CN109116968A

CN109116968A - Write policy control method, system and its applicable electronic equipment

Info

Publication number: CN109116968A
Application number: CN201810884580.3A
Authority: CN
Inventors: 马恺声
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2018-08-06
Filing date: 2018-08-06
Publication date: 2019-01-01
Anticipated expiration: 2038-08-06
Also published as: CN109116968B

Abstract

The application provides a write strategy control method, system and its applicable electronic equipment, and the policy control method of writing is applied in the electronic equipment with processor, the power cut-off information that the method passes through acquisition at least one moment of electronic equipment；And the following power-off time and power-off confidence level that prediction calculates the electronic equipment are carried out according to the power cut-off information；Then prediction is carried out according to the following power-off time and power-off confidence level and calculate to obtain writing policy instructions and writing tactful confidence level, execute write operation to enable the processor write policy instructions according to described in；Spare energy can be reduced by improving backup retention time whereby, and reasonably write strategy realizes energy-efficient control in turn is determined to the prediction of power-off time.

Description

Write policy control method, system and its applicable electronic equipment

Technical field

This application involves energy management technical fields, more particularly to a write strategy control method, write policy control system System, non-volatile processor, neural network chip, electronic equipment, electronic device and computer readable storage medium.

Background technique

The fast development of internet of things equipment and wireless communication technique, so that the wearable device of such as high-performance, small size Or the development that implantable devices are advanced by leaps and bounds, high performance demand mean the promotion of system energy consumption, and current battery Development speed lagged far behind the increases of energy requirements, and battery power supply still remains that volume weight is big and maintenance expense With expensive problem.For this purpose, wearable device or implantable devices would generally realize self energizing by acquisition outside energy, so And self energizing is there is finite energy, variation acutely and it is difficult to predict defect, therefore, Internet of things node needs pass through conjunction Reason stores and utilizes limited collecting energy, and carries out reasonable energy management according to the energy requirement of different loads, makes it possible to Utilization efficiency is measured to optimize.

In Internet of things node, other than in addition to the signal processing inside processor and controlling operation, there is also processors and periphery Equipment carries out the operation of data communication and information exchange, for example sensor information is read back into processor from sensor, memory The write-in and reading of data and data are transmitted and received etc. by radio frequency unit in chip.These can all cause processor There is very high demand to electric energy.In self-power supply system, each atomic operation all it has to be ensured that under enough energy ability It completes.Therefore, the energy management apparatus of system is required to provide such support.For this purpose, scientific and reasonable energy management is then shown It obtains particularly important.

Summary of the invention

In view of the disadvantage of existing correlation described above, the application's is designed to provide a write strategy control method, writes Policy controlling system, non-volatile processor, neural network chip, electronic equipment, electronic device and computer readable storage medium, To carry out energy management in the way of low cost.

In order to achieve the above objects and other related objects, the first aspect of the application provides a write strategy control method, Applied in the electronic equipment with processor, comprising the following steps: obtain the power-off at least one moment of electronic equipment Information；The following power-off time and power-off confidence level that prediction calculates the electronic equipment are carried out according to the power cut-off information；And It carries out prediction according to the following power-off time and power-off confidence level and calculates to obtain writing policy instructions and writing tactful confidence level, to enable The processor writes policy instructions according to described in and executes write operation；The policy instructions of writing include executing the write current of write operation And write time at least one information.

The second aspect of the application provides a write strategy control system, applied in the electronic equipment with processor, Including power-off fallout predictor and BACKUP TIME fallout predictor, wherein the power-off fallout predictor is used for according to the electronic equipment at least The power cut-off information at one moment carries out the following power-off time and power-off confidence level that prediction calculates the electronic equipment；The backup Versus time estimator is used to carry out prediction calculating according to the following power-off time and power-off confidence level to obtain to write policy instructions and write Tactful confidence level executes write operation to enable the processor write policy instructions according to described in；The policy instructions of writing include executing The write current and write time at least one information of write operation.

The third aspect of the application provides a kind of non-volatile processor, including writing policy control described in above-mentioned second aspect System and retention time controller, the retention time controller described write writing for policy controlling system output to receive When policy instructions, the write current for including in policy instructions is write according to described in and behaviour is write in the execution of write time at least one information Make.

The fourth aspect of the application provides a kind of neural network chip, including writing policy control described in above-mentioned second aspect System.

The 5th aspect of the application provides a kind of electronic equipment, including writing plan described in processor and above-mentioned second aspect Slightly control system.

The 6th aspect of the application provides a kind of electronic device, including writing policy control system described in above-mentioned second aspect System.

The 7th aspect of the application provides a kind of computer readable storage medium, is stored with the computer journey of energy management Sequence, the computer program are performed described in the above-mentioned first aspect of realization and write policy control method.

As described above, the application's writes policy control method, writes policy controlling system, non-volatile processor, neural network It is a large amount of standby in data backup operation consumption to improve system for chip, electronic equipment, electronic device and computer readable storage medium The case where with energy, is determined by the prediction to power-off time and reasonably writes strategy to improve backup retention time, Jin Ershi Existing energy-efficient control.

Detailed description of the invention

What Fig. 1 was shown as the application writes the flow chart of policy control method in one embodiment.

Fig. 2 is shown as the circuit block diagram of the characteristic extracting module of the application in one embodiment.

Fig. 3 is shown as a kind of the application mind neural network schematic diagram in one embodiment.

Fig. 4 is shown as the application mind another neural network schematic diagram in one embodiment.

What Fig. 5 was shown as the application writes the flow chart of policy control method in another embodiment.

Fig. 6 is shown as the application and writes the write current of strategy and the relationship signal of write pulse width in one embodiment Figure.

Fig. 7 is shown as the flow chart of the application write in the another embodiment of policy control method.

What Fig. 8 was shown as the application writes the schematic diagram of policy controlling system in one embodiment.

What Fig. 9 was shown as the application writes the write operation circuit diagram of policy controlling system in one embodiment.

What Figure 10 was shown as the application writes the schematic diagram of policy controlling system in another embodiment.

What Figure 11 was shown as the application writes schematic diagram of the policy controlling system in another embodiment

What Figure 12 was shown as the application writes the multiplexing hardware structure schematic diagram of policy controlling system in one embodiment.

Figure 13 is shown as the application and writes the multiplexing hardware structure schematic diagram of policy controlling system in another embodiment.

Figure 14 is shown as the configuration diagram of herein described non-volatile processor in one embodiment.

Figure 15 is shown as the configuration diagram of herein described neural network chip in one embodiment.

Figure 16 is shown as the schematic diagram of the electronic equipment of the application in one embodiment.

Figure 17 is shown as the schematic diagram of the electronic equipment of the application in another embodiment.

Figure 18 is shown as the application approximate calculation configuration diagram in one embodiment.

Specific embodiment

Presently filed embodiment is illustrated by particular specific embodiment below, those skilled in the art can be by this explanation Content disclosed by book understands other advantages and effect of the application easily.In described below, with reference to attached drawing, attached drawing description Several embodiments of the application.It should be appreciated that other embodiments also can be used, and can be without departing substantially from spirit herein With carried out in the case where range composition and operational change.Following detailed description should not be considered limiting, And the range of embodiments herein is only limited by claims of the patent of the application.Term used herein is only In order to describe specific embodiment, and it is not intended to limit the application.

Although term first, second etc. are used to describe various elements herein in some instances, these elements It should not be limited by these terms.These terms are only used to distinguish an element with another element.For example, the first threshold Value can be referred to as second threshold, and similarly, and second threshold can be referred to as first threshold, without departing from various described Embodiment range.

Furthermore as used in herein, singular " one ", "one" and "the" are intended to also include plural number shape Formula, unless there is opposite instruction in context.It for example in the application then include moment and more to " at least one moment " The case where a moment.Used herein, the phrase for quoting from the "at least one" in a list of items refers to appointing for these projects What is combined, including single member.As an example, " 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。

It will be further understood that term "comprising", " comprising " show that there are the feature, step, operation, element, groups Part, project, type, and/or group, but it is not excluded for one or more other features, step, operation, element, component, project, kind Presence, appearance or the addition of class, and/or group.Term "or" and being interpreted as including property of "and/or" used herein, or meaning Taste any one or any combination.It should be understood that the terms "and/or", only a kind of association for describing affiliated partner is closed System indicates may exist three kinds of relationships, for example, A and/or B, can indicate: individualism A exists simultaneously A and B, individualism These three situations of B.In addition, character "/" herein, typicallys represent the relationship that forward-backward correlation object is a kind of "or".

It is to be understood that " instruction " can also be construed broadly as into mean instruction, data, information, letter in this application Number or any combination thereof etc..Described " non-volatile " or " non-volatile " is represented as identical concept, and corresponding English is Nonvolatile。

In internet of things equipment such as wearable device or implantable devices, for example, the cost of replacement battery, cell safety Property, battery compartment volume, charging time and timeliness are all factors in need of consideration, and many equipment wish the volume of battery in system It is small even without battery as far as possible, therefore met the tendency of including the self-contained electric system that environmental energy collects power supply or data acquisition device And give birth to, especially with the development of non-volatile memory (Nonvolatile Processors, abbreviation NVP), so that environment Energy scavenging source is popularized in the use of wearable device.Non-volatile processor can be located by backup calculating state Unstable inputing power is managed, compared with battery, it can be ensured that in very short time range, allow to use these processors System work in the case where no battery or supercapacitor.

Although non-volatile processor may insure that program connects in the case where unstable electric energy supply to a certain extent It is continuous to execute, but when the power supply shakiness of data acquisition device, the processor in available data acquisition device can not be accomplished to handle Most freshly harvested data guarantee data precision, restore to calculate or back up calculating etc..Certainly, make full use of the energy that can increase standby The quantity of part operation, but also cause to waste more energy in unnecessary backup and recovery operation simultaneously, and according to section The strategy of the energy is saved, then may cause the unnecessary leakage of capacitor again, in addition capacitor can not be deposited in full electric situation The energy that Chu Xin is collected, the service response time that will also postpone.This just needs to carry out it optimization of energy management, such as in advance Following energy input is surveyed to be preferably that a follow-up work distributes resource, and prediction interruption duration is used for reduce The retention time and electric energy of backup operation.

Although many electronic equipments are realized with the help of non-volatile elements by unstable power supply (such as self-powered system Energy collecting device in system) it is powered, but the backup operation of electronic equipment at work still consumes a large amount of electric energy, especially It is to consume a large amount of electric energy when power supply is intermittent high.It therefore, if can be by obtaining power cut-off information, by NVM (Non- Volatile memory, nonvolatile storage, abbreviation NVM) retention time shortens to from long time (such as 10 years or more) It is only more a little longer than power-off time, the necessary energy can be saved during data backup operation.In addition, due in electronic equipment Usually there is block/distribution nonvolatile memory (NVM), this makes system usually a large amount of spare in data backup operation consumption Energy can reduce spare energy for these unnecessary consumption by improving backup retention time, and then realize energy conservation Control.For this purpose, in this application, can be determined by the prediction to power-off time and reasonably write strategy, in other words, then for At what time using great write current and the time backed up using the write time how long come determination data.

