CN103778468A - RRAM-based new type neural network circuit - Google Patents
RRAM-based new type neural network circuit Download PDFInfo
- Publication number
- CN103778468A CN103778468A CN201410021568.1A CN201410021568A CN103778468A CN 103778468 A CN103778468 A CN 103778468A CN 201410021568 A CN201410021568 A CN 201410021568A CN 103778468 A CN103778468 A CN 103778468A
- Authority
- CN
- China
- Prior art keywords
- neuron
- rram
- layer
- ground floor
- circuit
- 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
Images
Abstract
The invention provides a neural network circuit. The neural network circuit is characterized in that the neural network circuit comprises a plurality of sensors, a plurality of first layer of neuron branches and second layer of neuron branches. Each first layer of neuron branches comprises a plurality of RRAM devices and a first layer of neurons. The sensors are used to convert the picture colors into voltage signals, and transmit the voltage signals to the RRAM devices. The RRAM devices are used to generate current signals according to the receive voltage signals and transmit the current signals to the first layer of neurons. The first layer of neurons is used to perform summation on the received current signals, and if the neurons are activated, then a voltage pulse is transmitted to the backward stage. Each second layer of neuron branches comprises weight RRAM devices and a second layer of neurons. The weight RRAM devices are used to connect the first layer of neurons and the second layer of neurons. The second layer of neurons is used to collect the current signals of the plurality of first layer of neurons, and then the final judgement result is generated through calculation.
Description
Technical field
The present invention relates to technical field of semiconductors, be specifically related to a kind of new neural network circuit based on RRAM.
Background technology
Resistive formula storer RRAM had attracted to pay close attention to widely in recent years.(<5ns), low operating voltage (<1V) at a high speed, high storage density, is easy to the advantages such as integrated, makes RRAM become the strong rival of semiconductor memory of future generation.RRAM device generally has the structure of metal-insulator-metal type, between double layer of metal electrode, adds one deck to have the dielectric thin-film material of resistive characteristic, and these resistive materials are generally metal oxides, common are TiO
2, HfO
2, ZrO
2, WO
3, Ta
2o
5etc..By impressed voltage, can between low-resistance and high resistant, change the resistance of device.Utilize the circuit that is similar to cerebral nerve network forming based on resistance-variable storing device (RRAM) array, can accomplish extremely low power consumption and good fuzzy diagnosis function.It is an important development direction in the fields such as future image identification, voice recognition.
The principle of work of the existing image recognition circuit based on RRAM is roughly as described below: circuit structure is roughly divided into three parts, and Part I is sensor construction.Change the color signal of picture into voltage signal (voltage swing difference, or pulse length difference), this voltage is used for the RRAM array of rear class to operate.Part II is multiple RRAM arrays.RRAM device in these arrays receives the voltage signal of prime, and generation current signal, is delivered to rear one-level.The reason of utilizing multiple RRAM arrays is in use, and the RRAM device in each array can have different resistances to distribute, and the resistance of different arrays distributes and makes its response intensity to certain picture the highest.Device in these arrays corresponds to nervous system and is the huge cynapse of quantity.Part III is the neuron module of final stage, and each neuron is connected with RRAM devices all in an array, to the electric current summation above each device.When electric current and exceed certain threshold value, or when maximal value in multiple neuron, this neuron is activated, and also a picture has been made to corresponding reaction.
Although the neuron circuit based on RRAM of mentioning at present relevant work is all made up of multiple RRAM arrays, each array will be accepted the voltage producing from all pixels of whole picture.Sort circuit underaction, cannot tackle the change of shape of input picture.Owing to utilizing a RRAM array to receive the signal of whole picture, then sue for peace this process by the omissions of detail of picture through late-class circuit, while making available circuit cannot process " comprising logic ", recognition image, cannot distinguish primary and secondary.Meanwhile, with respect to the nervous system on biology, current structure is too simple, only has one-level neuron, can only process the simplest situation.
