CN117743927A - Time sequence signal analysis method based on multi-mode sensing input library network - Google Patents
Time sequence signal analysis method based on multi-mode sensing input library network Download PDFInfo
- Publication number
- CN117743927A CN117743927A CN202311759196.8A CN202311759196A CN117743927A CN 117743927 A CN117743927 A CN 117743927A CN 202311759196 A CN202311759196 A CN 202311759196A CN 117743927 A CN117743927 A CN 117743927A
- Authority
- CN
- China
- Prior art keywords
- mode sensing
- time sequence
- fusion
- library network
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 21
- 230000004927 fusion Effects 0.000 claims abstract description 41
- 230000000946 synaptic effect Effects 0.000 claims abstract description 28
- 230000004044 response Effects 0.000 claims abstract description 27
- 230000003287 optical effect Effects 0.000 claims abstract description 22
- 238000005286 illumination Methods 0.000 claims abstract description 12
- 238000000034 method Methods 0.000 claims abstract description 12
- 230000015654 memory Effects 0.000 claims abstract description 10
- 239000002135 nanosheet Substances 0.000 claims description 14
- 230000007547 defect Effects 0.000 claims description 11
- SYYDNLBOFQOSGT-UHFFFAOYSA-N [Bi]=O.[Se] Chemical group [Bi]=O.[Se] SYYDNLBOFQOSGT-UHFFFAOYSA-N 0.000 claims description 9
- 239000000758 substrate Substances 0.000 claims description 9
- KRHYYFGTRYWZRS-UHFFFAOYSA-N Fluorane Chemical compound F KRHYYFGTRYWZRS-UHFFFAOYSA-N 0.000 claims description 6
- 230000005284 excitation Effects 0.000 claims description 6
- 229910000449 hafnium oxide Inorganic materials 0.000 claims description 6
- 238000003475 lamination Methods 0.000 claims description 4
- 230000014759 maintenance of location Effects 0.000 claims description 4
- 229910052751 metal Inorganic materials 0.000 claims description 4
- 239000002184 metal Substances 0.000 claims description 4
- WIHZLLGSGQNAGK-UHFFFAOYSA-N hafnium(4+);oxygen(2-) Chemical compound [O-2].[O-2].[Hf+4] WIHZLLGSGQNAGK-UHFFFAOYSA-N 0.000 claims description 3
- 239000010703 silicon Substances 0.000 claims description 3
- 229910052814 silicon oxide Inorganic materials 0.000 claims description 3
- 238000001039 wet etching Methods 0.000 claims description 3
- 230000001443 photoexcitation Effects 0.000 claims 1
- 238000004364 calculation method Methods 0.000 abstract description 9
- 230000033764 rhythmic process Effects 0.000 abstract description 7
- 238000013528 artificial neural network Methods 0.000 abstract description 4
- 230000010354 integration Effects 0.000 abstract description 4
- 230000000737 periodic effect Effects 0.000 abstract description 4
- 230000000306 recurrent effect Effects 0.000 abstract description 4
- 241000255581 Drosophila <fruit fly, genus> Species 0.000 abstract description 3
- 238000004422 calculation algorithm Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000006403 short-term memory Effects 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 238000003909 pattern recognition Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 239000008358 core component Substances 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000002964 excitative effect Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000010365 information processing Effects 0.000 description 2
- 239000002055 nanoplate Substances 0.000 description 2
- 230000001242 postsynaptic effect Effects 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 230000000638 stimulation Effects 0.000 description 2
- 229940060587 alpha e Drugs 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000004791 biological behavior Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000002060 circadian Effects 0.000 description 1
- 230000027288 circadian rhythm Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 239000002064 nanoplatelet Substances 0.000 description 1
- 230000009022 nonlinear effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 210000000225 synapse Anatomy 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Landscapes
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention discloses a time sequence signal analysis method based on a multi-mode sensing input library network, and belongs to the technical field of sensing and memory calculation integration and intelligent recognition. The method utilizes a multi-mode sensing input library network to correspond to nonlinear responses of a plurality of same-type multi-mode sensing fusion synaptic devices biased by different temperatures; the input time sequence signals are respectively applied to the multi-mode sensing fusion synaptic devices in the form of optical signals, photo-generated current response is generated along with the application of illumination, channel current values of the devices are used as library node states in a multi-mode sensing input library network, when illumination and temperature are changed simultaneously, channel current changes in real time along with the input of the two modes, the expression of fusion characteristics of the photo-thermal signals is generated, and the fitted time sequence signals are output through linear weight weighting. The invention successfully simulates the biological rhythm of the drosophila when the drosophila is in the coexistence of periodic illumination, periodic temperature and the periodic temperature, and is better than the fitting performance of the same-scale multi-layer perceptron and the recurrent neural network.
