A macro-topological collective mode calculation system and an implementation method thereof

Abstract

The application relates to the technical field of information processing and computing architecture, and discloses a macroscopic topological collective mode computing system and an implementation method thereof. The system comprises a state unit, a write-in and read-out module, a coupling structure and a limited geometric path, wherein the state unit is used for carrying distinguishable, maintainable and modulatable collective states, the coupling structure is used for realizing the propagation and interaction of the states in the limited path, and the write-in and read-out module is used for completing state establishment, output collection and state discrimination. Through the design of structural parameters such as path direction, curvature, width gradient, local energy barrier and boundary condition, the propagation direction, propagation time, residence probability, threshold trigger condition and final state output distribution of the modulated state can be modulated, so that the computing process is realized through state propagation, competition, coupling and final state judgment. Preferably, the system can be combined with a reference unit for differential calibration or drift compensation, so as to improve the repeatability and stability of state discrimination. The application also discloses a path selection primitive, a threshold modulation primitive, a propagation delay primitive, a final state trigger primitive and an arrayed implementation mode thereof, which can be applied to magnetic systems, electronic systems, optical systems, mechanical systems, thermal diffusion systems and modeling implementation thereof. The technical scheme is favorable for the collaborative organization of state storage, state propagation, condition modulation and result judgment, and can be used for constructing a storage-computing integrated computing architecture.

Description

Technical Field

[0001] This invention relates to the field of information processing and computing architecture technology, and in particular to an in-memory computing system and its implementation method based on collective state evolution, geometric constraint modulation and electrical readout mechanism.

[0002] The "macro-topological collective mode computing system" described in this invention is not an abstract computing model, but a computing architecture built on an implementable physical system. The collective mode is preferably carried by a physical structure with topological stability, and its state modulation and electrical readout are achieved through external excitation. Background Technology

[0003] Existing computing systems mainly rely on charge flipping or voltage state switching to perform logic operations, and their basic calculation process depends on discrete switching behavior and Boolean logic structure.

[0004] As device size shrinks, power consumption, interconnect bandwidth, and manufacturing complexity continue to increase, putting traditional computing architectures at a bottleneck.

[0005] In recent years, solutions such as memristors, phase-change memories, and neuromorphic computing have introduced analog states or weighted modulation, but they still mainly rely on local state changes or parameter updates, lacking a mechanism to directly undertake computational functions by utilizing the internal state propagation process of the physical system.

[0006] Therefore, it is necessary to propose a new computing architecture that enables the generation, propagation, coupling, and discrimination of states to be completed in a unified physical mechanism, thereby facilitating the integration of storage and computing. Purpose of the invention

[0007] The purpose of this invention is to provide a macroscopic topological collective module computing system. By constructing a collective state unit with topological stability and modulating the propagation trajectory of the collective state using geometric constraint paths, the computing process can be realized through the evolution of the state in the spatial path, thereby achieving the integration of the computing process and the physical evolution process. Technical solution

[0008] This invention provides a macroscopic topological collective module computing system and its implementation method. Instead of using a single voltage flip or a single-point charge switch as the basic computational event, this system uses a collective state with a stable window as the state carrier, state propagation and coupling evolution within a constrained geometric structure as the computational process, and a stable distribution formed at the evolution endpoint as the output result. This achieves a unified approach to state storage, state propagation, condition modulation, result decision-making, and array-level expansion.

[0009] The technical solution of this invention can be summarized as follows: First, a collective state that is rewritable, persistent, and readable is established in one or more state units; then, through a predefined geometric constraint structure and coupling relationship, the state undergoes controlled propagation, contention, delay, threshold offset, or nonlinear combination in a constrained path; finally, the calculation result is obtained by judging the final state distribution, output proxy quantity, arrival order, or threshold triggering result. In this process, the geometric structure is not merely a device boundary, but a computational component involved in defining the transfer function, determining path selection, and constraining the final state distribution.

[0010] For ease of explanation, the technical solution of the present invention can be divided into the following nine interrelated parts.

[0011] 1. State Unit The state unit is used to carry the basic state variables in this invention and is the physical basis for achieving unified storage and computation. The state unit includes an active region, constraint boundaries, a write interface, and a read interface. If necessary, it may also include a local bias structure, a reference structure, a thermal management structure, a shielding structure, or a calibration structure.

[0012] like Figure 2 As shown, a state unit can be composed of an active region, a constraint boundary, a write interface, and a read interface, and carries state variables that can be repeatedly established, maintained, and determined within the active region.

[0013] The states within the state unit are not defined by the independent flipping of a single particle, but rather by a collective state formed by multiple local response regions, multiple mode components, or multiple local structures. This collective state can be a stable mode, a metastable state, a confined oscillatory state, a composite occupied state, or other reproducible and discriminable overall distribution.

[0014] In a preferred embodiment, the state is a collective mode with topological stability, which can be characterized by topological load, chirality, polarity, mode occupancy, spatial distribution, or a combination thereof. More preferably, the state can be represented by the physical configuration corresponding to a non-zero topological load, or can be distinguished by empty and non-empty states, multistable intervals, or continuous windows.

[0015] In this invention, the state unit satisfies at least the following three conditions: First, definability. A state unit can provide one or more distinguishable state windows, which can be discrete or continuous; they can correspond to multiple stable states or to discriminable sub-intervals within a continuous interval.

[0016] Second, controllability. The state unit can transition from the initial state to the target state under external stimuli, or can repeatedly switch between multiple candidate state windows.

[0017] Third, observability. A state unit has at least one observable physical quantity used to map its internal state to a measurable output. This observable physical quantity can be resistance, current, voltage, magnetic response, light intensity, phase, frequency, amplitude, temperature, or an equivalent proxy.

[0018] In a preferred embodiment, if the state unit is implemented by a magnetic topology, its topological properties can be characterized by the following topological charge:

[0019] This is a normalized vector field. This expression is used to illustrate that the preferred state in this invention is not a common local flip, but a topologically stable state determined by the overall spatial distribution. However, this invention does not limit the state definition to be uniquely given by the above expression; as long as the state meets the requirements of being definable, controllable, and observable, it can be used as a state unit in this invention.

[0020] In this invention, a "discriminable state window" refers to a state range that can be distinguished from adjacent state intervals by at least one observable physical quantity under preset measurement conditions; a "stable distribution" refers to the distribution state of the relevant state that remains within the discriminable state window within a preset time window and a preset disturbance range; an "equivalent surrogate quantity" refers to a measurable output quantity that can reflect the internal state changes of a state unit and is used to determine its state window, including but not limited to resistance, current, voltage, magnetic response, Hall response, light intensity, phase, frequency, amplitude, or temperature response; and "topological stability" refers to the ability of the relevant state to remain within the discriminable state window within a preset disturbance range, and can be characterized by topological charge, mode occupancy, chirality, or equivalent stable distribution.

[0021] In this invention, the preset measurement conditions, preset time window, preset disturbance range, and state window judgment threshold can be preset according to the specific device type, material system, measurement noise level, readout resolution, target task accuracy, and working environment, or determined through reference unit calibration, statistical testing, parameter backscanning, or closed-loop optimization. The specific values ​​of the above parameters may differ for different physical implementation methods, but as long as adjacent state intervals can be distinguished under the corresponding measurement conditions and can maintain repeatable discrimination within the preset disturbance range, they can be considered to meet the state window definition and stable distribution judgment requirements described in this invention.

[0022] The “collective state” described in this invention primarily refers to a state determined by multiple local degrees of freedom, multiple mode components, or multiple local response regions, and expressed in the form of overall distribution, overall mode, or overall occupancy, rather than an isolated state determined solely by a single local switching event.

[0023] In the preferred magnetic implementation, the topological stability can be characterized by topological charge or equivalent topological configuration; in other embodiments, it can be embodied as an equivalent stable mode that remains within a discriminable state window within a preset perturbation range.

[0024] The "stable distribution" includes, but is not limited to, the occupancy status distribution in the target output region, the relative occupancy relationship between multiple output regions, the spatial distribution corresponding to the proxy window, or the final output mode that can be repeatedly judged within a preset time window.

[0025] The “equivalent surrogate quantity” refers to an output quantity that, although it does not necessarily have the same physical form as the internal state variable itself, can stably reflect the changes in the internal state under preset measurement conditions and is used to determine the state window to which it belongs.

[0026] 2. Write and Read Mechanism This invention establishes an initial state through a writing mechanism and completes state discrimination through a reading mechanism; the two together constitute a state closed loop.

[0027] like Figure 2 and Figure 9 As shown, the writing module can be used to establish or switch the initial state in the state unit, and the reading module can perform differential comparison and drift compensation on the output of the target state unit in conjunction with the reference unit, thereby completing the state discrimination.

[0028] like Figure 9 As shown, the reference unit and the target state unit are connected to the differential comparison and compensation module to establish a dynamic reference and reduce the impact of drift.

