CN113867639B - Qualification trace calculator based on phase change memory - Google Patents
Qualification trace calculator based on phase change memory Download PDFInfo
- Publication number
- CN113867639B CN113867639B CN202111141322.4A CN202111141322A CN113867639B CN 113867639 B CN113867639 B CN 113867639B CN 202111141322 A CN202111141322 A CN 202111141322A CN 113867639 B CN113867639 B CN 113867639B
- Authority
- CN
- China
- Prior art keywords
- phase change
- change memory
- conductance
- qualification
- data
- 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.)
- Active
Links
- 230000015654 memory Effects 0.000 title claims abstract description 83
- 230000008859 change Effects 0.000 title claims abstract description 71
- 238000012797 qualification Methods 0.000 title claims abstract description 70
- 230000002787 reinforcement Effects 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000002093 peripheral effect Effects 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 20
- 238000005265 energy consumption Methods 0.000 abstract description 8
- 238000011161 development Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002269 spontaneous effect Effects 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000012782 phase change material Substances 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
- G06F3/0601—Interfaces specially adapted for storage systems
- G06F3/0602—Interfaces specially adapted for storage systems specifically adapted to achieve a particular effect
- G06F3/0626—Reducing size or complexity of storage systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
- G06F3/0601—Interfaces specially adapted for storage systems
- G06F3/0602—Interfaces specially adapted for storage systems specifically adapted to achieve a particular effect
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Human Computer Interaction (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Algebra (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- Analogue/Digital Conversion (AREA)
Abstract
The invention discloses a qualification trace calculator based on a phase change memory, which comprises a phase change memory array and a result converter. By utilizing the multi-value characteristic of the phase change memory, qualification trace data is stored in the memory cells in a conductive form, and compared with the traditional binary storage mode, the quantity of the memory cells can be effectively reduced, so that high-density storage is realized; the attenuation operation along with time is spontaneously realized by utilizing the conductance drift effect of the phase change memory, and other operation circuits are not needed, so that the hardware cost of operation is effectively reduced; and the storage and the attenuation operation of the qualification trace data are carried out in the phase change memory, so that huge energy consumption caused by frequent data carrying is avoided. In addition, by adjusting parameters in the result converter, the decay rate of the qualification trace can be flexibly adjusted, thereby being suitable for reinforcement learning tasks with different requirements. The invention can break through the limit of the storage wall in the traditional computing architecture and promote the further development of reinforcement learning.
Description
Technical Field
The invention belongs to the technical field of novel calculation, and particularly relates to a qualification trace calculator based on an intrinsic conductivity drift effect of a phase change memory.
Background
The reinforcement learning algorithm is focused by a plurality of scientific researchers in recent years due to excellent performance, a strategy for solving the problem can be generated based on rewards and punishments of task environments, complex tasks in a plurality of fields can be effectively completed through the optimization strategy generated by multiple rounds of iteration, and guidance or supervision from the outside is not needed. The reinforcement learning algorithm which is continuously optimized can achieve performance which is close to or even surpasses the human level in the fields of automatic driving, game fight and the like at present. The strong function of the reinforcement learning algorithm is not separated from the support of an effective and common mechanism, namely qualification trace, the state trace of an intelligent body in reinforcement learning in one round of training can be recorded in a manner of time attenuation, and the updating amplitude of strategies corresponding to different states is guided based on the amplitude of the trace, so that the formation of an optimal strategy is accelerated, the cost of the reinforcement learning training process is reduced, and the final training effect is improved.
The qualification trace implemented on the traditional computing platform is obtained by calculating a large number of exponential decay functions, which not only requires a large number of multiplication operations, but also requires frequent data handling between the calculator and the memory, and has high energy consumption, thereby severely limiting the implementation of complex reinforcement learning algorithms. Phase change memories are a new type of nonvolatile memory that rely on the significant difference in conductance between crystalline and amorphous states of the internal phase change material to achieve high-speed, high-density data storage, while the interior of unstable amorphous materials spontaneously disintegrates in structure, creating a lower conductance glassy state, so that the conductance state of the phase change memory decays over time, known as conductance drift. By reasonably utilizing the conductance drift of the phase change memory, the attenuation mechanism of the qualification trace can be automatically realized in an in-memory calculation mode, so that a large number of data handling and multiplication operations are avoided, and the expenditure of a large-scale reinforcement learning algorithm is effectively reduced.
Disclosure of Invention
In order to solve the problem that the energy consumption of the qualification trace calculation in the complex reinforcement learning algorithm is too high, the invention provides a qualification trace calculator based on the multi-value characteristic and the conductivity drift characteristic of a phase change memory, which can spontaneously realize the attenuation of the qualification trace in an in-memory calculation mode, thereby greatly reducing the energy consumption of the qualification trace calculation. By utilizing the spontaneous conductance drift effect of the phase change memory, the invention can automatically realize the attenuation operation of the qualification trace without a complex operation circuit, thereby effectively reducing the hardware cost; in addition, the storage and operation of the qualification trace are completed in the phase change memory, so that frequent data handling is avoided, and the energy consumption of operation is further reduced. The present invention thus provides significant advantages in terms of energy and hardware overhead over conventional qualification trace implementations.
The qualification trace calculator of the present invention is made up of two parts, referring to FIG. 1, the first part is a programmable phase change memory array including peripheral circuitry for generating programming pulses and reading device conductance and co-operatively connected phase change memory array cells; each phase change memory array unit consists of a phase change memory and a transistor, one end of the phase change memory is connected with the transistor, the other end of the phase change memory is grounded, and the transistor controls the on-off of the phase change memory and a peripheral circuit; each phase change memory stores corresponding qualification trace data in a conductive form and spontaneously carries out attenuation operation; the second part is a result converter comprising a comparator and a linear operator capable of converting conductance data read from the phase change memory array into qualification trace data for reinforcement learning.
The principle of the qualification trace calculator is that attenuation calculation is realized based on spontaneous conductivity drift of the phase change memory, and the rule of the conductivity drift is as follows: g (t) =g (t) 0 )(t/t 0 ) -v Wherein G (t) is the conductance of the phase change memory at time t, t 0 For the first time the conductance is measured, v is the conductance drift factor, which ranges from about 0.01 to 0.1. The reinforcement learning qualification trace updating process comprises two steps, wherein the first step is to update the qualification trace corresponding to the current state-action to 1: e (s, a) =1 (where E represents a qualification trace matrix, s represents a state number, a represents an action number), which can be achieved by a programming operation for a device at a corresponding location in the phase change memory array; the second step is to carry out attenuation operation on all qualification trace data, and the traditional implementation mode is as follows: e=αe (α<1, decay amplitude), whereas the qualification trace data G stored in the phase change memory in conductive form in the present invention spontaneously decays: g=g (t/t) 0 ) -v Without requiring external manipulation. Conductance data read from the phase change memory array is mapped to a range of 0-1 through a result converter for reinforcement learning, and the specific conversion process is as follows: first, the conductance G (s, a) is sent to a comparator and an upper and lower threshold G U 、G D Comparing, if G (s, a)>G U Then the corresponding eligibility E (s, a) =1; if G (s, a)<G D E (s, a) =0; if G D <G(s,a)<G U The data is sent to a linear operator for calculation: e (s, a) =k (G (s, a) -G D ) Where k=1/(G) U –G D ) Is the amplification factor; after the result converter, the conductance data is converted into qualification trace data ranging from 0 to 1, and the qualification trace data can further participate in measurement updating of reinforcement learning: q=q+δe, which isWhere Q is the policy table and delta is the update error.
Preferably, the result converter comprises two analog comparators and a linear operator, the conductance data G read from the phase change memory array is first fed into the first analog comparator and the upper conductance limit G U By comparison, if G>G U Then the corresponding qualification trace is determined directly as e=1; if G<G U Then G is fed into a second analog comparator and the lower conductance limit G D Continuing to compare; if G<G D Then the corresponding qualification trace is determined directly as e=0; if G>G D G is sent to a linear operator for conversion: e=k (G-b), where b=g D ，k＝1/(G U -G D ) The method comprises the steps of carrying out a first treatment on the surface of the Thereby converting the conductance data read from the phase change memory array to a range of 0,1]Is defined by the qualification trace data of the computer.
The invention provides a qualification trace calculator based on a conductivity drift effect of a phase change memory, which firstly utilizes the multi-value characteristic of the phase change memory, and a floating point type qualification trace data can be stored in a memory unit in a conductivity mode; the decay operation over time is then spontaneously implemented using the conductance drift effect of the phase change memory: g=g (t/t) 0 ) -v The hardware cost of operation can be effectively reduced without using other operation circuits; by adjusting the comparison threshold in the result converter, the decay speed of the qualification trace can be flexibly adjusted so as to be used in reinforcement learning tasks with different requirements; the storage and attenuation operation of the qualification trace data are completed in the phase change memory, so that frequent data handling is avoided, the energy consumption in the operation process can be effectively reduced, the limitation of a storage wall in a traditional computing architecture can be broken through, and the further development of reinforcement learning is promoted.
Drawings
FIG. 1 is a schematic diagram of a phase change memory-based qualification trace calculator according to the present invention.
FIG. 2 is a flowchart illustrating the operation of the phase change memory-based eligibility trace calculator of the present invention.
FIG. 3 is a schematic diagram showing the effect of the resulting converter on the modulation of the conductance drift effect according to the present invention.
Detailed Description
The present invention will be further described in detail with reference to the accompanying drawings, in order to more clearly clarify the objects, technical solutions and advantages of the present invention. The description herein is only for the purpose of illustrating the invention and is not to be construed as limiting the invention.
The invention provides a qualification trace calculator based on a conductance drift effect of a phase change memory, which not only can realize high-density storage of qualification trace data in the phase change memory, but also can automatically realize attenuation operation by means of the conductance drift effect. Compared with the traditional qualification trace calculation mode, the method reduces the hardware cost in the operation process and also avoids high energy consumption caused by carrying the data back and forth.
FIG. 1 is a schematic diagram of the overall structure of the present invention, and the qualification trace calculator is composed of two parts. The first part is a programmable phase change memory array, as shown on the left side of fig. 1, for storing qualification trace data and performing an automatic decay operation. Each unit in the array is composed of a phase-change memory and a transistor, each phase-change memory stores corresponding qualification trace data in a conductive mode, and the transistor is used for controlling connection and disconnection of the phase-change memory and the outside; one end of the phase change memory is connected with the transistor, and the other end of the phase change memory is grounded. Updating qualification trace data according to a reinforcement learning algorithm: when E (s, a) =1, the control circuit at the periphery of the array turns on the transistors of the row where the phase change memory G (s, a) is located, and applies a step-like programming current to the corresponding column, so that the programming current can be applied to the phase change memory G (s, a), thereby slightly improving the conductance thereof. The signal application scheme as shown in fig. 1 enables programming of phase change memories within the dashed box while also not affecting other phase change memories. In the subsequent decay operation, the qualification trace data stored in the phase change memory in the form of conductance spontaneously decays with time due to the conductance drift effect: g=g (t/t) 0 ) -v The operation process does not need external accessThe rows operate.
The second part of the invention is a result converter for converting the conductance data read in the phase change memory array into qualification trace data ranging from 0 to 1 as shown on the right side of fig. 1. The part is composed of two analog comparators and a linear arithmetic unit, the conductance data G read from the memory array is firstly sent into the first analog comparator and the upper limit G U By comparison, if G>G U Then the conductance data exceeds the upper limit and the corresponding qualification trace is determined directly as e=1, if G<G U Then G is fed into the next comparator and the lower conductance limit G D Continuing to compare; if G<G D Then the conductance data exceeds the lower limit and the corresponding qualification trace is determined directly as e=0, if G>G D G is described as being within a specified range, G being fed into a linear operator for conversion: e=k (G-b), where b=g D ，k＝1/(G U -G D ) The parameters are set so as to ensure that the calculated qualification trace data satisfies 0<E<1, a step of; by conversion by the result converter, the conductance data stored in the phase change memory is converted to a range of 0,1]And thus can be used for subsequent steps in reinforcement learning.
FIG. 2 is a flow chart of the phase change memory-based eligibility trace calculator of the present invention, wherein the process of calculating eligibility traces mainly comprises the following steps:
(1) According to the current state and action (s, a) in the reinforcement learning algorithm, selecting a corresponding device G (s, a) in the phase-change memory array, turning on a transistor of a unit of the device G, and applying programming current I_program to the phase-change memory to slightly improve the conductivity state of the phase-change memory.
(2) All conductivity data in the phase change memory array is read out and fed into the result converter.
(3) Selecting one G from the current conductivity data, and comparing the selected G with the upper conductivity limit G U Comparing, if G<G U Step (4) is entered, otherwise step (6) is entered.
(4) The conductance data G and the lower conductance limit G D Comparing, if G>G D Step (5) is entered, otherwiseAnd (7) a step (7).
(5) The conductance data G are linearly transformed: e=k (G-b), where b=g D ，k＝1/(G U -G D ) Outputting the qualification trace E and proceeding to step (8).
(6) Output eligibility e=1 and go to step (8).
(7) Output eligibility e=0 and go to step (8).
(8) And (3) judging whether all the conductivity data are converted, if so, entering a step (9), otherwise, returning to the step (3).
(9) And (5) completing the calculation of the qualification trace.
Since the rate of conductance drift of the phase change memory is substantially fixed, the phase change memory is controlled by the ratio of g=g (t/t 0 ) -v The way of (c) decays with time, characterized by a distance from the programming time t 0 The closer the conductance decays faster, and vice versa; however, the qualification trace attenuation rates required for different reinforcement learning tasks are different, so in the present invention, the upper and lower limits G of the conductance in steps (3) - (5) can be adjusted D 、G U The final qualification trace attenuation speed is adjusted in a mathematical operation mode, so that the final qualification trace attenuation speed is suitable for different reinforcement learning tasks without changing other hardware parts.
FIG. 3 is a schematic diagram showing the effect of the result converter on the modulation of the conductance drift effect according to the present invention, wherein (a) shows the comparison of the qualification trace attenuation effect achieved by the present invention with the conventional exponential attenuation effect before the result converter, wherein the dotted line clusters are the conductance drift effect (the conductance drift coefficient is in the range of 0.01-0.03 of the conventional phase change memory), and the solid line clusters are the exponential attenuation effect (the attenuation base is in the range of 0.8-0.9 of the conventional reinforcement learning). As can be seen by comparison, the difference of attenuation effects generated by the two operations is obvious, the qualification trace attenuation realized by the conductance drift is slower than the exponential attenuation, so that the direct use of the conductance drift in strengthening the learning effect is not ideal, and therefore, the conversion of the result of the conductance drift is indispensable. Fig. 3 (b) shows the effect of the decay of the conductance drift after the adjustment by the resulting converter, and by choosing the appropriate parameters k and b, the decay rate of the conductance drift can reach an exponential decay level, which also ensures that the qualification trace produced by the present invention can be used in reinforcement learning.
Compared with the traditional qualification trace calculation mode, the invention does not need a complex multiplication operator, effectively reduces the hardware cost in the operation process, stores qualification trace data and performs the attenuation operation in the phase change memory, and avoids huge energy consumption caused by frequent data carrying. In addition, by adjusting the parameters in the resulting converter of the present invention, the attenuation of the eligibility trace can be adjusted to a suitable rate, thereby adapting to different reinforcement learning tasks. The invention can effectively break through the limitation of the storage wall in the traditional computer architecture on the complex reinforcement learning algorithm, and has important significance for promoting reinforcement learning and further development of other artificial intelligence.
The above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and those skilled in the art may modify or substitute the technical solution of the present invention without departing from the spirit and scope of the present invention, and the protection scope of the present invention shall be subject to the claims.
Claims (5)
1. A qualification trace calculator based on a phase change memory, comprising a programmable phase change memory array and a result converter, wherein the programmable phase change memory array comprises peripheral circuits for generating programming pulses and reading device conductance and phase change memory array units connected in a common mode; each phase change memory array unit consists of a phase change memory and a transistor, one end of the phase change memory is connected with the transistor, the other end of the phase change memory is grounded, and the transistor controls the on-off of the phase change memory and a peripheral circuit; each phase change memory stores corresponding qualification trace data in a conductive form and spontaneously carries out attenuation operation; the result converter includes a comparator and a linear operator to convert conductance data read from the phase change memory array into qualification trace data.
2. The eligibility calculator of claim 1 wherein the result converter includes two analog comparators and a linear operator, conductance data G read from the phase change memory array being first fed to the first analog comparator and the upper conductance limit G U By comparison, if G>G U Then the corresponding qualification trace is determined directly as e=1; if G<G U Then G is fed into a second analog comparator and the lower conductance limit G D Continuing to compare; if G<G D Then the corresponding qualification trace is determined directly as e=0; if G>G D G is sent to a linear operator for conversion: e=k (G-b), where b=g D ，k＝1/(G U -G D ) The method comprises the steps of carrying out a first treatment on the surface of the Thereby converting the conductance data read from the phase change memory array to a range of 0,1]Is defined by the qualification trace data of the computer.
3. A method of performing a qualification computation in reinforcement learning using the qualification calculator of claim 1 or 2, comprising the steps of:
1) Selecting a corresponding array unit from the phase change memory array according to the current state and action in the reinforcement learning algorithm, opening a transistor of the unit, applying programming current I_program to the phase change memory, and slowly improving the conductivity state of the phase change memory;
2) Reading all conductivity data in the phase change memory array and sending the conductivity data into a result converter;
3) Selecting one G from the current conductivity data, and comparing the selected G with the upper conductivity limit G U Comparing, if G<G U Step 4) is entered, otherwise step 6) is entered;
4) The conductance data G and the lower conductance limit G D Comparing, if G>G D Enter step) otherwise enter step 7);
5) The conductance data G are linearly transformed: e=k (G-b), where b=g D ，k＝1/(G U -G D ) Outputting qualification trace E and entering step 8);
6) Outputting qualification trace e=1 and proceeding to step 8);
7) Outputting qualification trace e=0 and proceeding to step 8);
8) Judging whether all the conductivity data are converted, if yes, entering a step), otherwise, returning to the step 3);
9) And (5) completing the calculation of the qualification trace.
4. A method according to claim 3, wherein the programming current i_program in step 1) is of the staircase type.
5. A method according to claim 3, characterized in that the upper and lower limits G of conductance in steps 3) to 5) are adjusted according to different reinforcement learning task requirements D 、G U An adjustment of the final eligibility trace decay rate is achieved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111141322.4A CN113867639B (en) | 2021-09-28 | 2021-09-28 | Qualification trace calculator based on phase change memory |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111141322.4A CN113867639B (en) | 2021-09-28 | 2021-09-28 | Qualification trace calculator based on phase change memory |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113867639A CN113867639A (en) | 2021-12-31 |
| CN113867639B true CN113867639B (en) | 2024-03-19 |
Family
ID=78991734
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111141322.4A Active CN113867639B (en) | 2021-09-28 | 2021-09-28 | Qualification trace calculator based on phase change memory |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113867639B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104932267A (en) * | 2015-06-04 | 2015-09-23 | 曲阜师范大学 | Neural network learning control method adopting eligibility trace |
