WO2012006468A1 - Methods and systems for replaceable synaptic weight storage in neuro-processors - Google Patents
Methods and systems for replaceable synaptic weight storage in neuro-processors Download PDFInfo
- Publication number
- WO2012006468A1 WO2012006468A1 PCT/US2011/043254 US2011043254W WO2012006468A1 WO 2012006468 A1 WO2012006468 A1 WO 2012006468A1 US 2011043254 W US2011043254 W US 2011043254W WO 2012006468 A1 WO2012006468 A1 WO 2012006468A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- neuro
- processor chip
- weights
- removable memory
- memory
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Definitions
- Certain embodiments of the present disclosure generally relate to neural system engineering and, more particularly, to a method for storing synaptic weights separately from a neuro-processor chip into replaceable storage.
- synaptic weights which control strengths of connections between neurons.
- the synaptic weights are typically stored in non-volatile, on-chip memory in order to preserve the processor functionality after being powered down.
- the electrical circuit generally includes a neuro-processor chip with a plurality of neuron circuits and synapses, wherein each synapse connects a pair of the neuron circuits, and a removable memory connected to the neuro-processor chip storing weights of the synapses, wherein the weights determine a function of the neuro- processor chip.
- Certain embodiments of the present disclosure provide a method for implementing a neural system.
- the method generally includes using a removable memory to store weights of synapses, wherein each synapse connects two of a plurality of neuron circuits of a neuro-processor chip, and wherein the weights determine a function of the neuro-processor chip, and connecting the removable memory to the neuro-processor chip.
- inventions of the present disclosure provide an apparatus for implementing a neural system.
- the apparatus generally includes means for using a removable memory to store weights of synapses, wherein each synapse connects two of a plurality of neuron circuits of a neuro-processor chip, and wherein the weights determine a function of the neuro-processor chip, and means for connecting the removable memory to the neuro-processor chip.
- FIG. 1 illustrates an example neural system in accordance with certain embodiments of the present disclosure.
- FIG. 2 illustrates an example of neuro-processor interfaced with an external synaptic weight memory in accordance with certain embodiments of the present disclosure.
- FIG. 3 illustrates example operations for implementing the synaptic weight memory external to the neuro-processor in accordance with certain embodiments of the present disclosure.
- FIG. 3A illustrates example components capable of performing the operations illustrated in FIG. 3.
- FIG. 4 illustrates examples of non-volatile memories that may be used for implementing the external synaptic weight memory in accordance with certain embodiments of the present disclosure.
- FIG. 1 illustrates an example neural system 100 with multiple levels of neurons in accordance with certain embodiments of the present disclosure.
- the neural system 100 may comprise a level of neurons 102 connected to another level of neurons 106 though a network of synapse connections 104.
- a network of synapse connections 104 For simplicity, only two levels of neurons are illustrated in FIG. 1 , although more levels of neurons may exist in a typical neural system.
- each neuron in the level 102 may receive an input signal 108 that may be generated by a plurality of neurons of a previous level (not shown in FIG. 1).
- the signal 108 may represent an input current of the level 102 neuron. This current may be accumulated on the neuron membrane to charge a membrane potential. When the membrane potential reaches its threshold level, the neuron may fire and generate an output spike to be transferred to the next level of neurons (e.g., the level 106).
- the transfer of spikes from one level of neurons to another may be achieved through the network of synaptic connections (or simply "synapses") 104, as illustrated in FIG. 1.
- the synapses 104 may receive output signals (i.e., spikes) from the level 102 neurons, scale those signals according to adjustable synaptic weights ,...,
- ⁇ ( +I) wnere is a total number of synaptic connections between the neurons of levels 102 and 106
- wnere is a total number of synaptic connections between the neurons of levels 102 and 106
- wnere is a total number of synaptic connections between the neurons of levels 102 and 106
- Every neuron in the level 106 may generate an output spike 110 based on the corresponding combined input signal.
- the output spikes 110 may be then transferred to another level of neurons using another network of synaptic connections (not shown in FIG. 1).
- the neural system 100 may be emulated by a neuro-processor and utilized in a large range of applications, such as pattern recognition, machine learning and motor control.