The application provide it is a kind of write policy control method applied in the electronic equipment with processor, to power off The write operation of data backup is carried out in period, using the energy of storage so as to the processor described when next power up cycle arrives Continuation operation can be carried out based on the data of the write-in.In embodiment, the processor is, for example, non-volatile memory (NVP), however, it is not limited to this, in other examples, without departing substantially from inventive concept disclosed herein and thought In the case of, the processor is also possible to common processor, for example can be any commercially available processor, controller, micro-control Device or state machine processed.Processor is also implemented as calculating the combination of equipment, such as the combination, multiple of DSP and microprocessor Microprocessor, the one or more microprocessors cooperateed with DSP core or any other such configuration.

In embodiment, the electronic equipment is internet of things equipment, for example, wearable device or implantable devices, such as Wearable electronic may include any kind of electronic equipment that can be worn on the limbs of user.The wearable electronic is set It is standby to can be fixed on limbs such as wrist, ankle, arm or the leg of the mankind.This class of electronic devices includes but is not limited to health Or body-building assistant devices, digital music player, smart phone, calculate equipment or display take exercise or other active monitors, The equipment that can be given the correct time, equipment of biological characteristic parameter that wearer or user can be measured etc..The implantable devices are for example For blood sugar test equipment etc..

As an example, wearable electronic can be implemented as the form of wearable health-care aid, this is wearable strong (real-time or non real-time) provide of information relevant to health is arrived user, authorized third party and/or is associated by Kang assistant Supervision equipment.The equipment can be configured to provide information or data relevant to health, such as, but not limited to heart rate data, blood Data, temperature data, blood oxygen saturation data, diet/nutritional information, medical alert, prompt relevant to health or information are pressed, Or other data relevant to health.Associated supervision equipment can help for such as tablet computing device, phone, individual digital Reason, computer etc..

As another example, electronic equipment can be configured to the form of wearable communication equipment.Wearable communication equipment May include the processor for being coupled to the memory or being communicated, one or more communication interface, output equipment (such as display and Loudspeaker) and one or more input equipments.One or more communication interfaces can provide communication equipment and any PERCOM peripheral communication Electronic communication between network, equipment or platform, the communication interface such as, but not limited to wireless interface, blue tooth interface, USB connect Mouth, Wi-Fi interface, TCP/IP interface, network communication interface or any conventional communication interface.Other than communication, it can wear Wear communication equipment can provide the equipment about time, health, state or external connection or the equipment communicated and/or The information of the software run on these devices, message, video, operational order etc. (and can be received from external equipment above-mentioned Any one of).

Referring to Fig. 1, be shown as the application writes the flow chart of policy control method in one embodiment, as shown, It is described write policy control method the following steps are included:

Step S10 is first carried out, obtains power cut-off information (the Power Outage at least one moment of electronic equipment Sensing)；In embodiment, at least one moment of electronic equipment is obtained when detecting electronic equipment power-off Power cut-off information.

In embodiment, it can be in different time periods at the time of described, can be divided into according to different demands multiple Moment grade, such as 10ms, 100ms, 1s, 2s, 3s, 10s, 1 minute 10 minutes, 1 hour, 1 day etc., need to be stated clearly, above-mentioned The citing at moment is exemplary, under different status of implementation, it is not limited to this.

In embodiment, the power cut-off information is the Huo Zheren because energy supply is insufficient or depleted of energy in electronic equipment The information interrupted is inputted for electric energy caused by the factors such as factor (such as artificial setting or artificial damage) or unpredictalbe accident, Such as the information such as duration of time and power-off powered off.In one embodiment, the power-off time example of 10 ranks can be set For example 10ms, 100ms, 1s, 2s, 3s, 10s, 1 minute, 10 minutes, 1 hour, the different period such as 1 day.

In embodiment, calculate by the pressure drop that electric device both ends in the moment are let out in acquisition one and obtain the power-off Information, referring to Fig. 2, the circuit block diagram of the characteristic extracting module of the application in one embodiment is shown as, as shown, one In a example, the power cut-off information can be acquired by the power-off sensing unit of characteristic extracting module shown in Fig. 2, such as DC-DC converter, LDO device, charge breaker (Charge Breaker) in Fig. 2, leakage capacitance C2 and ADC converter structure At resolving electric induction unit, by during the recovery operation controlled by charge breaker (Charge Breaker) every time to letting out As shown in Fig. 2, checking the voltage of capacitor when electronic equipment is from service restoration, the system cut-off time can be with for drain capacitor charging It is calculated by voltage drop that ADC is detected.In details of the words, as shown in Fig. 2 power-off sensing unit part.When each system electrification (before recovery operation), leakage capacitor can all be charged with the voltage of complete stability.When power failure occurs, leakage capacitance Voltage on device declines as time go on, can by the residual voltage measured in the leakage capacitor after next time powers on To obtain the power-off time of detection.In the present embodiment, the power cut-off information at the received electronic equipment one or more moment (Power Outage Sensing) is stored into non-volatile shift unit.In one example, the received electricity For the continuous power cut-off information of nearest 10 of sub- equipment, by a counter (not shown) to calculate power-off time, the counting The power-off time finally calculated is pushed to non-volatile shift unit to replace earliest power-off time by device.

Then step S11 is executed, when carrying out the following power-off of the prediction calculating electronic equipment according to the power cut-off information Between (Power-off Time) and power-off confidence level (Confidence)；In embodiment, it is predicted by a neural network It calculates to obtain the following power-off time of the electronic equipment and power-off confidence level, such as configures one for the neural network and break Electric fallout predictor carries out prediction and calculates to obtain the following power-off time (Power-off Time) of the electronic equipment and power-off and set Reliability (Confidence).

In embodiment, the neural network is, for example, feedforward neural network (Feedforward Neural Network), the feedforward neural network is the nerve net of neural network or online backpropagation training through off-line training Network.A kind of neural network shown in Fig. 3 is please referred to, Fig. 3 is shown as a kind of the application neural network schematic diagram in one embodiment, As shown, in the present embodiment, the neural network is, for example, feedforward neural network, the feedforward neural network includes 1 Input layer (Input Layer), 2 hidden layers (Hidden Layer 1,2), 1 output layer (Output Layer), every layer There are 10 neurons, each neuron there are 10 outputs, is based on from non-volatile shift unit (Nonvolatile shifter, NV Shifter) to the power cut-off information at 10 moment (from On time_1 to On time_10) received, that is, the neural network Input layer be used for from non-volatile shift unit receive such as be 10 history power-off times power cut-off information, before described 2 hidden layers of feedback neural network predict the future of for example, 10 power off time levels by output layer output after calculating The power-off confidence level (Confidence) of power-off time (Power-off Time) and corresponding each following power-off time, Jin Ershi It is current to be predicted in the following possible power supply interruption or the power supply break time having occurred and that.But not limitation and this, In other possible embodiments, in another neural network structure as indicated at 4, Fig. 4 is shown as the application mind in an embodiment Middle another kind neural network schematic diagram, the feedforward neural network as shown in the figure may include more hidden layer (Hidden Layer 1,2 ... N), each layer may also include more or fewer neurons (N).Nerve net shown in Fig. 3 or Fig. 4 Network is the neural network of neural network or online backpropagation training through off-line training.

Then step S12 is executed, carries out predicting to calculate to obtain according to the following power-off time and power-off confidence level writing plan It slightly instructs (Write Configuration) and writes tactful confidence level (Confidence), to enable the processor according to described in It writes policy instructions and executes write operation；The policy instructions of writing include executing the write current and write time at least one of write operation Information.The write time is the duration of data write-in, and in concrete implementation, the duration of the data write-in is by being written What pulse width determined, the reservation that the write current and write pulse width can influence the data of write-in (is kept or standby Part) time.In the present embodiment, it is described write policy instructions and be include that the retention time configuration of data is written or be retention time (Retention Time) strategy, by the policy control method of writing of the application, can to improve backup retention time spare to reduce Energy, and then realize energy-efficient control.

In one embodiment, described to carry out predicting to calculate to obtain according to the following power-off time and power-off confidence level writing plan The step S12 slightly instructed by a neural network carry out prediction calculate with obtain it is described write policy instructions and write tactful confidence level, The neural network is, for example, feedforward neural network (Feedforward Neural Network), and the feedforward neural network is The neural network of neural network or online backpropagation training through off-line training, the feedforward neural network is in such as Fig. 3 Or the network structure of Fig. 4.Such as time backup fallout predictor is configured by the neural network and carries out prediction calculating to be write It policy instructions (Write Configuration) and writes tactful confidence level (Confidence).

In embodiment, the processor receive it is described execute the operation for writing data when writing policy instructions, show in one kind It is, for example, the number of calculating or the processing of processor to the data that the memory of electronic equipment is written by the write operation in example According to, or the calculating state etc. of record backup.The memory such as NVM (Non-volatile memory, non-volatile memory Device, abbreviation NVM), for example be the memory of electronic equipment.

The policy control method of writing of the application can be by prediction power-cut time acquisition data transmission bit wide or data write-in On the opportunity of strategy and starting operation, and then the energy that may insure that the operation of processor is stored with it matches, and according to writing Tactful dynamic adjusts the retention time of non-volatile element and power levels match.

Referring to Fig. 5, be shown as the application writes the flow chart of policy control method in another embodiment, as schemed institute Show, it is in the present embodiment, described to carry out predicting that calculating acquisition is write strategy and referred to according to the following power-off time and power-off confidence level The step S11 of order further include:

Step S111 carries out prediction calculating according to the following power-off time and power-off confidence level；In the present embodiment, institute It states prediction to be calculated as carrying out prediction calculating using feedforward neural network, by the future of received for example, 10 power off time levels The power-off confidence level (Confidence) of power-off time (Power-off Time) and corresponding each following power-off time carries out mind It is calculated through neural network forecast.

Step S112 writes write current and write time in policy instructions described in tradeoff to obtain and described write strategy and refer to It enables.In the present embodiment, the purpose that write current and write time in policy instructions are write in the tradeoff is to find optimal write Strategy utilizes the smallest write current and the shortest write operation for writing time progress data.