[0029] The writing module is used to apply external stimuli to the state unit, causing the state unit to enter a preset state, switch from one state window to another, or enter an active state suitable for propagation. The external stimuli can be electrical signals, magnetic fields, optical signals, thermal stimuli, or combinations thereof. Specifically, writing can be achieved through pulsed current, pulsed voltage, localized magnetic fields, continuous bias, time-series pulses, laser pulses, heat source pulses, or other methods capable of inducing state transitions.

[0030] The readout module is used to acquire the output surrogate quantity of the state unit and convert it into a discriminable result. The surrogate quantity is preferably an electrical surrogate quantity, such as resistance, current, voltage, tunneling magnetoresistance, Hall response, impedance spectrum response, or other equivalent output signals. In non-electrical implementations, optical intensity, phase, frequency drift, mechanical displacement, amplitude, temperature distribution, etc., can also be used as readout quantities.

[0031] To ensure the system maintains stable decision-making capability under conditions of device parameter fluctuations, ambient temperature changes, material aging, or batch drift, this invention preferably includes at least one reference unit. The reference unit may have the same structure as the target state unit or a different structure, but its function is to provide a reference for target readout.

[0032] In a preferred embodiment, the readout module performs the following steps: Obtain the current output signal of the target state unit; Obtain the reference output signal of the reference unit; The target signal is compared with the reference signal using a differential method; Normalize, compensate, or calibrate the difference results; The window to which the target state belongs is determined based on the compensated result.

[0033] Therefore, this invention does not require every unit to be completely identical in absolute value. Instead, it establishes a dynamic benchmark through reference units, enabling the system to achieve online balance and stability discrimination capabilities. This mechanism is particularly suitable for systems operating at room temperature that experience noise, thermal fluctuations, and differences in device distribution.

[0034] In a preferred embodiment, the state window determination threshold, differential comparison threshold, and drift compensation parameters can be determined through pre-calibration, online calibration, or a combination of both. Specifically, the target state unit and the reference unit can be repeatedly measured to obtain their output distribution under different input conditions, different temperature conditions, or different disturbance conditions. Then, the determination threshold is set based on the separation, overlap rate, or misjudgment rate between the output distributions. Thus, the threshold is not arbitrarily set, but rather an engineering parameter that matches the device's statistical characteristics, noise level, and task determination accuracy.

[0035] 3. Coupling and Propagation Mechanisms In this invention, the multiple state units are not isolated from each other, but are interconnected through a coupling structure, and state propagation and interaction are realized within the coupling structure. The coupling structure can be local coupling or non-local coupling; it can be static coupling or dynamically adjustable coupling.

[0036] like Figure 5 As shown, multiple state units can form a propagation and modulation structure through a restricted geometric path. The propagation behavior of the state in the path is affected by the path width, curvature, local obstacles and boundary conditions, thereby forming a designable propagation result.

[0037] The basic functions of a coupling structure include, but are not limited to: Provides a feasible path for state propagation; Altering the energy landscape of adjacent or distant units; Modulate the propagation direction, propagation success rate, and dwell probability; Change the threshold, lifetime, or stability window of the target cell; Establish competitive, cooperative, or conditional relationships between multiple state units.

[0038] In this invention, state propagation does not occur arbitrarily in free space, but rather within a constrained geometric path. This geometric path can be a linear channel, a branching channel, a loop channel, an inter-island connection region, a two-dimensional network, a heterogeneous channel, or any combination thereof. The propagation behavior is modulated by the following structural parameters: Path direction and its topological connectivity; Path curvature and bending angle; Path width, width gradient, and cross-sectional changes; Distribution of local barriers or local obstacles; Boundary shape, boundary roughness, and boundary conditions; If necessary, phase delay, dielectric inhomogeneity, and local bias distribution may also be included.

[0039] In the preferred magnetic implementation, propagation can be characterized by the following dynamic relationship:

[0040] This characterizes the effective potential field determined by geometric boundaries, path structure, and local conditions. This relationship is not intended to limit the invention to a specific physical model, but rather to illustrate that the propagation in this invention can be modulated by the potential field introduced by the geometric structure, thereby possessing engineering designability.

[0041] In a more general sense, the propagation mechanism can be summarized as follows: the geometric structure determines the effective propagation constraints, the propagation constraints determine the state evolution trajectory, and the state evolution trajectory determines the subsequent calculation results.

[0042] 4. Computerized system The computation in this invention is not defined directly through traditional Boolean gate flipping, but through the propagation, competition, coupling and final state formation process of the state in a restricted path.

[0043] like Figures 3 to 6 As shown, the state propagation process can form a complete computational closed loop through path branch selection, dual-unit threshold modulation, restricted path propagation, and final state output decision. Among them, path selection, threshold change, propagation order, and final state distribution jointly determine the final calculation result.

[0044] Specifically, the calculation process includes the following stages: The first stage is input setup. External inputs are converted into initial states in one or more state units through the writing module. This input can be a single-point input, a parallel multi-point input, a timing input, or a weighted input.

[0045] The second stage is propagation and evolution. The initial state enters a predefined geometric path and coupled network, and under the combined effect of local barriers, path structure, and interactions, it propagates, stays, branches, delays, amplifies, suppresses, or shifts the threshold.

[0046] The third stage involves competition and combination. Multiple paths can compete with each other, multiple state units can form threshold modulation relationships, and multiple inputs can form nonlinear combinations in the endpoint region.

[0047] The fourth stage is the final state decision. At the end of the evolution, the states form a stable distribution, which can be represented by one output unit occupying the state, multiple output units distributed across the state, a certain path winning, a certain threshold being triggered, or a certain pattern being identified. This final state distribution is the computation result.

[0048] To further describe the "path-as-computation" principle in this invention, the path can be mapped to a spatial location label, and the sequence of events in the propagation process can be mapped to a temporal sequence label. The spatial location and the propagation sequence are not independent of each other, but rather coupled through geometric paths and coupling structures. Therefore, in this invention, a result is not only defined by "where it appears," but also by "in what order it appears, within what time window it appears, and whether it appears under a preset threshold condition." This mechanism differs from traditional systems that only represent logic using static node weights.

[0049] The propagation order in this invention is not determined solely by an independent external clock or predefined logic gate jumps, but rather by a combination of constrained geometric paths, local coupling conditions, local energy barrier distribution, and the state propagation process. Therefore, the spatial location label and propagation order label in this invention are not independent external codes, but rather the result of coupling together during the same physical propagation process. This mechanism differs from traditional synchronous clock logic, fixed-gate-level netlist logic, or calculation methods that rely solely on static node weights. The order formation in this invention depends on the physical propagation of the state carrier along the constrained path and the final state decision, rather than on the pure logical scheduling of discrete symbols within a fixed-gate-level structure.

[0050] In this invention, the propagation time window, arrival order determination condition, threshold trigger condition, and final state determination condition can all be preset according to the specific implementation platform or determined through experimental calibration. For example, corresponding criteria can be set according to the propagation success rate distribution, arrival time distribution, dwell time distribution, trigger probability distribution, or final state output distribution, so that different paths, different states, or different input combinations can be distinguished under a preset determination accuracy. Thus, the "order," "time window," "trigger," and "final state" in this invention all have an objective determination basis in physical implementation, rather than being purely abstract logical labels.

[0051] 5. Computational primitives The “propagation primitive structure” refers to the smallest structural unit that can realize state propagation, condition modulation and result determination in a restricted path, including but not limited to path selection primitive structure, threshold modulation primitive structure, propagation delay primitive structure and final state triggering primitive structure.

[0052] like Figure 3 As shown, path selection primitives can be implemented using Y-type or other branching path structures; such as Figure 4 As shown, the threshold modulation primitive can be implemented using a two-unit coupled structure; as Figure 5 As shown, propagation delay primitives can be implemented by changes in path length, path curvature, or local energy barrier; such as Figure 6 As shown, the final state triggering and output decision can be achieved by the stable distribution or triggering result in the target output region.

[0053] like Figure 3 As shown, after the input state enters the branch region, a selection can be made among multiple candidate channels based on differences in branch width, differences in local energy barriers, or path geometric asymmetry.

[0054] like Figure 4 As shown, the state change of control state unit C can change the effective threshold, triggering condition, or hold window of target state unit D.

[0055] like Figure 5 As shown, the settings of path width, path curvature, local contraction zone or obstacle zone can change the state propagation direction, propagation time, dwell probability or final state output distribution.

[0056] like Figure 6 As shown, the occupancy state, proxy window, or threshold trigger state in the target output region can serve as a direct basis for final state determination.

[0057] To structure the above-mentioned propagation computation, this invention preferably defines several minimal computational primitives. A computational primitive is a minimal propagation-decision unit that can be repeatedly constructed, repeatedly invoked, and used as the basis for array-level complex computations.

[0058] In a preferred embodiment, at least the following primitives are included: (1) Path selection primitives. Through branching structures, such as Y-shaped paths, T-shaped paths, or multi-branch paths, the input state selects from multiple candidate channels. This selection can be determined by geometric asymmetry, energy barrier differences, local bias, or differences in propagation conditions. The output is determined by the channel or channel distribution into which the state finally enters.

[0059] (2) Threshold modulation primitives. The state transition threshold, hold window, or propagation threshold of the target unit is changed by one or more control state units. When the input conditions are the same but the control unit states are different, the triggering result of the target unit changes, thus forming conditional calculation.

[0060] (3) Propagation delay primitives. The propagation time of the modulated state is determined by changes in path length, curvature, energy barrier, or medium conditions. The output can be defined by arrival time, time difference, or whether it falls within a certain time window.

[0061] (4) Final state trigger primitive. When the propagation state, superposition state or cumulative effect exceeds the threshold, the target unit output is triggered to flip or enter the target steady state.

[0062] (5) Nonlinear combination primitives. Multiple input states are nonlinearly superimposed in the same endpoint region or local network. The output is not a simple linear sum, but is determined by competition, suppression, saturation, phase relationship or local threshold.

[0063] The primitives mentioned above can be used individually, or combined in series, parallel, nested, or looped combinations to form more complex computational structures.

[0064] 6. Geometric constraints, i.e., computation A core technical point of this invention is that the geometric structure itself is not merely an encapsulation boundary, but directly participates in defining the computational logic.

[0065] In traditional computing architectures, geometry is typically used only for wiring, isolation, and fabrication. However, in this invention, the geometry itself plays a decisive role in the direction of state propagation, propagation time, path selection results, local threshold distribution, and final state output pattern, or participates in defining these behaviors. The propagable region of the state; The preferred direction of state propagation; Probability distribution of propagation along different paths; The time distribution of states reaching different endpoints; Local thresholds and triggering order at different locations; The reachable set of the final state distribution.

[0066] In other words, whether the path is wider or narrower, the corners are larger or smaller, where local obstacles are placed, and where boundary contractions are located are not merely map details, but rather part of the computational rules. Therefore, in this invention, geometric parameters not only have manufacturing significance, but also algorithmic and logical significance.

[0067] Preferably, the geometric parameters may include: channel length, channel width, width gradient, radius of curvature, branch angle, local energy barrier size, boundary shape, and multi-scale connection method. By predefining or adjusting these parameters, direct control over propagation probability, propagation time, and threshold behavior can be achieved.

[0068] 7. Multiscale structures like Figure 7 As shown, the present invention can adopt a hierarchical organization method of system-level structure, device-level active region and local structural scale region to respectively undertake the functions of external interconnection, propagation primitives and state variable carrying.

[0069] To ensure compatibility between state preservation, path propagation, read / write interfaces, and system integration, this invention preferably employs a hierarchical scale structure.

[0070] The hierarchical scale structure includes at least: (1) System-level architecture. This layer is used to implement external interconnection, signal input / output, power supply, thermal management, test interface, and large-scale organization. This layer does not directly carry the minimum state texture, but mainly undertakes the role of system organization and engineering support.

[0071] (2) Device-level active region. Used to implement computational primitives such as path selection, threshold offset, gated propagation, local contention, and final state decision. This layer is the layer with the most concentrated propagation and modulation behavior in this invention.

[0072] (3) Local structural scale region. Used to carry specific state variables, such as local topological texture, local phase distribution, local conductive channels, cavity modes or vibration modes. This layer determines the minimum physical expression of the state.

[0073] In the preferred engineering implementation, the system-level structure can be at the micrometer scale, the device-level active region can be at the submicrometer to micrometer scale, and the local structure scale can be at the nanometer to submicrometer scale. However, this invention is not limited to fixed values; the focus of protection lies in this hierarchical organizational logic.

[0074] The system-level structure, device-level active region, and local structural scale region are respectively used to carry external interconnects, propagation primitives, and state variables, thereby improving the compatibility and stability of state propagation computation at different scales.

[0075] The layered structure achieves the following technical effects through physical isolation: Decouple external interconnects from internal nonlinear evolution; Decouple thermal management from the state core; Decouple large-scale signal transmission from small-scale computation processes; Reduce the risk of the interface directly disturbing the core state.

[0076] The system-level architecture, device-level active region, and local structural scale region are respectively used to carry external interconnects, propagation primitives, and state variables, thereby improving the compatibility and stability of state propagation calculations at different scales. Through the above-mentioned hierarchical organization, on the one hand, the direct disturbance of the local state core by external interconnects, power supply, and test interfaces can be reduced; on the other hand, thermal management, read / write interfaces, and path propagation functions can be separated to improve the repeatability and engineering feasibility of state propagation, final state decision, and array-level expansion.

[0077] 8. Array Structure like Figure 8 As shown, multiple state units and their coupling paths can be further configured into linear arrays, ring arrays, two-dimensional grid arrays, or combinations thereof to support more complex propagation, modulation, and decision tasks.

[0078] Multiple state units and their coupling paths can be further configured into an array system to support more complex tasks.

[0079] The array structure can be a one-dimensional linear array, a closed ring array, a two-dimensional mesh array, a sparse graph array, a hierarchical array, or any combination thereof. The units in the array can be homogeneous or heterogeneous. Homogeneous units refer to multiple units that use the same physical implementation and similar state windows; heterogeneous units refer to the mixed use of different state units, different coupling mechanisms, or different readout methods within the same array.

[0080] The following organizational methods can be used during array operation: Unified writing and centralized reading; Partition write and local area read; Distributed calibration of the reference unit; Bad block mapping and redundancy compensation; Dynamic gating and phased propagation.

[0081] Array architectures can be used to extend a single propagation primitive into a reusable system-level resource, and support small-scale array validation, complex pattern recognition, constraint convergence, local decision-making, or dedicated acceleration tasks.

[0082] 9. Generalization Implementation This invention is not limited to a specific physical carrier. Any system that simultaneously satisfies the following three conditions can be considered an implementation of this invention: like Figure 10 As shown, the present invention can also implement the state nodes, path graph structure, coupling relationship and update rules equivalently through modeling, so as to reproduce the restricted path propagation and final state decision mechanism on digital processing platforms, dedicated acceleration platforms or other computing platforms.

[0083] like Figure 10 As shown, the modeling implementation can reproduce the restricted path propagation and final state decision process through state nodes, path graph structure, coupling relationships, and update rules.

[0084] It can form a distinguishable, maintainable, and moduloable collective state; This enables the state to propagate and couple within a restricted path; It can provide calculation results through final state distribution, triggering results, time distribution, or equivalent proxy quantity.

[0085] Therefore, this invention can be implemented by a magnetic system, electronic system, optical system, mechanical system, thermal diffusion system, hybrid physics system, or other system that satisfies the conditions for collective state formation, confined path propagation, and final state determination. Furthermore, this invention can also be implemented through computational methods that model the aforementioned propagation and coupling mechanisms, such as digital simulation, analog simulation, software digital twins, graphical model computation, or combinations thereof.

[0086] In this case, although the underlying medium is different, as long as the implementation process still corresponds to the technical process of state establishment, restricted path propagation, coupling modulation and final state decision, it still falls within the protection scope of this invention.

[0087] The modeling implementation includes not only numerical simulation of the physical propagation process, but also equivalent implementation of the geometric constraint path, coupling relationship and final state decision logic on a digital processing platform, dedicated acceleration platform or non-traditional computing platform. Effect

[0088] Compared with the prior art, the present invention has at least the following beneficial effects: This invention unifies state storage, state propagation, condition modulation, and result decision within the same physical evolution framework, which helps to overcome the data transfer overhead caused by the traditional separation of storage and computation.

[0089] This invention uses state propagation in a constrained geometric path as the basis for computation, transforming the geometric structure from a passive encapsulation factor into an active computation rule, which can provide layout-level design freedom for novel in-memory computing architectures.

[0090] This invention allows the use of discrete multistable states, continuous state intervals, or topologically stable states as state windows, thus being compatible with binary, multi-valued, and simulation calculation modes.

[0091] This invention improves the discriminability and robustness of the system under device fluctuations, environmental drift, and room temperature noise conditions through reference cells, differential calibration, and drift compensation mechanisms.

[0092] The technical solution of this invention can be implemented by a magnetic system, or by an electronic, optical, mechanical, thermal diffusion, or modeling system, and has strong generalization ability and expansion potential. Attached Figure Description

[0093] To more clearly illustrate the technical solution of the present invention, the invention can be further described in conjunction with the following accompanying drawings. The drawings are for illustrative purposes only and do not constitute a limitation on the scope of protection of the present invention.

[0094] Figure 1 This is a schematic diagram of the overall structure of the macroscopic topology collective module computing system of the present invention, used to illustrate the overall organizational relationship between the writing module, state unit, coupling and propagation structure and reading module.