- Each neuron of the neural system 100 may be implemented as a neuron circuit within the neuro-processor chip.
- the neuron membrane charged to the threshold level to initiate the output spike may be implemented within the neuron circuit as a capacitor which integrates an electrical current that flows through it.
- a nanometer feature- sized memristor element may be utilized as the integrating device instead of the capacitor.
- Functionality of the neuro-processor that emulates the neural system 100 may depend on weights of synaptic connections, which may control strengths of connections between neurons.
- the synaptic weights may be stored in a non-volatile memory in order to preserve functionality of the processor after being powered down.
- having this memory on the same chip with the neuro-processor may limit the processor functionality and flexibility.
- the on-chip synaptic memory may limit choices for the type of non- volatile memory being utilized, and may increase the area and implementation cost of the overall chip.
- Certain embodiments of the present disclosure support implementation of the synaptic weight memory on a separate external chip from the main neuro-processor chip.
- the synaptic weight memory may be packaged separately from the neuro- processor chip as a replaceable removable memory. This may provide diverse functionalities to the neuro-processor, wherein a particular functionality may be based on synaptic weights stored in a removable memory currently attached to the neuro- processor.
- FIG. 2 illustrates an example of neuromorphic architecture 200 in accordance with certain embodiments of the present disclosure.
- a synaptic memory 206 may be implemented as a separate and external removable memory, which may be connected to a neuro-processor 202 through an interface circuit 204.
- the neuro- processor 202 may emulate the neural system 100 illustrated in FIG. 1. It may comprise a large number of neuron circuits and synaptic connections.
- the interface 204 may comprise a bus connecting the neuro-processor chip 202 and the external synaptic removable memory 206.
- the interface bus may be designed to carry the synaptic weights data in both directions, as well as commands such as "memory write,” "memory read,” and "address.”
- the neuro-processor 202 may typically comprise, for example, about 10,000 neuron circuits with about 100 synapses per neuron, which brings the total number of synapses in the neuro-processor 202 to approximately 10 6 .
- the strength of each synaptic connection may be associated with a weight represented with a certain number of bits according to a desired precision. Typically, up to ten bits may be required per synaptic weight to provide sufficient precision for a large variety of applications. If, for example, every weight is represented with ten bits, then the memory of approximately lOMbits may be required to store the synaptic weights for the neuro-processor with approximately 10 6 synapses.
- the number of neurons and synapses within a neuro-processor is expected to increase substantially in the near future for supporting even more complex neural system engineering applications.
- the required size of synaptic weight memory may be much larger than lOMbits.
- Implementation of the large synaptic memory as a removable memory external to the neuro-processor may provide more efficient die utilization of both the neuro-processor and the synaptic memory.
- fabrication processes of the neuro-processor and memory may be uniquely tailored to the needs of these separate chips in order to provide better performance and lower cost.
- functionality of the neuro-processor 202 may depend on weights of the synaptic connections between the neuron circuits.
- training of the synaptic weights may need first to be performed within the neuro-processor.
- the synaptic weights may be stored and loaded to/from the external memory 206 through the interface 204. Once the learning process is finished, all trained synaptic weights may be fully stored into the external memory chip 206.
- duration of the weight-training process within the neuro-processor may last a long time. However, once the trained synaptic weights are fully stored in the external removable memory 206, they may be then quickly replicated to another removable memory. In this way, it may be possible to simply "clone" a functionality of the neuro-processor 202 from one memory chip to another. The time- and power-consuming weight-training process within another neuro-processor chip may be then fully avoided, and the other neuro-processor chip may be able to execute the same function as the neuro-processor 202 without performing the weight-training.
- the external memory 206 may be implemented as a replaceable removable memory.
- the same neuro-processor 202 may have different functionalities depending on the synaptic removable memory attached to it.
- the replaceable removable memory may be shared between users, and a library of different functionalities (i.e., different weight values of same synapses) may be stored in different removable memorys. These synaptic removable memorys with diverse functionalities may be designed fully independently from the neuro-processor 202.