Summary of the invention
(1) technical matters solving
For the deficiencies in the prior art, the invention provides a kind of nerve network circuit, can guarantee, on the basis of recognition capability, to retain as far as possible the detailed information of picture.
(2) technical scheme
In order to realize above object, the present invention is achieved by the following technical programs:
A kind of nerve network circuit, comprising: several sensors, several ground floor neuron branch roads and several second layer neuron branch roads;
Each ground floor neuron branch road comprises: several the first resistive formula storer RRAM and ground floor neurons;
It is voltage signal by the color conversion of pixel that described sensor is used for, and this voltage signal is transferred to a described RRAM; The voltage signal generation current signal that a described RRAM basis receives, and transfer to described ground floor neuron; Described ground floor neuron is for suing for peace to the current signal receiving, if peripheral sensory neuron is activated, emitting voltage pulse is to described the 2nd RRAM;
Each second layer neuron branch road comprises: several the 2nd RRAM and second layer neurons;
Described the 2nd RRAM device couples together described ground floor neuron and described second layer neuron; Described second layer neuron is for gathering the current signal producing after ground floor neuronal activation described in several.
Wherein, a sensor is corresponding to a pixel, and a sensor is corresponding to a RRAM device simultaneously.
Wherein, the resistance of a described RRAM is inverse ratio with the size of the voltage signal that receives sensor.
Wherein, described ground floor neuron, comprising: CMOS neuron, feedback circuit.
Wherein, described feedback circuit is for changing the resistance of the RRAM that described ground floor neuron is corresponding.
Wherein, described second layer neuron is CMOS neuron.
(3) beneficial effect
The present invention at least has following beneficial effect:
1, owing to having adopted the structure of piecemeal, being divided into several parts by picture identifies respectively, single little RRAM array is responsible for a part for picture, as each the ground floor neuron branch road in circuit structure is responsible for the part of picture, by the concurrent working of multiple junior units or many branch roads, just can within a shorter work period, process the large picture of a width like this.The mode of this concurrent working has improved work efficiency.
2, the structure that adopts layering, different with existing technology, in circuit structure of the present invention, there is two-layer neuron circuit, the information of each fritter can be processed again, circuit just can be processed more complicated logical relation like this.
3, prior art can be solved and the shortcoming of " comprising logic " identification cannot be carried out, as shown in Figure 1, we suppose that white portion is effective information, the circuit of prior art is in training process, white part is carried out to SET(and changes high resistant into low-resistance) operation, black part is divided and is carried out RESET(and change low-resistance into high resistant) operation, the resistance of corresponding with white so RRAM device should SET to smaller value, its electric current is larger, but scheme for (a), its white portion comprises the white portion with (b) figure completely, the condition being activated according to neuron, cause the neuron of correspondence (b) in the time of identification (a) to be activated.But the present invention has adopted the mode of piecemeal, be divided into several parts by picture and identify respectively, the signal of detail section can be preserved, and then judge, rather than simple summation.
Accompanying drawing explanation
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these figure other accompanying drawing.
Fig. 1 is the schematic diagram of " comprising logic " problem;
Fig. 2 is the one-piece construction schematic diagram of the nerve network circuit of the RRAM based on one embodiment of the invention;
Fig. 3 is the circuit theory diagrams of the nerve network circuit based on RRAM in specific embodiment of the present invention.
Embodiment
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these figure other accompanying drawing.
Fig. 2 is the one-piece construction schematic diagram of the nerve network circuit of the RRAM based on one embodiment of the invention, and this circuit comprises: sensor 201, RRAM array 202, ground floor neuron 203, the 2nd RRAM204 and second layer neuron 205.Wherein, RRAM array is made up of several RRAM devices.