| CN108614422A (en) * | 2018-05-23 | 2018-10-02 | 中国农业大学 | Dissolved oxygen optimal control method, apparatus and system in the cultivation of continental rise plant circulation water |
| CN109343532A (en) * | 2018-11-09 | 2019-02-15 | 中国联合网络通信集团有限公司 | A kind of paths planning method and device of dynamic random environment |
| WO2020234658A1 (en) * | 2019-05-22 | 2020-11-26 | International Business Machines Corporation | Closed loop programming of phase-change memory |
| WO2021011923A1 (en) * | 2019-07-17 | 2021-01-21 | Mentium Technologies Inc. | Flash memory arrays for computation having digital input and analog output |
-
2021
- 2021-09-28 CN CN202111141322.4A patent/CN113867639B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104932267A (en) * | 2015-06-04 | 2015-09-23 | 曲阜师范大学 | Neural network learning control method adopting eligibility trace |
| CN108614422A (en) * | 2018-05-23 | 2018-10-02 | 中国农业大学 | Dissolved oxygen optimal control method, apparatus and system in the cultivation of continental rise plant circulation water |
| CN109343532A (en) * | 2018-11-09 | 2019-02-15 | 中国联合网络通信集团有限公司 | A kind of paths planning method and device of dynamic random environment |
| WO2020234658A1 (en) * | 2019-05-22 | 2020-11-26 | International Business Machines Corporation | Closed loop programming of phase-change memory |
| WO2021011923A1 (en) * | 2019-07-17 | 2021-01-21 | Mentium Technologies Inc. | Flash memory arrays for computation having digital input and analog output |
Non-Patent Citations (1)
| Title |
|---|
| Yigit Demirag,et al..PCM-Trace:Scalable Synaptic Eligibility Traces with Resistivity Drift of Phase-Change Materials.《2021 IEEE International Symposium on Circuits and Systems（ISCAS）》.2021,第1-4页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113867639A (en) | 2021-12-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112151091B (en) | 8T SRAM unit and memory computing device | |
| CN112348168B (en) | Ultra-short-term load prediction method and system considering data loss and feature redundancy | |
| CN208547942U (en) | Simulate vector-matrix multiplication operation circuit | |
| CN113867639B (en) | Qualification trace calculator based on phase change memory | |
| US6292398B1 (en) | Method for the in-writing verification of the threshold value in non-volatile memories | |
| CN110033797B (en) | Storage system and storage method | |
| Qiao et al. | Online-growing neural network control for dissolved oxygen concentration | |
| CN111079888B (en) | Water quality dissolved oxygen prediction method and system based on hybrid QPSO-DE optimization | |
| Wu et al. | Bulk‐Switching Memristor‐Based Compute‐In‐Memory Module for Deep Neural Network Training | |
| CN109635915A (en) | A kind of iterative learning control method based on balanced single evolution cuckoo algorithm | |
| CN115755589B (en) | Optimization method of fuzzy control system in reaction kettle based on genetic algorithm | |
| CN113672855A (en) | Memory operation method, device and application thereof | |
| CN114186667B (en) | Mapping method of cyclic neural network weight matrix to memristor array | |
| CN116243604A (en) | Self-adaptive neural network sliding mode control method, device and medium for sewage denitrification process | |
| Fan et al. | Optimization of Controller for Microbial Fuel Cell: Comparison between Genetic Algorithm and Fuzzy Logic | |
| CN115762600A (en) | Memristor array-based n-bit s-word multi-state programmable storage circuit and use method | |
| CN115239546A (en) | Forward gradient regression accelerator based on nonvolatile memory and operation method | |
| Topalov et al. | Fast genetic on-line learning algorithm for neural network and its application to temperature control | |
| CN118866050A (en) | Writing verification method based on finite state machine for memristive device | |
| CN118762733A (en) | Weight writing method and circuit system of multi-resistance state resistance change memory | |
| Mittelholzer et al. | Rewritable channels with data-dependent noise | |
| CN117236391A (en) | Bayesian neural network storage and calculation integrated method based on random calculation by utilizing MRAM | |
| CN114048968B (en) | Real-time blocking adjustment method, device and medium for key section of power grid | |
| CN118760422A (en) | Oscillation convergence programming method and device for multi-value resistive random access memory | |
| CN118428428B (en) | Neuron self-learning circuit based on memristor threshold characteristic |
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 | ||
| GR01 | Patent grant |