The reservation that the write current and write pulse width can influence the data of write-in (keeps or backs up) time. Referring to Fig. 6, be shown as the application writes the write current of strategy and the relation schematic diagram of write pulse width in one embodiment, As shown, abscissa is expressed as write pulse width, (ordinate is expressed as the size of write current, data are written in diagram Storage device medium be STT-RAM for, the said write time be data write-in duration, in concrete implementation, institute The duration for stating data write-in is determined by write pulse width, and the write current and write pulse width can influence to write The reservation of the data entered (keeps or backs up) time.

As shown in fig. 6, being made of using in Fig. 6 side's point when the retention time for writing the determining write-in data of strategy is 10ms Curve distribution write current and write pulse width, described in the curve tradeoff write current and the write time be then then Using the upper right corner for the small square illustrated in Fig. 6 vertex (i.e. the point of a indicated by arrow in Fig. 6) coordinate write current and Write pulse width；Correspondingly, it when the retention time for writing the determining write-in data of strategy is 1 day, is made of using in Fig. 6 dot Curve distribution write current and write pulse width, described in the curve tradeoff write current and the write time be then then Using the write current of the apex coordinate point (i.e. the point of b indicated by arrow in Fig. 6) in the upper right corner for the big square illustrated in Fig. 6 With write pulse width；Similarly, when write strategy determine write-in data retention time be 1 minute or 1 second when, using in Fig. 6 by The write current and write pulse width of the curve distribution of positive triangle point and inverted triangle point composition, weigh described in the curve and write Entering electric current and write time then is that the vertex in the upper right corner for the square illustrated in Fig. 6 is respectively adopted (i.e. in Fig. 6 indicated by arrow C point or d point) coordinate points write current and write pulse width.

Since the write current and write pulse width of data can influence the reservation (keep or back up) of the data of write-in Time, therefore a kind of write current provided herein, referring to Fig. 9, be shown as the application in one embodiment write behaviour Make circuit diagram, as shown, being generated according to from the write current and write time that policy instructions determine is write by current mirror Different electric currents determine the different time by a counter, by the selection of the write-in data MUX antenna array control electric current of diagram, Data are write with great electric current with determination, than I as shown₁To I₈8 road circuits of display, are compared by the write time of diagram Device selects different durations, to determine with how long writing data.

In Fig. 9, I_refIt is the base current of current mirror, by the W/L ratio for changing the current mirror being made of PMOS transistor It (breadth length ratio of transistor channel, in that case it can be decided that the amplification factor of current mirror) and then generates from I₁To I₈Different write currents, this reality It applies in example, maximum current slew rate is from 1 day to 10ms.It, can be in MUX array according to the prediction write current configuration in diagram Different electric currents is selected, write current is connected to " Bit " or " Bit B " (can be by overturning " Bit " or the electric current side of " Bit B " Always change write-in data), it is specifically dependent upon the input of " write-in data " signal.The control of another row of " Bit " or " Bit B " is write The angle of incidence, by 4 digit counters (sub ns per cycle) of a high frequency, the gate time of the counter and according to figure Prediction write time configuration in showing is compared, and is disconnected once the time of counter reaches the preset write time and ground (GND) connection is to terminate write operation.In the present embodiment, by executing during the write operation writes data into, institute Stating memory is NVM (Non-volatile memory, nonvolatile storage, abbreviation NVM), such as STT-RAM (Shared Transistor Technology Random Access Memory)。

It is described according to the following power-off time and power-off confidence level to carry out prediction and calculate to obtain writing policy instructions and writing plan The slightly step S12 of confidence level further include: judge to write whether tactful confidence level meets a preset condition described；It is exported if meeting The policy instructions of writing are to the processor；If being unsatisfactory for policy instructions are not write to described in processor transmission.In this reality It applies in example, the preset condition is a preset threshold X, and the threshold X is a preset power-off confidence level, for example power-off is set Reliability is 80%, when the power-off confidence level of predicted acquisition is greater than preset power-off confidence level X, writes strategy to processor transmission Instruction.Under different implementation states, the threshold X can be configured to different values, these different implementation states include The medium of the storage device of data, such as STT-RAM (Shared Transistor Technology Random is written Access Memory) etc.；In addition, these different implementation states be also possible to be processor requirement record data content not With etc..

In one embodiment, when predict power-off time it is shorter when or power-off confidence level it is relatively low (small) when, such as Power-off time is less than 50ms or power-off confidence level is lower than 80%, then the energy that storage can be used in system is run without Backup operation does not write policy instructions to processor transmission.

In another embodiment, when the memory in electronic equipment uses such as low standby power loss static random access memory In the case where device (Static Random-Access Memory, SRAM), since the standby energy consumption of such memory is relatively low, It is, for example, that between 50ms to 0.2s, also may not need backup between when power is off, i.e., does not write policy instructions to processor transmission.

In one embodiment, the policy control method of writing of the application further includes updating in a power off periods for power information The step of.Such as upper primary power supply is updated in this power off periods and (powers on) information, and this is stored in for power information non- Volatibility shift unit (NV Shifter) for other next predictions, such as following conduction time and energization forecast confidence it is pre- It surveys.

Referring to Fig. 7, the flow chart of the application write in the another embodiment of policy control method is shown as, as schemed institute Show, in the present embodiment, step S10 ' is first carried out, obtains the power cut-off information (Power at least one moment of electronic equipment Outage Sensing)；In embodiment, detect the electronic equipment power-off when obtain the electronic equipment at least one The power cut-off information at moment.In embodiment, the description of the step S10 ' please refers to above-mentioned the retouching for step step S10 of Fig. 1 It states, it will not be described here.

Then step S11 ' is executed, when carrying out the following power-off of the prediction calculating electronic equipment according to the power cut-off information Between (Power-off Time) and power-off confidence level (Confidence).In embodiment, the description of the step S11 ' please be joined The above-mentioned description for step step S11 of Fig. 1 is read, it will not be described here.

Then execute step S12 ', according to it is described power-off confidence level and acquisition data bit width instruction (Bitwidth) into Row prediction, which calculates, obtains service quality predictive information (Predicted QoS)；In the present embodiment, the data bit width, which instructs, is Storage information (Stored according to the following conduction time of prediction, energization confidence level and the received electronic equipment Energy Sensing) carry out what prediction calculating obtained.

Please continue to refer to Fig. 2, in the present embodiment, the characteristic extracting module for front-end circuit includes for example, battery Or the power supply including charging device, power supply one end ground connection, the other end connect a Rs resistance, the Rs resistance as detecting element, It is obtained to the electronic equipment by acquiring the current value of detecting element Rs or the voltage value at its both ends that flow through as resistance and is mentioned Confession gives the neural network prediction device for power information, and Rs resistance and 6 ADC converters constitute the present embodiment in Fig. 2 Described in detection unit, for detect obtain it is described for power information (Input power sensing), i.e. Fig. 2 shows middle solid line Shown in arrow, in embodiment, it is described for power information be electronic equipment upper power information, this is by electronic equipment for power information Self-contained electric system generate, such as energy collecting device obtains energy from human motion, such as people walks or the pendulum of limbs It moves, jump, press movements or the behavior bands such as (such as pressure of small energy collector acquisition when running in implantation shoes), breathing The vibrational energy come, is converted to electric energy for the vibrational energy, in other cases, the energy can be from nature ring Border, such as solar energy etc..The electric energy that the energy collecting device is collected is needed from AC to DC or DC to DC processing, then by collection Energy is temporarily stored in outside piece or even in on-chip capacitance device, is mainly used for supporting data rather than storage energy.

In the present embodiment, the energy-storage units are to obtain storage by the voltage difference (pressure drop) at acquisition energy-storage travelling wave tube both ends Power information (Stored energy sensing), i.e., in diagram shown in dotted arrow, i.e. the remaining capacity of energy-storage travelling wave tube；Institute Stating energy-storage travelling wave tube is the ground capacity C1 in Fig. 2.Capacitor C1 and ADC converter constitutes inspection described in the present embodiment in Fig. 2 Survey unit.

In the present embodiment, via being configured as the neural network of future energy fallout predictor based on the confession of the electronic equipment Power information (Power Sensing) predicts the following conduction time (Power-on Time) and the energization confidence of the electronic equipment It spends (Confidence).In the present embodiment, the future energy fallout predictor has one or more non-volatile shift units, The non-volatile shift unit is, for example, non-volatile shift unit (NV Shifter), and the future energy fallout predictor will receive The electronic equipment one or more moment store for power information into the non-volatile shift unit.In the present embodiment In, the future energy fallout predictor is by 10 moment of the received electronic equipment for power information storage to described non-volatile In shift unit.In the present embodiment, with institute's future energy fallout predictor it is continuous to nearest 10 of the received electronic equipment on For the electric time, by a counter (not shown) to calculate conduction time, when the energization which will finally calculate Between push to non-volatile shift unit to replace earliest conduction time.

In the present embodiment, by configuring so that neural network has the Feedforward Neural Networks of 2 hidden layers being fully connected 10 neurons of network, 10 neurons of every layer of setting, output layer export 10 potential energy levels (energy grade), will be every One energy level converts a corresponding conduction time, and each output indicates an energization confidence level, when energization confidence level and its When his confidence level is very different, wherein peak is output as the energy level and its confidence level of prediction.In a kind of situation Under, if the energization confidence level of preceding several predictions in 10% section, predicts that the energy level of the following conduction time is added Weight average, and corresponding energization confidence level is also averaged.

In the present embodiment, 10 potential energy levels are received via the neural network for being configured as bit wide fallout predictor When the energization confidence level of (energy grade) and each corresponding energy level, the number is calculated by the prediction of 2 hidden layers According to bit wide instructions.

In the present embodiment, the bit wide fallout predictor is used for according to following conduction time, the energization confidence level and connects The storage information (Stored Energy Sensing) for the electronic equipment received carries out approximate calculation to predict outputs data bits Wide instruction (Bitwidth) and starting threshold value (System Start Stored Energy Threshold).The bit wide prediction Device dynamically collect it is each booting interim dynamic prediction processor required bit wide, whereby with reduce processor calculate and Backup cost shortens the response time.

In the present embodiment, the bit wide fallout predictor is predicted by a feedforward neural network, with output data bit wide Instruction and starting threshold value, the feedforward neural network of the bit wide fallout predictor are led to by the energy level of the reception following conduction time For the storage energy level for including in electric confidence level and storage information as input, the feedforward neural network includes 1 input Layer, 2 hidden layers, 1 output layer, every layer has 10 neurons, and each neuron has 10 outputs, the output layer output 2 Kind information, i.e. output data bit wide instructions and starting threshold value, wherein 8 outputs of the output layer is used to refer to as data bit width The output of order uses 1 of the output layer to export the output as starting threshold value.

In the present embodiment, it is pre- to receive the future energy for the input layer of the feedforward neural network of the bit wide fallout predictor When surveying 10 potential energy levels (energy grade) of device output and corresponding to the energization confidence level of each energy level, by Output starting threshold value is calculated by the prediction of 2 hidden layers, with the instruction whether started to the processor.

For example, and there is no enough storages in buffer condenser if the potential input energy level of prediction is relatively low Energy then enables processor not start；If the potential input energy of prediction is high, but the energization confidence level received is relatively low, then Only when the energy level of storage it is sufficiently high to reach prediction threshold value when just indicate processor starting；If prediction is potential defeated Enter energy height, and the confidence level that is powered is relatively high, even if the energy of storage is lower, nevertheless indicates that the processor starting (prediction Device exports low-down threshold value) to obtain better forward progress and QoS satisfaction.In a kind of example, the energization confidence Spending relatively low is, for example, less than 30%, and the relatively high energization confidence level is, for example, to be greater than 70%.But not limitation and this, knowing In the case where knowing the application innovative idea, the threshold value of energization confidence level can be adjusted according to actual conditions dynamic.

In the present embodiment, the feedforward neural network of the bit wide fallout predictor is also used to by predicting with decision output data Bit wide, that is, the future energy fallout predictor is received according to the input layer of the feedforward neural network of the bit wide fallout predictor and is exported 10 potential energy levels (energy grade) and the energization confidence level of each corresponding energy level carry out approximate calculation, Output data bit wide appropriate is determined with prediction, to determine the processor under any energy level using which type of essence Degree carries out operation.

In the present embodiment, the feedforward neural network of bit wide fallout predictor is configured as according to following conduction time, logical Electric confidence level and the storage information of the electronic equipment carry out prediction and calculate the approximate configuration data (Approx of acquisition Config)；In one example, 10 potential energy levels are received via the neural network for being configured as bit wide fallout predictor When the energization confidence level of (energy grade) and each corresponding energy level, calculated by the prediction of 2 hidden layers described close Like configuration data, in the approximation configuration data comprising the bit wide fallout predictor through neural network according to following conduction time, logical Electric confidence level and storage information prediction calculate bit wide information obtained, for example processor is counted with how many bit wides It calculates, with determination is correct or the configuration of suitable bit wide.

The feedforward neural network of bit wide fallout predictor is configured as according to described in preset approximate unlabeled data (ACEN) identification Approximate configuration data (Approx Config) carries out approximate when being judged as that the approximate configuration data can be by approximate calculation It calculates to generate data bit width instruction (Bitwidth).In one example, the approximate unlabeled data (ACEN) is preset, tool For body, be via programmer identify setting can approximate data, these data include data buffer storage (data buffer) Or image (image), but do not include basic variable, such as the index in " for " circulation.

Figure 18 is please referred to, the application approximate calculation configuration diagram in one embodiment is shown as, as shown, described close It include the processor architecture of 5 level production lines like computing architecture, by a dynamic schema approximation control unit to the assembly line The control of processing framework progress approximate calculation.As shown, by one be configured as the neural network of future energy fallout predictor according to The following conduction time and the energization confidence that prediction calculates the electronic equipment are carried out for power information (Input Power) according to described Degree, then the neural network by being configured as bit wide fallout predictor is according to the following conduction time, energization confidence level and the electricity The storage information (Stored Energy) of sub- equipment carries out prediction and calculates the approximate configuration data (Approx Config) of acquisition, then Approximation control unit by being preset with approximate unlabeled data (ACEN) interface adds one to each approximate configuration data and is known as The position of ACEN, to identify whether the approximation configuration data can be approximate during operation, the dynamic schema approximation control unit Approximate configuration data (the Approx for reading approximate unlabeled data (ACEN) from two operators in an instruction and obtaining Config), if it is determined that operator can be approximate, but another operator can not be approximate, then will not the approximate instruction, such as Two operators of fruit can be approximate, then so that the processor is utilized described 5 to generate data bit width instruction (Bitwidth) The processor architecture of level production line carries out approximate calculation, and then realizes that processor calculates the control of (operation) precision.From the foregoing, it will be observed that The beginning for being configured as the predefined circulation of programmer of the bit wide fallout predictor in main program is activated, this is usually to be processed The circulation of new frame.This is configured as bit wide fallout predictor for determining bit wide appropriate to complete entire journey in this power up cycle The circulate operation of sequence.

In embodiment, the dynamic schema approximation control unit be, for example, be arranged in it is close in processor or execution module Like bit wide controller, the approximation bit wide controller is configured as the number of the neural network output of bit wide fallout predictor to receive According to the precision for instructing control arithmetic operation when bit wide instructions (Bitwidth) according to the data bit width；In the present embodiment, institute Stating approximate bit wide controller has one or more non-volatile shift units, and the non-volatile shift unit is, for example, non-easy The property lost shift unit (NV Shifter), the approximation bit wide controller arrive the instruction storage of received data bit width described non-volatile In property shift unit.

The processor be calculated as sensing data that processor obtains electronic equipment or interaction data calculate and Processing.In some instances, heart rate data, blood pressure number of the processing such as wearable device of the sensing data by acquisition According to, temperature data, blood oxygen saturation data, diet/nutritional information, medical alert, to the relevant prompt of health or information, or Other data relevant to health, which carry out processing generation, can carry out transmitting by wireless module or be carried out by display equipment The user data of display.

In some instances, the processing of the interaction data such as can be by user's operation wearable device to by host The event notice that equipment generates makes a response.Wearable device can receive the notice of event from host equipment, and present for user Prompting and the prompt to response.If user makes a response prompt, response can be transmitted to host and set by wearable device It is standby.For example, user can make a response in the received call in host equipment place, text message or other communications.

Step S13 ' is finally executed, exports the Service Quality when the service quality predictive information meets a preset condition Predictive information is measured to the processor.In one embodiment, described low to avoid electronic equipment from generating low-quality data output The data of quality are such as the data lower than 20dB, correspondingly, the data of 20dB or more are typically considered reasonable quality Data.The present processes also predict the potential output quality of the program of electronic equipment.

In the present embodiment, by configuring service quality fallout predictor for neural network, service quality fallout predictor is in frame number According to as the approximate bit wide of acquisition during input processing and average interruption in power forecast confidence.In embodiment, the Service Quality Fallout predictor 101 is measured to carry out according to data bit width instruction (Bitwidth) and power cut-off information (Power Outage Sensing) Prediction, which calculates, obtains service quality predictive information (Predicted QoS)；And in the service quality predictive information (Predicted QoS) is exported when meeting a threshold value Y to the processor, and in starting threshold value (the System Start Stored Energy Threshold) when meeting the threshold value Y output enabled instruction (System Start) to the processing Device.Wherein, the power cut-off information includes power-off confidence level (Confidence), and the power-off confidence level can be pre- by a power-off Device is surveyed to generate.In the present embodiment, the threshold value Y is, for example, 20dB, and however, it is not limited to this, under different implementation states, Such as different types of electronic equipment, different threshold value Y can be set.In one embodiment, the service quality prediction Device realizes the prediction of service quality by one includes the feedforward neural network of 2 hidden layers.

Due to high-bit width operation (processor computational accuracy is high) electricity more than low-bit width (processor computational accuracy is low) dissipation Can, storage energy threshold value is started by control processor in this application, the Startup time of control processor is in turn in energy pipe It manages and achievees the purpose that alleviate low quality output problem in reasonable situation, postpone the processing by the parameter of control threshold Y The Startup time of device, until there are enough energy to restart the processor in energy storage capacitor, and then reasonable Ensure data operation or the output of high quality under the mechanism of energy management.

Referring to Fig. 8, be shown as the application writes the schematic diagram of policy controlling system in one embodiment, as shown, The application provide it is a kind of write policy controlling system applied in the electronic equipment with processor, to sharp in power off periods The write operation of data backup is carried out, with the energy of storage so that the processor described when next power up cycle arrives can be based on The data of the write-in carry out continuation operation.In embodiment, the processor 3 is, for example, non-volatile memory (NVP), but simultaneously It is not limited to this, it is in other examples, described without departing substantially from inventive concept and thought disclosed herein Processor is also possible to common processor, for example can be any commercially available processor, controller, microcontroller or state Machine.Processor is also implemented as calculating the combination of equipment, for example, the combination of DSP and microprocessor, multi-microprocessor, with The one or more microprocessors or any other such configuration of DSP core collaboration.

As shown in figure 8, the policy controlling system 1 of writing includes power-off fallout predictor 11 and BACKUP TIME fallout predictor 12.

The power-off fallout predictor 11 according to the power cut-off information at least one moment of electronic equipment by carrying out based on prediction Calculate the following power-off time and power-off confidence level of the electronic equipment.In embodiment, the electronic equipment power-off is being detected When obtain the power cut-off information at least one moment of electronic equipment.

In embodiment, the power-off fallout predictor 11 carries out predicting to calculate setting to obtain the electronics by a neural network The standby following power-off time and power-off confidence level, such as configure a power-off fallout predictor for the neural network and carry out prediction calculating To obtain the following power-off time (Power-off Time) and power-off confidence level (Confidence) of the electronic equipment.In reality It applies in example, the power-off fallout predictor 11 includes one or more for storing the non-volatile shift unit of the power cut-off information, The non-volatile shift unit is non-volatile displacement device.

In embodiment, the neural network is, for example, feedforward neural network (Feedforward Neural Network), the feedforward neural network is the nerve net of neural network or online backpropagation training through off-line training Network.A kind of neural network shown in Fig. 3 is please referred to, as shown, in the present embodiment, the neural network is, for example, the mind that feedovers Through network, the feedforward neural network includes 1 input layer (Input Layer), 2 hidden layers (Hidden Layer 1, 2), 1 output layer (Output Layer), every layer has 10 neurons, and each neuron has 10 outputs, is based on from non-volatile Property shift unit (Nonvolatile shifter, NV Shifter) is to 10 moment received (from On time_1 to On Time_10 power cut-off information), that is, it is such as 10 that the input layer of the neural network, which is used to receive from non-volatile shift unit, The power cut-off information of a history power-off time, by the output layer after 2 hidden layers of the feedforward neural network calculate Output prediction is, for example, the following power-off time (Power-off Time) of 10 power off time levels and corresponds to each following disconnected The power-off confidence level (Confidence) of electric time, and then realize for the following electricity that may power supply interruption or have occurred and that Source break time is predicted.But not limitation and this, it is neural just as another kind shown in 4 in other possible embodiments Network structure, the feedforward neural network as shown in Figure 4 may include more hidden layers (Hidden Layer 1,2 ... N), each layer may also include more or fewer neurons (N).Neural network shown in Fig. 3 or Fig. 4 is through off-line training Neural network or online backpropagation training neural network.

