TW202011196A - Data storage device and control method for non-volatile memory - Google Patents

Abstract

High-efficiency control technology for non-volatile memory. A data storage device includes a non-volatile memory and a controller. The controller allocates an active block from a plurality of spare blocks of the non-volatile memory to fill in write data requested by the host. When the number of the spare blocks is less than a threshold block amount, the controller regards the active block as the destination for collecting the valid data from a first source block. When there is a second source block satisfies an endangered condition, the controller regards the active block as the destination for collecting the valid data from the second source block.

Description

Data storage device and non-volatile memory control method

This case is about the control of non-volatile memory.

There are many forms of non-volatile memory-for example, flash memory (flash memory), magnetoresistive random access memory (Magnetoresistive RAM), ferroelectric random access memory (Ferroelectric RAM), resistive random access Memory (Resistive RAM), spin transfer torque random access memory (Spin Transfer Torque-RAM, STT-RAM), etc., used for long-term data storage, can be used as a storage medium to achieve a data storage device.

Non-volatile memory usually has its special storage characteristics. The technical field needs to develop corresponding control technologies for the storage characteristics of the corresponding non-volatile memory.

A data storage device implemented according to an embodiment of the present case includes a non-volatile memory and a controller. The controller configures an active block from a plurality of idle blocks of the non-volatile memory to fill in write data requested by the host. When the number of idle blocks is less than a critical number, the controller uses the active block as the destination of the effective data of the source block. The controller uses the active block as a destination for the transfer of valid data of the second source block when a second source block meets an endangered condition.

In one embodiment, the controller allows valid data of one source block to be completely moved to the active block before allowing valid data of another source block to be moved to the active block.

In one embodiment, the controller moves the valid data of a source block to the active block in multiple batches, and allows write data sent by the host to be filled into the active block between different batches.

In one embodiment, the controller estimates a ratio between the effective data amount of a source block and the free space of the active block, and sets the effective data amount of each batch moved according to the ratio. According to this ratio, the controller can further set the amount of write data that allows the host to fill the active block between two batches.

In one embodiment, the controller estimates a ratio between the amount of effective data in a source block and the amount of written data requested by the host, and sets the amount of effective data moved in each batch according to the ratio.

In one embodiment, the second source block that meets the endangered condition is an error correction failure.

In one embodiment, the second source block that meets the endangered condition is for preventive removal.

In one embodiment, the second source block that satisfies the endangered condition is moved for average wear.

In one embodiment, after the controller transfers all the valid data of a source block to the active block, the source block is released as an idle block before the active block finishes writing the end.

In one embodiment, the controller operates a flag. After the controller transfers all the valid data of a source block to the active block, the controller marks the active block with the flag to allow the valid data of other source blocks to be saved.

The operation of the above controller on the non-volatile memory can also be realized by other structures. In this case, the non-volatile memory control method can be implemented by the aforementioned concept, including: operating a non-volatile memory according to the requirements of a host; configuring a plurality of idle blocks from the non-volatile memory with an active block to fill in the host Required write data; when the number of these idle blocks is less than a critical number, the active block is used as the destination of the effective data of the first source block; and a second source block matches one When the condition is in jeopardy, the active block is used as the transfer destination of valid data of the second source block.

The following describes the embodiments in detail and the accompanying drawings to explain the content of the present invention in detail.

The following description lists various embodiments of the present invention. The following description introduces the basic concept of the present invention and is not intended to limit the content of the present invention. The actual scope of invention shall be defined in accordance with the scope of patent application.

Non-volatile memory can be flash memory (Flash Memory), magnetoresistive random access memory (Magnetoresistive RAM), ferroelectric random access memory (Ferroelectric RAM), resistive memory (Resistive RAM, RRAM ), spin transfer torque random access memory (Spin Transfer Torque-RAM, STT-RAM), etc., provide storage media for long-term data storage. The following uses flash memory as an example for discussion.

Today's data storage devices often use flash memory as a storage medium to store user data from the host. There are many types of data storage devices, including products such as memory cards, USB flash devices, solid state drives (SSD), etc. One application is to use a multi-chip package to package the flash memory and its controller—called an embedded flash memory module (such as eMMC).

A data storage device using flash memory as a storage medium can be applied to various electronic devices. The electronic device includes a smart phone, a wearable device, a tablet computer, a virtual reality device, etc. The computing module of the electronic device can be regarded as a host, which operates the data storage device used to access the flash memory.

A data storage device using flash memory as a storage medium can also be used to construct a data center. For example, a server can operate an array of solid-state drives (SSDs) to form a data center. The server can be regarded as a host, operating the connected solid-state drive to access the flash memory.

FIG. 1 illustrates the data storage device 100 according to an embodiment of the present case. The data storage device 100 includes a flash memory 102 and a controller 104. The host 106 indirectly accesses the flash memory 102 through the controller 104. The controller 104 not only receives and executes the write command from the host (Host) 106, but also actively moves the user data stored in the flash memory 102.

Flash memory has its special storage characteristics, described below.

The host 106 distinguishes user data by logical address (for example, logical block address LBA or global host page number GHP... etc.). The physical space of the flash memory 102 is divided into a plurality of blocks (Blocks) configuration use. Each block (Block) includes plural pages (Pages). Each page includes N sectors (Sectors), N is an integer greater than 1, such as: 4. The page of 16KB space can be divided into four sectors, each sector is 4KB. In one embodiment, a block stores user data according to the page number, from low to high number.

In one embodiment, the data storage device adopts multi-channel technology, which can virtualize multiple blocks of different channels into a super block, and multiple pages can be virtualized into a super page, and use super blocks and super pages to perform data Storage space management can speed up the data throughput of data storage devices.

The data storage device records the correspondence between the logical address and the physical address of the user data in a logical-physical mapping table (Logical-Physical Mapping Table, L2P Table).

The storage space of the flash memory needs to be erased before it can be used again. The smallest unit of erase is the block. Blocks can be divided into data blocks, active blocks and idle blocks. Idle blocks can be used as active blocks to write user data. After the active block is filled with user data, after the closing process (writing EOB (End of Block) information), the active block is changed to a data block. With the update of user data, some user data stored in the data block will be changed from valid data to invalid data. When the user data stored in the data block is all invalid data, it will be changed to an idle block after being erased. In another embodiment, the data block filled with invalid data will be changed to an idle block, and the erasing process will not be executed until the idle block is used as the active block.

The use of flash memory involves data movement procedures, which can be divided into garbage collection procedures and non-garbage collection procedures. When the number of idle blocks is insufficient, garbage collection (Garbage Collection) can be performed on the storage space. For example, when multiple data blocks (also known as source blocks) only store sporadic valid data, garbage collection can be performed to collect valid data into one active block (also known as destination block) to recover multiple data Blocks, increase the number of idle blocks.

There are many types of non-garbage recycling programs, which are judged based on endangered conditions. For example, data blocks (source blocks) that generate ECC failed also need to be moved to rescue user data that is still readable. In addition, data blocks (source blocks) that are read too frequently also need to be moved to avoid damage to user data caused by the reduced data storage capacity of the data blocks. This operation is also called preventive move (Early Move). In addition, data movement may also be initiated in response to wear leveling considerations between blocks. For example, all user data (including valid and invalid data) of data blocks (source blocks) with low reading frequency Move to the active block (destination block) with a higher erase count to recover the data block. In addition, the wear leveling process can also be combined with the garbage collection process, that is, moving the effective data of multiple data blocks (source blocks) to the active block (destination block) with a higher erase count in the wear leveling process.

It should be noted that the data movement is preferably achieved by means of data copying.

In the figure, the idle blocks of the flash memory 102 are assigned to the idle block pool 108, and the data blocks are assigned to the data block pool 110. When a host 106 issues a write command, or the controller 104 initiates a data movement procedure, the controller 104 selects an idle block from the idle block pool 108 as the active block A0. At this time, the idle area of the idle block pool 108 The number of blocks will be reduced by one. After that, the user data is written into the active block A0. When the active block A0 is closed and becomes a data block, the number of data blocks will increase by one.

This case proposes a high-efficiency data storage method that can meet the needs of various data storage. In addition to storing user data from the host (host 106 requests to write to flash memory 102 with a write command), initiate data movement procedures, such as garbage collection, error correction failure movement, preventive movement, and loss average movement , Or other, you can use the high-performance data storage method in this case to store user data in the source box (data block).

The controller 104 generally receives user data from the host in an active block (labeled A0), which is usually provided by a write command. In particular, compared to storing user data from the source block in another active block (called A1 to distinguish it from A0), the controller 104 in this case makes the active block A0 also serve as the destination of the data migration process, and Instead of using active block A1. Once there is a data movement requirement, for example, garbage collection, error correction failure movement, preventive movement, loss average movement, or other, the user data of the source block in the data movement process can be collected to the active block A0. Compared with the traditional technique of configuring the active block A0 and the active block A1 at the same time, the design of this case has several advantages and is described as follows.

First of all, in this case, the active block A0 can not only store the user data written by the host by the write command, but also store the user data from the source block. Therefore, the usage of idle blocks can be reduced.

In the traditional example of using the active block A1, in response to a sudden power failure event, the Sudden Power Off Recovery (SPOR) procedure must specifically consider the reliability of the data, and will discard the unclosed initiative. Block A1 is still subject to the user data on the source block. Therefore, as long as the active block A1 has not been closed, all source blocks in the data movement process must be retained and cannot be released. The above design obviously hinders the recovery of source blocks, resulting in the number of idle blocks not increasing in time, and even leading to the start of different types of data movement procedures.