When circuit is trained, the picture of needs identification is divided into 4 and carries out respectively image recognition, sensor 201 receives the signal of picture pixel, produce potential pulse, potential pulse transfers to the RRAM device in RRAM array 202, and generation current signal, different owing to being added in voltage swing on each RRAM device 202, therefore the size of current producing is also different, the current signal that each RRAM array 202 produces gathers to ground floor neuron 203, if be aggregated into the electric current of ground floor neuron 203 and exceed certain threshold value or be the maximal value in multiple neurons, ground floor neuron 203 is activated, and feed back with its corresponding RRAM array 202 simultaneously, the feedback here refers to that the resistance of the RRAM device in pair array is carried out SET or RESET operates, after suitable number of times (20 left and right of every pictures), the resistance of the RRAM device in RRAM array 202 can form a distribution corresponding with picture, in addition, also possesses the function that the resistance of the 2nd RRAM204 to coupled rear class is adjusted, so just make the size of current difference of different ground floor neuron 203 through weight RRAM device 204, finally be summarised in second layer neuron 205, produce last judged result by computing, wherein, each ground floor neuron 203 is connected with second layer neuron 204.
In order to further illustrate above-described embodiment, Fig. 3 is the circuit theory diagrams based on above-described embodiment.As shown in Figure 3, this circuit comprises sensor 301, a RRAM302, ground floor neuron 303, the two RRAM304 and second layer neuron 305.
This circuit comprises several ground floor neuron branch roads, and more than one of second layer neuron in second layer neuron 305(practical application).Wherein, every ground floor neuron branch road comprises several sensors 301, several RRAM302, a ground floor neuron, a weight RRAM device.In the time carrying out circuit training, be voltage signal by sensor 301 by the color transition of picture, voltage signal transfers to a RRAM302, and generation current signal, the resistance of the one RRAM302 is inverse ratio with the size of voltage signal that receives sensor, according to this rule adjustment RRAM resistance, after making in use, just can produce maximum electric current, be equivalent to a maximum response.Owing to being added in, voltage swing on each RRAM302 is different, the difference that resistance of a RRAM becomes, and the size of current therefore producing is also different, and the current signal that several RRAM302 produce gathers to ground floor neuron 303.As shown in Figure 3, ground floor neuron 303 is by CMOS neuron, feedback circuit 306 and resistance Circuit tuning 307 form, wherein, described CMOS neuron is prior art, utilize threshold value comparator circuit wherein, make in the time that the current signal sum of inflow ground floor neuron 303 exceedes certain specific threshold value, ground floor neuron 303 is activated, feedback circuit 306 wherein feeds back corresponding several RRAM devices 302, the feedback here refers to that the resistance of the RRAM device in pair array is carried out SET or RESET operates, after suitable number of times (20 left and right), the resistance of several in RRAM array RRAM302 can form a distribution corresponding with picture, meanwhile, the resistance of two RRAM304 of its resistance Circuit tuning to coupled rear class is adjusted, and so just makes the size of current difference of different ground floor neurons 303 through the 2nd RRAM304, and circuit training is complete.In the time that circuit uses, add a width picture at input end, sensor 301 produces potential pulse, in the one RRAM302, there is electric current to flow through, flow into peripheral sensory neuron 303 and sue for peace, corresponding peripheral sensory neuron 303 is activated, to the pulse of rear class emitting voltage, this potential pulse produces different electric currents on the 2nd different RRAM304, finally gathers to nervus opticus unit 305.
Owing to having adopted the structure of piecemeal, being divided into several parts by picture identifies respectively, single little RRAM array is responsible for a part for picture, as each the ground floor neuron branch road in circuit structure is responsible for the part of picture, by the concurrent working of multiple junior units or many branch roads, just can within a shorter work period, process the large picture of a width like this.The method of this concurrent working has improved work efficiency.
The structure that adopts layering, different with existing technology, in circuit structure of the present invention, there is two-layer neuron circuit, the information of each fritter can be processed again, circuit just can be processed more complicated logical relation like this.
Prior art can be solved and the shortcoming of " comprising logic " identification cannot be carried out, the present invention has adopted the mode of piecemeal, is divided into several parts identifies respectively by picture, the signal of detail section can be preserved, and then judge, rather than simple summation.
Above embodiment only, in order to technical scheme of the present invention to be described, is not intended to limit; Although the present invention is had been described in detail with reference to previous embodiment, those of ordinary skill in the art is to be understood that; Its technical scheme that still can record aforementioned each embodiment is modified, or part technical characterictic is wherein equal to replacement; And these modifications or replacement do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.