The BACKUP TIME fallout predictor 12 is used to carry out prediction calculating according to the following power-off time and power-off confidence level Acquisition writes policy instructions and writes tactful confidence level, to enable the processor 3 execute write operation according to the policy instructions of writing；It is described Writing policy instructions includes executing the write current and write time at least one information of write operation.The write time is data The duration of write-in, in concrete implementation, the duration of the data write-in is determined by write pulse width, the write-in The reservation that electric current and write pulse width can influence the data of write-in (keeps or backs up) time.In the present embodiment, described It writes policy instructions and is and include that the retention time configuration of data is written or be retention time (Retention Time) strategy, pass through The application's writes policy control method and can improve backup retention time to reduce spare energy, and then realizes energy-efficient control.

In embodiment, the BACKUP TIME fallout predictor 12 receives the following power-off time by a neural network and breaks Electric confidence level carries out prediction calculating acquisition and writes policy instructions and write tactful confidence level, specifically, before the neural network is, for example, It presents neural network (Feedforward Neural Network), the feedforward neural network is the neural network through off-line training Or the neural network of online backpropagation training, the feedforward neural network are in the network structure such as Fig. 3 or Fig. 4.Such as Time backup fallout predictor is configured by the neural network to carry out predicting to calculate writing policy instructions (Write to obtain Configuration it) and writes tactful confidence level (Confidence).

In embodiment, the processor 3 receive it is described execute the operation for writing data when writing policy instructions, show in one kind It is, for example, the number of calculating or the processing of processor to the data that the memory of electronic equipment is written by the write operation in example According to, or the calculating state etc. of record backup.The memory such as NVM (Non-volatile memory, non-volatile memory Device, abbreviation NVM), for example be the memory of electronic equipment.

The policy controlling system 1 of writing of the application can be by prediction power-cut time acquisition data transmission bit wide or data write-in On the opportunity of strategy and starting operation, and then the energy that may insure that the operation of processor is stored with it matches, and according to writing Tactful dynamic adjusts the retention time of non-volatile element and power levels match.

In one embodiment, the BACKUP TIME fallout predictor 12 is also used to according to the following power-off time and power-off confidence Degree carries out prediction calculating, is obtained by writing write current in policy instructions and write time described in tradeoff and described write strategy and refer to It enables.In the present embodiment, the purpose that write current and write time in policy instructions are write in the tradeoff is to find optimal write Strategy utilizes the smallest write current and the shortest write operation for writing time progress data.

Since the write current and write pulse width of data can influence the reservation (keep or back up) of the data of write-in Time, therefore a kind of write current provided herein, referring to Fig. 9, being shown as the application writes policy controlling system one Write operation circuit diagram in embodiment, as shown, when according to from the write current and write-in for writing policy instructions determination Between, different electric currents are generated by current mirror, the different time is determined by a counter, pass through the write-in data MUX array of diagram The selection of electric current is controlled, data are write with great electric current with determination, than I as shown₁To I₈8 road circuits of display, pass through diagram Write time comparator select different durations, to determine with how long writing data.

The BACKUP TIME fallout predictor 12 is also used to judge to write whether tactful confidence level meets a preset condition described；If Satisfaction writes policy instructions to the processor described in then exporting；It does not write strategy to described in processor transmission if being unsatisfactory for and refers to It enables.In the present embodiment, the preset condition is a preset threshold X (X i.e. in diagram), and the threshold X is one default Power-off confidence level, such as power-off confidence level be 80%, when predicted acquisition power-off confidence level be greater than preset power-off confidence When spending X, policy instructions are write to processor transmission.Under different implementation states, the threshold X can be configured to different Value, these different implementation states include the medium that the storage device of data is written, such as STT-RAM (Shared Transistor Technology Random Access Memory) etc.；In addition, these different implementation states be also possible to be The difference of data content etc. of processor requirement record.

In another embodiment, when the memory in electronic equipment uses such as low standby power loss static random access memory In the case where device (Static Random-Access Memory, SRAM), since the standby energy consumption of such memory is relatively low, It is, for example, that between 50ms to 0.2s, also may not need backup between when power is off, i.e., does not write strategy to execution module transmission and refer to It enables.

In one embodiment, to avoid electronic equipment from generating low-quality data output, the low-quality data are such as For the data lower than 20dB, correspondingly, the data of 20dB or more are typically considered the data of reasonable quality.The side of the application Method also predicts the potential output quality of the program of electronic equipment.

Referring to Fig. 10, be shown as the application writes the schematic diagram of policy controlling system in another embodiment, as schemed institute Show, the policy controlling system 1 of writing of the application further includes service quality fallout predictor 13, and the service quality fallout predictor 13 is used for foundation The data bit width instruction of the power-off confidence level and acquisition carries out prediction and calculates acquisition service quality predictive information, and described Service quality predictive information exports the service quality predictive information to the processor 3 when meeting a threshold value, wherein the number It is the storage information of the following conduction time according to prediction, energization confidence level and the received electronic equipment according to bit wide instructions (Stored Energy Sensing) carries out prediction and calculates acquisition.

Please continue to refer to Fig. 2, in the present embodiment, the characteristic extracting module for front-end circuit includes for example, battery Or the power supply including charging device, power supply one end ground connection, the other end connect a Rs resistance, the Rs resistance as detecting element, It is obtained to the electronic equipment by acquiring the current value of detecting element Rs or the voltage value at its both ends that flow through as resistance and is mentioned Confession gives the neural network prediction device for power information, and Rs resistance and 6 ADC converters constitute the present embodiment in Fig. 2 Described in detection unit, for detect obtain it is described for power information (Input power sensing), i.e. Fig. 2 shows middle solid line Shown in arrow, in embodiment, it is described for power information be electronic equipment upper power information, this is by electronic equipment for power information Self-contained electric system generate.

In the present embodiment, by configuring service quality fallout predictor 13 for neural network, service quality fallout predictor 13 exists Frame data are as the approximate bit wide of acquisition during input processing and average interruption in power forecast confidence.In embodiment, the clothes Quality predictor 13 be engaged according to data bit width instruction (Bitwidth) and power cut-off information (Power Outage Sensing) It carries out prediction and calculates acquisition service quality predictive information (Predicted QoS)；And in the service quality predictive information (Predicted QoS) is exported when meeting a threshold value Y (Y i.e. in diagram) to the processor 3, and in the starting threshold value (System Start Stored Energy Threshold) exports enabled instruction (System when meeting the threshold value Y Start the processor 3) is given.Wherein, the power cut-off information includes power-off confidence level (Confidence), the power-off confidence Degree can be generated by above-mentioned power-off fallout predictor 11.In the present embodiment, the threshold value Y is, for example, 20dB, and however, it is not limited to this, Under different implementation states, such as different types of electronic equipment, different threshold value Y can be set.In an embodiment In, the service quality fallout predictor 13 realizes the prediction of service quality by one includes the feedforward neural network of 2 hidden layers.

Due to high-bit width operation (processor computational accuracy is high) electricity more than low-bit width (processor computational accuracy is low) dissipation Can, storage energy threshold value is started by control processor 3 in this application, the Startup time of control processor 3 is in turn in energy Achieve the purpose that alleviate low quality output problem in the case where rational management, postpones the place by the parameter of control threshold Y The Startup time for managing device, until there are enough energy to restart the processor 3 in energy storage capacitor, and then reasonable Energy management mechanism under ensure data operation or the output of high quality.

Figure 11 is please referred to, the application is shown as and writes schematic diagram of the policy controlling system in another embodiment, as shown, In the present embodiment, via being configured as the neural network of future energy fallout predictor 14 based on the electronic equipment for power information (Power Sensing) predicts the following conduction time (Power-on Time) and the energization confidence level of the electronic equipment (Confidence).In embodiment, it can be in different time periods at the time of described, can be divided into according to different demands Multiple moment grades, such as 10ms, 100ms, 1s, 2s, 3s, 10s, 1 minute 10 minutes, 1 hour, 1 day etc., need to be stated clearly, The citing at above-mentioned moment is exemplary, under different status of implementation, it is not limited to this.

In embodiment, it is described for power information be electronic equipment upper power information, this is by electronic equipment for power information What self-contained electric system generated, such as energy collecting device obtains energy from human motion, such as people walks or the swing of limbs, The movements such as jump, pressing (such as pressure of small energy collector acquisition when running in implantation shoes), breathing or behavior are brought Vibrational energy, which is converted into electric energy, in other cases, the energy can be from nature ring Border, such as solar energy etc..The electric energy that the energy collecting device is collected is needed from AC to DC or DC to DC processing, then by collection Energy is temporarily stored in outside piece or even in on-chip capacitance device, is mainly used for supporting data rather than storage energy, acquisition supplies The example of power information is in the description such as the above-mentioned corresponding portion being related to Fig. 2.

In the present embodiment, via being configured as the neural network of future energy fallout predictor 14 based on the electronic equipment It predicts the following conduction time (Power-on Time) of the electronic equipment and is powered to set for power information (Power Sensing) Reliability (Confidence).In the present embodiment, the future energy fallout predictor 14 has one or more non-volatile displacements Unit, the non-volatile shift unit are, for example, non-volatile shift unit (NV Shifter), the future energy fallout predictor 14 storing into the non-volatile shift unit for power information by the received electronic equipment one or more moment.? In the present embodiment, by taking institute's future energy fallout predictor 14 is to the received continuous power-on time of nearest 10 of the electronic equipment as an example, By a counter (not shown) to calculate conduction time, which pushes to the conduction time finally calculated non-easy The property lost shift unit is to replace earliest conduction time.

In the present embodiment, make neural network that there is the complete of 2 hidden layers by configuring future energy fallout predictor 14 10 neurons of the feedforward neural network of connection, 10 neurons of every layer of setting, output layer export 10 potential energy waters Each energy level is converted a corresponding conduction time by flat (energy grade), and each output indicates an energization confidence level, When energization confidence level and other confidence levels are very different, wherein peak is output as the energy level predicted and it is set Reliability.In one case, if the energization confidence level of preceding several predictions predicts the following conduction time in 10% section Energy level be weighted average, and corresponding energization confidence level is also averaged.

In the present embodiment, the bit wide fallout predictor 15 is predicted by a feedforward neural network, with outputs data bits Wide instruction and starting threshold value, the energy water that the feedforward neural network of the bit wide fallout predictor 15 passes through the reception following conduction time Flat, for the storage energy level for including in energization confidence level and storage information as input, the feedforward neural network includes 1 Input layer, 2 hidden layers, 1 output layer, every layer has 10 neurons, and each neuron has 10 outputs, and the output layer is defeated 2 kinds of information out, i.e. output data bit wide instructions and starting threshold value, wherein use 8 of the output layer to export as data bit width The output of instruction uses 1 of the output layer to export the output as starting threshold value.

