumption is only about 38% of that of the static allocation scheme. More importantly, thanks to the intervention of the power consumption trajectory prediction and voltage steady-state compensation module, the instantaneous peak current of reconstruction in Example 1 was significantly suppressed to 315.6mA, which is about 67% lower than that in Comparative Example 2. This directly resulted in the voltage drop of the power supply rail decreasing from 8.2% to 1.8%, greatly improving the physical stability of the chip during the reconstruction process.

[0059] Regarding real-time performance, this invention reduces the average reconstruction switching latency from 850.5 microseconds to 42.8 microseconds by configuring a bitstream compression engine and a multi-channel direct memory access controller. This order-of-magnitude improvement enables the system to handle the extremely high frequency of algorithm switching requirements in edge computing. Thanks to the introduction of virtual partitioning technology and resource topology mapper, the overall utilization rate of hardware resources reached 91.2%, effectively solving the problem of resource fragmentation. The significant reduction in the task deadline violation rate (from 5.2 percent to 0.3 percent) demonstrates the determinism and efficiency of the adaptive scheduling strategy in handling hard real-time tasks. The combined application of differential temperature sensing architecture and hotspot monitoring matrix effectively controls local temperature rise. The temperature rise in Example 1 is only 12.8 degrees Celsius, which is much lower than other solutions. This is of great engineering significance for extending the service life of edge devices in closed or harsh environments.

[0060] Specifically, when handling concurrent tasks of image recognition and encrypted transmission, the operation flow of the system of the present invention is as follows: The task semantic perception module first identifies the incoming image and marks it as a high bandwidth requirement. The scheduling core then calls the resource topology mapper and allocates logical operators in the partitions close to the video interface. When subsequent encryption tasks arrive, the adaptive scheduling strategy generator finds that the current logical resources are full, but learns through the monitoring module that the image processing task is about to end, so it starts the pre-reconstruction mechanism to preload the encrypted bit stream into the shadow register; At the moment the image task is completed, the logical partition is switched to encryption mode within nanoseconds through a global switching pulse. At the same time, the voltage compensation controller is activated to prevent voltage fluctuations caused by the high-frequency switching of the encryption logic. The entire process is transparent to upper-layer applications, demonstrating a high degree of intelligence.

[0061] In terms of physical layout design, this invention particularly strengthens the electromagnetic isolation between analog and digital circuits. The charge pump capacitors in the voltage sag compensation controller employ deep trench capacitor technology; these capacitors are deeply integrated into the silicon substrate and isolated from the digital switch logic through a ring protection structure. It saves expensive packaging area and utilizes the electrical characteristics of silicon wafers to achieve energy release with ultra-low equivalent series resistance; In the on-chip interconnect network wiring, a three-layer shielding routing technique is used to spatially separate the configuration bus and the task data bus, and to insert grounded shielding wires to prevent the high-frequency flipping of the reconstructed bit stream from causing electromagnetic interference to the task data being transmitted.

[0062] In the implementation of the software protocol stack, the resource scheduling API of this invention supports asynchronous task submission and callback mechanisms; Developers call hardware functions through a standardized operator library, while the underlying reconfiguration timing, voltage compensation intensity, and frequency scaling ratio are entirely determined autonomously by the hardware scheduling system. The system also has an online self-calibration function. The power consumption prediction model will automatically correct the weighting coefficients in the internal instantaneous current model library based on the actual monitored current waveform. For example, when a chip experiences semiconductor aging due to prolonged operation, leading to an increase in leakage current, the monitoring module will capture this trend and feed it back to the scheduling core. This will automatically increase the weight of static power consumption in subsequent power consumption cost assessments, guiding the system to select a more energy-efficient operating strategy.

[0063] The system described in this invention demonstrates excellent flexibility in resource borrowing mechanisms under multi-task concurrency. For example, when a sudden vibration analysis task (high priority) requires more resources than the capacity of a single partition, the scheduling core will quickly merge two adjacent partitions that are executing low-priority environmental monitoring tasks through the logic isolation gate control unit, and virtualize the three physical partitions into a super-large logic cluster by reconstructing the mask generator to support complex vibration analysis algorithms. After the task is completed, the system uses the snapshot stored in the status holding register to quickly restore the original environmental monitoring logic configuration; This hardware-capability-based virtualization management enables reconfigurable chips at the edge to achieve unprecedented flexibility in task processing.