Description
Technical Field
The invention relates to a time sequence signal analysis (multimodal temporal signal analysis algorithm based on parallel multisensory reservoir computing) method based on a multi-mode sensing input library network, and belongs to the technical field of sensing and memory calculation integration and intelligent recognition.
Background
The multi-mode sensing technology realizes comprehensive sensing of complex environments by integrating information acquired by various sensors. And processing the multi-source time sequence data, and mining the internal association among the multi-mode data through joint analysis to acquire more comprehensive context description on environmental characteristics, so as to improve the understanding and decision-making force of the system on the event. The traditional time sequence signal analysis technology mainly focuses on single-mode time sequence data, adopts algorithms such as decision trees, support vector machines, random forests and the like to classify the data, or uses models such as linear regression, ARIMA models, long-short-term memory networks and the like to model and predict the data, is limited by data isomerism and redundancy when multi-mode input is processed, cannot fully extract nonlinear characteristics among cross-mode data and mine potential information among different-mode data, and greatly limits the calculation efficiency and processing performance of an information processing system on large-scale multi-mode data streams. In contrast, the multi-mode sensing synaptic device is used for collecting and fusion analysis of environmental data in real time, so that association mining and collaborative analysis between space-time concurrent data can be promoted, and a solid foundation is provided for subsequent feature extraction and pattern recognition. However, to accomplish analysis of the multi-modal fusion features of the fuzzy expression, the information processing algorithm is required to have feature analysis capabilities which are strongly related to hardware. The library network calculation realizes the function of extracting high-dimensional nonlinear characteristics from input through the nonlinear conversion capability in the development dynamics system, and has the surprise potential in the field of time sequence data processing, so that the novel hardware and architecture design based on the library network algorithm is urgently required to be explored to realize the efficient analysis of the multi-mode time sequence signals.
Disclosure of Invention
The invention provides a time sequence signal analysis method realized by utilizing a multi-mode fusion sensing synaptic device, which realizes high-efficiency integration and deep analysis of multi-mode time sequence data.
The technical scheme provided by the invention is as follows:
a time sequence signal analysis method based on a multi-mode sensing input library network comprises the following specific steps:
1) The multi-mode sensing input library network corresponds to nonlinear responses of a plurality of same-type multi-mode sensing fusion synaptic devices biased by different temperatures;
2) The input time sequence signals are respectively applied to the multimode sensing fusion synaptic devices in the form of optical signals, the devices generate photo-generated current response along with the application of illumination, and the channel current of the devices is expressed as I (t) =I (t-1) ·αe -t/τ +f (u (t), I (t-1)), where u (t) represents the optical signal input at the current time t, I (t) represents the device channel current under optical excitation at the current time t, I (t-1) represents the device channel current at the time immediately before optical excitation, αe -t/τ A memory retention factor of e exponential decay, f represents nonlinear response current of the device to the current light input based on a channel current state at a time t-1;
3) And taking the channel current value I (t) of the device as a library node state in the multi-mode sensing input library network, generating the expression of fusion characteristics of the photothermal signals by changing the channel current value in real time along with the input of two modes when illumination and temperature are changed simultaneously, and outputting a fitted time sequence signal through linear weight weighting.
Further, the multi-mode sensing fusion synaptic device comprises a back gate substrate with a dielectric layer covered on the surface and a nanosheet channel layer for sensing, an interface defect layer is arranged between the nanosheet channel layer and the dielectric layer, and source electrodes and drain electrodes are arranged on two sides of the nanosheet channel layer.