- a local working memory with temporary data may be implemented within the neuro-processor chip 202 to provide faster processor operations.
- the local memory may be also utilized during the aforementioned weight-training process.
- a permanent memory comprising all trained synaptic weights fully determining the processor functionality may be external and implemented as the separate memory chip 206.
- FIG. 3 illustrates example operations 300 for implementing a synaptic removable memory external to a neuro-processor chip in accordance with certain embodiments of the present disclosure.
- a removable memory may be connected to the neuro-processor chip.
- the removable memory may be used to store synapse weights, wherein each synapse may connect two of a plurality of neuron circuits of the neuro-processor chip, and wherein the weights may define, at least in part, a function of the neuro-processor chip.
- An implementation area of one neuron circuit may be in the order of
- CMOS complementary metal-oxide-semiconductor
- 6 4 correspond to approximately 10 synapses for the exemplary processor comprising 10 neuron circuits.
- the implementation area per synapse may be in the order of
- 202 may be approximately equal to 110 mm (e.g., the die area of 10.5mmx l0.5mm ).
- the fastest firing rate of a neuron may be equal to one spike per 5ms.
- the maximum of about 10% of all neurons (or approximately 1000 neuron circuits in this exemplary case) may spike simultaneously in any given 5ms time period. Therefore, the maximum of 10 5 synaptic weights may need to be read every 5 ms from the synaptic weight memory 206 through the interface 204 into the neuro-processor 202. In other words, one synaptic weight may need to be read every 50ns, if only one synaptic weight at a time may be loaded from the external memory 206 to the neuro-processor 202.
- a memory write time may be determined based on the number of eligible synapses that may need to be updated when a reward signal arrives. In the worst-case scenario, the memory write time may be equal to a memory read time.
- the synaptic memory chip 206 may be typically required to store approximately 10 6 synaptic weights. If, for example, six bits are utilized per synaptic weight, then a total storage capacity of 6Mbits may be required.
- Magneto-resistive Random Access Memory MRAM
- Resistive Random Access Memory RRAM
- MRAM Magneto-resistive Random Access Memory
- RRAM Resistive Random Access Memory
- FIG. 4 illustrates a graph 400 with examples of non-volatile memories that may be used for the external synaptic memory 206 in accordance with certain embodiments of the present disclosure.
- Wide choices of non-volatile memory types include flash, ferroelectric, magnetic tunnel junction, spin-transfer torque devices, phase change memories, resistive/memristive switches, and so on. All these choices may represent possible candidates for the external synaptic memory 206.
- a portion 402 of the graph 400 may correspond to an operational region of a local working on-chip memory, which may store a portion of synaptic weights for faster processor operations.
- a Ferroelectric Random Access Memory FeRAM
- MRAM Magneto-resistive Random Access Memory
- SRAM Static Random Access Memory
- DRAM Dynamic Random Access Memory
- PRAM Phase-change Random Access Memory
- RRAM Resistive Random Access Memory
- PRAM, FeRAM and MRAM memories are all nonvolatile memories that do not require data to be erased before writing operations.
- the RRAM memory is the non-volatile memory that requires erasing before writing operation.
- DRAM and SRAM represent examples of volatile memories.
- a portion 404 of the graph 400 may correspond to an operational region of an external memory for storing all synaptic weights associated with an application executed by a neuro-processor interfaced with the external memory. It can be observed from FIG. 4 that a NAND flash memory, a NOR flesh memory and a PRAM may be possible choices for the external synaptic memory. While the NAND flash memories and NOR flesh memories are non- volatile memories that may require data to be erased before writing, the PRAM is the example of non-volatile RAM that does not require erasing before writing.
- the various operations of methods described above may be performed by any suitable means capable of performing the corresponding functions.
- the means may include various hardware and/or software component(s) and/or module(s), including, but not limited to a circuit, an application specific integrate circuit (ASIC), or processor.
- ASIC application specific integrate circuit
- those operations may have corresponding counterpart means-plus-function components with similar numbering.
- operations 300 illustrated in FIG. 3 correspond to components 300A, 302A and 304A illustrated in FIG. 3A.