[0095] Figure 2 This is a schematic diagram of the state unit structure, used to illustrate the state unit structure consisting of the active region, constraint boundary, write interface, and read interface.

[0096] Figure 3 This is a schematic diagram of a path selection unit based on Y-shaped geometric branches, used to illustrate the path selection primitives and their candidate channel selection behavior caused by differences in branch structure.

[0097] Figure 4 This is a schematic diagram of a dual-unit threshold modulation structure, used to illustrate the modulation relationship between the control state unit and the target state unit threshold, triggering condition, or hold window.

[0098] Figure 5 This is a schematic diagram of state propagation modulated by a restricted geometric path, used to illustrate the influence of geometric path parameters on propagation direction, propagation time, dwell probability, or final state distribution.

[0099] Figure 6 This diagram illustrates the relationship between the final state distribution and the output decision in state propagation. It is used to explain the correspondence between the occupied state, proxy window, or threshold-triggered state in the target output region and the final state decision.

[0100] Figure 7 This is a schematic diagram of multi-scale structure organization, used to illustrate the hierarchical organizational relationship between system-level structure, device-level active regions, and local structural scale regions.

[0101] Figure 8 These are schematic diagrams of linear arrays, ring arrays, and two-dimensional grid arrays, used to illustrate the array organization methods formed by multiple state units and their coupling paths.

[0102] Figure 9 This diagram illustrates the connection between the reference unit, the target state unit, and the differential comparison compensation module, and is used to explain the connection relationship between the reference unit, the target state unit, and the differential comparison compensation module.

[0103] Figure 10 This is a flowchart illustrating the modeling or software simulation implementation of the present invention, used to explain the modeling implementation process consisting of state nodes, path graph structure, coupling relationships, and update rules. Figure 10 It can also be used to illustrate the equivalent modeling relationships between different physical carriers, including the state mapping, path constraint mapping and final state decision mapping relationships between magnetic realization, electronic realization, optical realization, thermal diffusion realization and software simulation realization. Detailed Implementation

[0104] The present invention will be further described below with reference to embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. All equivalent substitutions, simplifications, modifications or combinations made within the spirit and principles of the present invention should fall within the scope of protection of the present invention.

[0105] In various embodiments of the present invention, the preferred main implementation method is as follows: a magnetic topology structure is used as the state unit, a restricted geometric path is used as the propagation and modulation structure, a Y-type branch path selection primitive and a dual-unit threshold modulation primitive are used as the basic computation primitives, a reference unit differential calibration is used as the readout stabilization mechanism, and a small-scale array is used as the system-level verification carrier. In a preferred embodiment, state variables can be preferentially determined using occupied states or surrogate windows corresponding to occupied states; readout methods can preferentially use electrical surrogate readouts; and array verification can preferentially examine indicators such as threshold distribution, hold time distribution, readout window overlap rate, and propagation success rate. In the preferred main implementation method, the path selection primitive, threshold modulation primitive, and final state decision primitive correspond to identifiable device structures, coupling structures, and output structures, respectively, thereby enabling the computation process to be implemented by specific physical components rather than abstract rules. In this preferred main implementation method, state establishment, path propagation, conditional modulation, final state decision, and array verification are interconnected and correspond to the preferred implementation chain of the present invention in terms of physical implementation, readout determination, and array expansion.

[0106] Example 1: Magnetic Topology State Unit like Figure 2 As shown, this embodiment provides a state unit that uses magnetic topological texture within a limited geometric region as a state carrier.

[0107] This state unit includes: a magnetic active region, a geometric constraint boundary, a write electrode, a read electrode, and a local bias structure if necessary. The magnetic active region can be composed of a single-layer magnetic film, a multi-layer magnetic heterostructure, a heavy metal / ferromagnetic / heavy metal interlayer, a compensated antiferromagnetic structure, a synthetic antiferromagnetic structure, or other material systems capable of supporting topological magnetic textures. The geometric constraint boundary is used to define the existence area, propagation path, or stability window of the magnetic texture; its shape can be disk-shaped, elongated, island-shaped, ring-shaped, Y-shaped, T-shaped, or other arbitrarily predetermined topologically connected structures.

[0108] In this embodiment, the state can be represented by the presence or absence of topological magnetic textures, by different topological texture types, or by topological charge, chirality, polarity, positional distribution, or the number of composite objects. Preferably, the empty state can be defined as... The existence state of a single topological object can be defined as follows: Multiple or composite object states can be defined as intervals of different integer or fractional topological loads. Furthermore, instead of directly using the topological load values ​​as external logical variables, the measurable proxy window corresponding to the topological state can be used as the state identifier.

[0109] The topological properties of this state element can be described by the following topological load expression:

[0110] This represents the normalized magnetic moment vector field. The purpose of this expression is to illustrate that the state in this embodiment is not a typical local magnetic domain flip, but rather represented by a collective magnetic configuration with topological stability.

[0111] The stability of this state unit can be described by the effective energy landscape, and preferably can be expressed as:

[0112] This reflects the limiting effect introduced by geometric constraint boundaries, local structural defects, edge barriers, or artificially defined layout constraints. This term does not limit the specific physical source, but rather serves to illustrate the fundamental idea of ​​"geometric constraints as computational constraints" in this invention.

[0113] The writing method can be pulsed current, local magnetic field, spin-orbit torque, spin-transfer torque, thermal pulse, optical pulse, or a combination thereof. For example, topological state nucleation can be achieved in the active region by short pulsed current; empty states can be transformed into single topological object states by local magnetic field bias; and different topological charge states can be switched between multiple windows by pulse sequence.

[0114] The preferred readout method is electrical readout, such as tunneling magnetoresistance, Hall signal, abnormal Hall voltage, anisotropic magnetoresistance, or other equivalent surrogate quantities. Calibration can also be performed during the research and development phase using magneto-optical imaging, electron microscopy, or scanning probe imaging, but electrical surrogate quantities are preferred for online operation.

[0115] The technical significance of this embodiment lies in establishing a fundamental state unit that can be stored, propagated, and further involved in coupled computation by constructing a topologically stable magnetic collective state within a finite geometric region. This embodiment primarily corresponds to the main physical implementation route of MTCC.

[0116] Optionally, the topological magnetic texture may include, but is not limited to, skyrmion, meron, antiskyrmion, skyrmion bag, magnetic bubble, composite chiral texture, or other magnetic collective states with equivalent topological stability. The above examples are merely illustrations and do not limit the scope of protection of this invention.

[0117] Example 2: Memristor-type analog state unit This embodiment provides a state unit that uses a memristor device or an equivalent variable conductivity device as the state carrier.

[0118] This state unit includes: top and bottom electrodes, a functional layer, a conductive path formation region, and a read / write interface. The functional layer can be made of metal oxides, chalcogenides, organic thin films, two-dimensional materials, ion-migrating materials, or other dielectric layers with tunable conductivity. The state is represented by the device's conductivity state, which can be discrete multi-level or continuous intervals.

[0119] In this embodiment, the state window is preferably defined as one or more intervals corresponding to the device's conductivity, equivalent resistance, conductivity, or current response amplitude. For example, low conductivity can be defined as an empty state, several intermediate conductivity intervals can be defined as intermediate states, and high conductivity intervals can be defined as high-occupancy states. To be compatible with the "collective state" expression in the parent patent, the "collective behavior" in this embodiment can be understood as the overall state of conductive filament distribution, ion migration distribution, interface barrier reconstruction, or current-carrying channel network formation, rather than a single electronic transition.

[0120] The preferred writing methods are voltage pulses, current pulses, pulse sequences, or combinations thereof. By changing the pulse amplitude, duration, duty cycle, or number of pulses, the device state can be written into different conductivity windows. For continuous state windows, iterative writing or closed-loop feedback writing can be used to gradually approach the target conductance range.

[0121] The preferred readout method is current measurement, resistance measurement, impedance spectrum measurement, or other equivalent electrical measurement under low disturbance voltage. To avoid readout disturbances, the readout pulse should be lower than the write threshold.

[0122] The computational significance of this embodiment lies in expressing the state through continuous or multi-level conductance windows, making it a fundamental unit for simulation calculations, threshold calculations, weighted propagation, or nonlinear responses. In a coupled structure, multiple such units can form a transfer function network, whose propagation and modulation behavior can simulate the "path modulation + steady-state distribution" framework in the MTCC parent case.

[0123] This embodiment is primarily used to illustrate that the present invention is not limited to magnetic implementation. As long as a certain type of state unit can provide an adjustable state window, rewritable, readable, and further coupled propagating state basis, it can fall within the protection framework of the present invention.

[0124] Optionally, a reference unit can be added to perform differential calibration, normalization compensation, or drift correction on the outputs of multiple memristor units to support the balance logic and online robust design in this invention.

[0125] Example 3: Phase Change Material State Unit This embodiment provides a state unit that uses the phase distribution of a phase change material as the state variable.

[0126] This state unit includes: upper and lower electrodes, a phase change layer, a thermal isolation layer, a local heating structure, and a read / write interface. The phase change layer can be made of GST-type materials, oxide phase change materials, perovskite materials, or other material systems capable of reversibly switching between different phase states.

[0127] The state is represented by the crystalline state, amorphous state, partially crystalline state, localized multiphase distribution state, or equivalent physical proxy of the phase change layer. Since the phase change material has multiple stable intermediate states, it can form multistable states or continuous approximate windows. Preferably, the empty state, partially occupied state, and highly occupied state can correspond to the conductive regions formed by different crystallization ratios.

[0128] The preferred writing method is electrothermal pulse or photothermal excitation. Different pulse widths, pulse peak values, and cooling rates correspond to different state windows. Amorphous states can be formed by rapid quenching with a single pulse, while partially crystalline states can be formed by longer pulses. The state window can be gradually modulated using pulse sequences.

[0129] The preferred readout method is resistance measurement, impedance measurement, optical transmittance measurement, reflectance measurement, or other phase-related proxy quantities.

[0130] In this embodiment, the "collective mode" can be understood as the overall state formed by the distribution of phase regions, phase boundary morphology, and local phase transition networks within the material. Although it does not necessarily have a direct magnetic topological charge, under the framework of the parent patent, its state is also determined by multiple microscopic degrees of freedom, and it can participate in propagation and decision-making under appropriate coupling structures.

[0131] The technical significance of this embodiment is that by representing the state through the material phase distribution, it is possible to realize multi-stable state storage, threshold triggering, delayed response and path-dependent nonlinear calculation, thus serving as one of the equivalent state unit embodiments of the MTCC parent patent.

[0132] Example 4: Optical Cavity State Unit This embodiment provides a state unit that uses the standing wave mode, phase distribution, field strength distribution, or mode occupancy status in an optical cavity or microcavity as the state carrier.

[0133] This state unit includes: an optical resonant cavity, an input coupling terminal, an output coupling terminal, a modulation structure, and a detection structure. The resonant cavity can be a ring resonant cavity, a Fabry-Perot cavity, a photonic crystal cavity, a microdisk cavity, a microring cavity, or other structures capable of supporting standing wave modes.

[0134] In this embodiment, the state is represented by the occupancy of the intracavity mode, the position of the standing wave node, the field intensity distribution, the polarization state, the phase state, or a combination of multiple modes. The state window can be characterized by the intracavity transmitted light intensity, the reflection spectrum, the phase delay, the spectral peak position, or the intensity window at the output end.

[0135] The preferred writing method is incident light pulse, continuous light modulation, phase modulation, electro-optic modulation, thermo-optic modulation, or mechanical modulation. By changing the incident frequency, pulse timing, modulation depth, or coupling coefficient, the cavity state can be switched to different mode windows.

[0136] The readout method can be transmitted light intensity detection, reflected light intensity detection, phase measurement, polarization analysis, or other equivalent optical measurement methods.

[0137] In this embodiment, the "collective state" can be understood as the overall mode formed by the cavity field distribution and boundary conditions. Multiple cavity state units can interact with each other through waveguides, couplers, or shared cavity regions, and can form path selection, phase modulation, and final state discrimination structures in an appropriate layout, thereby mapping to the path calculation framework of this invention.

[0138] This embodiment is used to illustrate that the present invention does not require the state to be carried by a magnetic object. As long as the stable mode in a certain physical system can propagate, couple and form a discernible final state in the geometric path, it can be used as the state unit of the present invention.

[0139] Example 5: Mechanical Oscillation State Unit This embodiment provides a state unit that uses the vibration mode, displacement mode, resonant frequency window, or phase window of a mechanical oscillation structure as state variables.

[0140] This state unit includes: a mechanical oscillating element, an excitation structure, a detection structure, and a necessary constraint framework. The mechanical oscillating element can be a cantilever beam, a thin-film oscillator, a microelectromechanical resonator, a coupled oscillator network, or other mechanical structures capable of forming stable modes.

[0141] In this embodiment, the state can be represented by vibration amplitude, vibration phase, dominance of the dominant mode, resonant frequency position, or a combination of multiple vibration modes. The state window can be divided into low amplitude, medium amplitude, and high amplitude regions, or it can be characterized by the frequency bands corresponding to different intrinsic modes.

[0142] The preferred writing methods are electrical excitation, piezoelectric excitation, thermal excitation, optical excitation, or magnetic excitation. By adjusting the excitation frequency, pulse duration, and excitation amplitude, the oscillator can be made to enter different vibration states.

[0143] The readout method can employ capacitance detection, piezoresistive detection, piezoelectric detection, optical displacement detection, or other equivalent methods.

[0144] In this embodiment, "collective behavior" can be understood as a stable mode shape formed by multiple degrees of freedom within a mechanical structure. Multiple such units, after coupling, can exhibit phase-locking, contention propagation, threshold triggering, and delay-coding behaviors, thus serving as a non-magnetic embodiment of the present invention.

[0145] Example 6: Local Coupling Mechanism This embodiment provides a local coupling mechanism, in which adjacent state units form coupling through direct interaction.

[0146] The local coupling can occur between two or more spatially adjacent state units. For magnetic implementations, local coupling can be achieved through exchange interactions, proximity magnetostatic interactions, boundary coupling, or shared local texture regions; for memristor implementations, it can be achieved through shared current paths, coupling of adjacent conductive regions, local thermal diffusion, or charge accumulation; for optical implementations, it can be achieved through near-field coupling, direct-coupled waveguides, or local mode overlap; for mechanical implementations, it can be achieved through elastic beam connections, local stress transfer, or shared support structures.

[0147] The effect of local coupling is that a state change in one state cell directly alters the energy landscape, threshold window, propagation direction, or readout response of adjacent cells. For example, when cell A enters an active state, it may lower the write threshold of adjacent cell B; it may also change the propagation initiation probability of B; or it may cause a phase drift in B.

[0148] The core of this embodiment is that the coupling strength is mainly determined by the relative position, spacing, local boundary and interface conditions, so it is suitable for constructing minimum condition calculation primitives, such as two-unit threshold offset structure or nearest neighbor propagation structure.

[0149] Example 7: Nonlocal Coupling Mechanism This embodiment provides a non-local coupling mechanism, in which multiple state units separated by a certain distance can exert remote influence through a shared field, a shared medium, or an intermediate propagation medium.

[0150] In magnetic implementations, nonlocal coupling can be achieved through spin waves in a shared magnetic matrix, long-range magnetostatic fields, shared antiferromagnetic substrates, shared magnetic layers, or shared topological stress fields; in optical implementations, it can be achieved through waveguide networks, shared cavities, mode buses, or photonic crystal channels; in electronic implementations, it can be achieved through shared power networks, floating nodes, long-range capacitive coupling, or transmission line coupling; and in mechanical implementations, it can be achieved through substrate propagation waves, acoustic resonant cavities, or coupled beam networks.

[0151] The technical significance of nonlocal coupling is that computation is no longer limited by geometric adjacency, but can map the state of a unit to a remote unit through an intermediate medium, thereby supporting global constraints, remote modulation, and cross-regional collaborative computation.

[0152] This embodiment is particularly suitable as a supporting mechanism for array-level consistency control, remote conditional modulation, and multipath contention computation.

[0153] Example 8: Dynamic Coupling Mechanism This embodiment provides a dynamic coupling mechanism in which the coupling strength, coupling direction, or coupling channel changes with time, state, or external control signals.

[0154] The dynamic coupling can be achieved by: opening or closing the coupling path through timing pulses; dynamically changing the energy barrier through local bias; or changing the propagation conditions between units through external time-varying magnetic fields, time-varying voltages, phase modulation signals, or time-varying mechanical tension.

[0155] For example, an adjustable gating region can be set between two state units. When the gating region is in a low-barrier state, the state can propagate; when the gating region is in a high-barrier state, the propagation is suppressed. Another example is in optical systems, where the coupling strength between two cavities can be changed by dynamically modulating the refractive index.

[0156] The advantage of this embodiment is that it can realize timing control, gated propagation, staged routing and programmable coupled networks, and is therefore suitable as the basis for higher-level primitives and small-scale arrays.

[0157] Example 9: Geometric Path Modulation Mechanism This embodiment provides a geometric path modulation mechanism. This embodiment is one of the core implementation directions of the present invention, namely, changing the state propagation trajectory and transfer function through the structural shape itself.

[0158] Specifically, geometrically constrained channels are set between state units. The width, curvature, branching angle, cross-sectional shape, length, local contraction region, barrier region, or boundary roughness of the channels are pre-designed so that the states are modulated by different effective potential fields during propagation. For magnetic realization, the effective potential can be written as:

[0159] It reflects the changes in geometric potential caused by path width gradient, curvature variation, and local obstacles.

[0160] State propagation can be described by the following dynamic form:

[0161] The results of geometric modulation include, but are not limited to: prioritizing a certain path, delaying arrival at a certain region, increasing or decreasing the probability of dwell time, and changing the final state distribution. This embodiment directly corresponds to the core idea of ​​"path is algorithm, layout is logic".

[0162] Example 10: Phase Coupling Mechanism This embodiment provides a phase coupling mechanism, which affects state propagation and calculation results through phase difference, phase locking, or phase competition.

[0163] In optical systems, phase coupling can be achieved by coherent optical fields, coupled cavity modes, or phase shifters; in mechanical systems, it can be achieved by phase synchronization or unlocking between oscillators; in electronic or memristor networks, it can be achieved by oscillating neurons, periodic pulse sources, or phase-coded signals; and in magnetic systems, it can also be expressed by spin wave phase or oscillating mode phase.

[0164] The output of the state unit is no longer determined solely by the amplitude, but can also be defined by a phase difference window. For example, when two propagation paths arrive at their endpoints simultaneously, the final decision can be determined by whether their phases are constructive or destructive. Multiple phase-coupled units can be used to implement oscillation calculations, synchronization selection, and nonlinear combinations.

[0165] The significance of this embodiment lies in extending "propagation" from relying solely on path occupancy to relying on phase structure, thereby enhancing the expressive power and scalability of the present invention.

[0166] Example 11: Two-unit competition primitives like Figure 3 As shown, this embodiment provides a two-unit competition primitive, that is, two propagation candidate paths or two state units form a two-choice decision through competition.

[0167] like Figure 3 As shown, after the input state enters the branch region, it can compete for propagation between multiple candidate directions based on differences in branch width, differences in local energy barriers, or path geometric asymmetry.

[0168] This primitive includes an input region and two candidate output regions, or two parallel propagation channels. After the input state is written, the state propagates competitively along the two candidate directions. Due to differences in the geometric parameters, coupling strength, local energy barriers, or initial conditions of the two directions, only one direction will eventually reach the endpoint first, or the two directions will be occupied with different probabilities.

[0169] The output can be defined as follows: if the final state distribution is biased towards the first output region, the output logic value is 0; if it is biased towards the second output region, the output logic value is 1. Alternatively, the result can be presented as a probability distribution in simulation calculations.

[0170] This primitive is a higher-level generalization of the Y-type path selection primitive and can serve as the basis for minimal selectors, classifiers, or competitive decision-making devices.

[0171] Example 12: Three-unit door structure primitives This embodiment provides a three-unit gate structure primitive, which consists of two input state units and one output state unit, and an intermediate coupling unit can also be further provided.

[0172] In a preferred implementation, the three-unit gate structure can utilize Figure 4 The threshold modulation relationship shown is implemented, wherein the state change of the control state unit C can change the effective threshold, trigger condition or hold window of the target state unit D; in the three-unit gate structure, the above threshold modulation relationship can be implemented by the two input state units A and B acting together on the output state unit C, or by acting on the target output unit through the intermediate coupling unit.

[0173] Input units A and B each receive write pulses, and the state of output unit C is influenced by both A and B. Specific influence methods may include: A and B simultaneously lowering the threshold of C; A or B acting alone producing only a partial offset; and the superposition of the states of A and B creating an over-threshold response in C.

[0174] By designing different coupling methods and threshold windows, equivalent behavior to AND, OR, NAND, NOR, or more general nonlinear gate structures can be achieved. However, this invention does not require strict replication of traditional CMOS logic gates, but emphasizes achieving gate behavior through collective state propagation and steady-state formation.

[0175] The significance of this embodiment is that it proves that multiple input states can be collectively coupled and converge into a single output steady state, thereby forming a composable minimal logic primitive.

[0176] Example 13: Path Delay Coding Primitives like Figure 5 As shown, this embodiment provides a path delay coding primitive, which represents information through state arrival time.

[0177] In this primitive, the input state is written by a starting unit and propagates along at least two paths with different lengths or different energy barriers. The time to reach the destination satisfies:

[0178] Different paths have different geometric structures, corresponding to different arrival time windows. The system can determine the output result based on the arrival time, the time difference, or whether the arrival occurs within a preset time window.

[0179] This primitive is applicable to time encoding, priority sorting, temporal classification, and constraint satisfaction problems. It is also an important manifestation of "order-address coupling" at the implementation level, because the spatial path structure directly defines the temporal order label.

[0180] Example 14: Threshold Triggered Primitives This embodiment provides a threshold trigger primitive, which triggers the output unit to flip or enter the target state when a certain state variable, propagation probability, cumulative amplitude, phase combination, or energy offset exceeds a preset threshold.

[0181] The threshold can be preset based on material properties, geometric constraints, reference element calibration results, or external control parameters. The triggering condition can be that the threshold is reached by a single path, or that the threshold is reached by the cumulative effect of multiple inputs.

[0182] For example, in a dual-cell threshold offset structure, control unit A causes the effective threshold of target cell B to shift from...

[0183] If the external write pulse is fixed at \(T_1) This primitive is suitable for implementing conditional control, event triggering, gating selection, and critical behavior decision-making.

[0184] Example 15: Nonlinear Combinatorial Primitives This embodiment provides a nonlinear combination primitive, in which multiple state variables, path contributions, or phase components are combined in the endpoint region. The combination result is not a simple linear superposition, but presents a nonlinear response.

[0185] For example, three propagation paths carry state contributions with different weights. After reaching the same output region, due to local energy landscape, threshold effect, phase continuation or cancellation, and material nonlinear response, the final output may exhibit saturation, threshold switching, competitive suppression, or multi-peak response.

[0186] This primitive can express more complex computational functions, such as pattern fusion, nonlinear classification, context-dependent output, and local optimal selection.

[0187] This embodiment illustrates that the present invention covers not only basic logic primitives, but also combinatorial primitives that depend on physical nonlinearity.

[0188] Example 16: Linear Array System This embodiment provides a linear array system, which consists of multiple state units arranged in a one-dimensional sequence, with adjacent or near-neighbor units connected by local coupling or restricted channels.

[0189] Input state can be written from one end of the array or from multiple locations simultaneously. The state propagates along the array direction, and threshold offsets, delay accumulation, distributed updates, or local contention occur at different nodes. Output can be read from the end of the array or from a middle tap.

[0190] Linear arrays are suitable for sequence propagation, dynamic filtering, delayed chain coding, temporal pattern recognition, and pipelined local decision making.

[0191] Example 17: Ring Array System This embodiment provides a ring array system in which multiple state units are connected end to end to form a closed loop.

[0192] Due to the existence of loops, states can propagate cyclically within the loop, forming loop lap probabilities, dwell times, number of loops, or distribution patterns on the loop. By setting local gating regions, readout regions, and injection regions on the loop, cyclic memory, oscillation calculation, period determination, and constrained convergence can be achieved.

[0193] Circular arrays are particularly suitable for tasks requiring repeated iterations, backtracking, or cyclical competition. Their technical advantage lies in the fact that the same set of cells can reuse the propagation process multiple times without having to be moved step by step by an external clock.

[0194] Example 18: Two-dimensional grid array system This embodiment provides a two-dimensional grid array system in which multiple state units are arranged in a matrix and connected by horizontal, vertical or diagonal coupling channels.

[0195] Two-dimensional meshes support more complex path competition, local constraint propagation, and multi-source information fusion. Input states can be injected from edge or internal nodes, and the state propagates within the mesh according to geometry and local energy barrier distribution. The final state distribution can be defined by a global occupancy pattern, locally winning cells, or a combination of multi-point outputs.

[0196] Two-dimensional meshes are suitable for template matching, local consistency judgment, approximate solution of graph problems, pattern classification, and computation of problems oriented towards physical constraints.

[0197] Example 19: Heterogeneous Array System This embodiment provides a heterogeneous array system, wherein the state units in the array are not all of the same type, but are composed of a mixture of units with different physical implementations, different state windows or different coupling mechanisms.

[0198] For example, the input of the array uses magnetic topology units to carry the propagation state, the intermediate layer uses memristors or phase-change units as threshold modulation units, and the output uses optical or electrical high-sensitivity readout units. Furthermore, fast propagation units and high-retention memory units can coexist in the same array.

[0199] The advantage of heterogeneous arrays lies in their ability to combine the strengths of different physical platforms, thereby enhancing the system's flexibility, robustness, and task adaptability. This embodiment also further supports the cross-platform coverage capability of the parent patent of this invention.

[0200] Example 20: Small-scale array verification system like Figure 8 and Figure 9 As shown, this embodiment provides a small-scale array verification system that verifies state existence, path propagation, final state decision, and array consistency through array structure and reference cell calibration mechanism.

[0201] like Figure 9 As shown, the reference unit and the target output unit can be connected to the differential comparison and compensation module to reduce the drift effect and improve the consistency of output discrimination during array-level verification.

[0202] The small-scale array can be, for example, a 4×4 array, a 2×N array, a ring array, or a combined array consisting of several dual-units and Y-type primitives.

[0203] This small-scale array does not pursue general computation but is used to verify the following minimal closed loops: state presence, writing, holding, reading, coupling, path modulation, and final state decision. Threshold distribution, holding time distribution, readout window overlap rate, and propagation success rate of multiple cells can be statistically analyzed within the same process batch to evaluate array consistency.

[0204] This embodiment directly corresponds to the most realistic stage of MTCC's transition from cells to arrays. The focus is not on scale, but on proving that multiple cells can exhibit sufficiently similar windows in the same batch and support minimum cascading.

[0205] Example 21: Pure Circuit Equivalent Implementation This embodiment provides a pure circuit equivalent implementation, in which the state unit, coupling structure, path competition and final state discrimination are all composed of analog or mixed signal circuits.

[0206] For example, capacitor networks, resistor networks, controlled sources, comparators, adjustable threshold units, and integrators can be used to simulate state windows, coupling strength, path barriers, and final state discrimination, respectively. Geometric path modulation can be achieved by impedance distribution, delay distribution, or gain distribution in a multi-branch analog network.

[0207] The significance of this embodiment is that even without using magnetic or other native physical state units, as long as the system still performs calculations through "state propagation - path modulation - final state distribution", it is an equivalent implementation of the present invention.

[0208] Example 22: Equivalent Implementation of Digital Simulation This embodiment provides an equivalent implementation of digital simulation, which updates the state propagation and coupling process in discrete time steps in digital circuits, FPGAs, dedicated digital processors, or programmable logic systems.

[0209] The system can abstract each state unit as a node, the coupled path as an edge, and the geometric modulation as edge weights, path constraints, or transition probabilities. At each time step, the system updates the state for the next time step based on the current state, the neighborhood state, and the path rules, and finally outputs the result based on the final state distribution.

[0210] Although this embodiment may not necessarily retain the original continuous physical process, it retains the core causal structure of the present invention, and therefore can serve as direct support for the "modeling implementation" in the claims.

[0211] Example 23: Implementation of Optical Equivalence This embodiment provides an equivalent system that uses light propagation to achieve state propagation and computation.

[0212] In this system, state units consist of optical modes, intensity windows, or phase windows; coupling paths consist of waveguides, couplers, splitters, and cavity structures; geometric path modulation is achieved by waveguide width, refractive index gradient, splitting angle, and phase delay units. The final state distribution is determined by the intensity distribution, phase distribution, or mode occupancy at the output.

[0213] This embodiment demonstrates that the core of the present invention is not magnetism itself, but rather the architectural concept of "modulating state propagation through geometric constraints and characterizing the computational results with the final state distribution".

[0214] Example 24: Equivalent Realization of Thermal Diffusion This embodiment provides an equivalent system based on a thermal diffusion process.

[0215] In this system, state units can be represented by local temperature windows, thermal capacity states, or phase transition thresholds; coupling is achieved through thermal conductivity networks or diffusion channels; and geometric paths are defined by thermal conductivity layer thickness, thermal conductivity distribution, thermal insulation barriers, and thermal diffusion boundaries. Inputs are written via local heating, and outputs are read via temperature sensors or thermal response surrogate quantities.

[0216] With proper design, different thermal paths correspond to different diffusion rates and final state distributions, thus enabling path competition, threshold triggering, and final state determination. This embodiment further supports the invention's coverage of "non-magnetic but isomorphic" physical systems.

[0217] Example 25: Software Simulation Implementation like Figure 10 As shown, this embodiment provides a software simulation implementation that runs on a general-purpose computer, server, cloud platform, or embedded computing device.

[0218] like Figure 10 As shown, software simulation can reproduce the restricted path propagation and final state decision process through state nodes, path graph structure, coupling relationships, and update rules.

[0219] The software defines multiple state nodes, each with a state window, threshold, drift compensation parameters, and coupling relationships; a path graph structure is defined to describe geometric constraints; and update rules are defined to simulate state propagation, competition, delay, and final state convergence along the path. The output is determined by the stable distribution of the evolution endpoint.

[0220] Software simulation can be implemented using continuous-time models, discrete-time iterative models, random sampling models, probability transition models, or partial differential equation approximation models. Its purpose can be hardware pre-verification, task mapping, parameter optimization, digital twins, or pure software computation.

[0221] This embodiment is used to support the following: as long as a system can reproduce the state propagation and coupling of the present invention through modeling. The combination and final state decision mechanism can be regarded as one of the modeling implementations within the scope of protection of this invention.

[0222] Example 26: A combined calculation example based on Y-type path selection, dual-unit threshold modulation, and reference unit calibration. like Figure 3 , Figure 4 , Figure 5 and Figure 9 As shown, this embodiment provides a combined calculation embodiment based on Y-type path selection, dual-unit threshold modulation, and reference unit calibration, which illustrates how state establishment, restricted path propagation, conditional modulation, final state decision, and output calibration are interconnected and complete the calculation process in the same system.

[0223] (1) Structural composition The system of this embodiment includes at least: an input state unit A; a restricted propagation channel connected to the input state unit A; a Y-shaped branch structure formed by the restricted propagation channel; a first candidate output region B1 and a second candidate output region B2 respectively disposed at the ends of the Y-shaped branch; a control state unit C; a target state unit D coupled to the control state unit C; a reference unit R; and a readout module connected to the first candidate output region B1, the second candidate output region B2, the target state unit D, and the reference unit R.

[0224] In a preferred embodiment, the input state unit A, control state unit C, target state unit D, and reference unit R can be implemented using magnetic topology state units; the Y-shaped branch structure can be formed by a restricted geometric path; and the readout module can be implemented using an electrical proxy readout method. The first candidate output region B1 and the second candidate output region B2 can correspond to different output state windows or different output logic results, respectively.

[0225] In an optional embodiment, the input state unit A, control state unit C, target state unit D, and reference unit R can also be implemented by memristor-type state units, phase-change state units, optical cavity state units, mechanical oscillation state units, or other state units that satisfy the state definition conditions of this invention. The restricted geometric path can also be composed of waveguide structures, heat-conducting channels, vibration coupling paths, or equivalent propagation paths.

[0226] (2) Initialization and state establishment In this embodiment, an initialization process is preferably performed before the system operates. The initialization process includes at least the following steps: Set the input state unit A to the preset initial state; Set the control state unit C to either the first control state or the second control state; Set the target state unit D to the state to be triggered or the state to be written to the reference. Set the reference unit R to the preset reference state; The readout module is subjected to reference sampling to obtain the initial output signal of the reference unit R.

[0227] In a preferred embodiment, the state of input state unit A can correspond to the existence, occupation, or activation state of the object to be propagated; the state of control state unit C can correspond to different threshold modulation conditions; the state of target state unit D can correspond to different trigger threshold windows; and the state of reference unit R can be used to provide a reference output under the current working environment.

[0228] After initialization, an external stimulus is applied to input state unit A via the writing module, causing it to enter the propagation state. The external stimulus can be a current pulse, voltage pulse, local magnetic field, thermal pulse, optical pulse, or a combination thereof. For different state unit implementations, the above stimulus form can be selected based on the material system and read / write interface. (3) Y-shaped path selection process

[0229] like Figure 3 and Figure 5 As shown, after the state to be propagated established by input state unit A enters the Y-shaped branch structure, it can choose among multiple candidate channels. The path selection behavior can be determined by one or more of the following factors: The difference in width between the first branch and the second branch; The curvature difference between the first and second branches; The difference in path length between the first branch and the second branch; Differences in the setting of local obstacle zones or local contraction zones; Differences in the distribution of local energy barriers at branch points; Differences in boundary roughness or boundary conditions at the branch points; If necessary, it also includes phase differences, dielectric inhomogeneities, or local bias differences.

[0230] In a preferred embodiment, if the width of the first branch is greater than the width of the second branch, or if the first branch corresponds to a lower local energy barrier, the probability of the state to be propagated entering the first candidate output region B1 is greater than the probability of entering the second candidate output region B2; conversely, if the second branch has a lower effective potential field constraint, the state to be propagated can preferentially enter the second candidate output region B2.

[0231] For the preferred magnetic implementation, the propagation behavior of the state in the confined path can be characterized by the following nonlinear dynamic relationship, which illustrates the constraint effect of the effective potential field on the propagation direction, propagation speed, dwell probability, or final state distribution:

[0232] The effective potential field is determined by the boundary, path shape, local obstacles, and local bias. Therefore, the propagation direction, propagation speed, dwell probability, and final state distribution of states can all be influenced by the geometric configuration of the Y-shaped branching structure.

[0233] In this embodiment, the Y-shaped path selection process corresponds to the path selection primitive in this invention. The path selection result can be characterized by any of the following: the state preferentially arrives at the first candidate output region B1; the state preferentially arrives at the second candidate output region B2; the probability distribution of the state arriving at B1 and B2 respectively; the time distribution of the state arriving at B1 and B2 respectively; and the residence time distribution of the state in the branch region or candidate output region.

[0234] (4) Dual-unit threshold modulation process like Figure 4 As shown, in addition to the Y-shaped path selection process, this embodiment further sets up a coupling structure between the control state unit C and the target state unit D to form a dual-unit threshold modulation primitive.

[0235] Specifically, changes in the state of control state unit C can alter the effective threshold, triggering condition, or hold window of target state unit D. This modulation relationship can be achieved through local coupling, non-local coupling, shared medium coupling, local energy landscape alteration, local bias alteration, thermal coupling, magnetic coupling, electrical coupling, optical coupling, or other equivalent coupling methods.

[0236] In a preferred embodiment, when control state unit C is in the first control state, the effective threshold of target state unit D is the first threshold.

[0237] Within the corresponding trigger interval, whether the target state unit D enters the trigger state will be determined by the current state of the control state unit C.

[0238] For example, in a preferred embodiment, when control state unit C is in a state that lowers the threshold of target state unit D, target state unit D can be triggered; while when control state unit C is in a state that raises the threshold of target state unit D, target state unit D may not be triggered. Thus, control state unit C and target state unit D form a conditional triggering relationship.

[0239] In this embodiment, the triggering condition of the target state unit D can be determined by one or more of the following factors: whether the input state arrives first candidate output region B1 or second candidate output region B2 first; the cumulative occupancy degree of the input state in the target region; whether the time of the input state arriving in the target region falls within a preset time window; the threshold offset degree of the control state unit C to the target state unit D; and the current reference signal provided by the reference unit R.

[0240] (5) Reference unit calibration and output discrimination like Figure 9 As shown, in this embodiment, a reference unit R is set, and the reference unit R and the target output signal are differentially compared, normalized, and drift compensated by the readout module to improve the reliability of state discrimination.

[0241] In a preferred embodiment, the readout process includes at least the following steps: Read the output proxy value of the first candidate output region B1; Read the output proxy value of the second candidate output region B2; Read the output proxy quantity of the target state unit D; Read the reference output signal of the reference unit R; The difference between the target output and the reference output is calculated using the difference comparison module; The difference is corrected by a normalization or drift compensation module; The status window is determined based on the corrected output.

[0242] The output proxy quantity can be resistance, current, voltage, tunneling magnetoresistance, Hall response, impedance spectrum response, or other equivalent proxy quantities. For non-electrical implementations, the output proxy quantity can also be light intensity, phase, frequency, amplitude, displacement, or temperature response.

[0243] In a preferred embodiment, the state window determination threshold, differential comparison threshold, and drift compensation parameters can be preset based on the device statistical test results or determined through an online calibration process. Therefore, the output determination in this embodiment is not based on a single absolute measurement result, but rather on a relative determination based on a dynamic benchmark established by the reference unit R.

[0244] (6) Definition of output results In this embodiment, the calculation result can be determined by combining the Y-type path selection result, the threshold triggering result of the target state unit D, and the output proxy quantity after calibration by the reference unit.

[0245] In a preferred definition, the output results can be given according to the following rules: If the state to be propagated enters the first candidate output region B1 first, and the target state unit D is not triggered, then the first result is output; If the state to be propagated enters the second candidate output region B2 first, and the target state unit D is triggered, then the second result is output; If the output proxy of the state to be propagated enters B1 or B2 and falls into the corresponding preset window, and the output proxy of the target state unit D meets the preset threshold condition, then it is determined to be a valid output. If the output proxy quantity does not fall into the preset window, or the target state unit D does not meet the preset trigger condition, it can be determined as a retry output, an invalid output, or an intermediate state output.

[0246] In another preferred definition, the state combination of the first candidate output region B1, the second candidate output region B2, and the target state unit D can also be defined as a multi-valued output. For example, "B1 is dominant and D is not triggered", "B1 is dominant and D is triggered", "B2 is dominant and D is not triggered", and "B2 is dominant and D is triggered" can each correspond to different output results, thereby forming a multi-valued or conditional output mode.

[0247] (7) Technical function of this embodiment This embodiment combines path selection primitives, threshold modulation primitives, and reference cell calibration mechanisms within the same system, forming a seamless implementation chain from input state establishment, restricted path propagation, conditional modulation, final state decision to output calibration. Therefore, this embodiment illustrates that the present invention does not merely comprise isolated state units or single propagation phenomena, but rather forms a minimum functional computational combination structure; the geometric path structure and the local coupling structure jointly determine the output result; output determination can be further refined to obtain repeatable outputs through reference cell calibration; and the minimum combination structure can be further extended to small-scale arrays to support more complex computational tasks.

[0248] (8) Optional deformation Without departing from the spirit of the invention, this embodiment may also have the following modifications: The Y-shaped branch structure can be replaced by a T-shaped branch structure, a multi-branch structure, or a loop branch structure; The coupling method between the control state unit C and the target state unit D can be replaced by local magnetic coupling, electrical coupling, thermal coupling, optical coupling, mechanical coupling, or a combination thereof; The reference unit R can be set as a single reference unit or as multiple distributed reference units; The output determination can be completed based on a single target state unit D, or it can be completed jointly by multiple target state units. The system can be further extended to a linear array, a ring array, a two-dimensional grid array, or a heterogeneous array.

Claims

1. A macroscopic topological collective module computing system, characterized in that, include: A state unit, comprising an active region, a constraint boundary, a write interface, and a read interface, has an adjustable state window, and its state is characterized by a topologically stable mode formed by the collective behavior of multiple basic units, and has observable physical quantities to characterize the state; a write and read module, used to modulate the state through external excitation and obtain electrical or equivalent surrogate quantities; a coupling structure, comprising at least one constrained geometric path, used to realize state propagation and interaction between multiple state units; further characterized in that: the calculation is achieved by modulating the state propagation path through geometric constraints, the modulation including changes in path curvature, width gradient, local energy barrier, or boundary conditions to form a nonlinear transfer function; and the state propagation path and its combination relationship correspond to a preset propagation primitive structure, and the output result is determined by the path occupancy state, arrival order, or threshold triggering result in the target output region; wherein, the output result is determined by the stable distribution formed by the occupancy state, surrogate quantity window, or threshold triggering state in the target output region at the evolution endpoint.

2. A computational method based on collective state evolution, characterized in that, include: Initialize or modulate the state unit; Implement state propagation and evolution in geometrically constrained paths; The calculation process is formed by the selection of propagation paths, propagation order, or threshold changes; Obtain the output proxy quantity of the target state unit; obtain the reference output signal of the reference unit, or obtain a preset reference output; calibrate the output proxy quantity through differential comparison, normalization, or drift compensation; The window to which the target state belongs is determined based on the calibration results, and the calculation results are obtained; wherein, the calculation results are determined by the stable distribution of the state at the evolution endpoint.

3. The system or method according to claim 1 or 2, characterized in that: The state is a collective mode with topological stability and includes discrete multistable or continuous state intervals, preferably characterized by non-zero topological loads.

4. The system or method according to claim 1 or 2, characterized in that: The writing is achieved through electrical signals, magnetic signals, optical signals, or thermal excitation, and the reading is obtained through electrical or equivalent surrogate quantities.

5. The system or method according to claim 1 or 2, characterized in that: The readout module establishes a dynamic benchmark through a reference unit and performs differential calibration or real-time drift compensation on the state readout to achieve repeatable or deterministic discrimination of the state.

6. The system or method according to claim 1 or 2, characterized in that: The coupling structure modulates state propagation through a restricted geometric path, wherein the modulation parameters of the restricted geometric path include path direction, curvature, width gradient, and local energy barrier; and the modulation parameters are used to modulate the phase, amplitude, temporal order, dwell probability, or path selection result of state propagation.

7. The system or method according to claim 1 or 2, characterized in that: The calculation is achieved through the coupling relationship between the spatial location defined by the restricted geometric path and the sequence label formed during the state propagation process. The sequence label is determined by the path structure, local coupling conditions, or local energy barrier distribution, and is used to map the calculation result into a deterministic or discriminable steady-state distribution.

8. The system or method according to claim 1 or 2, characterized in that: The system adopts a hierarchical structure, including a system-level structure, a device-level active region, and a local structural scale region. The hierarchical structure achieves decoupling between signal transmission and the computing core through physical isolation.

9. The system or method according to claim 1 or 2, characterized in that: The system may be constructed by a physical system or its model that satisfies the following conditions: capable of forming a distinguishable, maintainable, and moduloable collective state; capable of propagating and coupling the collective state in a restricted path; capable of providing computational results through final state distribution, triggering results, time distribution, or equivalent proxy quantities; the physical system or its model that satisfies the conditions for collective state formation, restricted path propagation, and final state determination may be magnetic, electronic, optical, or other implementation methods.

Images