The nano-sheet channel layer is made of bismuth-oxygen selenium material, and the thickness is in the range of 5-20 nm. The substrate is a heavily doped silicon/hafnium oxide substrate wherein the hafnium oxide layer has a thickness in the range of 10-20 nm. The source electrode and the drain electrode are composed of Ti/Au metal lamination layers, the total thickness is not more than 100nm, and the interface defect layer is introduced by hydrofluoric acid wet etching.
The invention has the following advantages:
1) According to the invention, the multi-mode fusion sensing synaptic device is utilized, the characteristic analysis capability of the library network for calculating the strong correlation with hardware is exerted, the correlation mining is carried out on space-time concurrency data, the extracted fusion characteristic is expressed in a fuzzy manner, and the efficient integration and the deep analysis of the multi-mode time sequence data are realized.
2) The multi-modal fusion sensing synaptic device has multi-modal channel input capability, allows parallel input of multi-source information, fully allows potential nonlinear characteristics among space-time concurrent data to be subjected to joint analysis, greatly improves calculation parallelism, and has extremely high instantaneity by adopting a multi-library network parallel calculation mode in a library network calculation architecture.
3) The multi-mode fusion sensing synaptic device has wide response capability for multi-mode stimulation, so that the multi-mode fusion sensing synaptic device can dynamically and flexibly adapt to common input of different sub-modes or multiple modes, and the output captures the remarkable characteristics of each mode, so that the multi-mode fusion sensing synaptic device can easily cope with extreme cases such as mode deletion and the like.
Drawings
FIG. 1 is a schematic diagram of a multi-modal fusion sensing synaptic device for use with the present invention;
FIG. 2 is a flow chart of the present invention, wherein P represents an input optical signal, M represents adding a time-sharing mask to the input, T 1 ～T N Representative of different temperatures τ 1 ～τ N Representing relaxation times of the device to light input at different temperatures;
FIG. 3 is a current response of a multi-modal fusion sensing synaptic device for the same single pulse of light at different temperatures, EPSC is an abbreviation for excitatory (excitatory postsynaptic current, EPSC) postsynaptic current, I ph Representing photocurrent, wherein (a) is the change of channel current of the multi-mode fusion sensing synaptic device with time under the same light pulse at different temperatures, and (b) is the value of current relaxation time and pulse first point photocurrent at different temperatures extracted from (a);
FIG. 4 is a graph of current response and photocurrent response of a multi-modal fusion sensing synaptic device at different temperatures and different light intensities, with (a) being the graph of current response at different temperatures and different light intensities and (b) being the photocurrent response at different temperatures and different light intensities;
FIG. 5 shows the input and output results of the present invention in a waveform classification task, and the response currents of the device to the optical input of the classification task at different temperatures, with FIG. (a) showing the response currents of the device to the optical input of the waveform classification task at different temperatures, and FIG. (b) showing the input, target output and fitting results of the task;
fig. 6 is a schematic diagram of input, implementation and output of the present invention in a biorhythm simulation task. Wherein CLR of the graph (c) is an abbreviation of circadian rhythm (Circadian Locomotion Rhythm, CLR). The graph (a) is input in a multi-mode environment such as photo-thermal environment, the graph (b) is used for achieving multi-mode fusion characteristic extraction and fusion signal analysis, and the graph (c) is target output of a biological rhythm simulation task, namely the biological rhythms of the drosophila under different periodic photo-thermal combinations.
FIG. 7 is a comparison of the performance of the present invention with a conventional Multi-modal fusion algorithm, wherein MLP is an abbreviation for Multi-Layer perceptron (MLP), RNN is an abbreviation for recurrent neural network (Recurrent Neural Network, RNN), LSTM is an abbreviation for Long-Short-Term Memory (LSTM), and RC is an abbreviation for library network computation (Reservoir Computing, RC). FIG. A shows the comparison of the fitting error of the present invention with the conventional multi-modal fusion algorithm, and FIG. B shows the specific fitting waveform output of the present invention with the conventional multi-modal fusion algorithm;
in the figure: 1-a dielectric layer; 2-a substrate; 3-a source electrode; 4-a drain electrode; 5-two-dimensional nanoplatelet channel layers.