- determining encompasses a wide variety of actions. For example, “determining” may include calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, a database or another data structure), ascertaining and the like. Also, “determining” may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory) and the like. Also, “determining” may include resolving, selecting, choosing, establishing and the like.
- a phrase referring to "at least one of a list of items refers to any combination of those items, including single members.
- "at least one of: a, b, or c" is intended to cover: a, b, c, a-b, a-c, b-c, and a-b-c.
- DSP digital signal processor
- ASIC application specific integrated circuit
- FPGA field programmable gate array signal
- PLD programmable logic device
- a general purpose processor may be a microprocessor, but in the alternative, the processor may be any commercially available processor, controller, microcontroller or state machine.
- a processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
- a software module may reside in any form of storage medium that is known in the art. Some examples of storage media that may be used include random access memory (RAM), read only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM and so forth.
- RAM random access memory
- ROM read only memory
- flash memory EPROM memory
- EEPROM memory EEPROM memory
- registers a hard disk, a removable disk, a CD-ROM and so forth.
- a software module may comprise a single instruction, or many instructions, and may be distributed over several different code segments, among different programs, and across multiple storage media.
- a storage medium may be coupled to a processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor.
- the methods disclosed herein comprise one or more steps or actions for achieving the described method.
- the method steps and/or actions may be interchanged with one another without departing from the scope of the claims.
- the order and/or use of specific steps and/or actions may be modified without departing from the scope of the claims.
- a storage media may be any available media that can be accessed by a computer.
- Such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
- Disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray ® disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.
- certain embodiments may comprise a computer program product for performing the operations presented herein.
- a computer program product may comprise a computer readable medium having instructions stored (and/or encoded) thereon, the instructions being executable by one or more processors to perform the operations described herein.
- the computer program product may include packaging material.
- Software or instructions may also be transmitted over a transmission medium.
- a transmission medium For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio and microwave are included in the definition of transmission medium.
- DSL digital subscriber line
- modules and/or other appropriate means for performing the methods and techniques described herein can be downloaded and/or otherwise obtained by a user terminal and/or base station as applicable.
- a user terminal and/or base station can be coupled to a server to facilitate the transfer of means for performing the methods described herein.
- various methods described herein can be provided via storage means (e.g., RAM, ROM, a physical storage medium such as a compact disc (CD) or floppy disk, etc.), such that a user terminal and/or base station can obtain the various methods upon coupling or providing the storage means to the device.
- storage means e.g., RAM, ROM, a physical storage medium such as a compact disc (CD) or floppy disk, etc.
- CD compact disc
- floppy disk etc.
- any other suitable technique for providing the methods and techniques described herein to a device can be utilized.
Abstract
Description
Claims
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013518841A JP2013534017A (en) | 2010-07-07 | 2011-07-07 | Method and system for interchangeable synaptic load storage in a neuroprocessor |