In the present embodiment, the input layer of the feedforward neural network of the bit wide fallout predictor 15 receives the future energy The energization confidence level of 10 potential energy levels (energy grade) and each corresponding energy level that fallout predictor 14 exports When, output starting threshold value is calculated by the prediction of 2 hidden layers, with the instruction whether started to the processor.

In the present embodiment, the feedforward neural network of the bit wide fallout predictor 15 is also used to by predicting to export number with decision According to bit wide, that is, it is defeated to receive the future energy fallout predictor according to the input layer of the feedforward neural network of the bit wide fallout predictor The energization confidence level of the potential energy level (energy grade) of 10 out and each corresponding energy level carries out approximate meter It calculates, output data bit wide appropriate is determined with prediction, to determine the processor under any energy level using what Precision carry out operation.

As shown in figure 18, the approximate calculation framework includes the processor architecture of 5 level production lines, close by a dynamic schema The control of approximate calculation is carried out to the processing framework of the assembly line like control unit.As shown, being configured as not by one The neural network for carrying out energy predicting device carries out the prediction calculating electronic equipment for power information (Input Power) according to described The following conduction time and energization confidence level, then the neural network by being configured as bit wide fallout predictor according to it is described it is following be powered when Between, the storage information (Stored Energy) of energization confidence level and the electronic equipment carries out prediction and calculates obtaining approximation and matching It sets data (Approx Config), then the approximation control unit by being preset with approximate unlabeled data (ACEN) interface is to each Approximate configuration data adds the position for being known as ACEN, to identify whether the approximation configuration data can be approximate during operation, The dynamic schema approximation control unit reads approximate unlabeled data (ACEN) from two operators in an instruction and obtains The approximate configuration data (Approx Config) obtained, if it is determined that an operator can be approximate, but another operator can not Approximation, then will not the approximate instruction, if two operators can be approximate, to generate data bit width instruction (Bitwidth) So that the processor is carried out approximate calculation using the processor architecture of 5 level production lines, and then realizes that processor calculates (fortune Calculate) control of precision.From the foregoing, it will be observed that being configured as the beginning of the predefined circulation of programmer of the bit wide fallout predictor in main program It is activated, this is usually the circulation of new frame to be processed.This be configured as bit wide fallout predictor for determine bit wide appropriate so as to The circulate operation of entire program is completed in this power up cycle.

In embodiment as shown in figure 11, the processor 3 further includes retention time controller 31, the retention time control What device 31 processed was used to receive BACKUP TIME fallout predictor output writes policy instructions, and writing of including in policy instructions is write according to described in Enter electric current and write time at least one information executes write operation.In the present embodiment, the retention time controller 31 has One or more non-volatile shift units, the non-volatile shift unit is, for example, non-volatile shift unit (NV Shifter), the retention time controller 31 writes policy instructions storage into the non-volatile shift unit for received. The retention time controller 31 is according to the operation write policy instructions execution and write data received, in a kind of example, by The write operation is, for example, the data of calculating or the processing of processor, or record to the data that the memory of electronic equipment is written Calculating state of backup etc..

In the present embodiment, then as shown in figure 11, the processor includes: starting controller 30, approximate bit wide controller 32 and retention time controller 31.

Data bit width instruction of the approximation bit wide controller 32 to receive the neural network prediction device output (Bitwidth) precision of control arithmetic operation is instructed when according to the data bit width；In the present embodiment, the approximate bit wide Controller 32 has one or more non-volatile shift units, and the non-volatile shift unit is, for example, non-volatile displacement The instruction storage of received data bit width is arrived the non-volatile displacement by device (NV Shifter), the approximation bit wide controller 32 In unit.The processor 3 be calculated as sensing data that processor obtains electronic equipment or interaction data calculate and Processing.In some instances, heart rate data, blood pressure number of the processing such as wearable device of the sensing data by acquisition According to, temperature data, blood oxygen saturation data, diet/nutritional information, medical alert, to the relevant prompt of health or information, or Other data relevant to health, which carry out processing generation, can carry out transmitting by wireless module or be carried out by display equipment The user data of display.

The starting controller 30 is described non-to start when receiving the enabled instruction of the output of bit wide fallout predictor 15 The work of volatile processor 3；In one embodiment, the starting controller 30 has one or more non-volatile displacements single Member, the non-volatile shift unit are, for example, non-volatile shift unit (NV Shifter), and the starting controller 30 will connect The enabled instruction of receipts is stored into the non-volatile shift unit.In the present embodiment, the starting controller 30 is, for example, NVP startup trigger (NVP Start Trigger Controller) be used for control non-volatile processor whether starting work Make.

The processor 3 also receives the service quality predictive information (Predicted that the service quality fallout predictor 13 exports QoS), so that the processor 3 executes the approximate Bu Tong approximation in dynamic data backing-up retention time based on dynamic bit wide Method predicts the potential output quality of electronic equipment operation program, so that the result of entire Energy Management System has service Quality control.

In one embodiment, power-off fallout predictor 11 and the BACKUP TIME write in policy controlling system shown in fig. 8 Fallout predictor 12 can be multiplexed hardware structure by one to realize the following power-off time and power off the prediction of confidence level and calculate, and It writes policy instructions and writes the prediction calculating of tactful confidence level.

In one embodiment, power-off fallout predictor 11, the BACKUP TIME write in policy controlling system shown in Figure 10 Fallout predictor 12 and service quality fallout predictor 13 can realize the following power-off time and power-off by a multiplexing hardware structure The prediction of confidence level calculates, write policy instructions and write tactful confidence level prediction calculate and service quality predictive information it is pre- It surveys and calculates.

In one embodiment, power-off fallout predictor 11, the BACKUP TIME write in policy controlling system shown in Figure 11 Fallout predictor 12, service quality fallout predictor 13, future energy fallout predictor 14 and approximate bit wide fallout predictor 15 can be multiple by one The following power-off time is realized with hardware structure and powers off the prediction calculating of confidence level, is write policy instructions and is write tactful confidence level Prediction calculates and the prediction of service quality predictive information calculates, and the prediction of the following conduction time and energization confidence level calculates, number It is calculated according to bit wide instructions and the prediction for starting threshold value.

Figure 12 is please referred to, the application is shown as and writes the multiplexing hardware structure signal of policy controlling system in one embodiment Figure, as shown, in the present embodiment, the neural network module 10 in the multiplexing hardware structure includes neural network unit 101 and single predicted state machine 102, wherein the neural network unit 101 includes neuron register 1011, is stored with It is the weight register 1012 of multiple weights, multiple for selecting data to input or the selector 1013 and multiply-accumulate of output Unit 1014.The single predicted state machine 102 receives at least one letter for controlling the neural network unit 101 Breath carries out the timing that single prediction calculates.Neural network module 10 shown in Figure 12 is a serial frame, and the single is pre- Input of the state machine 102 according to the neural network unit 101 is surveyed, controls corresponding selector 1013 from the weight register 1012 one weight of selection and source neuron and the target nerve member for needing to be activated, transfer to a multiplication tired again after calculating Add (Multiply-and-Accumulate, MAC) unit 1014, then writes back the neuron in neural network unit until place All neurons in the input layer, hidden layer and output layer are managed.The weight prestored in the weight register 1012 is Trained acquisition.

In the present embodiment, the weight register 1012 includes for storing the non-of the weight of corresponding each prediction calculating Volatile shift unit or nonvolatile memory cell, wherein the non-volatile shift unit is non-volatile shift unit (Nonvolatile shifter, NV Shifter), the nonvolatile memory cell are nonvolatile storage (Non- Volatile memory, abbreviation NVM).

In the present embodiment, the single predicted state machine 102 has non-volatile shift unit or nonvolatile memory cell, For storing timing control program, specifically, being used to control the defeated of each selector 1013 for the timing control program Timing out.Wherein, the non-volatile shift unit is non-volatile shift unit, and the nonvolatile memory cell is non-volatile memory Device.

The time-sequence control module 102 is for controlling the neural network module 10 based at least one received moment For power information (Power Sensing), storage information (Stored Energy Sensing) and power cut-off information (Power Outage Sensing) at least one of information output data bit wide instructions (Bitwidth), enabled instruction (System Start at least one of policy instructions (Write Configuration) instruction and/or service quality predictive information are write) or The timing that the prediction of (Predicted QoS) calculates, and then may insure that above-mentioned a variety of predictions calculating can share a mind Through network module, in other words, it can use prediction hardware (a neural network framework) and complete all these fallout predictors Function.

More specifically, please refer to Figure 13, being shown as the application, to write the multiplexing of policy controlling system in another embodiment hard Part configuration diagram, as shown, in the present embodiment shown in Figure 13, in order to enable a variety of predictions calculating can be hard at one It completes to calculate in different times in part framework, and then has standardized hardware structure in the application.As shown, the neural network Module 10 further comprises 104 (i.e. Figure 13 of softmax state machine 103 and the judging unit being configured in multiply-accumulator Shown in OR==0 part).In the present embodiment, for specification hardware framework, in the neural network module in net Many virtual links are constructed in network, to pass through 0 connection weight (i.e. weight register shown in Figure 13 of insertion Weights 1, Weights 2 ... Weights 5) neural network topology is standardized.

In the embodiment shown in fig. 13, the single predicted state machine 102 includes for storing timing control program Non-volatile shift unit.The single predicted state machine 102 controls the information progress that the neural network unit receives input The signal dotted arrow as shown in Figure 13 for the timing that single prediction calculates indicates；Time-sequence control module controls the mind in figure The signal of timing through network module dotted lines arrow as shown in Figure 13 indicates.

Once predicted when the neural network module executes (such as prediction or the following power-off time of the following make-and-break time Prediction) when, single predicted state machine 102 controls the source nerve that selected weight and needs are activated by the selector Member and target nerve member export after once calculating to the judging unit 104, and by judging unit 104, any one input is No is 0, if it is any input be 0, bypass multiplier, if it is any input not be 0, via multiplier multiplied by and add up, Then the neuron write back in neural network unit 101 is all in the input layer, hidden layer and output layer until having handled Neuron.Finally, executing softmax layers under the control of softmax state machine 103.In all steps by the neural network mould After block executes, the output of the neural network module is then stored in from the non-volatile shift unit in processor 3 by selection, And updated in some output nonvolatile storages shown in Figure 13 (NVM) (for example the time-sequence control module also controls The neural network module updates the power cut-off information in a power up cycle；Or time-sequence control module controls the nerve Network module updates described for power information in a power off periods), for other next predictions, such as power off forecast confidence Or energization confidence level.In the present embodiment, multiplier and adder are floating-point multiplier and floating-point in the multiply-accumulator Adder.

In the present embodiment, the time-sequence control module includes the non-volatile shift unit for storing timing control program Or nonvolatile memory cell, wherein the non-volatile shift unit is non-volatile shift unit, and the nonvolatile memory cell is non- Volatile memory.

In one embodiment, the neural network module includes one or more non-volatile shift units and non-volatile deposits Storage unit, for store the electronic equipment one or more moment obtained received from characteristic extracting module for power information, Storage information and power cut-off information are respectively stored in these non-volatile shift units.As shown, described for power information (Power Sensing) and power cut-off information (Power Outage Sensing) are stored in these non-volatile shift units In, the storage information (Stored Energy Sensing) is directly extracted from characteristic extracting module 2 to neural network module, The information of update is stored in the nonvolatile storage, for example the time-sequence control module also controls the neural network module The power cut-off information is updated in a power up cycle；Or time-sequence control module controls the neural network module and breaks at one Update is described in the electric period such as powers off forecast confidence or energization confidence level for other next predictions for power information.

The application realizes that a variety of predictions calculate by the hardware structure of a neural network prediction device in different time period, In other words, the application realizes that the prediction of multiple small-scale neural networks calculates using the different periods, reaches saving whereby The purpose of hardware cost and area.

The application also provides a kind of non-volatile processor (Nonvolatile Processors, abbreviation NVP), please refers to figure 14, it is shown as the configuration diagram of herein described non-volatile processor in one embodiment, as shown, the non-volatile place Reason device 4 includes writing policy controlling system 40 and retention time controller 41.Under different implementation scenes, the non-volatile place A variety of encapsulating structures can be presented according to the demand being applied in distinct electronic apparatuses in reason device.

The policy controlling system 40 of writing according to the following power-off time and power-off confidence level for carrying out prediction and calculating acquisition Write policy instructions and write tactful confidence level, with enable the retention time controller 41 according to it is described write policy instructions execution write behaviour Make, please refers among the above for the description in embodiment involved by Fig. 8 to Figure 13, it will not be described here.

The retention time controller 41 to receive it is described write when writing policy instructions of policy controlling system output, according to The write current for including in policy instructions and write time at least one information execution write operation are write according to described.In embodiment, The retention time controller 41 is in such as above-mentioned description for Fig. 9 according to writing policy instructions to execute the description of write operation, herein It will not go into details.

In the present embodiment, the retention time controller has one or more non-volatile shift units, described non- Volatibility shift unit is, for example, non-volatile shift unit (NV Shifter), and the retention time controller writes plan for received Slightly instruction storage is into the non-volatile shift unit.The retention time controller is held according to the policy instructions of writing received The operation of row write data is, for example, to the data that the memory of electronic equipment is written by the write operation in a kind of example The data of calculating or the processing of processor, or the calculating state etc. of record backup.

The non-volatile processor 4 writes the service quality predictive information that policy controlling system 40 exports described in also receiving (Predicted QoS) is executed in approximate and dynamic data backing-up retention time so that the processor is based on dynamic bit wide Different approximation methods predict the potential output quality of electronic equipment operation program so that the entire Energy Management System Result with service quality control.

The application also provides a kind of neural network chip, connects an execution module, the execution module includes retention time Controller please refers to Figure 15, is shown as the configuration diagram of herein described neural network chip in one embodiment, as schemed institute Show, the neural network chip 5 includes writing policy controlling system 50.

The policy controlling system 50 of writing according to the following power-off time and power-off confidence level for carrying out prediction and calculating acquisition Write policy instructions and write tactful confidence level, with enable the retention time controller 41 according to it is described write policy instructions execution write behaviour Make, please refers among the above for the description in embodiment involved by Fig. 8 to Figure 13, it will not be described here.

The retention time controller 60 of the execution module 6 to receive it is described write policy controlling system output write plan When slightly instructing, the write current for including in policy instructions is write according to described in and write time at least one information executes write operation. In embodiment, the retention time controller 60 is in as above-mentioned for Fig. 9 according to writing policy instructions to execute the description of write operation Description, it will not be described here.

In the present embodiment, the retention time controller 60 has one or more non-volatile shift units, described Non-volatile shift unit is, for example, non-volatile shift unit (NV Shifter), and the retention time controller 60 will be received Policy instructions storage is write into the non-volatile shift unit.The retention time controller 60 writes strategy according to what is received Instruction execution writes the operation of data, in a kind of example, data from the write operation to the memory of electronic equipment that be written by The for example, data of calculating or the processing of processor, or the calculating state etc. of record backup.In the present embodiment, the execution Module is, for example, processor, such as non-volatile processor (Nonvolatile Processors, abbreviation NVP).

In embodiment, the neural network chip can use a prediction hardware (a neural network framework) and complete There are many neural network prediction device function.The hardware structure of the neural network prediction device please refer to it is above-mentioned for Figure 12 extremely State shown in Figure 13, it will not be described here.Under different implementation scenes, the neural network chip foundation is applied to not A variety of encapsulating structures can be presented with the demand in electronic equipment.

The application also provides a kind of electronic device (not illustrated), and the electronic device includes institute in the various embodiments described above That states writes policy controlling system.In one embodiment, the electronic device is for example laid with the circuit board of integrated circuit or chip Or board.The circuit board is, for example, double-layer PCB board or multi-layer PCB board.

The application also provides a kind of electronic equipment, and in embodiment provided by the present application, the electronic equipment is Internet of Things Equipment, for example, wearable device or implantable devices, such as wearable electronic may include the limb that can be worn on user Any kind of electronic equipment on body.The wearable electronic can be fixed limbs such as wrist, ankle, the hand of the mankind On arm or leg.This class of electronic devices includes but is not limited to health or body-building assistant devices, digital music player, intelligence electricity Words, calculate equipment or display take exercise or other active monitors, can give the correct time equipment, wearer or user can be measured The equipment etc. of biological characteristic parameter.The implantable devices are, for example, blood sugar test equipment etc..

Figure 16 is please referred to, the schematic diagram of the electronic equipment of the application in one embodiment is shown as, as shown, the electricity Sub- equipment 7 includes characteristic extracting module 71, writes policy controlling system 72 and processor 73.

The characteristic extracting module 71 is used to extract the power cut-off information at least one moment of electronic equipment, such as The embodiment of above-mentioned Fig. 2 description, it will not be described here.

The policy controlling system 72 of writing according to the following power-off time and power-off confidence level for carrying out prediction and calculating acquisition Write policy instructions and write tactful confidence level, with enable the retention time controller 72 according to it is described write policy instructions execution write behaviour Make, please refers among the above for the description in embodiment involved by Fig. 8 to Figure 13, it will not be described here.

The processor 73 to receive it is described write when writing policy instructions of policy controlling system output, write according to described in The write current and write time at least one information for including in policy instructions execute write operation.In embodiment, the processing Device 73 is in such as above-mentioned description for Fig. 9 according to writing policy instructions to execute the description of write operation, and it will not be described here.

In the present embodiment, the processor 73 has one or more non-volatile shift units, described non-volatile Shift unit is, for example, non-volatile shift unit (NV Shifter), and the processor 73 is write received policy instructions storage and arrived In the non-volatile shift unit.The processor 73 executes according to the policy instructions of writing received and writes the operations of data, It is, for example, the calculating or processing of processor to the data that the memory of electronic equipment is written by the write operation in a kind of example Data, or the calculating state etc. of record backup.In the present embodiment, the such as non-volatile processor of the processor 73 (Nonvolatile Processors, abbreviation NVP).

Figure 17 is please referred to, the schematic diagram of the electronic equipment of the application in another embodiment is shown as.As shown, one In embodiment, the electronic equipment 7 further includes the power supply device 70 for generating or storing electric energy.In the present embodiment, described Power supply device 70 is, for example, battery or self-contained electric system, and the self-contained electric system includes energy collecting device, is obtained from human motion Take energy, such as people walks or the swing of limbs, jump, (for example small energy collector when running in implantation shoes obtains for pressing The pressure taken), breathing etc. movements or behavior bring vibrational energy, which is converted into electric energy, in other situations Under, the energy can be from natural environment, such as solar energy etc..The electric energy that the power supply device 70 is collected is needed from AC It is handled to DC or DC to DC, then the energy of collection is temporarily stored in outside piece or even in on-chip capacitance device, is mainly used for Support data rather than storage energy.In one embodiment, the characteristic extracting module can be a part of power supply device.

In one embodiment, as shown in figure 17, the electronic equipment 7 further includes one or more sensing devices 75, described One or more sensing devices 75 are for sensing geographical location information, ambient light information, environmental magnetic field information, acoustic information, temperature Spend information, humidity information, pressure sensitive information, acceleration information, ultraviolet light information, blood glucose information, alcohol concentration information, pulse At least one of information, heart rate information, respiration information, amount of exercise information information.

In embodiment, the sensor 75 may include various electronic equipments, mechanical equipment, electromechanical equipment, optical device, Or provide the other equipment of information relevant to the external condition around wearable device.In some embodiments, sensor can Digital signal is provided to processing subsystem, such as needed based on stream transmission or in response to being carried out by processing subsystem Poll.The combination of any kind of environmental sensor and environmental sensor can be used；Show by way of example accelerometer, Magnetometer, gyroscope and GPS receiver.

Some environmental sensors can provide the position in relation to wearable device and/or the information of movement.For example, accelerometer The acceleration (relative to free-falling) along one or more axis can be sensed, for example, combining using piezoelectric part or other component Associated electronic device generates signal.Magnetometer can sense environmental magnetic field (for example, magnetic field of the earth) and generate and can be solved It is interpreted as the correspondence electric signal in compass direction.Gyrosensor can sense rotary motion in one or more directions, such as Use one or more MEMS (MEMS) gyroscope and relevant control and sensing circuit.Global positioning system (GPS) receiver can determine position based on from GPS satellite received signal.

In addition to or instead of these examples, it may also include other sensors.For example, sound transducer in combination with microphone together with Associated circuit and/or program code may also include temperature sensor, close biography with the decibel level of determining such as ambient sound Sensor, ambient light sensor, biometric sensor/physiological characteristic sensor, such as heartbeat, breathing, pulse, blood glucose, alcohol Concentration detection sensor etc..In some embodiments, physiology or biosensor can be used for verifying the wearer of wearable device Identity.

In one embodiment, as shown in figure 17, the electronic equipment further includes storage device 74, for storing the processing The data of device device output.In some instances, the storage device 74 be, for example, NVM (Non-volatile memory, it is non-easily Lose memory, abbreviation NVM), read-only memory (ROM, Read-OnlyMemory), random access memory (RAM, RandomAccessMemory), EEPROM, CD-ROM or disk storage device or other magnetic storage apparatus, flash memory or energy Being enough in storage has the desired program code of instruction or data structure form and can be accessed by computer any Other media.In addition, any connection can be properly termed as computer-readable medium.