Detailed Description
Specific implementations and capabilities of the present invention are described in detail below by way of example with reference to the accompanying drawings.
The realization of the invention relies on a multi-mode fusion sensing synaptic device, the structure of the multi-mode fusion sensing synaptic device is shown in figure 1, a nano-sheet channel layer positioned on a substrate and a medium layer is a core component of the multi-mode fusion synaptic sensor for sensing external light and heat stimulus, an interface defect layer positioned between the nano-sheet channel and the medium layer is a core component for realizing short-time memory of external light and heat stimulus, and the nano-sheet channel layer is a layered bismuth oxygen selenium nano-sheet with the thickness within the range of 5-20 nm. The source drain (S/D) electrodes on two sides of the bismuth oxygen selenium nano-plate are used for collecting response current of the device reflecting external stimulus, the two-dimensional layered bismuth oxygen selenium nano-plate has sensitivity to optical and thermal environment stimulus, when the optical stimulus is applied to a bismuth oxygen selenium channel layer, light provides energy for electrons in the channel to break an original equilibrium state, and electrons are excited to transition to enter the bismuth oxygen selenium conduction band bottom, so that fermi energy level is moved upwards. Therefore, under illumination, defects of higher energy levels are filled with electrons, and when the illumination is removed, the newly filled electrons are slowly released, thereby realizing short-term memory relaxation of the illumination signal. When thermal stimulus is applied to the bismuth oxygen selenium channel layer, according to the relationship between the fermi level of the n-type heavily doped semiconductor and the temperature, the fermi level is reduced due to the temperature rise, so that electrons originally filled in the higher defect level are released at high temperature, and when the temperature is restored to the initial state, the released electrons are captured by the defect again along with the gradual rise of the fermi level, so that the short-term memory relaxation of thermal signals is realized. And because the defect of interacting with electrons under thermal stimulus is located at a deeper energy level, the memory time of the device for thermal signals is longer than that of optical signals.
The schematic diagram in the multimode bias operation mode is shown in fig. 2, the multimode library network is composed of a plurality of multimode sensing fusion synaptic devices of the same type biased by different temperatures, the optical time-varying signals applied to the global are collected by all devices simultaneously, and thus nonlinear transformation is performed by means of the dynamic characteristics of the devices in different states, and then the obtained transformation results are subjected to linear weighting to fit the target output.
FIG. 3 is a graph of current responses of a multi-modal fusion sensing synapse device for the same single light pulse at different temperatures. From the graph, under the bias of different temperatures, light pulses with the same intensity and duration are applied to the device, the channel current of the device shows significantly different dynamic responses, the relaxation time of the current response pulse is continuously shortened along with the rise of the temperature, and the photocurrent is gradually reduced. Figure 4 further discusses the nonlinearity exhibited by the response amplitude of the device as a function of the applied excitation intensity. From this graph, it can be seen that the device current response exhibits different power exponent relationships for different light intensities at different temperature biases.
The multi-mode sensing fusion synaptic device provided by the embodiment of the invention adopts the back gate substrate to heavily dope silicon/hafnium oxide (p++ Si/HfO) 2 ) The thickness of the hafnium oxide layer is within the range of 10-20nm, the source electrode and the drain electrode are composed of Ti/Au metal lamination layers, the total thickness is not more than 100nm, the thickness of the bismuth oxygen selenium nano-sheet of the channel layer is within the range of 5-20nm, and the interface defect layer is introduced by hydrofluoric acid wet etching.
The illumination signals are applied to the multi-modal sensing fusion synaptic devices through devices such as a single-color LED light source, a random signal generator and the like, the temperature signals are applied by using a thermoelectric system consisting of a semiconductor refrigerating sheet and a high-temperature module, and the multi-modal sensing fusion synaptic devices of the same type in the multi-modal sensing input library network are set to be biased at different temperatures.
Taking a waveform classification task as an example, a waveform u (t) to be classified consists of cosine waves and square waves, each waveform consists of 8 data points, and the output y (t) to be fitted is that the classification result represents the square waves by using a label 1, and the label 0 represents the cosine waves. The library network that accomplishes this task is 4 x 10 in size, where 10 represents the overall library network comprising 10 sub-libraries, in this task 10 sub-libraries correspond to the nonlinear responses of 2 devices biased at 5 temperatures, respectively, 20 ℃,30 ℃,40 ℃,50 ℃,60 ℃ etc., and 4 represents 4 virtual library nodes contained in each sub-library, the library nodes being obtained by multiplying the task input by a mask of length 4, which is a random combination of 0, 1.