CN201180033657.2A CN102971754B (en) | 2010-07-07 | 2011-07-07 | For the method and system that the alternative synapse weight in neuron processor stores |
KR1020137003298A KR101466251B1 (en) | 2010-07-07 | 2011-07-07 | Methods, apparatuses and computer-readable medium for implementing a neural system, and an electrical circuit thereby |
EP11733755.0A EP2591449A1 (en) | 2010-07-07 | 2011-07-07 | Methods and systems for replaceable synaptic weight storage in neuro-processors |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US12/831,484 US8676734B2 (en) | 2010-07-07 | 2010-07-07 | Methods and systems for replaceable synaptic weight storage in neuro-processors |
US12/831,484 | 2010-07-07 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2012006468A1 true WO2012006468A1 (en) | 2012-01-12 |
Family
ID=44584777
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2011/043254 WO2012006468A1 (en) | 2010-07-07 | 2011-07-07 | Methods and systems for replaceable synaptic weight storage in neuro-processors |
Country Status (6)
Country | Link |
---|---|
US (1) | US8676734B2 (en) |
EP (1) | EP2591449A1 (en) |
JP (3) | JP2013534017A (en) |
KR (1) | KR101466251B1 (en) |
CN (1) | CN102971754B (en) |
WO (1) | WO2012006468A1 (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104854602A (en) * | 2012-12-03 | 2015-08-19 | Hrl实验室有限责任公司 | Generating messages from the firing of pre-synaptic neurons |
US10339041B2 (en) | 2013-10-11 | 2019-07-02 | Qualcomm Incorporated | Shared memory architecture for a neural simulator |
US10387298B2 (en) | 2017-04-04 | 2019-08-20 | Hailo Technologies Ltd | Artificial neural network incorporating emphasis and focus techniques |
US11221929B1 (en) | 2020-09-29 | 2022-01-11 | Hailo Technologies Ltd. | Data stream fault detection 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 |
US11237894B1 (en) | 2020-09-29 | 2022-02-01 | Hailo Technologies Ltd. | Layer control unit instruction addressing safety mechanism in an artificial neural network processor |
US11263077B1 (en) | 2020-09-29 | 2022-03-01 | Hailo Technologies Ltd. | Neural network intermediate results safety mechanism in an artificial neural network processor |
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 |
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 |
Families Citing this family (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8977583B2 (en) * | 2012-03-29 | 2015-03-10 | International Business Machines Corporation | Synaptic, dendritic, somatic, and axonal plasticity in a network of neural cores using a plastic multi-stage crossbar switching |
CN104809501B (en) * | 2014-01-24 | 2018-05-01 | 清华大学 | A kind of computer system based on class brain coprocessor |
CN104809498B (en) * | 2014-01-24 | 2018-02-13 | 清华大学 | A kind of class brain coprocessor based on Neuromorphic circuit |
US20150324691A1 (en) * | 2014-05-07 | 2015-11-12 | Seagate Technology Llc | Neural network connections using nonvolatile memory devices |
EP3035249B1 (en) * | 2014-12-19 | 2019-11-27 | Intel Corporation | Method and apparatus for distributed and cooperative computation in artificial neural networks |
US11100397B2 (en) | 2015-05-21 | 2021-08-24 | Rochester Institute Of Technology | Method and apparatus for training memristive learning systems |
US11226840B2 (en) | 2015-10-08 | 2022-01-18 | Shanghai Zhaoxin Semiconductor Co., Ltd. | Neural network unit that interrupts processing core upon condition |
US10664751B2 (en) | 2016-12-01 | 2020-05-26 | Via Alliance Semiconductor Co., Ltd. | Processor with memory array operable as either cache memory or neural network unit memory |
US11216720B2 (en) | 2015-10-08 | 2022-01-04 | Shanghai Zhaoxin Semiconductor Co., Ltd. | Neural network unit that manages power consumption based on memory accesses per period |
US11221872B2 (en) | 2015-10-08 | 2022-01-11 | Shanghai Zhaoxin Semiconductor Co., Ltd. | Neural network unit that interrupts processing core upon condition |