Claims (6)
1. the nerve network circuit based on RRAM, is characterized in that, comprising: several sensors, several ground floor neuron branch roads and several second layer neuron branch roads;
Each ground floor neuron branch road comprises: several RRAM and ground floor neurons;
It is voltage signal by the color conversion of pixel that described sensor is used for, and this voltage signal is transferred to a described RRAM; The voltage signal generation current signal that a described RRAM basis receives, and transfer to described ground floor neuron; Described ground floor neuron is for suing for peace to the current signal receiving, if peripheral sensory neuron is activated, emitting voltage pulse is to described the 2nd RRAM;
Each second layer neuron branch road comprises: several the 2nd RRAM and second layer neurons;
Described the 2nd RRAM device couples together described ground floor neuron and described second layer neuron; Described second layer neuron is for gathering the current signal producing after ground floor neuronal activation described in several.
2. circuit as claimed in claim 1, is characterized in that,
A sensor is corresponding to a pixel, and a sensor is corresponding to a RRAM device simultaneously.
3. circuit as claimed in claim 1, is characterized in that,
The resistance of a described RRAM is inverse ratio with the size of the voltage signal that receives sensor.
4. circuit as claimed in claim 1, is characterized in that,
Described ground floor neuron, comprising: CMOS neuron, feedback circuit.
5. circuit as claimed in claim 6, is characterized in that,
Described feedback circuit is for changing the resistance of the RRAM that described ground floor neuron is corresponding.
6. circuit as claimed in claim 1, is characterized in that,
Described second layer neuron is CMOS neuron.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410021568.1A CN103778468B (en) | 2014-01-16 | 2014-01-16 | A kind of new neural network circuit based on RRAM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410021568.1A CN103778468B (en) | 2014-01-16 | 2014-01-16 | A kind of new neural network circuit based on RRAM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103778468A true CN103778468A (en) | 2014-05-07 |
| CN103778468B CN103778468B (en) | 2016-09-07 |
Family
ID=50570678
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410021568.1A Active CN103778468B (en) | 2014-01-16 | 2014-01-16 | A kind of new neural network circuit based on RRAM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103778468B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104376362A (en) * | 2014-11-21 | 2015-02-25 | 北京大学 | Synaptic device for artificial neural network and artificial neural network |
| CN105184366A (en) * | 2015-09-15 | 2015-12-23 | 中国科学院计算技术研究所 | Time-division-multiplexing general neural network processor |
| CN106875005A (en) * | 2017-01-20 | 2017-06-20 | 清华大学 | Adaptive threshold neuronal messages processing method and system |
| CN107273972A (en) * | 2017-05-11 | 2017-10-20 | 北京大学 | It is a kind of based on resistive device and to adapt to excite the neuromorphic system and implementation method of neuron |
| CN107533459A (en) * | 2016-03-31 | 2018-01-02 | 慧与发展有限责任合伙企业 | Use the data processing of resistive memory array |
| CN108681772A (en) * | 2018-04-02 | 2018-10-19 | 北京大学 | Multi-modal neuron circuit and neuron implementation method |
| CN109146070A (en) * | 2017-06-16 | 2019-01-04 | 华为技术有限公司 | A kind of peripheral circuit and system of neural network training of the support based on RRAM |
| CN110462643A (en) * | 2017-04-04 | 2019-11-15 | 海露科技有限公司 | The neural network processor of calculating and memory component including multilevel hierarchy polymerization |
| CN110797067A (en) * | 2019-10-21 | 2020-02-14 | 杭州闪亿半导体有限公司 | Storage array module and control method, device and module thereof |