Task inputs u (t) are applied to the devices in the form of optical signals, respectively, the devices generate photo-generated current responses with the application of illumination, and the current device channel current can be expressed as I (t) =i (t-1) ·αe due to the short-term memory of the optical signals possessed by the devices -t/τ +f (u (t), I (t-1)), where u (t) represents the optical signal input at the current time tI (t) represents the current of the device channel under the light excitation at the current time t, I (t-1) represents the current of the device channel at the time before the light excitation, and alpha e -t/τ And (2) the memory retention factor of e exponential decay, wherein f represents the nonlinear effect of the device on the channel current on the response of the input optical signal u (t) based on the channel current state at the time t-1. Memory retention factor αe when the device is biased at different temperatures -t/τ Will change with temperature because the temperature rise provides energy for electrons to cross the defect barrier such that the time constant τ decreases with temperature T. Meanwhile, temperature will affect the mobility of channel carriers, according to the current density formula j=qμne, where q is the single electron charge amount, μ is the carrier mobility, n is the carrier concentration, and E is the channel electric field strength, so temperature will also significantly affect the channel current I (t).
Taking the device channel current value I (t) as the library node state in the library network, the library node state will become more abundant with the addition of different temperature biases, and the 4 x 10 library nodes obtained via device current sensing will be weighted via linear weights in the linear sense layer to fit the output y (t). Fig. 5 shows the input and output results of the waveform classification task according to the embodiment of the present invention, and the response currents of the device to the waveform classification task optical input at different temperatures, as known from the response currents of the device to the waveform classification task optical input at different temperatures in fig. (a), the diversity of nonlinear transformation of the optical input signal is enriched by different temperature biases, and the task input, target output and fitting results shown in fig. (b) prove that the present invention has excellent performance in classical pattern recognition task waveform classification for library network calculation, and the error value nrmse=0.08, which is lower than all reports at present.
FIG. 6 shows a schematic diagram of a biological rhythm simulation task performed in a multi-modal fusion mode of operation using the present invention. The living beings receive the stimulation of the external environment, especially the input of the periodical fluctuation environment variables such as photo-thermal and the like, and a set of biological rhythms which synchronously change with the external environment is constructed in the living beings, so that the self-behaviors are optimally coordinated and adapted to the external environment. Similarly, the invention can accept the same light and heat multi-mode input as biology, and the fitting of the target optimal biological behavior rhythm is completed through a multi-mode sensing input library network consisting of a plurality of multi-mode fusion sensing synaptic devices of the same type.
The richness of nonlinear transformation which can be realized by the method is increased, the diversity of library network nodes is greatly expanded, the method is compared with the traditional multi-modal fusion algorithm, and the result of fig. 7 shows that the method is superior to the multi-layer perceptron and recurrent neural network with the same scale in fitting performance and slightly inferior to a long-term and short-term memory network which is far higher than the algorithm in network complexity. The method has remarkable advantages for improving the pattern recognition accuracy and the chaotic sequence prediction accuracy of the library network calculation in the time sequence signal processing.
While the invention has been described in terms of preferred embodiments, it is not intended to be limiting. Any person skilled in the art can make many possible variations and modifications to the technical solution of the present invention or modifications to equivalent embodiments using the methods and technical contents disclosed above, without departing from the scope of the technical solution of the present invention. Therefore, any simple modification, equivalent variation and modification of the above embodiments according to the technical substance of the present invention still fall within the scope of the technical solution of the present invention.
