Sensing, storing and computing integrated nerve probe and probe system
By integrating the induction electrode layer, MTJ array and base processing module in the neural probe, combined with neural network training, detailed recording of neural activities and fast time-domain response are achieved, solving the problems of signal delay and transmission loss in the prior art, and improving the efficiency of neural signal processing.
Patent Information
- Application Number
- CN202510454873.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-11
- Publication Date
- 2025-07-11
AI Technical Summary
The existing neural probes mainly focus on the perception and recording of neural signals, lack detailed records of neural signal movement and generation process, and there are problems of signal delay and transmission loss.
A sensor-memory integrated neural probe is designed, including a handle and a base. The handle area is recorded through densely arranged induction electrode layers and MTJ arrays, and the base area is secondaryly processed and stored, and integrated filtering, amplification, analog-to-digital conversion modules are combined with the learning and training of the neural network to realize real-time feedback and processing of signals.
The entire process of neural activity is recorded, the signal processing efficiency is improved, the signal delay and transmission loss problems of traditional probes are overcome, and the neural signals can be analyzed and processed in real time in the probe area.
Smart Images
Figure CN120297338A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of neural signal sensing and fast time-domain response, and particularly relates to an integrated sensing and computing neural probe and a probe system. Background Art
[0002] In biological tissues, the brain and the entire nervous system are one of the most complex and precise organs and systems in the body, and are highly efficient and low-power "processor + memory" products of continuous natural evolution and selection over tens of millions of years. However, human understanding of the brain and neural signals is still in a very preliminary stage of exploration. The research on biological brain tissues and neural mechanisms has always been a research hotspot and difficulty. Developing new types of neural probes to detect and regulate the activities of single or multiple nerves and further understand the operating mechanism of the brain is of great significance for promoting basic biological research, developing neural disease treatment methods, and realizing high-performance brain-like computing systems.
[0003] MEMS probes are detectors or probes manufactured based on microelectromechanical system (MEMS) technology, which is a technology that integrates tiny mechanical components, sensors, actuators, and electronic circuits onto a chip of tiny size. MEMS probes utilize the advantages of this technology to make the probes have the characteristics of miniaturization, integration, and high performance. Currently, silicon-based neural MEMS probes with high-density electrical leads manufactured using nanotechnology on this basis can achieve simple recording of neural signals after being implanted into organisms.
[0004] IMEC provides a neural probe using CMOS circuits. It uses a high-density electrode array and integrates in-situ buffering under each electrode. By activating different electrode parts of the probe, it can sense nerves in different regions of the brain. Currently, it can achieve high-precision recording of neural signal impulses. This neural probe only stays on the function of recording neural signal impulses. Although it can achieve high-precision recording of neural signal impulses, it cannot record the specific generation activities and operation processes of neural signals in detail, thus being hindered in detecting and regulating the activities of single or multiple neurons and studying the operating mechanism of the brain.
[0005] A research team from Tsinghua University proposed a multifunctional implantable probe based on thin-film light-emitting devices, using thin-film LEDs as wave sources to stimulate biological tissues, and then making perceptions through functional devices such as photodetectors and electrochemical sensors. This sensing method that relies on thin-film light-emitting devices to stimulate the generation of biological signals and then senses neural responses through functional devices first faces the problem of unstable sensing. In addition, due to the use of functional devices such as electrochemical sensors and optical sensors, this sensing method also has the problem of large time delay and cannot achieve fast time-domain response.
[0006] Generally speaking, current neural probes focus on the sensing and recording of neural signals, and various methods such as high-density electrode arrays and thin-film light-emitting devices are involved to achieve the functions of sensing and recording. However, in actual applications, most of them still have the inherent defects of low sensing accuracy and few achievable functions, and lack the detailed recording of the movement and generation process of neural signals.