| CN112165384A (en) * | 2020-10-15 | 2021-01-01 | 清华大学 | Data encryption method and decryption method, and data encryption device and decryption device |
| US11221929B1 (en) | 2020-09-29 | 2022-01-11 | Hailo Technologies Ltd. | Data stream fault detection mechanism in an artificial neural network processor |
| US11237894B1 (en) | 2020-09-29 | 2022-02-01 | Hailo Technologies Ltd. | Layer control unit instruction addressing safety mechanism in an artificial neural network processor |
| US11238334B2 (en) | 2017-04-04 | 2022-02-01 | Hailo Technologies Ltd. | System and method of input alignment for efficient vector operations in an artificial neural network |
| US11263077B1 (en) | 2020-09-29 | 2022-03-01 | Hailo Technologies Ltd. | Neural network intermediate results safety mechanism in an artificial neural network processor |
| TWI768174B (en) * | 2018-10-24 | 2022-06-21 | 旺宏電子股份有限公司 | Device and method of in-memory data pooling for machine learning |
| US11544545B2 (en) | 2017-04-04 | 2023-01-03 | Hailo Technologies Ltd. | Structured activation based sparsity in an artificial neural network |
| US11551028B2 (en) | 2017-04-04 | 2023-01-10 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network |
| US11562229B2 (en) | 2018-11-30 | 2023-01-24 | Macronix International Co., Ltd. | Convolution accelerator using in-memory computation |
| US11615297B2 (en) | 2017-04-04 | 2023-03-28 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network compiler |
| US11811421B2 (en) | 2020-09-29 | 2023-11-07 | Hailo Technologies Ltd. | Weights safety mechanism in an artificial neural network processor |
| US11861429B2 (en) | 2018-04-30 | 2024-01-02 | Hewlett Packard Enterprise Development Lp | Resistive and digital processing cores |
| US11874900B2 (en) | 2020-09-29 | 2024-01-16 | Hailo Technologies Ltd. | Cluster interlayer safety mechanism in an artificial neural network processor |
| US11934480B2 (en) | 2018-12-18 | 2024-03-19 | Macronix International Co., Ltd. | NAND block architecture for in-memory multiply-and-accumulate operations |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102412827A (en) * | 2011-11-02 | 2012-04-11 | 北京大学 | Method for realizing logic operation by utilizing RRAM devices |
| CN102611424A (en) * | 2011-12-22 | 2012-07-25 | 北京大学 | Method for realizing integral operation by using resistance changing device |
| CN103337253A (en) * | 2013-05-29 | 2013-10-02 | 北京大学 | Cascade system and method of RRAM logic device |
-
2014
- 2014-01-16 CN CN201410021568.1A patent/CN103778468B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102412827A (en) * | 2011-11-02 | 2012-04-11 | 北京大学 | Method for realizing logic operation by utilizing RRAM devices |
| CN102611424A (en) * | 2011-12-22 | 2012-07-25 | 北京大学 | Method for realizing integral operation by using resistance changing device |
| CN103337253A (en) * | 2013-05-29 | 2013-10-02 | 北京大学 | Cascade system and method of RRAM logic device |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104376362B (en) * | 2014-11-21 | 2017-10-03 | 北京大学 | Cynapse device and artificial neural network for artificial neural network |
| CN104376362A (en) * | 2014-11-21 | 2015-02-25 | 北京大学 | Synaptic device for artificial neural network and artificial neural network |
| CN105184366A (en) * | 2015-09-15 | 2015-12-23 | 中国科学院计算技术研究所 | Time-division-multiplexing general neural network processor |
| CN105184366B (en) * | 2015-09-15 | 2018-01-09 | 中国科学院计算技术研究所 | A kind of time-multiplexed general neural network processor |
| CN107533459A (en) * | 2016-03-31 | 2018-01-02 | 慧与发展有限责任合伙企业 | Use the data processing of resistive memory array |
| US10942673B2 (en) | 2016-03-31 | 2021-03-09 | Hewlett Packard Enterprise Development Lp | Data processing using resistive memory arrays |
| CN107533459B (en) * | 2016-03-31 | 2020-11-20 | 慧与发展有限责任合伙企业 | Data processing method and unit using resistance memory array |