Claims (7)
1. A time sequence signal analysis method based on a parallel multi-mode sensing input library network comprises the following specific steps:
1) The multi-mode sensing input library network corresponds to nonlinear responses of a plurality of same-type multi-mode sensing fusion synaptic devices biased by different temperatures;
2) The input time sequence signals are respectively applied to the multimode sensing fusion synaptic devices in the form of optical signals, the multimode sensing fusion synaptic devices generate photoproduction current response along with the application of illumination, and the channel current of the devices is expressed as I (t) =I (t-1) ·αe -t/τ +f (u (t), I (t-1)), where u (t) represents the optical signal input at the current time t, I (t) represents the device channel current under optical excitation at the current time t, I (t-1)) Representing the device channel current, αe, at the time immediately prior to photoexcitation -t/τ A memory retention factor of e exponential decay, f represents nonlinear response current of the device to the current light input based on a channel current state at a time t-1;
3) And taking the channel current value I (t) of the device as a library node state in the multi-mode sensing input library network, generating the expression of fusion characteristics of the photothermal signals by changing the channel current value in real time along with the input of two modes when illumination and temperature are changed simultaneously, and outputting a fitted time sequence signal through linear weight weighting.
2. The method for analyzing time sequence signals based on the parallel multi-mode sensing input library network according to claim 1, wherein the multi-mode sensing fusion synaptic device comprises a back gate substrate with a dielectric layer covered on the surface and a nano-sheet channel layer for sensing, an interface defect layer is arranged between the nano-sheet channel layer and the dielectric layer, and source electrodes and drain electrodes are arranged on two sides of the nano-sheet channel layer.
3. The method for analyzing time sequence signals based on the parallel multi-mode sensing input library network according to claim 2, wherein the back gate substrate is a heavily doped silicon/hafnium oxide substrate.
4. The time sequence signal analysis method based on the parallel multi-mode sensing input library network according to claim 2, wherein the channel layer of the nano-sheet is bismuth oxygen selenium nano-sheet, and the thickness of the channel layer is in the range of 5-20 nm.
5. The method for analyzing time sequence signals based on the parallel multi-mode sensing input library network according to claim 2, wherein the source electrode and the drain electrode are composed of Ti/Au metal lamination layers, and the thickness of the Ti/Au metal lamination layers is not more than 100nm.
6. The method for analyzing time sequence signals based on the parallel multi-mode sensing input library network according to claim 2, wherein the interface defect layer is introduced by hydrofluoric acid wet etching.
7. The method for analyzing time sequence signals based on the parallel multi-mode sensing input library network according to claim 3, wherein the thickness of the hafnium oxide layer is in the range of 10-20 nm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311759196.8A CN117743927B (en) | 2023-12-20 | Time sequence signal analysis method based on multi-mode sensing input library network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311759196.8A CN117743927B (en) | 2023-12-20 | Time sequence signal analysis method based on multi-mode sensing input library network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117743927A true CN117743927A (en) | 2024-03-22 |
| CN117743927B CN117743927B (en) | 2024-06-21 |
Family
ID=
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190692A (en) * | 2018-08-21 | 2019-01-11 | 北京师范大学 | The moving object recognition methods and system of mechanism are recognized and chosen based on biological brain |
| US20210406660A1 (en) * | 2018-01-24 | 2021-12-30 | The Regents Of The University Of California | Synaptic Resistors for Concurrent Parallel Signal Processing, Memory and Learning with High Speed and Energy Efficiency |
| CN113901132A (en) * | 2021-09-06 | 2022-01-07 | 北京许继电气有限公司 | Method and system for synchronously processing historical data among nodes of massive time sequence database |
| CN114093438A (en) * | 2021-10-28 | 2022-02-25 | 北京大学 | Based on Bi2O2Se multi-mode library network time sequence information processing method |
| CN114141813A (en) * | 2021-11-01 | 2022-03-04 | 北京大学 | Based on Bi2O2Se multi-modal feeling storage and calculation integrated system and implementation method thereof |
| US20230030949A1 (en) * | 2021-07-16 | 2023-02-02 | Korea Advanced Institute Of Science And Technology | Synaptic device, reservoir computing device including the synaptic device, and reservoir computing method using the computing device |