[0007] Moreover, current neural probes mainly focus on the functions of sensing, processing, and transmitting neural signals. The extracted neural signals are mainly handed over to the next-level processing module to complete, and no signal analysis function is integrated in the probe area. This will form a traditional neural signal detection and regulation mode in which the probe area senses and transmits neural signals, the external module analyzes and processes neural signals, and the final terminal forms the result. This mode will inevitably lead to problems such as signal delay and large transmission loss. Summary of the Invention
[0008] The purpose of the present invention is to solve the above problems existing in the prior art, and proposes a neural probe and probe system integrating sensing, storage, and computing functions. It is a neural probe system that integrates sensing, storage, and computing functions based on detecting biological neural signals. The probe consists of a handle part and a base part. The handle part senses signals and records the movement of neural signals through a magnetic tunnel junction (MTJ) array. The base part performs secondary processing on neural signals, overcoming the drawback that conventional probes are only limited to rough recording. In addition, through the learning and training of neural networks, the obtained eigenvalue can be timely fed back to the probe area, and signal processing and storage can be realized in the probe area, greatly improving the integration and functionality of the probe. This probe system provided by the present invention has great significance in many aspects such as detecting, studying neural activities, and analyzing neural signals.
[0009] The technical solution of the present invention is as follows:
[0010] The present invention provides a neural probe integrating sensing, storage, and computing functions, comprising:
[0011] A handle part, the handle part area serves as a detection and recording unit, including two-layer interconnection structures, namely an upper sensing electrode layer and a lower signal recording and holding layer; the sensing electrode layer includes a densely arranged sensing electrode array and an in-situ amplification circuit, and the signal recording and holding layer includes an MTJ array and a signal holding circuit;
[0012] The base, where the base region serves as a storage and computing unit, includes a three-layer interconnection structure, namely an upper-layer circuit, a storage layer, and an addressing and computing layer; the upper-layer circuit processes signals from the handle region, including an amplification module, a filtering module, and an analog-to-digital conversion module; the storage layer includes storage units for storing neural signal eigenvalues, and the addressing and computing layer includes an addressing unit, a computing unit, and a control unit for querying and processing eigenvalues.
[0013] Further, the size of the sensing electrode is smaller than the size of nerve cells; the condition satisfied by the size of the sensing electrode is that the size of each nerve cell is greater than three times the size of a single electrode.
[0014] Further, the size of a single sensing electrode is any size within the range of 1μm * 1μm to 4μm * 4μm; for example, it can be 1 * 1μm, 2 * 2μm, 3 * 3μm, or 4 * 4μm, or any other value within the above range.
[0015] Further, for the electrode material, the sensing electrode is made of a material with low impedance and biocompatibility, including materials such as TiN, Pt, and iridium oxide.
[0016] Further, after the sensing electrode senses a neural signal, it first performs in-situ amplification on the tiny neural signal, then sends it into the MTJ array to record neural activities, and sends the discharge pattern signal into the signal holding circuit to hold the neural signal, causing a certain amount of delay compared to the initial state to better capture the signal and achieve a fast time-domain response to neural signals.
[0017] Further, the signal processed in the handle region is sent into the base region. First, the signal is amplified, then sent into the filtering module to remove clutter, distinguish local field potential and action potential signals, and then the signal is sent into the analog-to-digital conversion module for digital signal conversion.
[0018] The principle of using the filtering module to distinguish the two signals: LFPs (local field potentials) are in the low frequency range (<1khz), and Aps (action potentials) are in the high frequency range (0.3 - 10khz). Just filter them with high-pass and low-pass filters respectively.
[0019] Further, the storage unit can be a magnetoresistive random access memory device based on a magnetic tunnel junction, or a static random access memory or a dynamic random access memory device can also be used.
[0020] The present invention also provides a neural probe system integrating sensing and computing functions, including the neural probe described in any one of the above, and further including an external operation and storage module; the operation and storage module performs operation and analysis on the obtained neural signals, or performs training and learning of neural networks and brain-like algorithms on the obtained neural signals, extracts relevant feature values and training results, and feeds them back to the base region of the probe.