| CN106875005B (en) * | 2017-01-20 | 2019-09-20 | 清华大学 | Adaptive threshold neuronal messages processing method and system |
| CN106875005A (en) * | 2017-01-20 | 2017-06-20 | 清华大学 | Adaptive threshold neuronal messages processing method and system |
| US11461614B2 (en) | 2017-04-04 | 2022-10-04 | Hailo Technologies Ltd. | Data driven quantization optimization of weights and input data in an artificial neural network |
| US11514291B2 (en) | 2017-04-04 | 2022-11-29 | Hailo Technologies Ltd. | Neural network processing element incorporating compute and local memory elements |
| CN110462643A (en) * | 2017-04-04 | 2019-11-15 | 海露科技有限公司 | The neural network processor of calculating and memory component including multilevel hierarchy polymerization |
| US11675693B2 (en) | 2017-04-04 | 2023-06-13 | Hailo Technologies Ltd. | Neural network processor incorporating inter-device connectivity |
| US11615297B2 (en) | 2017-04-04 | 2023-03-28 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network compiler |
| US11551028B2 (en) | 2017-04-04 | 2023-01-10 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network |
| US11544545B2 (en) | 2017-04-04 | 2023-01-03 | Hailo Technologies Ltd. | Structured activation based sparsity in an artificial neural network |
| US11263512B2 (en) | 2017-04-04 | 2022-03-01 | Hailo Technologies Ltd. | Neural network processor incorporating separate control and data fabric |
| US11238334B2 (en) | 2017-04-04 | 2022-02-01 | Hailo Technologies Ltd. | System and method of input alignment for efficient vector operations in an artificial neural network |
| US11354563B2 (en) | 2017-04-04 | 2022-06-07 | Hallo Technologies Ltd. | Configurable and programmable sliding window based memory access in a neural network processor |
| US11216717B2 (en) | 2017-04-04 | 2022-01-04 | Hailo Technologies Ltd. | Neural network processor incorporating multi-level hierarchical aggregated computing and memory elements |
| US11461615B2 (en) | 2017-04-04 | 2022-10-04 | Hailo Technologies Ltd. | System and method of memory access of multi-dimensional data |
| US11238331B2 (en) | 2017-04-04 | 2022-02-01 | Hailo Technologies Ltd. | System and method for augmenting an existing artificial neural network |
| CN107273972B (en) * | 2017-05-11 | 2019-09-13 | 北京大学 | It is a kind of based on resistive device and adaptation-excitation neuron neuromorphic system and implementation method |
| CN107273972A (en) * | 2017-05-11 | 2017-10-20 | 北京大学 | It is a kind of based on resistive device and to adapt to excite the neuromorphic system and implementation method of neuron |
| CN109146070B (en) * | 2017-06-16 | 2021-10-22 | 华为技术有限公司 | Peripheral circuit and system for supporting neural network training based on RRAM |
| US11409438B2 (en) | 2017-06-16 | 2022-08-09 | Huawei Technologies Co., Ltd. | Peripheral circuit and system supporting RRAM-based neural network training |
| CN109146070A (en) * | 2017-06-16 | 2019-01-04 | 华为技术有限公司 | A kind of peripheral circuit and system of neural network training of the support based on RRAM |
| CN108681772A (en) * | 2018-04-02 | 2018-10-19 | 北京大学 | Multi-modal neuron circuit and neuron implementation method |
| CN108681772B (en) * | 2018-04-02 | 2020-09-29 | 北京大学 | Multi-mode neuron circuit and neuron implementation method |
| US11861429B2 (en) | 2018-04-30 | 2024-01-02 | Hewlett Packard Enterprise Development Lp | Resistive and digital processing cores |
| TWI768174B (en) * | 2018-10-24 | 2022-06-21 | 旺宏電子股份有限公司 | Device and method of in-memory data pooling for machine learning |
| US11636325B2 (en) | 2018-10-24 | 2023-04-25 | Macronix International Co., Ltd. | In-memory data pooling for machine learning |
| US11562229B2 (en) | 2018-11-30 | 2023-01-24 | Macronix International Co., Ltd. | Convolution accelerator using in-memory computation |
| US11934480B2 (en) | 2018-12-18 | 2024-03-19 | Macronix International Co., Ltd. | NAND block architecture for in-memory multiply-and-accumulate operations |