| CN117062450A (en) * | 2023-08-31 | 2023-11-14 | 山东云海国创云计算装备产业创新中心有限公司 | Electro-optical artificial synapse device and method for manufacturing electro-optical artificial synapse device |
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210406660A1 (en) * | 2018-01-24 | 2021-12-30 | The Regents Of The University Of California | Synaptic Resistors for Concurrent Parallel Signal Processing, Memory and Learning with High Speed and Energy Efficiency |
| CN109190692A (en) * | 2018-08-21 | 2019-01-11 | 北京师范大学 | The moving object recognition methods and system of mechanism are recognized and chosen based on biological brain |
| US20230030949A1 (en) * | 2021-07-16 | 2023-02-02 | Korea Advanced Institute Of Science And Technology | Synaptic device, reservoir computing device including the synaptic device, and reservoir computing method using the computing device |
| CN113901132A (en) * | 2021-09-06 | 2022-01-07 | 北京许继电气有限公司 | Method and system for synchronously processing historical data among nodes of massive time sequence database |
| CN114093438A (en) * | 2021-10-28 | 2022-02-25 | 北京大学 | Based on Bi2O2Se multi-mode library network time sequence information processing method |
| CN114141813A (en) * | 2021-11-01 | 2022-03-04 | 北京大学 | Based on Bi2O2Se multi-modal feeling storage and calculation integrated system and implementation method thereof |
| CN117062450A (en) * | 2023-08-31 | 2023-11-14 | 山东云海国创云计算装备产业创新中心有限公司 | Electro-optical artificial synapse device and method for manufacturing electro-optical artificial synapse device |
Non-Patent Citations (1)
| Title |
|---|
| ITSUKI KANEMURA 等: "Association between dif f erent sensory modalities based on concurrent time series data obtained by a collaborative reservoir computing model", 《NATURE》, 4 February 2023 (2023-02-04), pages 1 - 16 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xie et al. | SNAS: stochastic neural architecture search | |
| US11599782B2 (en) | Self-powered analog computing architecture with energy monitoring to enable machine-learning vision at the edge | |
| Przyczyna et al. | In-materio neuromimetic devices: Dynamics, information processing and pattern recognition | |
| Ruiz-Euler et al. | Dopant network processing units: towards efficient neural network emulators with high-capacity nanoelectronic nodes | |
| Katte | Recurrent neural network and its various architecture types | |
| Zhang et al. | A survey on reservoir computing and its interdisciplinary applications beyond traditional machine learning | |
| Wang et al. | Synaptic transistor with multiple biological functions based on metal-organic frameworks combined with the LIF model of a spiking neural network to recognize temporal information | |
| CN117743927B (en) | Time sequence signal analysis method based on multi-mode sensing input library network | |
| Rouhani et al. | TinyDL: Just-in-time deep learning solution for constrained embedded systems | |
| Wu et al. | Photonic bayesian neural network using programmed optical noises | |
| Bai et al. | Design strategies and applications of reservoir computing: Recent trends and prospects [feature] | |
| Li et al. | Deep echo state networks with multi-span features for nonlinear time series prediction | |
| CN117743927A (en) | Time sequence signal analysis method based on multi-mode sensing input library network | |
| Putra et al. | Forecasting air quality using massive-scale wsn based on convolutional LSTM network | |
| Zhang et al. | Composing recurrent spiking neural networks using locally-recurrent motifs and risk-mitigating architectural optimization | |
| Zeddini et al. | Experimental validation of an advanced metaheuristic algorithm for maximum power point tracking of a shaded photovoltaic system: A comparative study between three approaches | |
| Lin et al. | Realization of the trajectory propagation in the mm-sqc dynamics by using machine learning | |
| CN115690906A (en) | Human body action recognition method based on self-attention mechanism and Bi-GRU | |
| Zhang et al. | Obstacle‐transformer: A trajectory prediction network based on surrounding trajectories | |
| Tao et al. | Monostable stochastic resonance activation unit-based physical reservoir computing | |
| Liu | [Retracted] Intelligent Physical Training Data Processing Based on Wearable Devices | |
| Deepti et al. | Implementation of an adaptive neural network-based controller for improving the conversion efficiency of solar photovoltaic module | |
| AlDahoul et al. | Feature fusion: H-ELM based learned features and hand-crafted features for human activity recognition | |
| Ma et al. | Representation learning from time labelled heterogeneous data for mobile crowdsensing | |
| Morie et al. | A 2-dimensional Si nanodisk array structure for spiking neuron models |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |