TW201719656A - Method for writing data from memory back to storage device and data accessing system using the same - Google Patents

Abstract

A method for writing data from a memory back to a storage device is provided. The memory includes a plurality of data units for storing data. The method includes steps of: calculating the corresponding write-back energy for each data unit; classifying the data units into a plurality of groups according to the calculated write-back energy, wherein each group corresponds to a write-back energy level; selecting a target group from the groups, wherein one of the groups with a lower write-back energy level is assigned with higher possibility to be chosen as the target group; choosing a target data unit from the target group, and writing the data of the target data unit back to the storage device.

Description

Method for writing data from memory back to storage device and data access system using same

The present invention relates to a method of writing data from a memory back to a storage device and a data access system using the same.

Phase-Change Memory (PCM) has been widely used in various non-Volatile storage devices because of its high-speed access and retention of data after power is turned off. In practice, when writing/programming the phase change memory, the crystal state of the phase change material in the memory cell can be changed by heating the electrode so as to present the resistance value corresponding to the written data. For example, a multi-level cell (MLC) of phase change memory can be programmed to four possible resistance value distributions to represent "11", "10", "00", "01, respectively. "Equivalent two-dimensional data.

However, although multi-level memory cells of phase-change memory can store more bits of data, when programming, it takes more energy than programming a single-level cell (SLC). , so that the system energy consumption increases.

Therefore, how to reduce the energy required to write data to the storage device to reduce the overall system energy consumption is one of the current issues in the industry.

The invention relates to a method for writing data from a memory back to a storage device and a data access system using the same, which can make the data with lower write back energy have a higher probability of being written back to the storage device, thereby reducing the pair. The energy required to store data in the storage device.

According to an aspect of the present invention, a method for writing data from a memory back to a storage device is provided. The memory includes a plurality of data units for storing data. The method includes the following steps: calculating, for each data unit, writing data back to the storage device Corresponding write back energy; classifying the data units into a plurality of groups according to the write back energy corresponding to each data unit, and each of the groups corresponds to a write back energy level; selecting one from the groups a target group, wherein the group corresponding to the lower write energy level in the groups is given a higher probability of being selected as the target group; selecting the target data unit from the target group, and the target data unit The information is written back to the storage device.

According to an aspect of the present invention, a data access system is provided that includes a processing unit, a memory, and a storage device. The memory includes a plurality of data units to store data. The storage device is coupled to the memory. The processing unit is coupled to the memory, and is configured to calculate, for each data unit, the write back energy corresponding to the data to be written back to the storage device, and classify the data units into multiple groups according to the write back energy corresponding to each data unit. Each of the groups corresponds to a writeback energy level, wherein the processing unit executes a writeback procedure when the memory space is full, the writeback procedure includes: selecting a target group from the groups, wherein the group The group corresponding to the lower write energy level in the group is given a higher probability to be selected as the target group; and the target data unit is selected from the target group, and the data of the target data unit is written back to the storage. Device.

In order to better understand the above and other aspects of the present invention, the preferred embodiments are described below, and in conjunction with the drawings, the detailed description is as follows:

In the present description, some embodiments of the invention are described in detail with reference to the drawings, but not all embodiments are illustrated in the drawings. In fact, these inventions may use a variety of different variations and are not limited to the embodiments herein. In contrast, the present disclosure provides these embodiments to meet the statutory requirements of the application. The same reference symbols are used in the drawings to refer to the same or similar elements.

1 is a simplified block diagram of a data access system 100 in accordance with an embodiment of the present invention. The data access system 100 includes a processing unit 102, a memory 104, and a storage device 106. The processing unit 102 is, for example, a central processing unit (CPU), a processor, a controller, or other arithmetic control circuit. The memory 104 is coupled to the processing unit 102. The memory 104 is, for example, the main memory of the system, and is available for the processing unit 102 to store/stage data required for the operation. The memory 104 can be implemented by a dynamic random access memory (DRAM) or other form of volatile memory (Volatile) memory, or by non-volatile (Non-Volatile) memory. The storage device 106 is coupled to the memory 104. The storage device 106 can be implemented, for example, by a non-volatile memory such as a phase change memory (PCM), a NAND flash memory, or a magnetoresistive random access memory (MRAM). ), Resistive Random Access Memory (ReRAM), etc., for permanent storage of data.

In operation, data access system 100 can be viewed as an asymmetric access architecture. For example, the program will be stored first in the storage device 106, and then the program will be loaded into the memory 104 for execution by the processing unit 102 (as indicated by path A1). The result of the correlation operation generated during the execution of the processing unit 102 may be temporarily stored in the memory 104. When the memory 104 is filled, or if certain data must be stored back in the storage device 106, the processing unit 102 will perform a write back action to write the data from the memory 104 back to the storage device 106 (eg, path) A2)). Since writing data back to the storage device 106 requires a certain amount of write back energy (for example, the energy required to apply thermal energy to the phase change memory for programming), if the writeback energy required to perform the write back operation can be reduced, Can effectively reduce system energy consumption.

According to an embodiment of the present invention, the processing unit 102 may first evaluate the write back energy of each data unit in the memory 104, and then classify the data units according to their corresponding write back energy to corresponding different write back energy levels. Group. Then, in conjunction with the probability selection method, the group corresponding to the low energy level has a higher probability to be selected, and then the appropriate write back data is found from the selected group. In this way, the data selected to be written back to the storage device 106 is mostly corresponding to low write back energy, so that the system energy consumption can be effectively reduced.

FIG. 2 is a flow chart of a method for writing data from the memory 104 back to the storage device 106 according to an embodiment of the invention. At step 202, the processing unit 102 calculates, for each data unit of the memory 104, the write back energy corresponding to writing the data back to the storage device 106. The data unit may be a memory page or other data management unit, depending on the access architecture of the system.

In an embodiment, the number of high-level data in each data unit can be counted to determine the write back energy corresponding to each data unit. The high energy level data refers to data that requires more energy to be programmed to reach the target data level, for example, corresponding to one or more specific bit patterns. For example, when programming a memory cell of a phase change memory, the first bit of the data may be written based on a similarly programmed single-level cell (SLC) to form a corresponding The resistance value distribution of the data "1" (or "11"), "0" (or "00"). If you want to write the second bit again, you need to perform more precise programming control to further separate the resistance value distribution of the corresponding data "10" and "01". Therefore, the energy required to write the data "10", "01" will be significantly higher than the energy required to write the data "11", "00". Therefore, in this example, the data of the bit pattern of "10" and "01" will be regarded as high-level data, and the processing unit 102 can transmit the number of bit patterns included in the counting data unit to "10" and "01". The data is used to estimate the write back energy corresponding to the data unit.

It can be understood that the present invention is not limited to the above examples, and the data corresponding to different memory cell distributions may be adjusted according to different designs, so that the bit pattern corresponding to the high energy level data is different. In other examples, processing unit 102 may further consider other factors to calculate the writeback energy corresponding to the data unit. The related description will be described with reference to Fig. 3.

In step 204, the processing unit 102 may classify the data units into a plurality of groups according to the write back energy corresponding to each data unit, wherein each of the groups corresponds to a write back energy level. For example, the processing unit 102 can classify the data unit whose write back energy is located in the first write back energy interval (eg, 50 joules to 150 joules) to the first group, and write the write back energy in the second write back energy interval. Data units (eg, 150 joules to 250 joules) are classified into the second group, and so on, until all of the data units are sorted into the appropriate group. At this time, the first group corresponds to a lower write energy level, which includes a data unit whose write back energy falls between 50 joules and 150 joules, and the second group corresponds to a corresponding higher one. Write back the energy level, which contains the data unit that writes back energy between 150 joules and 250 joules. In other words, the group corresponding to the relatively low write energy level means that the write back energy of the data unit included in the group falls in a relatively low write back energy interval; and the corresponding relatively high write back energy level The group refers to the write back energy of the data unit contained in the group is in a relatively high write back energy interval.

At step 206, the processing unit 102 selects a target group from the groups, wherein the groups corresponding to the lower write energy levels in the groups are given a higher probability of being selected as the target group. . For example, if the processing unit 102 selects a target group in a random manner, the number of samples corresponding to the low write back energy level group may be increased in the randomly selected sample space, and the corresponding high write back energy level group may be reduced. The number of samples, thereby increasing the probability that the low write back energy level group is selected as the target group. For example, suppose there are three groups A, B, and C, and the corresponding write back energy ranges are, for example, 0 to 100 joules, 200 to 300 joules, 300 to 400 joules, and the three groups A and B. C and C are given different weights, for example, 4, 2, and 1, respectively. At this time, the probability of picking to group A is 4/7, and the probability of picking to group B is 2/7, and the group is selected. The probability of C is 1/7. Therefore, the probability that group A with relatively low write back energy will be picked up will be higher than other groups B, C. In this manner, the storage device 106 will have a greater chance of being written back to the lower write back energy, thereby reducing the amount of energy required to perform a write back operation on the storage device 106.

In an embodiment, the weight selection of the probability picking group may be started from a group corresponding to the highest write back energy level, and the weight is set to 1, the second highest is 2, the third highest is 4, and so on. Then, these weights are added together as a denominator, and the numerator is the weight of the corresponding group (ie, the weight of the corresponding group is divided by the sum of the weights) to obtain the selected probability corresponding to the group. As a result, groups with relatively low writeback energy will have a higher probability of being picked. It is to be understood that the weight values set forth in the above description are for illustrative purposes only and are not intended to limit the invention. In other embodiments, the setting of the weight value can also be adjusted depending on the application.

In addition, through the spread of probability, the impact of cold data (Cold Data) and hot data (Hot Data) on energy consumption can be eliminated. Further, although some materials have low write back energy, they are written back at a high frequency (thermal data), so that the overall write back energy is still high. Through the probability selection method, some data (cold data) with high energy but low write frequency can be written back, which can slow down the influence of cold and hot data on energy consumption.

At step 208, the processing unit 102 selects the target data unit from the selected target group and writes the data of the target data unit back to the storage device 106. Processing unit 102 may select the target data unit based on a random manner or a certain rule. Since each data unit has been classified, and the group corresponding to the low write back energy level is given a higher probability of being selected, the processing unit 102 does not need to compare the write back energy of each data unit to find a relatively low one. Write back the energy data and then write it back. Therefore, the processing unit 102 can quickly pick out the data to be written back to the storage device 106 in this way, thereby shortening the overall programming time.

In an embodiment, after performing a write operation on the memory 104, the processing unit 102 performs management and classification of the related data units as in steps 202 and 204, and when the memory 104 space is filled, The write back procedure of steps 206 and 208.

FIG. 3 is a schematic diagram showing the data structure of a memory according to an embodiment of the invention. For convenience of explanation, the memory system described herein is exemplified by the memory 104 of FIG. 1, but the present invention is not limited thereto.

In the example of FIG. 3, the memory 104 includes a plurality of data units DU, and each of the data units DU includes a plurality of cache lines CL. The data unit DU is, for example, a data page whose size is, for example, 4K Bytes or 4M bytes, and the size of one cache line CL is, for example, 64-bit or 128-bit, depending on the actual application. And set.

Each data unit DU is respectively assigned a data counter 302, and each data counter 302 is used to calculate how many high energy-consuming data (such as the bit pattern "10" and "01") in the corresponding data unit DU, The processing unit 102 estimates the write back energy corresponding to each data unit DU.

In an embodiment, the data counter 302 does not need to count the amount of high energy-consuming data contained in all the cache lines CL in the corresponding data unit DU, but only needs to count the high consumption in the cache line CL of the data that has been modified. The amount of information can be. Further, each cache line CL is assigned a dirty bit DB to indicate whether it has been modified/written. When a cache line CL is marked by the dirty bit DB (for example, the value of the dirty bit DB = 1), it indicates that the cache line CL has been modified. On the other hand, when the cache line CL is not marked by the dirty bit DB (for example, the value of the dirty bit DB = 0), it means that the cache line CL has not been modified. In general, only the modified data may have the need to write back to the storage device 106. Therefore, the data counter 302 can count the high-level data contained in the cache line CL marked by the dirty bit DB in each data unit DU. The number is used by the processing unit 102 to calculate the write back energy corresponding to each data unit DU. For example, the processing unit 102 can calculate the write back energy corresponding to the data unit DU based on the following formula:

(Formula 1)

Where Epf represents the write back energy corresponding to a data unit DU, Cpc represents the count result of the data counter 302, Cdirty represents the number of cache lines CL marked by the dirty bit DB in the data unit DU, and Ncl represents a cache line CL. In the number of bits in the box, Ehe represents the average write back energy written to high-level data (such as data "01" or "10"), and Ele represents the average written in low-level data (such as data "00" or "11"). Write back energy.

Further, the item Ncl ́Cdirty is the number of bits included in all cache lines CL marked by the dirty bit DB, and for multi-level memory cells (MLC), it is written for each write operation. The information of two bits, so in (Formula 1), the item Ncl ́Cdirty is divided by 2 to calculate the actual written action. In addition, the reason why the item Cpc ́Ehe is not divided by 2 in (Formula 1) is that Cpc itself records the number of high-energy data (such as data "01" or "10") instead of separately recording A few bits, so there is no need to divide by 2.

It can be understood that the present invention is not limited to the above examples. It is within the spirit of the present invention to evaluate the write back energy of the corresponding data unit based on the number of high energy level data in the data unit.

FIG. 4 is a schematic diagram of a group corresponding to different write back energy levels according to an embodiment of the invention. In the example of Fig. 4, there are n groups G1 to Gn, where n is a positive integer. The groups G1~Gn correspond to the write back energy levels LV1~LVn, respectively, and are sequentially arranged from low write back energy to high write back energy. The data groups corresponding to the same write back energy level/write back energy interval are classified into the same group. As shown in FIG. 4, the data units DU11~DU1i corresponding to the write back energy level LV1 are classified into the group G1; the data units DU21~DU2j corresponding to the write back energy level LV2 are classified into the group G2; The data units DUn1~DUnk written back to the energy level LVn are classified into the group Gn, where i, j, k are positive integers. For example, suppose that the write back energy level LV1 represents a write back energy level of 100 joules to 200 joules, and the write back energy level LV2 represents a write back energy level of 200 joules to 400 joules, at this time, if the first The energy required to write the data unit, the second data unit, and the third data unit back to the storage device 106 is 130 joules, 180 joules, and 300 joules, respectively, and the first and second data units are classified into the same group G1. The third data unit will be classified to group G2.

In an embodiment, the write back energy interval corresponding to each group G1 G Gn may be an average division. For example, if the maximum write back energy is E, the interval width of the write back energy interval corresponding to each group G1~Gn may be set to E/n. At this time, the group G1 corresponds to 0 to Write back energy interval, group G2 corresponds to Write back energy interval, group G3 corresponds to Write back to the energy interval, and so on.

In another embodiment, the write back energy interval corresponding to each group G1 G Gn may be a non-average partition. For example, if the maximum write back energy is E, the write back energy interval corresponding to each group G1~Gn may be divided based on a power of a specific base Q. Taking the base Q equal to 2 as an example, the group G1 corresponds to 0 to Write back energy interval, group G2 corresponds to Write back energy interval, group G3 corresponds to Write back to the energy interval, and so on. If the groups G1~Gn are sequentially arranged from low to high according to the level of the write back energy, the write back energy interval width corresponding to the low write back energy group is the write back energy interval width corresponding to the high write back energy group. Q ^R times, where Q represents the base of the maximum write back energy E, and R represents the ordinal difference between the two groups, and both Q and R are positive integers. Taking the foregoing example as an example, in the case where the base Q is equal to 2, the write back energy interval width group G3 (corresponding to the ordinal number 3) of the group G1 (corresponding to the ordinal number is 1) is 2 ⁽³ of the write back energy interval width). ^-1) times.

It is to be understood that the invention is not limited to the above examples. The writeback energy interval corresponding to each group can also be divided based on other rules, depending on different applications.

FIG. 5 is a flow chart showing the operation of the data access system according to an embodiment of the invention. For convenience of explanation, the data access system described herein is exemplified by the data access system 100 of FIG. 1, but the present invention is not limited thereto.

At step 502, the processing unit 102 first determines whether the data unit (e.g., material page) to be accessed is present in the memory 104. If so, then proceeding to step 504, processing unit 102 will determine to perform a read or write operation on the data unit. If not, then proceeding to step 516, processing unit 102 then determines if memory 104 is full. If so, proceeding to step 506, processing unit 102 executes a write back data picking process to select the appropriate write back data from memory 104 and write it back to storage device 106. An exemplary detail flow of the selection procedure will be described with reference to Fig. 6.

If the result of the determination in step 516 is no, that is, the space of the memory 104 has not been filled, the processing unit 102 will execute step 518 to establish a data unit to be accessed in the memory 104. For example, the processing unit 102 can read from the storage device 106 or directly create a memory page with empty content on the memory 104 for reading and writing.

At step 510, when the processing unit 102 determines that a read operation is to be performed on the data unit, the processing unit 102 will read the data from the data unit.

At step 508, when the processing unit 102 determines that a write operation is to be performed on the data unit, the processing unit 102 will write the data to the data unit. Then, as shown in steps 512 and 514, for the data unit after the writing, the processing unit 102 will calculate the corresponding write back energy, and classify it according to the obtained write back energy to the corresponding group, so as to write later. Back to the selection of information.

FIG. 6 is a flow chart showing the operation of the write back data selection program according to an embodiment of the invention. This operational flowchart can be used, for example, as a detailed flowchart of step 506 in FIG. 5, but the invention is not limited thereto.

At step 602, processing unit 102 first initializes the value of a loop counter, such as resetting it to zero.

At step 604, processing unit 102 selects the group in a random manner. Each group is given, for example, a different probability of being selected, such that groups of low energy levels are more easily selected than groups of high energy levels.

At step 606, processing unit 102 adjusts the value of the loop counter to count the number of times the processing unit 102 randomly selected the group. For example, the value of the loop counter can be adjusted based on the following formula:

i=i+1 (Formula 2)

Where i represents the value of the loop counter.

At step 608, processing unit 102 determines if there are any data units in the selected group. If not, it indicates that the processing unit 102 needs to reselect the group. As shown in step 612, the processing unit 102 will determine whether the value of the loop counter exceeds a predetermined threshold, and when it is determined that the value of the loop counter has not exceeded the predetermined threshold, perform step 604 again to re-select the group. . On the other hand, when the value of the loop counter exceeds the predetermined threshold, it indicates that the processing unit 102 has performed multiple random selection procedures and still cannot select a valid group (ie, a group in which the data unit is stored). At this time, the processing unit 102 will perform step 614 to search for a valid group in a fixed order, for example, starting from the group corresponding to the lowest write energy level to find the first group with the data unit.

The selected valid group can be regarded as the target group. After the target group is selected, the processing unit 102 will select the target data unit from the selected target group and write the target data unit back to the storage device 106, as shown in steps 610 and 616. Upon completion of the write back data picking procedure as shown in FIG. 6, processing unit 102 may, for example, continue with step 518 of FIG. 5 to perform subsequent operations on memory 104.

In summary, the method for writing data from a memory back to a storage device and the data access system using the same may first estimate the write back energy of each data unit in the memory, and then The data unit is classified according to its corresponding write back energy to a group corresponding to different write back energy levels. Then, in conjunction with the probability selection method, the group corresponding to the low energy level has a higher probability to be selected, and then the appropriate write back data is found from the selected group. In this way, not only can the system energy consumption be effectively reduced, but also the speed of selecting suitable data to be written back can be improved, thereby shortening the overall programming time.

While the invention has been described above in the preferred embodiments, it is not intended to limit the invention. A person skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the scope of the invention is defined by the scope of the appended claims.

100‧‧‧Data Access System
102‧‧‧Processing unit
104‧‧‧ memory
106‧‧‧Storage device
A1, A2‧‧‧ Path
202, 204, 206, 208, 502, 504, 506, 508, 510, 512, 514, 516, 518, 602, 604, 606, 608, 610, 612, 614, 616 ‧ ‧ steps
302‧‧‧ data counter
DU, DU11~DU1i, DU21~DU2j, DUn1~DUnk‧‧‧data unit
CL‧‧‧Cache line
DB‧‧‧ dirty bit
G1~Gn‧‧‧Group
LV1~LVn‧‧‧Write back energy level
E‧‧‧Maximum write back energy

1 is a simplified block diagram of a data access system in accordance with an embodiment of the present invention. FIG. 2 is a flow chart showing a method for writing data from a memory back to a storage device according to an embodiment of the invention. FIG. 3 is a schematic diagram showing the data structure of a memory according to an embodiment of the invention. FIG. 4 is a schematic diagram of a group corresponding to different write back energy levels according to an embodiment of the invention. FIG. 5 is a flow chart showing the operation of the data access system according to an embodiment of the invention. FIG. 6 is a flow chart showing the operation of the write back data selection program according to an embodiment of the invention.

202, 204, 206, 208‧‧ steps

Claims (10)

A method for writing data from a memory back to a storage device, the memory comprising a plurality of data units for storing data, the method comprising: calculating, for each of the data units, a write back energy corresponding to writing the data back to the storage device; Sorting the data units into a plurality of groups according to the write back energy corresponding to each of the data units, each of the groups corresponding to a write back energy level; selecting a target group from the groups, wherein the The group corresponding to the lower write energy level in the group is given a higher probability to be selected as the target group; and selecting a target data unit from the target group, and the data of the target data unit Write back to the storage device. The method of claim 1, further comprising: counting the number of high-level data in each of the data units to determine a write-back energy corresponding to each of the data units; wherein the high-energy data corresponds to one or more Specific bit styles. The method of claim 1, wherein the data units respectively comprise a plurality of cache lines, the method further comprising: counting cache lines marked by dirty bits in each of the data units The number of high-level data is included to determine the write-back energy corresponding to each of the data units; wherein the high-level data corresponds to one or more specific bit patterns. The method of claim 1, further comprising: classifying, in the data unit, a data unit whose energy is located in a first write back energy interval into a first group of the groups; And storing, in the data unit, the data unit whose write energy is located in a second write back energy interval to a second group of the groups; wherein the interval width of the first write back energy level is equal to the first Second, write back the interval width of the energy level. The method of claim 1, further comprising: classifying, in the data unit, a data unit whose energy is located in a first write back energy interval into a first group of the groups; And storing, in the data unit, the data unit whose write energy is located in a second write back energy interval to a second group of the groups; wherein the first write back energy interval and the second write back The energy interval is generated by dividing a maximum write back energy based on a power of a particular base. A data access system comprising: a memory comprising a plurality of data units for storing data; a storage device coupled to the memory; and a processing unit coupled to the memory for each of the data units Calculating the write back energy corresponding to the data back to the storage device, and classifying the data units into a plurality of groups according to the write back energy corresponding to each data unit, and each of the groups corresponds to a write back energy a level, wherein the processing unit executes a write back procedure when the memory space is full, the write back procedure includes: selecting a target group from the groups, wherein the groups are correspondingly lower write back The energy level group is assigned a higher probability to be selected as the target group; and a target data unit is selected from the target group, and the data of the target data unit is written back to the storage device. The data access system of claim 6, further comprising: a plurality of data counters for counting the number of high-level data in the data units, wherein the processing unit calculates the corresponding data unit Write back energy; wherein the high energy level data corresponds to one or more specific bit patterns. The data access system of claim 6, wherein the data units respectively comprise a plurality of cache lines, the data access system further comprising: a plurality of data counters for counting each of the data units, a number of high-level data included in a cache line marked by a dirty bit, for the processing unit to calculate a write-back energy corresponding to each of the data units; wherein the high-energy data corresponds to one or more A specific bit style. The data access system of claim 6, wherein the processing unit classifies the data units in the data units that are written in the first write back energy interval into one of the groups. a group, and classifying, in the data unit, a data unit whose energy is located in a second write back energy interval to a second group of the groups; wherein the first write back energy level interval The width is equal to the interval width of the second write back energy level. The data access system of claim 6, wherein the processing unit classifies the data units in the data units that are written in the first write back energy interval into one of the groups. a group, and classifying, in the data unit, a data unit whose energy is located in a second write back energy interval to a second group of the groups; wherein the first write back energy interval and the The second write back energy interval is generated by dividing a maximum write back energy by a power of a particular base.