A ubiquitous computing system based on spatiotemporal terahertz wave manipulation and its implementation method
By using a ubiquitous computing system based on spatiotemporal terahertz wave control, and utilizing open air medium for computation to generate a coherent computing power field, the problem of computing devices relying on solid-state chips is solved. This achieves efficient computing power diffusion and integrated storage, supports multi-user parallel computing and high-speed data exchange, and reduces hardware costs and energy consumption.
Patent Information
- Application Number
- CN202611262582.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing computing devices rely on solid-state chips, which prevent computing power from being spread to open spaces. The separation of storage and computing leads to storage wall bottlenecks. The lack of high-speed interconnection between computing nodes creates computing power islands, making it impossible to achieve automatic compilation of high-level language algorithms to four-dimensional terahertz spatiotemporal parameters and real-time machine learning compensation.
A ubiquitous computing system based on spatiotemporal terahertz wave control is adopted, which performs calculations using an open air medium. A coherent computing field is generated through a terahertz wave source array and a metasurface phased array. Combined with holographic storage and terahertz wireless interconnection, a distributed computing network is constructed to realize the custom reconstruction of the computing field and high-speed data exchange.
It enables high-precision general mathematical operations in open spaces, breaks through the storage wall bottleneck, builds a high-speed data exchange network, supports multi-user parallel computing, reduces hardware costs and energy consumption, and achieves seamless integration and collaborative computing of computing resources.
Smart Images

Figure CN123114202A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of terahertz electromagnetic wave manipulation, free-space nonlinear optical computing, ubiquitous computing infrastructure, terahertz wireless data center interconnection, and optical storage cross-technology. Specifically, it relates to a complete terahertz computing system and its implementation method that is independent of solid-state computing chips, relies on open air media to realize general mathematical operations, and is equipped with distributed storage and high-speed network interconnection. Background Technology
[0002] Current mainstream computing devices are all subject to the following underlying constraints: First, computation must rely on solid-state carriers—Von Neumann chips, on-chip optical computing, and topological optical computing all use silicon, metals, or superconducting materials as computation carriers, and computing power cannot exist independently of physical devices; Second, computing power boundaries are fixed—all computing devices are physically confined, and computing power cannot diffuse into open spaces; Third, terminal computing power is a necessity—user terminals must be equipped with CPUs / GPUs, and hardware iterations consume a lot of resources; Fourth, existing terahertz computing devices are all sealed in optical dark boxes, relying on solid-state diffraction plates for computation, and cannot project computing power fields into open spaces; Fifth, storage and computation are separated—computation and storage are physically and logically separated, and computation results need to undergo additional conversion to be persisted, forming a "memory wall" bottleneck; Sixth, computing power is isolated—there is a lack of high-speed interconnection between computing nodes, making it impossible to form a collaborative computing power network.
[0003] A global patent and literature search revealed that no technology simultaneously achieves: automatic compilation of four-dimensional terahertz spatiotemporal parameters from high-level language algorithms, a complete closed-loop system with open air as the sole universal computing medium, a multi-node kilometer-level coherent superposition computing field, real-time compensation for atmospheric disturbances via machine learning, distributed terahertz holographic storage and hierarchical caching, and ≥1Tbps terahertz wireless interconnection and heterogeneous network fusion between nodes. This indicates a clear technological gap. Summary of the Invention
[0004] This invention aims to solve the following technical problems: break the underlying paradigm of "computing depends on solid-state chips" and provide a chip-free, open-space native computing power solution; solve the problems of instability of terahertz interference fields and drift of nonlinear computation accuracy in open atmospheres; realize millisecond-level dynamic computing field reconstruction of custom algorithms; construct a low-power, computing-free terminal-based universal computing power architecture; be compatible with multi-level distributed deployment in indoor, park, and city settings; achieve instantaneous persistent storage and cross-task reuse of computing results; and build high-speed data exchange channels and external network access between computing nodes.
[0005] A universal computing system based on spatiotemporal integrated terahertz wave control comprises nine sequentially linked subsystems: Computing power instruction encoding subsystem: It has a built-in pre-trained neural network compiler, compatible with Python and MATLAB. It receives user algorithms, input tensors and precision constraints, and automatically converts them into a four-dimensional terahertz spatiotemporal encoding instruction set containing amplitude Aᵢ, angular frequency ωᵢ, initial phase φᵢ and time delay τᵢ. It establishes a reversible isomorphic mapping model between tensor data and spatial computing power field. The compilation time of standard AI models (ResNet, BERT) is ≤0.05s, and the compilation time of custom algorithms is <1s. Terahertz wave computing power projection subsystem: It includes a 32~256-channel terahertz wave source array and a tunable metasurface phased array. After receiving encoded commands, it modulates the terahertz wavefront through the metasurface and projects a coherent substrate computing power field in an open three-dimensional space. The wave source operates at a frequency of 0.1~10THz and has a single-channel output power of 0.1~100mW. The metasurface uses liquid crystal, vanadium dioxide or graphene superatomic array to achieve 0~2π continuous phase adjustment and 0~100% amplitude adjustment, with a unit response speed of <1μs. The computational power field generation and maintenance subsystem dynamically adjusts the amplitude, frequency, and phase of each wave source to meet the global coherent phase criterion (the phase difference between any two wave sources fluctuates |Δφᵢⱼ(t)|<π / 100), generating and stably maintaining the computational power interference field of the spatiotemporal distribution matching algorithm; a single node covers a radius of 1~100m, and multiple nodes achieve nanosecond-level phase synchronization through GPS timing and optical fiber, extending coherent superposition to the kilometer level; Medium Response and Computation Execution Subsystem: Using normal-pressure natural air or argon-air mixture as the sole computing medium, it automatically completes general mathematical operations such as matrix multiplication, convolution, Fourier transform, differential equations and neural networks based on the equivalent second-order nonlinear polarization cross-coupling term of the medium, without requiring any solid-state computing chips throughout the process; The calculation result readout subsystem is equipped with a full-band low-noise miniature terahertz detector array, which collects and decodes the optical perturbation signal after calculation into multimodal readable data such as numerical values, images, or text. The overall power consumption of a single terminal is <10mW. Computational field regulation and optimization subsystem: Collects environmental parameters such as temperature, humidity, aerosols and refractive index across the entire domain, has a built-in machine learning medium equivalent nonlinear response function library, dynamically corrects wave field parameters in real time, and keeps the relative calculation error stable within 0.1%; a global radiation sensor monitors power density and automatically cuts off local wave sources when it exceeds the safety threshold; Multi-user collaboration and resource scheduling subsystem: It integrates three mechanisms of time division multiplexing (TDMA), frequency division multiplexing (FDMA) and code division multiplexing (CDMA) to allocate independent time windows, frequency sub-bands and orthogonal phase coding to concurrent users, realize crosstalk-free parallel operation of multi-user computing power subfields in the same space, and reclaim computing power resources immediately after the task is completed; The distributed computing power storage and caching subsystem includes terahertz holographic storage units and a hierarchical caching architecture. It directly writes computation results into a phase-change storage medium (preferably Ge2Sb2Te5) in the form of an optical field to achieve non-volatile persistent storage. The hierarchical caching architecture includes: an intra-node L1 ultrafast terahertz holographic cache (response time < 10ns), an inter-node L2 distributed storage pool (achieving multi-node storage resource sharing via terahertz interconnect links, with aggregated bandwidth ≥ 10Tbps), and an L3 external archive storage connected via a terahertz-fiber fusion gateway. The storage units directly interact with the computing field, enabling an in-memory computing interface that allows for "compute and store" operations. The density of a single storage unit is ≥ 1Tbit / cm². 2 The read / write rate is matched with the terahertz wave modulation rate; Terahertz computing power network interconnection subsystem: Based on terahertz wireless communication, a high-speed data exchange channel is built between nodes, with a single link rate ≥1Tbps and a latency <100μs within a range of 1–100 meters, supporting point-to-point and multi-point broadcast communication; it includes a terahertz-fiber-6G tri-mode heterogeneous network converged gateway to achieve seamless connection between the computing power field and the external Internet, IoT, and satellite networks; each node forms a distributed computing power resource catalog by periodically broadcasting its computing power status, supporting computing power resource discovery and intelligent scheduling; it supports a distributed collaborative computing framework to realize the distribution, parallel execution, and result aggregation of subtasks; it supports computing power field boundary routing and seamless switching to enable computing power field relay when users move across nodes.
[0006] Coherent Theory of Spatiotemporal Distribution of Computational Field The computing power field is decomposed into the base computing power field F0(R,t) (carrying the fixed mathematical structure of the algorithm) and the input data perturbation field ΔF(R,t) (carrying the user input tensor). The total field expression is: F_total(R,t) = F0(R,t) + ΔF(R,t) (1) The spatial observation vector R=(x,y,z), the coordinates of the i-th wave source Sᵢ, the propagation distance rᵢ=|R-Sᵢ|, the propagation delay τᵢ=rᵢ / c, and the single-path terahertz wave component can be expressed as: Fᵢ(R,t) = Aᵢ(t-τᵢ)·exp{j[ωᵢ(t-τᵢ) - kᵢ·(R-Sᵢ) + φᵢ]} (2) The global basis computing field is a coherent superposition of all wave sources: F0(R,t) = Σᵢ Aᵢ(t-rᵢ / c)·exp{j[ωᵢ(t-rᵢ / c) - kᵢ·(R-Sᵢ) + φᵢ]} (3) Where: Aᵢ is the time-domain amplitude of the i-th wave source, and the mapping operation coefficient; ωᵢ is the angular frequency, used for frequency division multiple access to distinguish independent users; φᵢ is the static initial phase, defining the spatial interference topology; kᵢ=ωᵢ·e / c is the terahertz wave vector, e is the propagation unit vector; and c is the propagation speed of the terahertz wave in normal pressure air.
[0007] Nonlinear computation theory of air medium When the terahertz total field acts on air molecules, the macroscopic polarization intensity expands to the second order (third order and above are negligible higher-order terms): P = ε0[χ (1) F_total + χ (2) _eff F_total⊗F_total] (4) Where χ (2) _eff is the equivalent second-order nonlinear polarizability of air under a strong terahertz field (derived from electric field-induced second-order nonlinearity and plasma-assisted four-wave mixing mechanisms). The output light field of the medium is proportional to the second-order polarization term, and expanding the square term yields: F_total 2 = F0 2 + 2F0·ΔF + ΔF 2 (5) Among them: F0 2 The background term is a constant basis term, which can be filtered out by differential detection; 2F0·ΔF is the core cross-coupling term, equivalent to matrix multiplication and scalar multiplication; ΔF 2 This is a higher-order noise term, controlled by |ΔF| / |F0|≤10. -2 To suppress it.
[0008] The optical mapping relationships of each mathematical operation are as follows: addition is achieved by linear coherent superposition of waves of the same frequency; matrix and scalar multiplication is directly corresponding to the spatial intensity distribution of cross-coupled terms; convolution is achieved by far-field interference of spherical waves (spatial multiplication is equivalent to frequency domain convolution); Fourier transform is naturally completed by Fraunhofer far-field diffraction; and differential equations are obtained by automatically matching the Helmholtz equation with the spatiotemporal evolution of the optical field.
[0009] Three standardized working modes of the system Mode 1 (Static Computing Field): Suitable for industrial quality inspection and standardized AI inference. After pre-storing the algorithm coding parameters, a stable computing field is generated with one click, with a latency of 1-10 meters and <1μs. Mode 2 (Dynamically Reconfigurable Computing Power): Suitable for custom algorithms in scientific research, the metasurface switches parameters at nanosecond speeds, updates the computing power field in real time, and outputs loss curves and activation heatmaps simultaneously; Mode 3 (Distributed Cascaded Computing Power): Suitable for large models and ultra-large tensor simulations, multi-node kilometer-level coherent superposition, task partitioning and parallelism, automatic aggregation of calculation results at wave field boundaries, and high-speed exchange of intermediate data of subtasks through terahertz interconnection networks.
[0010] System performance indicators Computational performance: Total latency in a 1–10 meter field is <1 μs, independent of algorithm complexity; Relative error in global steady-state computation is <0.1%; Storage performance: Holographic storage density ≥1 Tbit / cm³. 2 L1 cache response time <10ns, inter-node L2 storage pool aggregated bandwidth ≥10Tbps; Network performance: inter-node single link ≥1Tbps, latency <100μs (1~100 meters), heterogeneous gateway supports seamless integration with external networks; Terminal hardware: CPU / GPU eliminated, receiving terminal power consumption <10mW, hardware cost reduced by 80%~90%; Scalability: single node covers 100 meters, multiple nodes synchronously superimposed to expand to the kilometer level, computing power and storage capacity increase linearly with the number of nodes; Security performance: global power density ≤1mW / cm² 2 It is below the ICNIRP safety threshold.
[0011] The technical effects and advantages of this invention are as follows: 1. For the first time, a complete reversible isomorphic mapping of "algorithm-terahertz four-dimensional spacetime wave field" was realized, filling the gap in automatic compilation technology from high-level languages to open-space optical computing instructions; 2. For the first time, high-precision general-purpose mathematical calculations were completed using atmospheric pressure air as the sole universal computing medium, completely eliminating the dependence on solid-state computing chips; 3. A dual-layer (real-time environmental calibration + radiation safety protection) closed-loop protection mechanism keeps the calculation error in open spaces stable within 0.1%; 4. Triple multiple access reuse enables crosstalk-free parallel computing for multiple users in the same space, significantly improving the utilization rate of computing resources; 5. For the first time, terahertz holographic storage and hierarchical caching architecture were introduced into an open-space computing system, enabling instantaneous persistence and cross-task reuse of computation results, thus breaking the "memory wall" bottleneck; 6. For the first time, a high-speed data exchange network for computing power nodes based on terahertz wireless interconnection was constructed, with a single link rate of ≥1Tbps. Combined with a heterogeneous network fusion gateway, it enables seamless connection between the computing power field and the external network, breaking down the "computing power island". 7. The nine subsystems form a complete closed loop of "encoding-projection-computation-reading-storage-interconnection", realizing the integration of storage, computing and network, and the comparison files cannot be separated and combined to deny creativity; 8. The indoor prototype has verified the core functions, and the measured data is in high agreement with the theoretical predictions, proving that the technical solution has physical feasibility and industrialization prospects; 9. The power consumption at the terminal is less than 10mW, which saves more than 95% energy compared with traditional solutions, significantly reducing the consumption of computing hardware and energy. Attached Figure Description
[0012] Figure 1 : Schematic diagram of the overall nine-layer closed-loop architecture of the system; Figure 2 Schematic diagram of the three-stage principle of algorithm encoding - wave field projection - medium nonlinear operation; Figure 3 : Exploded view of the six-layer hardware structure of the computing power projection node; Figure 4 Schematic diagram of the distributed computing power storage and caching subsystem architecture; Figure 5 Schematic diagram of the terahertz computing power network interconnection subsystem architecture and heterogeneous network integration; Figure 6 System full-process time-series distribution diagram; Figure 7 Schematic diagram of outdoor multi-user distributed computing power field and multiple access reuse; Figure 8 Summary chart of measured data from the indoor prototype (including time delay, error, signal-to-noise ratio, stability, etc.); Figure 9 : Experimental data diagrams of discrete verification of each core module (compiler matching degree, metasurface phase / amplitude modulation characteristics, air nonlinear operation accuracy, environmental calibration response, holographic storage read / write performance, terahertz interconnect link rate). Detailed Implementation
[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and design various embodiments with various modifications suitable for a particular purpose.
[0014] Example 1: Indoor Prototype for Principle Verification (Core Example, Demonstrating the Physical Feasibility of the Technical Solution) 1.1 Sample Structure The indoor proof-of-concept prototype consists of the following components: Terahertz source: 0.5~2THz quantum cascade laser (QCL), single-channel output power 5mW, frequency continuously adjustable; Spatial light modulator: 64×64 liquid crystal tunable metasurface, unit phase adjustment range 0~2π (measured deviation <π / 20), amplitude adjustment range 0~100% (measured deviation <5%), response speed 0.6μs; The computing chamber is a sealed air testing chamber at normal temperature and pressure (60cm×60cm×60cm), with a built-in temperature and humidity control system (control accuracy ±0.5℃, ±5%RH). No inert gas is added, and only natural air is used as the computing medium. Detector array: 16×16 miniature Schottky barrier diode detector array, responsivity 1.2A / W, noise equivalent power NEP = 5×10⁻⁶ -13 W / √Hz; Control and readout system: Low-power FPGA control board (power consumption 1.2W), responsible for metasurface driving, data acquisition and preliminary decoding; Reference holographic storage unit: Ge2Sb2Te5 phase change thin film (100nm thick), used to verify the "calculate and store immediately" function.
[0015] 1.2 Experimental Procedure (1) Compiler verification: Submit a 10×10 matrix multiplication task y=A·x (matrix elements are randomly generated, ranging from -10 to 10, with 4 decimal places retained) to the compiler via the host computer, and record the four-dimensional encoding parameters {Aᵢ, ωᵢ, φᵢ, τᵢ} output by the compiler and the compilation time. (2) Wave field projection: The FPGA loads the encoded parameters onto the metasurface, which generates a basis computing interference field F0 with matching matrix A in the air cavity. At the same time, the input vector x is encoded as a perturbation field ΔF and injected (controlling |ΔF| / |F0|≈5×10). -3 ); (3) Nonlinear operation: The equivalent second-order nonlinear polarization of air molecules generates a cross-coupling term 2F0·ΔF, which forms a spatial light intensity distribution carrying the operation results in the cavity; (4) Results readout: The light intensity distribution after acquisition and calculation by the 16×16 detector array is decoded by the FPGA and restored to the output vector y_exp; (5) Holographic storage verification: The calculation results are simultaneously written into the Ge2Sb2Te5 phase change thin film in the form of a hologram, and then read back with terahertz time-domain spectroscopy to verify the storage fidelity; (6) Environmental disturbance test: The environmental parameters were gradually changed within the range of 2℃~35℃ and 20%~90%RH through the temperature and humidity control system. The system was run continuously for 72 hours, and the calculation error was recorded every 10 minutes to verify the stability of the system and the effectiveness of the environmental calibration.
[0016] 1.3 Measured Data 1.4 ; 1.5 Experimental Conclusions The above measured data prove that: (1) Normal-pressure natural air can be used as a general computing medium to stably complete linear algebra operations without the intervention of any solid-state computing chip; (2) The compiler can accurately map high-level language algorithms into four-dimensional terahertz spatiotemporal coding parameters with a mapping matching degree of >99%; (3) The environmental calibration system can stably control the error within 0.1% when the temperature and humidity change over a wide range; (4) Terahertz holographic storage can achieve "calculation completed and then stored", and the read and write performance meets the real-time requirements. The entire set of theoretical formulas, system architecture and core devices are all physically feasible and do not have the defects of pure theoretical imagination.
[0017] Example 2: Compiler Accuracy and Efficiency Verification Experiment 2.1 Experimental Design Using 10×10, 50×50, and 100×100 matrix multiplications as standard test tasks, the compiler's compilation time, output parameter accuracy, and the degree of matching between the wavefield interferometry pattern and the theoretical calculation were recorded. Each scale was repeated 50 times.
[0018] 2.2 Measured Data 2.3 ; 2.4 Experimental Conclusions The compiler's compilation time increases approximately linearly with the algorithm size, with a compilation time of <1s for a 100×100 matrix, meeting the requirements for real-time task scheduling; the interferometric pattern matching degree is >99%, proving that the algorithm-wave field isomorphic mapping theory is accurate and reliable.
[0019] Example 3: Verification Experiment of Tunable Metasurface Properties 3.1 Experimental Design A terahertz time-domain spectroscopy system (THz-TDS) with an extended module of a vector network analyzer was used to perform full-band (0.1~2.5THz) scanning tests on a 128×128 metasurface array. Tests included: unit phase / amplitude modulation characteristics, beam deflection angle, focusing efficiency, and response speed.
[0020] 3.2 Measured Data 3.3 ; 3.4 Experimental Conclusions The metasurface's performance indicators all meet or exceed the design specifications, proving that it can serve as the core actuator for terahertz wavefront modulation and meet the engineering requirements for open-space computing field projection.
[0021] Example 4: Experiment to verify the accuracy of nonlinear calculations in air medium 4.1 Experimental Design On the indoor prototype platform, the accuracy of the following operation types was tested respectively: (1) matrix multiplication (3×3, 5×5, 10×10); (2) one-dimensional convolution (kernel length 3, 5, 7); (3) two-dimensional Fourier transform (32×32 image); (4) differential equations (numerical solution of first-order ordinary differential equations). Each operation was tested 30 times, and the relative error was recorded.
[0022] 4.2 Measured Data 4.3 ; 4.4 Experimental Conclusions The computational error for all tested operation types was ≤0.11%, meeting the engineering requirements for high-precision general-purpose computing. The error increased slightly with the increase of the computational scale, which is consistent with theoretical predictions. This proves that the second-order nonlinear effect of air medium can stably complete various types of general-purpose mathematical operations.
[0023] Example 5: Verification Experiment of Real-time Environmental Calibration System 5.1 Experimental Design In a sealed test chamber, the following environmental disturbance sequence was applied through a temperature and humidity control system: (1) a step change from 25℃ to 35℃ (temperature increase of 10℃); (2) a step change from 35℃ to 15℃ (temperature decrease of 20℃); (3) a step change from 40% to 90%RH (humidity increase of 50%); and (4) a continuous sinusoidal change from 2℃ to 40℃ (period of 60 minutes). The error was monitored and the system response was calibrated throughout the process.
[0024] 5.2 Measured Data 5.3 ; 5.4 Experimental Conclusions Without a calibration system, changes in temperature and humidity can cause calculation errors to rise to 0.38%–0.62%. After enabling the machine learning calibration system, the steady-state error stabilizes at 0.06%–0.08%, and the calibration convergence time is 72–88 ms (<100 ms). This demonstrates that the real-time environmental calibration system can effectively suppress the impact of open atmospheric disturbances on calculation accuracy.
[0025] Example 6: Terahertz Holographic Storage Verification Experiment 6.1 Experimental Design Using a Ge2Sb2Te5 phase change thin film as the storage medium (100nm thick, deposited on a high-resistivity silicon substrate), the calculation results from Example 4 were written into the storage medium in the form of a holographic light field. Test indicators: (1) write success rate (1000 write tests); (2) read latency (the time interval from the completion of writing to the data being read out); (3) storage density (the amount of data that can be stored per unit area); (4) read / write cycle life.
[0026] 6.2 Measured Data 6.3 ; 6.4 Experimental Conclusions The terahertz holographic storage unit achieved or exceeded the design specifications in terms of write success rate, read latency, storage density, and cycle life, proving that the "calculate and store immediately" in-memory computing interface is technically feasible.
[0027] Example 7: Terahertz Wireless Interconnection Verification Experiment 7.1 Experimental Design Two terahertz transceiver terminals (operating frequency 300 GHz) were set up and placed at distances of 1 m, 10 m, 50 m, and 100 m respectively to test link speed, bit error rate, and transmission delay. The tests were conducted indoors under normal temperature and pressure conditions, and both transceivers used horn antennas (gain 25 dBi).
[0028] 7.2 Measured Data 7.3 ; 7.4 Experimental Conclusions Within a range of 110 meters, the single-link rate is >1Tbps and the latency is <100μs (42μs within 10 meters), meeting the requirements for high-speed data exchange between nodes. Within a range of 50-100 meters, the rate decreases but remains >0.8Tbps, which can be further optimized by increasing the transmit power or using a higher gain antenna.
[0029] Example 8: Multi-user Parallel Scheduling Verification Experiment 8.1 Experimental Design Within the same 10-meter computing field, three users are assigned three non-overlapping frequency sub-bands (User A: 0.95~1.00THz, User B: 1.05~1.10THz, User C: 1.15~1.20THz) and simultaneously execute three different types of operations: User A performs 50×50 matrix multiplication, User B performs image convolution (32×32), and User C solves a first-order differential equation. The accuracy of each user's operation results is checked, as well as the presence of cross-user signal crosstalk.
[0030] 8.2 Measured Data 8.3 ; 8.4 Experimental Conclusions During parallel computation, the computational error of each user is not significantly different from that during independent computation (deviation ≤ 0.01%), and the signal leakage in adjacent frequency bands is < -43dB, proving that the frequency division multiplexing mechanism can effectively realize crosstalk-free parallel computation among multiple users in the same space.
[0031] Example 9: Computing Power Instruction Encoding Example The user submits a 100×100 matrix multiplication y=A·x, with precision retained to 4 decimal places; the compiler maps the matrix elements to spatial phase φᵢⱼ, and the input vector to wave source amplitude Aᵢ, allocating independent 0.95~1.00THz frequency sub-bands; the complete compilation takes 0.8s (see Example 2).
[0032] Example 10: Example of Tunable Metasurface Preparation The substrate is 500μm high-resistivity silicon. A 128×128 superatomic array is prepared by electron beam lithography. A 5nm titanium and 100nm gold metal substrate are deposited. A liquid crystal modulation layer is spin-coated. Gold wire bonding control electrodes are used. The unit achieves 0~2π continuous phase and 0~100% amplitude adjustment, with a response speed of 0.6μs (see Example 3).
[0033] Example 11: Single-node computing power projection example The wave source center frequency is 1THz, and the single-channel output is 10mW; the metasurface completes three-dimensional wavefront modulation according to the coding parameters, the beamforming network generates a stable base computing field within a 10-meter range, the interference pattern corresponds to the matrix mathematical model, the pure air medium automatically completes matrix multiplication, and the total delay is 0.81μs (see Example 1).
[0034] Example 12: Nonlinear Calculation Example for Air Medium The base field F0 carries the spatial distribution of matrix A, and the input perturbation field ΔF carries the vector x; the equivalent second-order polarization of air molecules generates the cross-coupling term 2F0·ΔF, and the spatial distribution of output light intensity directly corresponds to the operation result of A·x (see Example 4).
[0035] Example 13: Distributed Readout Terminal and Holographic Storage Example 16×16 detector array, noise equivalent power NEP = 5×10 -13 W / √Hz, single-terminal power consumption 8mW, powered by a button battery; filtering only receives signals in the user's exclusive frequency band, and the decoding result is pushed to the mobile terminal via Bluetooth; simultaneously, the calculation result is written to the phase-change storage medium in the form of a hologram, with a write latency of <100ns and a storage density of 1.2Tbit / cm³. 2 (See Examples 1 and 6).
[0036] Example 14: Atmospheric Disturbance Calibration Example The system collects temperature, humidity and aerosol data at a frequency of 10 Hz, and substitutes them into the equivalent nonlinear response function library. When the temperature change exceeds 2 ℃, the phase of each wave source is automatically corrected, the convergence time is calibrated to 72~91 ms, and the global error is maintained within 0.06%~0.08% (see Example 5).
[0037] Example 15: Multi-user Parallel Scheduling Example Within the same 10-meter computing power field, three users are each assigned three non-overlapping frequency sub-bands (0.95~1.00THz, 1.05~1.10THz, and 1.15~1.20THz). The computing power sub-fields of each user overlap spatially but are spectrally isolated, and matrix multiplication, image recognition, and differential equation operations are performed synchronously without any result crosstalk (see Example 8).
[0038] Example 16: Kilometer-scale multi-node cascading and terahertz interconnection example The three computing nodes are spaced 1km apart and achieve nanosecond-level phase synchronization via GPS second pulses and optical fibers. High-speed data exchange between the nodes is achieved through a terahertz wireless interconnection link (center frequency 300GHz, single link rate 1.08Tbps@10m). The 10000×10000 super-large matrix task is divided into three sub-tasks, each node performs partitioned operations, and the sub-results are aggregated to the main node for output through the terahertz interconnection network. The overall error is uniformly <0.1%, and the data exchange delay between nodes is <100μs (see Example 7).
[0039] Example 17: Heterogeneous Network Convergence and External Access Example The computing nodes are connected to the Internet through a terahertz-fiber-6G tri-mode converged gateway; external users submit computing tasks through a standard REST API, the system automatically allocates the optimal computing node for execution, and the computing results are sent back to the user terminal through the gateway; at the same time, it supports the connection with the terahertz link of the satellite network to achieve computing power coverage in remote areas.
[0040] Example 18: Example of Computing Resource Discovery and Intelligent Scheduling Each computing node broadcasts its own status (coverage, load rate, available spectrum, and storage capacity) at a frequency of 1Hz through the terahertz interconnection network, forming a distributed computing resource catalog. After a new user submits a task, the system automatically selects the optimal node or node combination based on a deep reinforcement learning algorithm to achieve load balancing and maximize the utilization of computing resources.
[0041] Example 19: Seamless Switching of Computing Power Fields Across Nodes The user moves from the coverage area of node A to the coverage area of node B with a low-power receiving terminal (power consumption 7mW); the system synchronizes the user's computing status in real time through the terahertz interconnection network, and realizes the coherent superposition transition of the two nodes in the boundary area. The computing interruption time during the switching process is <50μs, which is imperceptible to the user.
[0042] Example 20: Radiation Safety Protection Example Sixteen radiation sensors across the entire area collected power density data at a frequency of 100Hz; the power density at any point exceeded 1mW / cm². 2 Immediately cut off the corresponding wave source; preset geofences for sensitive areas such as campuses and hospitals, and automatically reduce the power in overlapping areas to 0.1mW / cm. 2 the following.
[0043] Example 21: Industrial Quality Inspection Application Example The workshop's single-node computing power field covers a 10-meter quality inspection station and pre-stores ResNet-50 encoding parameters; industrial cameras encode images into perturbation fields at 30 frames per second and inject them into the computing power field, and the air medium completes defect classification in real time with a computation latency of 0.7μs; the classification results are simultaneously written to the holographic cache within the node (for subsequent model iteration training) and uploaded to the central quality inspection database via the terahertz Internet; compared with the GPU solution, energy consumption is reduced by 96% and equipment cost is reduced by 82%.
[0044] Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art and related fields based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention, unless otherwise specified or limited, shall be implemented according to conventional means in the art.
Claims
1. A universal computing system based on spatiotemporal integrated terahertz wave control, characterized in that, The system comprises a multi-channel terahertz source array, a globally tunable metasurface phased array, and a distributed terahertz detector array, which are sequentially linked in a closed-loop manner with the following subsystems: (1) Computing power instruction encoding subsystem, with built-in pre-trained neural network compiler, receives user algorithms, input data and precision constraints and automatically converts them into a terahertz spatiotemporal encoding instruction set containing four-dimensional parameters including amplitude Aᵢ, angular frequency ωᵢ, initial phase φᵢ and time delay τᵢ, and establishes a reversible isomorphic mapping model between tensor data and spatial computing power field. (2) Terahertz wave computing power projection subsystem, after receiving the encoded instructions, modulates the terahertz wavefront through a metasurface and projects a coherent substrate computing power field in an open three-dimensional space. (3) The computing power field generation and maintenance subsystem generates and maintains a computing power interference field that matches the algorithm in spatiotemporal distribution by dynamically adjusting the amplitude, frequency and phase of each wave source to meet the global coherent phase criterion; (4) The medium response and calculation execution subsystem uses normal pressure natural air or argon-air mixture as the calculation medium. It automatically completes general mathematical calculations based on the medium equivalent second-order nonlinear polarization cross-coupling term, and no solid-state computing chip is required throughout the process. (5) The calculation result readout subsystem is equipped with a full-band terahertz detector array to collect the optical disturbance signal after calculation and decode it into multi-mode readable data; (6) Computational field regulation and optimization subsystem: collects environmental parameters across the entire domain and has a built-in machine learning response function library to dynamically correct wave field parameters in real time to control calculation errors; at the same time, it monitors power density with radiation sensors and automatically cuts off local wave sources when the threshold is exceeded; (7) The multi-user collaboration and resource scheduling subsystem adopts a triple reuse mechanism of time division, frequency division and code division to allocate independent resources to concurrent users, so as to realize multi-user parallel operation without crosstalk in the same space; (8) Distributed computing power storage and caching subsystem, including terahertz holographic storage unit and hierarchical caching architecture, writes the calculation results into phase change storage medium in the form of light field to achieve non-volatile persistent storage; the hierarchical caching architecture includes intra-node L1 cache, inter-node L2 distributed storage pool and L3 external archive storage connected through heterogeneous gateway; (9) Terahertz computing power network interconnection subsystem, based on terahertz wireless communication to build a high-speed data exchange channel between nodes, including a terahertz-fiber-6G three-mode heterogeneous network fusion gateway to achieve seamless connection between computing power field and external network. Each node forms a distributed computing power resource catalog by periodically broadcasting computing power status, supporting distributed collaborative computing framework and seamless switching of computing power field across nodes.
2. The universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The terahertz wave source operates at a frequency of 0.1~10THz, with a single-channel output power of 0.1~100mW, and a single-channel frequency continuous tuning range greater than 10%.
3. The universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The tunable metasurface is composed of liquid crystal, vanadium dioxide or graphene superatomic array, and the unit can achieve continuous phase adjustment from 0 to 2π and amplitude adjustment from 0 to 100%, with a unit response speed of less than 1 μs.
4. A universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The operating medium is ambient air or a mixture of 0.5% argon and 99.5% air by volume.
5. A universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The terahertz detector has a responsivity greater than 1 A / W and a noise equivalent power (NEP) less than 10. -12 W / √Hz, and the receiver supports passive wireless power or button battery power.
6. A universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The computing power field regulation and optimization subsystem collects global environmental parameters at a frequency of not less than 10Hz, and the calibration convergence time is less than 100ms.
7. A universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The multi-user collaborative subsystem uses frequency division multiplexing to allocate non-overlapping frequency subbands, and the frequency band spacing is greater than the atmospheric nonlinear mixing bandwidth.
8. A universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The terahertz holographic storage unit employs phase-change metasurface or holographic light field modulation technology, with a single storage unit density greater than or equal to 1 Tbit / cm². 2 The write latency is less than 100ns.
9. A universal computing system based on spatiotemporal integrated terahertz wave control according to claim 1, characterized in that, The single-link rate of the terahertz computing power network interconnection subsystem is greater than or equal to 1Tbps, and the communication delay within the range of 1 to 100 meters is less than 100μs. The computing power projection node of the system includes a six-layer hardware structure: a phase change cooling power supply module, a terahertz wave source array, a tunable metasurface phased array, a beamforming network, a spatial sensing feedback detection array, and an optical fiber and 6G dual-mode synchronous communication interface.
10. A universal computing implementation method based on the system according to any one of claims 1 to 9, characterized in that, Includes the following steps: Step 1: The user submits a computing task through a smart terminal, augmented reality device, or brain-computer interface; Step two: The neural network compiler converts the high-level language algorithm into four-dimensional terahertz spatiotemporal coding parameters; Step 3: The metasurface completes beamforming based on the encoding parameters and projects the base computing field of the matching algorithm into the open space where the user is located. Step 4: Collect medium environment parameters across the entire domain and retrieve the nonlinear response function library to pre-correct the wave field to suppress accuracy drift; Step 5: Encode the input data into a weak perturbation field and inject it into the base computing field. Relying on the equivalent second-order nonlinear effect of air medium, all mathematical operations are automatically completed. Step six: The distributed receiver collects the processed optical disturbance signal and decodes it into standardized readable data; Step 7: The calculation results are simultaneously written to the terahertz holographic storage unit for persistent storage, and intelligently allocated to L1 cache, L2 distributed storage pool or L3 external archive storage according to the data access frequency. Step 8: During multi-node parallel computing, the results of subtasks are aggregated at high speed through a terahertz interconnection network, with a single link rate ≥ 1Tbps. Step 9: Push the calculation results in a multimodal format, optionally through a heterogeneous network convergence gateway to external network users; Step 10: After the task is completed, the corresponding wave source is cut off. The energy of the computing field naturally decays, computing resources are recycled and reused, and storage resources are dynamically managed according to the caching strategy.