US10725934B2 (en) | 2015-10-08 | 2020-07-28 | Shanghai Zhaoxin Semiconductor Co., Ltd. | Processor with selective data storage (of accelerator) operable as either victim cache data storage or accelerator memory and having victim cache tags in lower level cache wherein evicted cache line is stored in said data storage when said data storage is in a first mode and said cache line is stored in system memory rather then said data store when said data storage is in a second mode |
US11029949B2 (en) | 2015-10-08 | 2021-06-08 | Shanghai Zhaoxin Semiconductor Co., Ltd. | Neural network unit |
CN114037068A (en) * | 2016-03-21 | 2022-02-11 | 杭州海存信息技术有限公司 | Neural computation circuit using three-dimensional memory to store activation function look-up table |
US11270193B2 (en) | 2016-09-30 | 2022-03-08 | International Business Machines Corporation | Scalable stream synaptic supercomputer for extreme throughput neural networks |
US10489702B2 (en) * | 2016-10-14 | 2019-11-26 | Intel Corporation | Hybrid compression scheme for efficient storage of synaptic weights in hardware neuromorphic cores |
CN108073982B (en) * | 2016-11-18 | 2020-01-03 | 上海磁宇信息科技有限公司 | Brain-like computing system |
US10423876B2 (en) * | 2016-12-01 | 2019-09-24 | Via Alliance Semiconductor Co., Ltd. | Processor with memory array operable as either victim cache or neural network unit memory |
US11580373B2 (en) | 2017-01-20 | 2023-02-14 | International Business Machines Corporation | System, method and article of manufacture for synchronization-free transmittal of neuron values in a hardware artificial neural networks |
US10197971B1 (en) | 2017-08-02 | 2019-02-05 | International Business Machines Corporation | Integrated optical circuit for holographic information processing |
JP6914342B2 (en) * | 2017-09-07 | 2021-08-04 | パナソニック株式会社 | Neural network arithmetic circuit using semiconductor memory element |
US11074499B2 (en) | 2017-11-20 | 2021-07-27 | International Business Machines Corporation | Synaptic weight transfer between conductance pairs with polarity inversion for reducing fixed device asymmetries |
WO2019141902A1 (en) * | 2018-01-17 | 2019-07-25 | Nokia Technologies Oy | An apparatus, a method and a computer program for running a neural network |
CN109886416A (en) * | 2019-02-01 | 2019-06-14 | 京微齐力(北京)科技有限公司 | The System on Chip/SoC and machine learning method of integrated AI's module |
CN112684977A (en) * | 2019-10-18 | 2021-04-20 | 旺宏电子股份有限公司 | Memory device and in-memory computing method thereof |
US11270759B2 (en) | 2019-10-21 | 2022-03-08 | Samsung Electronics Co., Ltd. | Flash memory device and computing device including flash memory cells |
KR20210047413A (en) | 2019-10-21 | 2021-04-30 | 삼성전자주식회사 | Flash memory device and computing device incuding flash meory cells |
US11264082B2 (en) | 2019-10-28 | 2022-03-01 | Samsung Electronics Co., Ltd. | Memory device, memory system and autonomous driving apparatus |
KR20210050634A (en) | 2019-10-28 | 2021-05-10 | 삼성전자주식회사 | Memory device, memory system and autonomous driving apparatus |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5448682A (en) * | 1992-05-30 | 1995-09-05 | Gold Star Electron Co., Ltd. | Programmable multilayer neural network |
Family Cites Families (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5214743A (en) * | 1989-10-25 | 1993-05-25 | Hitachi, Ltd. | Information processing apparatus |
US5329611A (en) * | 1990-05-22 | 1994-07-12 | International Business Machines Corp. | Scalable flow virtual learning neurocomputer |
JPH04182769A (en) * | 1990-11-17 | 1992-06-30 | Nissan Motor Co Ltd | Digital neuro processor |
IT1244912B (en) * | 1991-01-31 | 1994-09-13 | Texas Instruments Italia Spa | LEARNING SYSTEM FOR A NEURONIC NETWORK OF OPPORTUNITY ARCHITECTURE PHYSICALLY INSERABLE IN THE LEARNING PROCESS. |
IT1244911B (en) * | 1991-01-31 | 1994-09-13 | Texas Instruments Italia Spa | PHYSICALLY INSERABLE NEURONIC NETWORK ARCHITECTURE IN THE LEARNING PROCESS. |
JPH04320223A (en) * | 1991-04-19 | 1992-11-11 | Sharp Corp | Auto iris device |
JPH0561843A (en) * | 1991-05-15 | 1993-03-12 | Wacom Co Ltd | Neural network device |
JPH04346040A (en) * | 1991-05-23 | 1992-12-01 | Minolta Camera Co Ltd | Colorimeter |
JPH0561847A (en) * | 1991-09-02 | 1993-03-12 | Toshiba Corp | Neural network learning device |
US5278945A (en) * | 1992-01-10 | 1994-01-11 | American Neuralogical, Inc. | Neural processor apparatus |
KR0178805B1 (en) * | 1992-08-27 | 1999-05-15 | 정호선 | Multilayer neural net |
JPH06250997A (en) * | 1993-03-01 | 1994-09-09 | Toshiba Corp | High-speed learning device for neural circuit network |
JPH07192073A (en) * | 1993-12-27 | 1995-07-28 | Rohm Co Ltd | Programmable device and device using the programmable device |
EP0694854B1 (en) * | 1994-07-28 | 2002-06-05 | International Business Machines Corporation | Improved neural semiconductor chip architectures and neural networks incorporated therein |
JPH1021209A (en) * | 1996-06-27 | 1998-01-23 | Shimazu Yasumasa | General purpose digital neural computer |
JP2004295177A (en) * | 2003-03-25 | 2004-10-21 | Seiko Epson Corp | Controller of electronic equipment and printer controller |
MY138544A (en) * | 2003-06-26 | 2009-06-30 | Neuramatix Sdn Bhd | Neural networks with learning and expression capability |
JP2005098555A (en) * | 2003-09-22 | 2005-04-14 | Denso Corp | Neural network type air conditioner |
WO2006018898A1 (en) * | 2004-08-20 | 2006-02-23 | Fujitsu Limited | Wireless network system |
JP2006085557A (en) * | 2004-09-17 | 2006-03-30 | Toshiba Corp | Memory management device and data processing device |
US7533071B2 (en) * | 2005-06-28 | 2009-05-12 | Neurosciences Research Foundation, Inc. | Neural modeling and brain-based devices using special purpose processor |
US7904398B1 (en) * | 2005-10-26 | 2011-03-08 | Dominic John Repici | Artificial synapse component using multiple distinct learning means with distinct predetermined learning acquisition times |
US20070288410A1 (en) * | 2006-06-12 | 2007-12-13 | Benjamin Tomkins | System and method of using genetic programming and neural network technologies to enhance spectral data |
JP4842742B2 (en) * | 2006-09-05 | 2011-12-21 | 富士通株式会社 | Software management program, software management method, and software management apparatus |
US7814038B1 (en) * | 2007-12-06 | 2010-10-12 | Dominic John Repici | Feedback-tolerant method and device producing weight-adjustment factors for pre-synaptic neurons in artificial neural networks |
JP5154666B2 (en) * | 2008-03-14 | 2013-02-27 | ヒューレット−パッカード デベロップメント カンパニー エル.ピー. | Neuromorphic circuit |
-
2010
- 2010-07-07 US US12/831,484 patent/US8676734B2/en active Active
-
2011
- 2011-07-07 WO PCT/US2011/043254 patent/WO2012006468A1/en active Application Filing
- 2011-07-07 CN CN201180033657.2A patent/CN102971754B/en active Active
- 2011-07-07 EP EP11733755.0A patent/EP2591449A1/en not_active Ceased
- 2011-07-07 JP JP2013518841A patent/JP2013534017A/en not_active Withdrawn
- 2011-07-07 KR KR1020137003298A patent/KR101466251B1/en active IP Right Grant
-
2015
- 2015-04-01 JP JP2015075223A patent/JP2015167019A/en active Pending
-
2017
- 2017-08-14 JP JP2017156493A patent/JP2018014114A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5448682A (en) * | 1992-05-30 | 1995-09-05 | Gold Star Electron Co., Ltd. | Programmable multilayer neural network |
Non-Patent Citations (1)
Title |
---|
MISRA J ET AL: "Artificial neural networks in hardware: A survey of two decades of progress", NEUROCOMPUTING, ELSEVIER SCIENCE PUBLISHERS, AMSTERDAM, NL, vol. 74, no. 1-3, 1 December 2010 (2010-12-01), pages 239 - 255, XP027517200, ISSN: 0925-2312, [retrieved on 20101123], DOI: 10.1016/J.NEUCOM.2010.03.021 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104854602A (en) * | 2012-12-03 | 2015-08-19 | Hrl实验室有限责任公司 | Generating messages from the firing of pre-synaptic neurons |
CN104854602B (en) * | 2012-12-03 | 2017-04-19 | Hrl实验室有限责任公司 | Neural network unit and relevant system and method |
US10339041B2 (en) | 2013-10-11 | 2019-07-02 | Qualcomm Incorporated | Shared memory architecture for a neural simulator |
US11675693B2 (en) | 2017-04-04 | 2023-06-13 | Hailo Technologies Ltd. | Neural network processor incorporating inter-device connectivity |
US11216717B2 (en) | 2017-04-04 | 2022-01-04 | Hailo Technologies Ltd. | Neural network processor incorporating multi-level hierarchical aggregated computing and memory elements |