In one embodiment, as shown in figure 17, the electronic equipment further includes wireless communication module 76, described for sending The data of processor device output, or receive the data of external equipment wireless transmission.The communication interface of the wireless communication module 76 is all Such as, but not limited to, wireless interface, blue tooth interface, USB interface, Wi-Fi interface, TCP/IP interface, network communication interface or any Conventional communication interface.

In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit It divides, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or The mutual coupling, direct-coupling or communication connection discussed can be through some interfaces, the indirect coupling of device or unit The form closed or communicated to connect.

The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme 's.

It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit It is that each unit physically exists alone, can also be integrated in one unit with two or more units.

The application also provides a kind of computer readable storage medium, is stored with the computer program of energy management, the meter Calculation machine program, which is performed, realizes that earlier figures 1, Fig. 5 and the as described in the examples of Fig. 7 write policy control method.

It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product It is stored in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially in other words The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a People's computer, server or network equipment etc.) execute each embodiment the method for the application all or part of the steps.

In embodiment provided by the present application, the computer-readable storage medium of writing may include read-only memory (ROM, Read-OnlyMemory), random access memory (RAM, RandomAccessMemory), EEPROM, CD-ROM or Other optical disk storage apparatus, disk storage device or other magnetic storage apparatus, flash memory, USB flash disk, mobile hard disk or it can be used in Store any other Jie that there is the desired program code of instruction or data structure form and can be accessed by computer Matter.In addition, any connection can be properly termed as computer-readable medium.For example, if instruction is using coaxial cable, light The wireless technology of fine optical cable, twisted pair, digital subscriber line (DSL) or such as infrared ray, radio and microwave etc, from net Stand, server or other remote sources send, then the coaxial cable, optical fiber cable, twisted pair, DSL or such as infrared ray, The wireless technology of radio and microwave etc includes in the definition of the medium.It is to be understood, however, that computer-readable It writes storage medium and data storage medium does not include connection, carrier wave, signal or other fugitive mediums, and be intended to be directed to Non-transitory, tangible storage medium.As application used in disk and CD include compact disk (CD), laser-optical disk, CD, digital versatile disc (DVD), floppy disk and Blu-ray Disc, wherein disk usually magnetically replicate data, and CD is then With laser come optically replicate data.

In conclusion the policy control method of the application, system, non-volatile processor, neural network chip, electronics are set The power cut-off information that standby, electronic device and computer readable storage medium pass through acquisition at least one moment of electronic equipment；And foundation The power cut-off information carries out the following power-off time and power-off confidence level that prediction calculates the electronic equipment；Then according to described in not Carry out power-off time and power-off confidence level carries out prediction calculating acquisition and writes policy instructions and write tactful confidence level, to enable the processor Policy instructions are write according to described in executes write operation；Spare energy can be reduced by improving backup retention time whereby, and right The prediction of power-off time reasonably writes strategy to determine, and then realizes energy-efficient control.

The principles and effects of the application are only illustrated in above-described embodiment, not for limitation the application.It is any ripe Know the personage of this technology all can without prejudice to spirit herein and under the scope of, carry out modifications and changes to above-described embodiment.Cause This, those of ordinary skill in the art is complete without departing from spirit disclosed herein and institute under technical idea such as At all equivalent modifications or change, should be covered by claims hereof.

Claims

1. a write strategy control method, applied in the electronic equipment with processor, which is characterized in that including following step It is rapid:

Obtain the power cut-off information at least one moment of electronic equipment；

The following power-off time and power-off confidence level that prediction calculates the electronic equipment are carried out according to the power cut-off information；And

It carries out prediction according to the following power-off time and power-off confidence level and calculates to obtain writing policy instructions and writing tactful confidence level, Write operation is executed to enable the processor write policy instructions according to described in；The policy instructions of writing include executing the write-in of write operation Electric current and write time at least one information.

2. according to claim 1 write policy control method, which is characterized in that described to obtain the electronic equipment at least one The step of power cut-off information at a moment is to obtain at least one moment of the electronic equipment when detecting electronic equipment power-off Power cut-off information.

3. according to claim 1 write policy control method, which is characterized in that at the acquisition described electronic equipment one The step of power cut-off information at quarter calculate for the pressure drop of letting out electric device both ends in the moment by acquisition one obtain it is described disconnected Power information.

4. according to claim 1 write policy control method, which is characterized in that described to be carried out in advance according to the power cut-off information The step of surveying the following power-off time for calculating the electronic equipment and power-off confidence level includes by described in neural network foundation Power cut-off information carries out prediction and calculates to obtain the following power-off time of the electronic equipment and power-off confidence level.

5. according to claim 1 write policy control method, which is characterized in that it is described according to the following power-off time and Power-off confidence level carries out prediction and calculates the step of policy instructions are write in acquisition further include:

Prediction calculating is carried out according to the following power-off time and power-off confidence level；And

Write current and write time in policy instructions are write described in tradeoff to obtain and described write policy instructions.

6. writing policy control method according to claim 1 or 5, which is characterized in that it is described according to the following power-off when Between and power-off confidence level carry out prediction and calculate to obtain the step of writing policy instructions and writing tactful confidence level further include: judge described Write whether tactful confidence level meets a preset condition；Policy instructions are write described in exporting if meeting to the processor；If discontented It is sufficient then not to the processor transmission described in write policy instructions.

7. writing policy control method according to claim 1 or 5, which is characterized in that it is described according to the following power-off when Between and power-off confidence level carry out prediction and calculate obtaining the step of writing policy instructions and following powering off by a neural network according to described Time and power-off confidence level carry out prediction and calculate to obtain and described write policy instructions and write tactful confidence level.

8. according to claim 1 write policy control method, which is characterized in that further comprise the steps of:

Data bit width instruction according to the power-off confidence level and acquisition carries out prediction and calculates acquisition service quality predictive information； And

The service quality predictive information is exported when the service quality predictive information meets a preset condition to the processing Device.

9. according to claim 8 write policy control method, which is characterized in that the data bit width instruction is according to prediction The following conduction time, energization confidence level and the received electronic equipment storage information carry out prediction and calculate to obtain.

10. according to claim 9 write policy control method, which is characterized in that when the following energization according to prediction Between, the storage information of energization confidence level and the received electronic equipment carries out prediction and calculates to obtain the data bit width instruction It is to carry out Approximate prediction by a neural network to calculate acquisition.

11. a write strategy control system, applied in the electronic equipment with processor characterized by comprising

Fallout predictor is powered off, calculates the electronics for carrying out prediction according to the power cut-off information at least one moment of electronic equipment The following power-off time and power-off confidence level of equipment；And

BACKUP TIME fallout predictor writes strategy for carrying out prediction calculating acquisition according to the following power-off time and power-off confidence level Tactful confidence level is instructed and write, executes write operation to enable the processor write policy instructions according to described in；It is described to write policy instructions Write current and write time at least one information including executing write operation.

12. according to claim 11 write policy controlling system, which is characterized in that the power-off fallout predictor is by a feature Extraction module obtains the power cut-off information at least one moment of electronic equipment.

13. according to claim 12 write policy controlling system, which is characterized in that the characteristic extracting module detects institute State the power cut-off information that at least one moment of electronic equipment is obtained when electronic equipment power-off.

14. according to claim 12 write policy controlling system, which is characterized in that the characteristic extracting module passes through acquisition One pressure drop for letting out electric device both ends at least one moment, which calculate, obtains the disconnected of at least one moment of electronic equipment Power information.

15. according to claim 11 write policy controlling system, which is characterized in that the power-off fallout predictor include one or Multiple non-volatile shift units for being used to store the power cut-off information.

16. according to claim 11 write policy controlling system, which is characterized in that the power-off fallout predictor is by a nerve Network carries out the following power-off time and power-off confidence level that prediction calculates the electronic equipment according to the power cut-off information.

17. according to claim 11 write policy controlling system, which is characterized in that the BACKUP TIME fallout predictor is also used to Prediction calculating is carried out according to the following power-off time and power-off confidence level, by writing the write-in electricity in policy instructions described in tradeoff Stream and write time described write policy instructions to obtain.

18. writing policy controlling system described in 1 or 17 according to claim 1, which is characterized in that the BACKUP TIME fallout predictor is also For judging to write whether tactful confidence level meets a preset condition described；Policy instructions are write described in exporting if meeting to described Processor；If being unsatisfactory for policy instructions are not write to described in processor transmission.

19. according to claim 11 write policy controlling system, which is characterized in that the BACKUP TIME fallout predictor is by one Neural network carries out predicting to calculate to obtain to write policy instructions and write strategy setting according to the following power-off time and power-off confidence level Reliability.

20. according to claim 11 write policy controlling system, which is characterized in that further include service quality fallout predictor, use Prediction, which is carried out, in the data bit width instruction according to the power-off confidence level and acquisition calculates acquisition service quality predictive information, and The service quality predictive information is exported when the service quality predictive information meets a threshold value to the processor, wherein The data bit width instruction is the storage of the following conduction time according to prediction, energization confidence level and the received electronic equipment Power information carries out prediction and calculates acquisition.

21. according to claim 20 write policy controlling system, which is characterized in that the service quality fallout predictor is by one Neural network carries out prediction according to the data bit width instruction of the power-off confidence level and acquisition and calculates acquisition service quality prediction Information.

22. according to claim 11 write policy controlling system, which is characterized in that the processor further includes retention time Controller writes policy instructions for receive BACKUP TIME fallout predictor output, writes in policy instructions according to described in and includes Write current and write time at least one information execute write operation.

23. according to right want 22 described in write policy controlling system, which is characterized in that the retention time controller include one Or multiple non-volatile shift units.

24. according to claim 11 write policy controlling system, which is characterized in that the processor is non-volatile processing Device.

25. a kind of non-volatile processor, which is characterized in that write policy control including such as claim 11-24 is described in any item System and retention time controller, the retention time controller described write writing for policy controlling system output to receive When policy instructions, the write current for including in policy instructions is write according to described in and behaviour is write in the execution of write time at least one information Make.

26. a kind of neural network chip, which is characterized in that write policy control including such as claim 11-24 is described in any item System.

27. a kind of electronic equipment, which is characterized in that write tactful control including processor such as claim 11-24 is described in any item System processed.

28. a kind of electronic device, which is characterized in that write policy controlling system including such as claim 11-24 is described in any item.

29. a kind of computer readable storage medium is stored with the computer program of energy management, which is characterized in that the calculating Machine program is performed realization, and claim 1-10 is described in any item writes policy control method.