[0064] During the multiply-accumulate operation of convolution, these dedicated hardware paths significantly reduce the number of general-purpose logic gate flips, thereby reducing dynamic power consumption at the source. Each logic partition edge is equipped with an independent power-gated transistor, directly controlled by the multi-dimensional constraint integrated scheduling core module. When a partition is determined to have no tasks and no pre-reconfiguration requirements within a predicted period of time, the scheduling core will execute a hard shutdown instruction to completely cut off the power supply path of the partition and reduce the static leakage current to the picoampere level. When a new task arrives and needs to wake up the partition, the voltage steady-state compensation module will intervene first, controlling the power gate transistor to turn on slowly in a stepped manner to avoid the surge current from impacting the system power supply.

[0065] In summary, the low-power reconfigurable electronic component chip scheduling system for edge computing described in this invention constructs a logically rigorous, feedback-closed-loop, and high-performance hardware scheduling system through a series of technological innovations in the task perception layer, resource monitoring layer, power management layer, and scheduling execution layer. It not only theoretically breaks through the bottleneck of improving the energy efficiency ratio of reconfigurable chips, but also provides a reproducible and quantifiable low-power solution in engineering practice. Verification through measured data demonstrates the significant technical gains of this invention in reducing power consumption, reducing latency, improving stability, and increasing resource utilization, thus opening up a new technical path for the development of heterogeneous acceleration chips in the field of edge computing.

[0066] The system of this invention has an extremely wide range of applications. It can not only meet the requirements of intelligent industrial IoT gateways for real-time conversion of multiple protocols, but also adapt to the rigid needs of intelligent monitoring terminals for dynamic switching of various complex algorithms. Its highly reliable fault-tolerant mechanism and anomaly rollback strategy ensure that the system can still maintain the continuity of business logic in extremely harsh edge physical environments. This integrated design scheme from the physical layer to the system architecture represents the cutting-edge development direction of the deep integration of reconfigurable computing technology and edge computing, and has extremely high commercial value and technical promotion significance.

[0067] In future expansion implementations, this scheduling system can further introduce environmental sensing parameters such as radiation intensity and vibration frequency by increasing the input dimensions of the multi-dimensional constraint integrated scheduling core module. This enables more precise hardware scheduling in aerospace or special industrial environments. The modular design of the system allows for rapid portability to different process nodes. Whether in advanced 5nm or mature 28nm processes, it can maintain its excellent energy efficiency control characteristics by adjusting the model library parameters. This invention fully discloses all the necessary technical details for implementing this efficient scheduling system through detailed circuit structure description, rigorous logic flow construction, and real data comparison, ensuring that those skilled in the art can achieve stable and reliable engineering mass production based on this.

[0068] Furthermore, when handling large-scale concurrent tasks, the adaptive scheduling strategy generator inside the multi-dimensional constraint integrated scheduling core module adopts a predictive reconstruction algorithm based on load trends. When the system detects that the frequency of a certain type of operator (such as FFT) in the task queue is increasing within the sliding window, even if there are no idle logical resources, the scheduling core will prepare the corresponding bit stream address pointer in the non-volatile storage unit in advance and instruct the direct storage access controller to prefetch some configuration data to the local cache. Once a logical partition finishes the current task and releases resources, the system will immediately skip the regular task parsing and feature extraction stages and directly enter the reconstruction execution stage. This predictive scheduling mechanism further reduces the overall response latency of the task by more than 15%.

[0069] In the voltage compensation logic at the physical level, the voltage drop compensation controller not only supports static voltage injection, but also has dynamic fine-tuning capability based on negative feedback. During the microseconds of the reconfiguration process, the real-time voltage value fed back by the monitoring module is compared with the reference at high speed. If it is found that the pre-compensation amount is insufficient to completely offset the inductive drop, the charge pump will automatically increase the switching frequency and increase the energy injection rate. This dual closed-loop (pre-compensation and real-time feedback) power supply guarantee mechanism enables the system to maintain the signal integrity of the system bus when dealing with extremely complex reconfiguration sequences (such as multiple large-size partitions being reconfigured at the same time).

[0070] The system described in this invention also fully considers compatibility with existing industry standards in its interface design. Its task semantic perception module supports hardware-level parsing of model description files exported from common neural network frameworks. Through the supporting component configuration driver library, the upper-layer software can achieve precise control of the underlying reconstruction logic through simple function calls. This hardware and software co-design approach greatly reduces the threshold for developers to deploy complex algorithms on the edge.

[0071] During the configuration bitstream writing process, the hardware automatically calculates the checksum of each frame of data and compares it with the expected value in the message in real time. If a bit flip caused by environmental interference is detected, the reconstruction execution module will immediately trigger a retransmission instruction instead of waiting until the entire reconstruction is completed before performing error detection. This write-and-check mechanism reduces the error recovery latency to the nanosecond level, further enhancing the system's robustness under complex electromagnetic conditions.