| CN110797067A (en) * | 2019-10-21 | 2020-02-14 | 杭州闪亿半导体有限公司 | Storage array module and control method, device and module thereof |
| CN110797067B (en) * | 2019-10-21 | 2021-10-22 | 上海闪易半导体有限公司 | Storage array module and control method, device and module thereof |
| US11263077B1 (en) | 2020-09-29 | 2022-03-01 | Hailo Technologies Ltd. | Neural network intermediate results safety mechanism in an artificial neural network processor |
| US11811421B2 (en) | 2020-09-29 | 2023-11-07 | Hailo Technologies Ltd. | Weights safety mechanism in an artificial neural network processor |
| US11221929B1 (en) | 2020-09-29 | 2022-01-11 | Hailo Technologies Ltd. | Data stream fault detection mechanism in an artificial neural network processor |
| US11874900B2 (en) | 2020-09-29 | 2024-01-16 | Hailo Technologies Ltd. | Cluster interlayer safety mechanism in an artificial neural network processor |
| US11237894B1 (en) | 2020-09-29 | 2022-02-01 | Hailo Technologies Ltd. | Layer control unit instruction addressing safety mechanism in an artificial neural network processor |
| CN112165384A (en) * | 2020-10-15 | 2021-01-01 | 清华大学 | Data encryption method and decryption method, and data encryption device and decryption device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103778468B (en) | 2016-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103778468A (en) | RRAM-based new type neural network circuit | |
| Wu et al. | Homogeneous spiking neuromorphic system for real-world pattern recognition | |
| CN106981567B (en) | A kind of artificial synapse device and its modulator approach based on photoelectric coupling memristor | |
| US20150088797A1 (en) | Synapse circuits for connecting neuron circuits, unit cells composing neuromorphic circuit, and neuromorphic circuits | |
| TWI728556B (en) | Neuron circuit and artificial neural network chip | |
| Hunsberger et al. | Spiking deep networks with LIF neurons | |
| CN106815636B (en) | A kind of neuron circuit based on memristor | |
| Park et al. | RRAM-based synapse for neuromorphic system with pattern recognition function | |
| US20200342301A1 (en) | Convolutional neural network on-chip learning system based on non-volatile memory | |
| CN108985447A (en) | A kind of hardware pulse nerve network system | |
| Linares-Barranco et al. | Exploiting memristance in adaptive asynchronous spiking neuromorphic nanotechnology systems | |
| CN103580668B (en) | A kind of associative memory circuit based on memristor | |
| CN107273972B (en) | It is a kind of based on resistive device and adaptation-excitation neuron neuromorphic system and implementation method | |
| WO2019034970A1 (en) | Neuromorphic processing apparatus | |
| EP3121767A1 (en) | Methods and systems for implementing deep spiking neural networks | |
| Brivio et al. | Extended memory lifetime in spiking neural networks employing memristive synapses with nonlinear conductance dynamics | |
| CN105913119B (en) | The heterogeneous polynuclear heart class brain chip and its application method of ranks interconnection | |
| WO2020099680A2 (en) | Resilient neural network | |
| US20210174173A1 (en) | Circuit Structure and Driving Method Thereof, Neural Network | |
| CN109034379A (en) | A kind of neuron and neuron circuit built by class brain device memristor | |
| CN104376362B (en) | Cynapse device and artificial neural network for artificial neural network | |
| Garbin et al. | Probabilistic neuromorphic system using binary phase-change memory (PCM) synapses: Detailed power consumption analysis | |
| Yu et al. | Design considerations of synaptic device for neuromorphic computing | |
| Zendrikov et al. | Brain-inspired methods for achieving robust computation in heterogeneous mixed-signal neuromorphic processing systems | |
| Douglas et al. | Hybrid analog-digital architectures for neuromorphic systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |