Vector information sensing system and sensing method

Abstract

The invention discloses a vector information sensing system and method, and the system comprises at least one group of sensing units, the output end of each group of sensing units is sequentially connected with a group of storage units and a group of calculation units, and each sensing unit comprises a scalar signal distribution input end, a horizontal vector output end and a vertical vector output end. The storage unit comprises a horizontal vector information storage subunit and a vertical vector information storage subunit which are respectively used for carrying out in-situ storage on horizontal vector signals and vertical vector signals to form two storage states, and the calculation unit comprises a horizontal vector information processing subunit and a vertical vector information processing subunit which are respectively used for carrying out in-situ storage on the horizontal vector signals and the vertical vector signals to form two storage states. The storage states of the horizontal vector signal and the vertical vector signal are respectively subjected to weighted calculation and summarized at an output end to obtain a calculation result, and the calculation result is used for classification or discrimination. According to the method, the problems of low calculation efficiency, high data redundancy and the like in the processing process of distribution of scalar signals such as images and the like are solved on the hardware level.

Description

A vector information sensing system and sensing method Technical Field

[0001] This invention relates to the field of integrated hardware technology for information sensing and in-memory computing, specifically to a vector information sensing system and sensing method. Background Technology

[0002] In various sensing and intelligent processing scenarios, system inputs are often not single scalar values, but rather scalar signal sets composed of a large number of scalar sampling points, such as two-dimensional pixel intensity distributions, pressure distributions, temperature distributions, potential or current distributions, and multi-point sound intensity distributions. With increasing sensor array density, sampling frequency, and real-time requirements, the size of scalar signal sets is growing rapidly, causing significant bottlenecks in bandwidth, storage, and energy consumption in traditional processing chains based on "sampling-transmission-storage-centralized computation." Existing methods typically read out the scalar signal distribution point by point and store it in scalar form, then perform feature extraction and recognition operations on an external processor. However, scalar signal distributions exhibit significant local correlations in space, with much effective information reflected in the differences and trends between adjacent sampling points, such as structural features like local gradients, boundaries, abrupt change locations, contours, or dynamic change directions. Using only a large number of raw scalar samples as input for subsequent calculations not only results in redundant data storage and handling but also increases the complexity of subsequent calculations and limits the system's real-time processing capabilities under high-resolution or high-frequency sampling conditions.

[0003] While some studies have attempted to vectorize scalar signal distributions at the algorithmic level to more compactly represent the aforementioned structural information, existing methods largely rely on software implementation and lack targeted hardware support and optimization, limiting their effectiveness in practical applications. Therefore, designing a vector information sensing system capable of sensing, vectorizing, storing, and computing scalar signal distributions such as images at the hardware level, and outputting the processing results, has become an important direction for further development of information processing technology. Vectorizing scalar signal distributions can reduce redundant data while improving processing speed and accuracy, bringing higher performance and broader application prospects to information processing technology. Summary of the Invention

[0004] Purpose of the Invention: The purpose of this invention is to provide a vector information sensing system and method that solves the problems of low computational efficiency and high data redundancy in the processing of scalar signal distribution at the hardware level.

[0005] Technical Solution: To achieve the above objectives, the vector information sensing system of the present invention includes at least one set of sensing units. Each set of sensing units has its output terminal sequentially connected to a set of storage units and a set of computing units. Each sensing unit includes an input terminal for a scalar signal distribution, a horizontal vector output terminal, and a vertical vector output terminal. The storage unit includes a horizontal vector information storage subunit and a vertical vector information storage subunit, which respectively store the horizontal and vertical vector signals in situ, forming two storage states. The computing unit includes a horizontal vector information processing subunit and a vertical vector information processing subunit, which respectively perform weighted calculations on the storage states of the horizontal and vertical vector signals and summarize the calculation results at the output terminal. The calculation results are used for classification or discrimination.

[0006] Preferably, the scalar signal includes, but is not limited to, optical pixel intensity, electrical signal amplitude, pressure value, temperature value, or sound intensity; the scalar signal distribution is the arrangement of the scalar signal at several discrete sampling positions, and the arrangement is a one-dimensional, two-dimensional, or three-dimensional distribution; the scalar signal distribution includes, but is not limited to, pixel light intensity distribution, electrical signal distribution, pressure distribution, temperature distribution, or sound intensity distribution.

[0007] Preferably, the horizontal vector output terminal and the vertical vector output terminal are respectively composed of a positive response sensor and a negative response sensor, wherein the output of the positive response sensor and the output of the negative response sensor are superimposed on the same common port to further obtain vector signals in the horizontal and vertical directions.

[0008] Preferably, the sensors in the sensing unit include, but are not limited to, optical sensors, electrical sensors, pressure sensors, temperature sensors, or sound sensors.

[0009] Preferably, the memory within the storage unit includes, but is not limited to, memristors, floating gate devices, and ferroelectric devices.

[0010] Preferably, the computing unit applies a preset readout voltage or readout current to the storage device in the storage unit, converts the storage state into a current or voltage response, and then uses a differential path to achieve the superposition or equivalent combination of positive and negative signals.

[0011] The vector information sensing method of the present invention includes dividing a scalar signal distribution into several local regions. For each local region, a set of sensing units is used to collect signal differences between different positions to form horizontal and vertical vector signals, which are then stored in situ using a storage unit. Further, a computing unit performs a weighted summation operation on the in-situ stored states of the horizontal and vertical vector signals of the local region and outputs the summation results to obtain the calculation results of the local region. Finally, the calculation results of all local regions are summarized for classification or discrimination.

[0012] Preferably, the sensing unit includes an input terminal for a scalar signal distribution, a horizontal vector output terminal, and a vertical vector output terminal. The horizontal vector output terminal and the vertical vector output terminal are respectively composed of a positive response sensor and a negative response sensor. The outputs of the positive response sensor and the negative response sensor are superimposed on the same common port to further obtain vector signals in the horizontal and vertical directions.

[0013] Preferably, the storage unit includes a horizontal vector information storage subunit and a vertical vector information storage subunit, which are electrically connected to the horizontal vector output terminal and vertical vector output terminal of the corresponding connected sensing unit, respectively, to store the horizontal vector signal and the vertical vector signal in situ, forming two storage states.

[0014] Preferably, the computing unit applies a preset readout voltage or readout current to the storage device in the storage subunit, converts the storage state into a current or voltage response, and then uses a differential path to achieve the superposition or equivalent combination of positive and negative signals.

[0015] Beneficial Effects: This invention has the following advantages: 1. At the hardware level, this invention vectorizes scalar signal distributions, such as image pixel intensity distribution, to obtain vector signals containing amplitude and direction information. This allows for the explicit expression of structural features such as local differences, boundaries, and trends in the scalar signal distribution in a compact form. While preserving and highlighting structural information, the vectorization encoding reduces redundant representation of the original scalar data, lowering the scale of subsequent storage and computation. 2. This invention supports in-situ storage and in-situ weighted summation of vector signals at the hardware level, reducing data transfer overhead between the sensing end, storage end, and processing end, improving computational speed and energy efficiency, thereby achieving efficient perception of scalar signal distribution. 3. This invention can employ a parallel array architecture composed of multiple sets of sensing units, storage units, and computing units to achieve large-scale parallel processing of local areas of scalar signal distribution. While preserving the spatial structural features of the signal, it significantly improves the system's real-time processing capability for high-resolution or complex signals. Attached Figure Description

[0016] Figure 1 is an information flow diagram of the vector information perception system of the present invention;

[0017] Figure 2 is an optical microscope photograph of the vector information sensing system according to an embodiment of the present invention, wherein the sensing unit is an adjustable tungsten diselenide field-effect transistor array and the storage unit is a silicon-based field-effect transistor-memristor array.

[0018] Figure 3 is a diagram illustrating the information processing process of the vector information sensing system according to an embodiment of the present invention.

[0019] Figure 4 is a circuit structure diagram of the adjustable tungsten diselenide field-effect transistor and a schematic diagram of vector information encoding in the vector information sensing system according to an embodiment of the present invention.

[0020] Figure 5 is a circuit structure diagram of a silicon-based field-effect transistor-memristor unit and a schematic diagram of in-situ vector information processing in the vector information sensing system according to an embodiment of the present invention.

[0021] Figure 6 is a result diagram of the vector information perception system according to an embodiment of the present invention performing image information sensing and vector encoding tasks;

[0022] Figure 7 shows the result of in-situ storage of vector information by the vector information sensing system according to an embodiment of the present invention.

[0023] Figure 8 shows the result of in-situ weighted processing of vector information by the vector information perception system according to an embodiment of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be described in detail below with reference to the embodiments and accompanying drawings.

[0025] Example 1

[0026] This embodiment provides a vector information sensing system, including a sensing unit, a storage unit, and a computing unit, as shown in Figure 1, which is an information flow diagram of the vector information sensing system.

[0027] The sensing unit includes an input port for a scalar signal distribution and two vector output ports in different directions, namely a horizontal vector output port and a vertical vector output port, thereby forming two output signals V. x With V y V x With V y For two signed signals in mutually perpendicular directions, V x With V y The scalar signal is characterized by local variations in the horizontal and vertical directions, respectively. These two components together constitute a vectorized representation of the local differences, where the sign represents the direction of change and the amplitude represents the intensity of change. The vector output in each direction consists of a positive response sensor and a negative response sensor. The outputs of the positive and negative response sensors are superimposed at the same common port to obtain a signed vector component. Specifically, the positive and negative response sensors in the horizontal direction share a horizontal vector information output port, and the positive and negative response sensors in the vertical direction share a vertical vector information output port. Vector information encoding in the horizontal and vertical directions is achieved through the superposition of positive and negative electrical signals.

[0028] The sensing unit possesses tunable external signal sensitivity, capable of generating positive and negative responses to external scalar signals. It receives scalar signal distributions, performs vector encoding, and outputs vector information. These scalar signals include, but are not limited to, optical pixel intensity, electrical signal amplitude, pressure values, temperature values, or sound intensity. The output of the sensing unit changes with external physical stimuli, thereby achieving real-time sensing of environmental information. The sensing unit can employ any sensor material, structure, and device type with positive and negative responses or equivalent positive and negative output capabilities, including optical sensors, electrical sensors, pressure sensors, temperature sensors, or sound sensors.

[0029] The storage unit includes a horizontal vector information storage subunit and a vertical vector information storage subunit, which are electrically connected to the horizontal vector output port and vertical vector output port of the sensing unit, respectively, thereby respectively storing V x With V y In-situ storage is performed, resulting in two storage states, characterized by equivalent conductance or equivalent resistance, denoted as G respectively. x With G y The storage unit in this embodiment can be any material, structure, and type of memory with non-volatile storage characteristics, including but not limited to memristors, floating-gate devices, and ferroelectric devices. These storage devices have long storage times and high stability, enabling them to support long-term storage of large amounts of information.

[0030] The computing unit comprises a horizontal vector information processing subunit and a vertical vector information processing subunit, which respectively process the stored state G. x With G y The system performs a weighted summation operation and aggregates the results at the output. It then applies a threshold determination or selects the maximum response based on these results to obtain an output response for classification or discrimination. The calculation unit has the capability to apply continuously tunable positive and negative weights to perform weighted summation operations on stored vector information and output the calculation results required for recognition.

[0031] Example 2

[0032] As shown in Figure 2, this embodiment provides a vector information sensing system, including a sensing unit based on a tungsten diselenide transistor, a storage unit and a computing unit based on a silicon-based field-effect transistor-hafnium oxide memristor, and a signal interconnection circuit between the three units. The sensing unit, storage unit, and computing unit are connected one-to-one according to the unit index, thereby forming a complete information closed loop from input sensing, vector encoding, in-situ storage, in-situ calculation, and recognition output.

[0033] In this embodiment, the sensing unit may include N groups, and each group of sensing units is sequentially connected to a group of storage units and a group of computing units. The number of units N may correspond to the number of local regions into which the scalar signal distribution is divided. In this embodiment, taking an image signal as an example, the light intensity distribution of image pixels is divided into several two-dimensional local regions, and each local region corresponds to a sensing unit and subsequent storage and computing units.

[0034] Taking optical image information as the input as an example, the image is projected onto a sensing array composed of sensing units in the form of light intensity distribution. As shown in Figure 3, each sensing unit performs vector encoding on its corresponding two-dimensional local region and outputs two vector component signals. As shown in Figure 4a, in this embodiment, the sensing unit is implemented using an tunable tungsten diselenide field-effect transistor array, wherein the response characteristics of the transistor are affected by the gate voltage V. g Adjustment allows similar devices to exhibit a positive response V under the same input. oc + Or negative response V oc — As shown in Figure 4b, each sensing unit has two output ports: a horizontal vector output port and a vertical vector output port, corresponding to the output V. x With V y In the horizontal direction, the positive response sensor and the negative response sensor within the sensing unit generate positive electrical signals V, respectively. x + With negative electrical signal V x — The vectors are then superimposed at the horizontal vector output port to form a signed horizontal vector component V. x The same mechanism is used to obtain V in the vertical direction. y V x With V y The sign is used to characterize the direction of local change, and its magnitude is used to characterize the intensity of change, thus converting local differences into vector information (V). x V y This enables sensing and vectorization encoding of input images.

[0035] The storage unit is used for storing the vector information (V) output by the sensing unit. x V y ) Perform in-situ storage, denoted as G x With G y The computing unit is used to process the stored vector information G. x With G y After weighting, the final output is the weighted summed current vector (I). x I y ), corresponding to the image recognition result, are used to perform information processing tasks such as image recognition and information classification.

[0036] The processing procedure of the computing unit is as follows: by applying a preset readout voltage or readout current to the storage device within the storage unit, the storage state is converted into a current or voltage response. These responses are superimposed at the aggregation node to obtain a weighted aggregated output. To achieve positive and negative weighting, the computing unit uses a differential path to superimpose or equivalently combine positive and negative signals, allowing the symbol information of the vector components to participate in the calculation. The weighted aggregated output can further perform threshold determination or maximum response selection, thereby outputting category labels or response values ​​for each category. The computing unit in this system can be an information processor of any material, structure, and type with positive and negative weighting capabilities.

[0037] As shown in Figure 5, in this embodiment, the storage and computing unit is implemented using a 1T1R unit composed of a silicon-based field-effect transistor and a hafnium oxide memristor as an example. As shown in Figure 5a, the N-type field-effect transistor-memristor unit is used to receive and store positive vector information (V > 0), and the P-type field-effect transistor-memristor unit is used to receive and store negative vector information (V < 0). After the vector information is stored, a preset readout voltage V is applied to the memristor terminal. in As a readout excitation and weighting modulation method, each cell generates an output current I related to its storage state. out The current of each unit is summarized at the output terminal to obtain the calculation result. The calculation result can be further fed into a comparator or selection circuit to perform threshold determination or maximum response selection, thereby outputting the recognition result. In order to express the symbol information of the vector at the hardware level, each vector component is stored and calculated by a different 1T1R unit. As shown in Figure 5b, when the horizontal component is a positive signal, the N-type transistor-memristor unit will be activated to form G. x + When the horizontal component is a negative signal, it will activate the P-type transistor-memristor unit to form G. x — The vertical components are also written in the same way to form G. y + With G y — The G mentioned x + G x — G y + With G y — This is a non-volatile storage state, which in this embodiment is the memristor conductance. A readout voltage V is applied to the horizontal path. Rx Apply a readout voltage V to the vertical path Ry This causes the conductance state to generate a corresponding output current, and outputs a current vector (I). x , Iy This method completes the weighted modulation of vector information. It enables efficient information processing within the storage device, avoiding the latency and energy consumption problems caused by frequent data transfers in traditional computing systems.

[0038] Example 3

[0039] To verify the vector sensing and in-situ calculation process described in Examples 1 and 2, this example demonstrates using square, circular, and triangular images as inputs. As shown in Figure 6, the sensing unit array composed of tungsten diselenide transistors generates different V values ​​for different image inputs. oc Spatial distribution reflects the local light intensity gradient of the image. After dividing the image into several local regions, the corresponding in-situ vector coding results can be obtained. The output of each local region (V...) x V y This forms a directional vector distribution in space, and vector distributions corresponding to different shapes are distinguishable. Further, as shown in Figure 7, the vector components are written into and maintained in a conductance state G in the positive and negative 1T1R units. x + G x — G y + With G y — This enables in-situ storage of vector information. Furthermore, as shown in Figure 8, a preset readout voltage V is applied to each 1T1R unit during the readout phase. Rx With V Ry Then, the corresponding output current component I can be obtained. x with I y And the net output current distribution, wherein the output current characterizes the weighted result of the stored vector information, and can be used as input for identification strategies such as threshold determination or maximum response selection, to realize the output of category label or response value.

[0040] In this invention, "perception" includes two stages: "sensing" and "recognition." Sensing refers to the physical conversion of external input signals into electrical signal outputs; recognition refers to the process of classifying, discriminating, or selecting based on the electrical signal or its storage state and outputting the results. "Vector encoding" refers to the process of converting local information in the input signal (e.g., differences between pixels in an image, changes in adjacent values ​​in a signal) into vector form. "In-situ computation" means that computation is completed within the array of storage cells, without needing to transfer vector information to an external processor. "Applying weights" means using the storage state or equivalent conductance of the storage device as an adjustable weight carrier, mapping the weights to an analog electrical response through readout voltage or readout current during readout. "Weighted summation" means summing the currents of multiple storage cells at a common node, or superimposing multiple voltages, to obtain a weighted summary output.

[0041] The vector information sensing system proposed in this invention is a system integrating sensing, storage, and computation. The sensing unit is not limited to tungsten diselenide field-effect transistor arrays; any device capable of adjustable positive and negative responses to external input signals and generating signed vector component outputs can be used to implement vector encoding. The storage unit is not limited to memristors; any device with non-volatile storage capability and capable of state retention and readout can be used instead. The computation unit is not limited to a 1T1R structure; any circuit or device array capable of positive and negative weight modulation and weighted summation output, and capable of cooperating with a decision circuit to output the recognition result, can be used. All the above equivalent substitutions do not depart from the spirit and scope of protection of this invention.

[0042] This vector information sensing system transforms pixel differences in images into spatial vector information, effectively compressing redundant data while preserving the spatial structure of the image, thus significantly reducing storage requirements and computational complexity. Simultaneously, vector encoding allows information processing to proceed in parallel across multiple dimensions, greatly improving computational efficiency. Furthermore, integrating sensing, storage, and computation functions onto a single hardware platform avoids latency and energy consumption issues caused by frequent data transmission, enabling rapid information sensing, storage, and processing. This efficient signal encoding and processing method not only improves the overall system performance but also enhances adaptability to changes in the external environment, making it suitable for fields such as real-time image processing, environmental monitoring, and intelligent devices, thus driving the development of information sensing systems towards greater efficiency and lower redundancy.

Claims

1. A vector information sensing system, characterized in that, The system includes at least one set of sensing units. Each set of sensing units has its output terminal sequentially connected to a set of storage units and a set of computing units. Each sensing unit includes an input terminal for a scalar signal distribution, a horizontal vector output terminal, and a vertical vector output terminal. Each storage unit includes a horizontal vector information storage subunit and a vertical vector information storage subunit, which store the horizontal vector signal and the vertical vector signal in situ, respectively, forming two storage states. Each computing unit includes a horizontal vector information processing subunit and a vertical vector information processing subunit, which perform weighted calculations on the storage states of the horizontal vector signal and the vertical vector signal, respectively, and summarize the calculation results at the output terminal. The calculation results are used for classification or discrimination.

2. The vector information sensing system according to claim 1, characterized in that, The scalar signal includes, but is not limited to, optical pixel intensity, electrical signal amplitude, pressure value, temperature value, or sound intensity; the scalar signal distribution is the arrangement of the scalar signal at several discrete sampling positions, and the arrangement is a one-dimensional, two-dimensional, or three-dimensional distribution; the scalar signal distribution includes, but is not limited to, pixel light intensity distribution, electrical signal distribution, pressure distribution, temperature distribution, or sound intensity distribution.

3. The vector information sensing system according to claim 1, characterized in that, The horizontal vector output terminal and the vertical vector output terminal are respectively composed of a positive response sensor and a negative response sensor. The output of the positive response sensor and the output of the negative response sensor are superimposed on the same common port to further obtain vector signals in the horizontal and vertical directions.

4. The vector information sensing system according to claim 1, characterized in that, The sensors in the sensing unit include, but are not limited to, optical sensors, electrical sensors, pressure sensors, temperature sensors, or sound sensors.

5. The vector information sensing system according to claim 1, characterized in that, The memory within the storage unit includes, but is not limited to, memristors, floating gate devices, and ferroelectric devices.

6. The vector information sensing system according to claim 1, characterized in that, The computing unit applies a preset readout voltage or readout current to the storage devices in the storage unit, converts the storage state into a current or voltage response, and then uses a differential path to achieve the superposition or equivalent combination of positive and negative signals.

7. A vector information sensing method, characterized in that, The scalar signal distribution is divided into several local regions. For each local region, a set of sensing units collects the signal differences between different locations to form horizontal and vertical vector signals, which are then stored in situ using a storage unit. The computing unit then performs a weighted summation operation on the in-situ stored state of the horizontal and vertical vector signals of the local region and outputs the summation results to obtain the calculation results of the local region. Finally, the calculation results of all local regions are summarized for classification or discrimination.

8. The vector information sensing method according to claim 7, characterized in that, The sensing unit includes an input terminal for a scalar signal distribution, a horizontal vector output terminal, and a vertical vector output terminal. The horizontal vector output terminal and the vertical vector output terminal are respectively composed of a positive response sensor and a negative response sensor. The outputs of the positive response sensor and the negative response sensor are superimposed on the same common port to further obtain vector signals in the horizontal and vertical directions.

9. The vector information sensing method according to claim 8, characterized in that, The storage unit includes a horizontal vector information storage subunit and a vertical vector information storage subunit, which are electrically connected to the horizontal vector output terminal and vertical vector output terminal of the corresponding connected sensing unit, respectively, to store the horizontal vector signal and the vertical vector signal in situ, forming two storage states.

10. The vector information sensing method according to claim 9, characterized in that, The computing unit applies a preset readout voltage or readout current to the storage devices in the storage subunit, converts the storage state into a current or voltage response, and then uses a differential path to achieve the superposition or equivalent combination of positive and negative signals.

Images