[0072] In the data processing of the hotspot monitoring matrix of the logical unit, the system adopts a sliding window average filtering algorithm to eliminate transient noise that may occur from the sensor. When the predicted temperature of a certain partition is close to the preset thermal shutdown threshold, the scheduling core does not simply shut down the partition. Instead, it gradually reduces the operating frequency of the partition through the clock frequency pre-regulator and logically finds a backup partition with a lower temperature for task migration. This smooth performance degradation strategy avoids business interruption caused by hard shutdown and reflects the system's deep consideration in ensuring task continuity.

[0073] When the logic array performs high-frequency operations, applying a small periodic perturbation to the clock frequency can effectively disperse the spectral energy of electromagnetic interference, reduce the electromagnetic radiation impact of the system on the external environment, and improve the electromagnetic compatibility of the system itself. This detailed design, combined with the aforementioned physical isolation and voltage compensation scheme, jointly constructs an edge computing hard core platform with highly stable electrical characteristics.

[0074] Through the detailed description of the specific embodiments of the present invention above, it can be seen that the present invention is not a simple stacking of modules, but rather, based on a deep understanding of the physical characteristics of reconfigurable chips in edge computing scenarios, it achieves a logically rigorous, data-rich, and highly engineering-feasible technical solution through cross-level collaborative optimization. The implementation of the system described in this invention will significantly improve the energy efficiency and real-time processing capabilities of existing edge computing devices, providing an important reference paradigm for the evolution of hardware architecture in the era of the intelligent Internet of Things.

[0075] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A low-power reconfigurable electronic component chip scheduling system for edge computing, characterized in that, The scheduling system is connected to the underlying reconfigurable electronic component logic array via a physical bus and logic control link. The system includes: The task semantic awareness and feature extraction module is set at the input front end of the system. It is used to receive and parse heterogeneous task streams from the edge computing network layer, extract the instruction set features, operator types and real-time levels of the tasks, and generate task semantic feature vectors. The reconfigurable resource status real-time monitoring module is deployed at the junction of the physical layer and the abstraction layer of the chip. It is used to capture in real time the logic partition occupancy status, local temperature rise data, on-chip network congestion level and power physical parameters of the logic array of the reconfigurable electronic components. The power consumption trajectory prediction and voltage steady-state compensation module is used to predict current spikes during the reconstruction process based on the flip characteristics of the configuration bit stream, and to perform voltage pre-compensation and frequency pre-adjustment within a preset clock cycle before the reconstruction action occurs, so as to suppress power rail voltage fluctuations. The dynamic reconfiguration granularity control execution module is used to compress the configuration bit stream and perform local fine-grained hardware configuration reorganization on the reconfigurable electronic component logic array according to the scheduling decision, while implementing logical physical isolation during the reconfiguration process. The multi-dimensional constraint integrated scheduling core module generates scheduling strategies based on the task semantic feature vector and the physical parameters fed back by the real-time monitoring module of reconfigurable resource status, and coordinates resource mapping, energy efficiency assessment and reconfiguration execution through multi-objective optimization logic.

2. The low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 1, characterized in that, The task semantic perception and feature extraction module includes: The task message parser uses a high-speed hardware state machine to deeply decompose the heterogeneous task flow entering the system and extract the instruction set identifier, expected operator type, and data block start address from the metadata in the task packet header. The computational intensity evaluator uses a pre-built hardware feature library to perform quantization mapping on the parsed algorithm type, and evaluates the total amount of multiply-accumulate operations required for task execution, the predicted value of logic gate flip frequency, and the on-chip network access bandwidth requirements. The data dependency analysis unit constructs an operand flow graph within a task by maintaining a dynamic task flow tracking array, thereby identifying the parallelization potential within the task and the critical path of serial execution. The real-time level identifier classifies tasks into hard real-time, soft real-time, and non-real-time levels based on the priority field in the task descriptor, and generates a multi-dimensional task semantic feature vector by combining parsing parameters.

3. The low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 1, characterized in that, The real-time monitoring module for the status of reconfigurable resources includes: Configure the bit stream status register group and use a bit-mapped storage structure to record the occupancy status of each logic partition in the reconfigurable electronic component logic array, the currently loaded function ID, and the remaining space of the configuration memory; The logic cell hotspot monitoring matrix consists of multiple differential micro temperature sensors distributed inside the logic array. Each monitoring point includes a temperature-sensing diode located in the computing core area and a reference diode located in the chip reference low temperature area. It is used to accurately capture the microscopic thermal effects caused by local high-intensity reconstruction and eliminate ambient temperature common-mode interference. The on-chip network congestion sensor is used to monitor the buffer occupancy rate of interconnect resources within the logic array in real time. When the congestion level exceeds a preset threshold, it triggers a congestion signal to guide the scheduling core to avoid the congested path. The power management unit feedback interface obtains the current core voltage fluctuation value, current intensity, and inductor energy storage status in real time.

4. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 3, characterized in that, The power consumption trajectory prediction and voltage steady-state compensation module includes: The configuration transition Hamming distance calculator is used to compare the configuration bit stream of the current logic array with the bit stream of the target reconstruction scheme before reconstruction, calculate the total number of logic switch switching between the two, and obtain the configuration transition Hamming distance. A library of instantaneous current models is used to predict the slope of the current change during the reconstruction switching instant based on the configured jump Hamming distance; The voltage drop compensation controller, based on the predicted current spike, instructs the charge pump structure integrated inside the chip to inject compensation charge into the power supply rail within a preset clock cycle before the reconfiguration action occurs. A clock frequency pre-regulator is used to dynamically reduce the operating frequency of the affected region during reconfiguration, thereby helping to maintain power supply stability by suppressing dynamic switching power consumption.

5. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 1, characterized in that, The dynamic reconfiguration granularity control execution module includes: Configure the bitstream compression and decompression engine, which adopts a compression algorithm based on coordinate differential mapping, and only extracts and transmits the coordinates of the configuration storage unit that has undergone a state change and its new state value; The multi-channel direct memory access controller supports parallel loading of multiple configuration bit streams to shorten the reconstruction delay of large-capacity logic function modules. The reconstruction mask generator generates a write enable mask based on the boundary information of the logical partition, so that the reconstruction action is applied to the selected physical area, while keeping the logical state of other non-target areas unchanged. A logic-isolated gate control unit is used to physically isolate the output of the target region from the global interconnect network during the reconfiguration process.

6. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 1, characterized in that, The multidimensional constraint integrated scheduling core module includes: The resource topology mapper searches for idle logic clusters with physical locations in the reconfigurable electronic component logic array based on the task semantic feature vector, in order to minimize the interconnection path length between logic units. The power consumption-latency cost evaluation engine comprehensively evaluates the reconfiguration time overhead, reconfiguration energy loss, expected task execution time, and static leakage current power consumption to calculate the comprehensive energy efficiency score for candidate scheduling schemes. An adaptive scheduling strategy generator dynamically selects whether to execute tasks by waiting in place, reorganizing logic, pre-restructuring, or migrating them to a heterogeneous processor core integrated within the chip, based on current hardware activity, power margin, and task priority.

7. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 5, characterized in that, The multi-dimensional constraint integrated scheduling core module supports a resource borrowing mechanism when handling multi-task concurrency scenarios: When the logical resources required by a high-priority task exceed the capacity of the currently available partition, the scheduling core module evaluates the task urgency of adjacent partitions. If the adjacent partition is executing a low-priority or non-real-time task, the adjacent partition resources are temporarily preempted by the logical isolation gate control unit, and multiple physical partitions are integrated into a logical cluster to execute high-priority tasks through rapid reconfiguration. Once the task is completed, the system automatically releases the borrowed resources and restores the original logical configuration of adjacent partitions using the snapshot stored in the status register.

8. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 1, characterized in that, The reconfigurable electronic component logic array adopts a configuration memory architecture composed of a hybrid of non-volatile memory cells and SRAM cells: Non-volatile memory cells are embedded in the configuration storage nodes of the logic units to store core logic configuration operators used frequently, in order to support zero-power standby and instant wake-up; The SRAM cell is used to store temporary reconfigured bit streams for specific burst tasks, and combines a double-buffered synchronization mechanism with a shadow register to realize background bit stream loading and single-clock-cycle function switching.

9. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 1, characterized in that, The system also includes: The global clock management unit supports independent frequency scaling and spread spectrum clock technology for different logical partitions. The multi-dimensional constraint integrated scheduling core module dynamically adjusts the driving frequency of each partition according to the task's real-time requirements and power consumption constraints. The security verification unit, located on the configuration link, is used to perform digital signature verification and logic conflict check on the target configuration bit stream before performing the reconstruction action, to ensure that the reconstructed circuit structure does not generate power supply to ground short circuits or signal race hazards.

10. A low-power reconfigurable electronic component chip scheduling system for edge computing according to claim 4, characterized in that, The power consumption trajectory prediction and voltage steady-state compensation module has online learning and self-calibration functions: The scheduling system automatically corrects parameter deviations in the model library by comparing the actual power consumption waveform monitored by the power management unit feedback interface with the model trajectory predicted by the instantaneous current model library. When an increasing leakage current trend due to semiconductor aging is detected, the multi-dimensional constraint integrated scheduling core module automatically increases the static power consumption weight in the power consumption cost assessment to guide the system to execute a more energy-efficient partition shutdown strategy.

Images