US10387298B2 (en) | 2017-04-04 | 2019-08-20 | Hailo Technologies Ltd | Artificial neural network incorporating emphasis and focus techniques |
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 |
US11238331B2 (en) | 2017-04-04 | 2022-02-01 | Hailo Technologies Ltd. | System and method for augmenting an existing artificial neural network |
US11551028B2 (en) | 2017-04-04 | 2023-01-10 | Hailo Technologies Ltd. | Structured weight 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 |
US11615297B2 (en) | 2017-04-04 | 2023-03-28 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network compiler |
US11354563B2 (en) | 2017-04-04 | 2022-06-07 | Hallo Technologies Ltd. | Configurable and programmable sliding window based memory access in a neural network processor |
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 |
US11461615B2 (en) | 2017-04-04 | 2022-10-04 | Hailo Technologies Ltd. | System and method of memory access of multi-dimensional data |
US11514291B2 (en) | 2017-04-04 | 2022-11-29 | Hailo Technologies Ltd. | Neural network processing element incorporating compute and local memory elements |
US11544545B2 (en) | 2017-04-04 | 2023-01-03 | Hailo Technologies Ltd. | Structured activation based sparsity in an artificial neural network |
US11221929B1 (en) | 2020-09-29 | 2022-01-11 | Hailo Technologies Ltd. | Data stream fault detection mechanism in an artificial neural network processor |
US11263077B1 (en) | 2020-09-29 | 2022-03-01 | Hailo Technologies Ltd. | Neural network intermediate results 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 |
US11811421B2 (en) | 2020-09-29 | 2023-11-07 | Hailo Technologies Ltd. | Weights safety mechanism in an artificial neural network processor |
Also Published As
Publication number | Publication date |
---|---|
JP2013534017A (en) | 2013-08-29 |
CN102971754A (en) | 2013-03-13 |
EP2591449A1 (en) | 2013-05-15 |
KR20130036323A (en) | 2013-04-11 |
US8676734B2 (en) | 2014-03-18 |
JP2018014114A (en) | 2018-01-25 |
CN102971754B (en) | 2016-06-22 |
KR101466251B1 (en) | 2014-11-27 |
US20120011087A1 (en) | 2012-01-12 |
JP2015167019A (en) | 2015-09-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US8676734B2 (en) | Methods and systems for replaceable synaptic weight storage in neuro-processors | |
EP2776988B1 (en) | Method and apparatus for using memory in probabilistic manner to store synaptic weights of neural network | |
TWI705444B (en) | Computing memory architecture | |
US11361216B2 (en) | Neural network circuits having non-volatile synapse arrays | |
TWI515670B (en) | Appartus,system and computer product for reinforcement learning | |
US9727459B2 (en) | Non-volatile, solid-state memory configured to perform logical combination of two or more blocks sharing series-connected bit lines | |
US11620505B2 (en) | Neuromorphic package devices and neuromorphic computing systems | |
US11880226B2 (en) | Digital backed flash refresh | |
CN110751276A (en) | Implementing neural networks with ternary inputs and binary weights in NAND memory arrays | |
KR102409859B1 (en) | Memory cells configured to generate weighted inputs for neural networks | |
US10761773B2 (en) | Resource allocation in memory systems based on operation modes | |
CN114121086A (en) | Bayesian networks in memory | |
US11694065B2 (en) | Spiking neural unit | |
CN116075832A (en) | Artificial neural network training in memory | |
JP2024035145A (en) | IMC circuit, neural network device including IMC circuit, and method of operating IMC circuit |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 201180033657.2 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 11733755 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 10419/CHENP/2012 Country of ref document: IN |
|
ENP | Entry into the national phase |
Ref document number: 2013518841 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REEP | Request for entry into the european phase |
Ref document number: 2011733755 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2011733755 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 20137003298 Country of ref document: KR Kind code of ref document: A |