[0021] Further, the feature values obtained by the operation and storage module through algorithm training of neural signals are fed back to the base region, the control unit turns on the write control signal, and the feature value data is stored in the corresponding storage unit through the decoder; for the neural signals processed by the handle, the control unit turns on the write control signal, sends the target signal to the input addressing unit, and then sends it to the calculation unit together with the feature value data in the storage unit according to a certain weight to complete the arithmetic logic operation; finally, the data is output through the input addressing unit.
[0022] The above external operation and storage module uses an artificial neural network algorithm as the training model, including but not limited to a convolutional neural network algorithm. It also includes other artificial neural network algorithms or brain-like learning algorithms such as a feedforward neural network algorithm, a modular neural network algorithm, a radial basis function neural network algorithm, and a recurrent neural network algorithm.
[0023] Further, the operation and storage module uses a convolutional neural network algorithm as the training model. The neural signals processed by the probe form a parameter matrix, which serves as the input layer, forms a new feature map on the convolutional layer, and is output to the pooling layer for downsampling operation. The multi-dimensional feature maps extracted by the convolutional layer and the pooling layer are unfolded into one-dimensional vectors in the flattening layer and fed into the fully connected layer. After linearly combining them with the weight matrix, the output layer maps the final features to output feature values through the activation function.
[0024] For the external operation and storage module, the storage part can select DRAM and SRAM or other storage devices.
[0025] Advantages of the present invention:
[0026] (1) The present invention provides a neural probe system integrating sensing and computing functions. By utilizing the flipping of different MTJ devices caused by the difference in the firing order of neural signals during a single neural activity, a complete record of the neural activity is achieved through recording the flipping order of different MTJ devices, thereby realizing the function of recording neural activity. The dense sensing electrodes in the probe handle region contact and sense biological neural signals, and the in-situ amplification circuit below the electrodes amplifies the sensed neural signals. After amplification, the signals are sent into the MTJ array for recording the neural signal firing process. In addition, the signal holding circuit part delays the recorded neural pattern signals to better capture the signals, realizing a fast time-domain response to neural signals. After being processed in the handle region, the signals enter the base region. The base circuit first performs a filtering process on the signals to filter out clutter signals, and then sends them to the external operation and storage module through amplification and an analog-to-digital converter (ADC) for processing the obtained neural pattern signals, finally obtaining the detailed transmission process and operation process of the neural signals.
[0027] In addition, through the learning and training of neural networks, the eigenvalue of the relevant signal is extracted and timely fed back to the handle region. Therefore, after training and feedback are completed, the sensed neural pattern signal can be timely identified and analyzed in the probe handle region to determine whether it is the previously trained model, effectively improving the signal processing efficiency.
[0028] (2) The present invention also provides a training, learning, and feedback mechanism established between the external storage and operation module and the probe base region. It mainly feeds back the training results of the external module to the probe base region, enabling the relevant neural signals input from the handle to be directly processed in the base region. This solution realizes the feedback mechanism between the external module and the probe internal, enabling signal analysis and processing to be carried out inside the probe in the later stage without having to go through the external module for processing and directly reaching the terminal, greatly improving the transmission and operation efficiency.
[0029] (3) The technical solution of the present invention can record the whole process of a single neural activity, specifically depict the operation process of the neural activity, overcome the drawback that traditional neural probes can only perform simple impulse recording of neural signals, and also realize the retention of the read high-speed neural signals. Brief Description of the Drawings
[0030] Figure 1 is the structure diagram of the neural probe system;
[0031] Figure 2 is the structure diagram of the probe handle region;
[0032] Figure 3 is the structure diagram of the probe base region;
[0033] Figure 4 is the working flow chart of the probe;
[0034] Figure 5 Schematic diagrams of the operation of the storage unit, addressing unit, and computing unit;
[0035] Figure 6 Frame diagram of unit operation;
[0036] Figure 7 Convolutional neural network algorithm.
[0037] Figure 8 Schematic diagram of the neural activity discharge sequence. Detailed implementation manner
[0038] To further understand the present invention, the present invention will be further described below in conjunction with the accompanying drawings. The technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.
[0039] Neural probes play an important role in aspects such as neural activity recording, brain-computer interface research, and nervous system intervention. However, the current neural probes are still a device under traditional thinking, that is, they only play the role of recording neural signal impulses, and cannot completely record neural activities, nor can they perform analysis and processing of neural signals in the probe part.
[0040] Therefore, the present invention provides a neural probe and probe system integrating sensing, storage, and computing functions. For the neural probe, a high-density electrode array, high-frequency signal sampling, MTJ array, and signal holding circuit are used for the specific recording of neural activities.
[0041] At the same time, a storage layer and an addressing and computing layer are integrated on the handle area of the neural probe, and a filtering, amplification, ADC module, etc. are integrated on the base area. In addition, the main body part of the neural probe is connected to the subsequent operation and storage module, and the operation training results are fed back to the base area of the probe in real time. In this way, the neural probe integrating multiple modules can overcome many drawbacks of traditional neural probes and realize the efficient processing of biological neural signals.
[0042] Embodiment 1
[0043] This embodiment provides a neural probe and probe system integrating sensing, storage, and computing functions. The overall structure of the probe system is as Figure 1 shown.
[0044] The neural probe is mainly divided into two parts. First is the handle region which is inserted into the living organism as the detection and recording unit, and second is the base region which processes the neural signals sensed by the handle as the storage and operation unit.
[0045] For the handle region, as Figure 2 shown, it can be divided into two layers. The upper layer is the sensing electrode layer, whose size is smaller than that of nerve cells and serves as the sensing layer. The lower layer is the signal recording and holding layer, which mainly realizes the recording of a single neural activity by different flipping of the MTJ array caused by different firing sequences of neural signals. And a signal holding circuit is added to make it have a certain amount of delay compared with the initial state so that subsequent analysis and processing can be better carried out.
[0046] From Figure 2 it can be seen that the sensing electrode layer of the first layer includes a densely arranged sensing electrode array and an in-situ amplification circuit, and the size of the sensing electrode is smaller than that of nerve cells. Further, the size of a single sensing electrode is any size within the range of 1μm * 1μm to 4μm * 4μm, as long as it is ensured that the size of each nerve cell is more than 3 times the size of a single electrode. The material of the sensing electrode can be selected from materials such as TiN, Pt, iridium oxide, etc. with low impedance and biocompatibility.
[0047] The signal recording and holding layer of the second layer includes an MTJ array and a signal holding circuit. The in-situ amplified neural signal is sent into the MTJ array arranged perpendicular to the sensing electrode. A single neural activity will cause different sequences of neural signal discharges, thus resulting in different sequences of flipping of the MTJ array. As Figure 8 shown, the firing sequence diagram reflected by a single neural activity on the probe is given, which will cause the corresponding MTJ device to flip in this sequence. The MTJ device itself, as a storage device, records the sequence, that is, the neural activity, specifically and completely, and sends the recorded discharge mode signal into the signal holding circuit to hold the neural signal, making it have a certain amount of delay compared with the initial state so as to better capture the signal and realize the fast time-domain response of the neural signal.
[0048] For the base region, as Figure 3 shown, it can be divided into three layers. The first layer is the upper circuit which processes the signals from the handle region, including an amplification module, a filtering module. In addition, it has an analog-to-digital conversion module ADC to convert the analog signal into a digital signal for convenient subsequent processing.
[0049] The second and third layers are the storage layer and the addressing and calculation layer respectively, which serve as the training feedback area of the subsequent operation and storage module. The storage layer can be a magnetoresistive random access memory (MRAM) device based on magnetic tunnel junctions (MTJ), or a static random access memory (SRAM) or dynamic random access memory (DRAM) device.
[0050] The storage layer includes storage units for storing neural signal feature values. The addressing and computing layer includes an addressing unit, a computing unit, and a control unit for querying and processing feature values. As Figure 5 shown, it is a schematic diagram of the operation of the storage unit, the addressing unit, and the computing unit (ALU).
[0051] In addition to the probe, the probe system also includes an external operation and storage module, which can simply store the acquired neural signals and use relevant algorithms of neural networks and brain-like computing for training and learning, extract relevant feature values and training results, and timely feedback them to the probe area.
[0052] For the feature values obtained by the external operation and storage module through algorithm training on neural signals and fed back to the base area, as Figure 6 shown, the control unit turns on the write control signal, and stores the feature value data into the corresponding storage unit through the decoder. For the neural signals processed by the handle, the control unit turns on the write control signal, sends the target signal to the input addressing unit, and then sends it together with the feature value data in the storage unit to the ALU (arithmetic logic unit) according to a certain weight to complete the arithmetic logic operation; finally, the data is output through the input addressing unit.
[0053] For the external operation and storage module, the artificial neural network algorithm is used as the training model, taking the convolutional neural network algorithm as an example.
[0054] As Figure 7 shown, for the neural signals processed by the probe part, a parameter matrix of the neural signals is formed as the input layer. On the convolutional layer, through several convolutions and sliding operations on the input matrix, a new feature map is obtained. Further, the pooling layer performs downsampling on the output of the convolutional layer to reduce the size of the feature map and the amount of calculation. Then, the multi-dimensional feature maps extracted by the convolutional layer and the pooling layer are unfolded into one-dimensional vectors in the flattening layer and fed into the fully connected layer. The fully connected layer performs a linear combination of it with the weight matrix. Finally, the output layer maps the final features to output feature values through the activation function.
[0055] Embodiment 2
[0056] As Figure 1 shown is the overall structure of the probe system. This embodiment also provides the working process of the probe system as Figure 4 shown, and the specific process is as follows:
[0057] First, the induction electrodes in the handle area sense nerve signals. Since the electrode size is smaller than the cell size, multiple electrodes can be contained within a single cell. At this time, when a single nerve cell or multiple nerve cells are active, the generated electrical stimulation can be sensed by multiple electrodes simultaneously. A single nerve activity will cause single or multiple nerve signals to discharge in different sequences. After the induction electrodes sense this change, they first perform an in-situ amplification process on the tiny nerve signals, and then sending them into the MTJ array will cause the MTJ devices to flip in different sequences. As Figure 8 shown, a discharge sequence diagram reflected by a single nerve activity on the probe is given, which will cause the corresponding MTJ devices to flip in this sequence. The MTJ devices themselves, as storage devices, record the sequence, that is, the nerve activity, specifically and completely, and send the recorded discharge mode signal into the signal holding circuit to hold the nerve signal, resulting in a certain amount of delay compared to the initial state to better capture the signal and achieve a fast time-domain response to the nerve signal.
[0058] After being processed in the handle area, the signal is sent to the base area of the probe. The circuit in the base area first amplifies the signal and then sends it into the filtering module to remove clutter. Since they represent local field potential and action potential respectively, they need to be distinguished to facilitate subsequent separate processing. Secondly, the signal is sent into the ADC analog-to-digital conversion module for digital signal conversion. At this time, the processing of the nerve signal in the probe is completed.
[0059] The processed nerve signal is sent to an external operation and storage module, where two functions can be achieved. First, the obtained nerve signal is analyzed through operations, and the processed signal is sent to subsequent terminals (such as a PC). Second, the obtained nerve signal is used for training and learning of neural networks and related brain-like algorithms to extract the key feature values of the nerve signal and feedback them to the base unit of the probe. At this time, after the base receives the electrical signal processed in the handle, it can directly analyze the nerve signal based on the existing training results. If it conforms to the existing training model, the training results in the storage layer are directly transmitted to the subsequent terminal; if it does not conform to the existing training model, it is sent to the external operation and storage module.
[0060] The above description is only the preferred embodiment of the present invention and is not a limitation to the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, modifications, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.
Claims
1. A sensory-in-memory-integrated neural probe, characterized in that Comprising: A handle portion, the handle portion area serving as a detection and recording unit, including a two-layer interconnected structure, namely an upper sensing electrode layer and a lower signal recording and holding layer; the sensing electrode layer includes a densely arranged sensing electrode array and an in-situ amplification circuit, and the signal recording and holding layer includes an MTJ array and a signal holding circuit; A base portion, the base portion area serving as a storage and operation unit, including a three-layer interconnected structure, namely an upper circuit layer, a storage layer, and an addressing and calculation layer; the upper circuit processes the signals from the handle portion area, including an amplification module, a filtering module, and an analog-to-digital conversion module; the storage layer includes storage units for storing neural signal characteristic values, and the addressing and calculation layer includes an addressing unit, a calculation unit, and a control unit for querying and processing the characteristic values.
2. The integrated sensing, storage, and computing neural probe according to claim 1, wherein The size of the sensing electrode is smaller than the size of a nerve cell; the size of the sensing electrode satisfies that the size of each nerve cell is more than 3 times the size of a single electrode.
3. The integrated sensing and computing neural probe according to claim 2, wherein The size of a single sensing electrode is 1*1μm to 4*4μm.
4. The integrated sensing, storage, and computing neural probe according to claim 1, wherein The sensing electrode is made of a material with low impedance and biocompatibility, including TiN, Pt, and iridium oxide.
5. The integrated sensing, computing and storage neural probe according to claim 1, wherein, After the sensing electrode senses a neural signal, it first performs in-situ amplification on the tiny neural signal, and then sends it into the MTJ array to record neural activities, and sends the discharge pattern signal into the signal holding circuit to hold the neural signal, causing a certain amount of delay compared to the initial state to better capture the signal, achieving a fast time-domain response to neural signals.
6. The integrated sensing, storage, and computing neural probe according to claim 1, wherein The signal processed in the handle portion area is sent into the base portion area. First, the signal is amplified, then sent into the filtering module to remove clutter, distinguish local field potential and action potential signals, and then the signal is sent into the analog-to-digital conversion module for digital signal conversion.
7. A neural probe system integrating sensing and in-memory computing functions, characterized in that, Comprising the neural probe according to any one of claims 1-6, further including an external operation and storage module; the operation and storage module performs operation and analysis on the obtained neural signals, or performs training and learning of neural networks and brain-like algorithms on the already obtained neural signals, extracts relevant characteristic values and training results, and feeds them back to the base portion area of the probe.
8. The neuromorphic probe system integrating sensing and in-memory computing according to claim 7, wherein The characteristic values obtained by the operation and storage module through algorithm training of neural signals are fed back to the base portion area. The control unit turns on the write control signal, and stores the characteristic value data into the corresponding storage unit through a decoder; for the neural signals processed in the handle portion, the control unit turns on the write control signal, sends the target signal into the input addressing unit, and then sends it into the calculation unit together with the characteristic value data in the storage unit according to a certain weight to complete arithmetic logic operations; finally, the data is output through the input addressing unit.
9. The neuromorphic probe system integrating sensing and in-memory computing functions according to claim 7, characterized in that, The operation and storage module uses a convolutional neural network algorithm as the training model. The neural signals processed by the probe form a parameter matrix, which serves as the input layer, forms a new feature map on the convolutional layer, and is output to the pooling layer for downsampling operations. The multi-dimensional feature maps extracted by the convolutional layer and the pooling layer are unfolded into one-dimensional vectors in the flattening layer and passed into the fully connected layer. After linearly combining it with the weight matrix, the output layer maps the final feature to the output characteristic value through an activation function.