Compared with the traditional technology, this case takes the active block A0 as the target block in the data transfer procedure. The SPOR procedure will not completely discard the active block A0. After the data migration is completed, the source block in the data migration process can be recovered without the need to retain it for the SPOR process. Therefore, compared with the traditional technology in which the active block A1 is specifically used as the target block, the number of idle blocks in this case can be effectively increased to overcome the above-mentioned problems.

In addition, in order to close the active block A1 as the target block early, the traditional technology may fill in some dummy data, which will reduce the data storage capacity of the data block and increase the frequency of block erasure , Shorten the life of flash memory. In contrast, using the active block A0 as the destination of the data movement process can avoid writing false data and overcome the above-mentioned problems.

In one embodiment, the controller 104 allows the effective data of one source block (whether it is garbage collection or non-garbage collection) to be completely moved to the active block A0 before allowing another source block (whether it is garbage collection or non-garbage collection). The effective data of garbage collection is moved to the active block A0.

As mentioned above, the data movement procedure can be divided into garbage recycling procedures and non-garbage recycling procedures. When the number of idle blocks is less than the critical number TH1, the controller 104 starts the garbage collection process, and uses the active block A0 as the destination block in the garbage collection process, as indicated by the garbage collection path GC.

For non-garbage collection procedures, such as error correction failure movement, preventive movement, and loss average movement, etc., the controller 104 also uses the active block A0 as the destination block in the non-garbage collection procedure, such as the non-garbage collection path Non_GC.

In one embodiment, in addition to storing user data from the host 106, the active block A0 is also used as the destination block of the garbage collection process. After the active block A0 stores user data from the host 106, if the active block A0 is not closed, the active block A0 can be used as the destination block in the garbage collection process to store the user from the source block BLK#0 data. Under this setting, if all valid data of the source block (take BLK#0 as an example) has been moved, the source block BLK#0 can be recovered and become an idle block without waiting for the active block A0 to close.

In one embodiment, in addition to storing user data from the host 106, the active block A0 can be used as the destination block of the non-garbage collection process. After the active block A0 stores user data from the host 106, if the active block A0 is not closed, the active block A0 can be used as the destination block of the non-garbage collection process to store the source block (using BLK#1 as Example) User data. Under this setting, if all the valid data of the source block BLK#1 has been moved, the source block BLK#1 can be recovered as an idle block without waiting for the active block A0 to close.

In one embodiment, in addition to storing user data from the host 106, the active block A0 can be used as a destination block for garbage collection procedures and non-garbage collection procedures. After the active block A0 stores user data from the host, if the active block A0 is not closed, the controller 104 can start a garbage collection process to move the user data of the source block (take BLK#0 as an example) to the active Block A0. If the active block A0 is still not closed, the controller 104 may start a non-garbage collection process to move the user data of the source block BLK#0 to the active block A0. Furthermore, the active block A0 can be used as the destination block of the non-garbage collection process to store user data from the source block (take BLK#1 as an example). Under this setting, if all valid data of source block BLK#0 or source block BLK#1 has been moved, then source block BLK#0 or source block BLK#1 can be recovered and become an idle block. There is no need to wait for the active block A0 to close.

In addition, the user data from the host, garbage collection program or non-garbage collection program is preferably stored or moved to the active block A0 in segments (batch) and interleaved, where segment means storing or moving a certain amount User data to the active block A0, or store or move a certain amount of user data to the active block A0 at a preset time; interleaved means that the data is provided in turn by multiple data sources, where the data source can be host, junk Source block of the recycling process or non-garbage recycling process.

In one embodiment, the controller 104 estimates the ratio of the effective data amount of the source block and the user data amount from the host 106, and sets the amount of data to be moved and written in sections according to the ratio. In another embodiment, the controller 104 estimates the ratio of the effective data amount of the source block and the free space of the active block A0, and sets the amount of data to be moved and written in sections according to the ratio.

Figures 2A-2B are flowcharts illustrating a high-performance data storage method implemented according to an embodiment of the present case, and use a flag cleanflag to control the movement of data.

Step S202: The controller 104 configures the active block A0. The controller 104 configures one of the idle blocks from the idle block pool 108 as the active block A0.

Step S204: The controller 104 initializes the flag cleanflag to "DISABLE". The flag cleanflag status "DISABLE" indicates that the active block A0 can also be configured as the destination block of the data movement process.

Step S206: The controller 104 determines whether to execute the data transfer procedure? If yes, go to step S210, otherwise go to step S208. When the predetermined condition is satisfied, for example, the number of idle blocks is less than the critical number TH1, or an error correction failure occurs, the controller 104 starts (executes) the data movement process.

Step S208: The controller 104 determines whether to close the active block A0? If yes, go to step S214, otherwise go to step S212. If the active block A0 still has free space to store data, the controller 104 does not turn off the active block A0.

Step S212: The controller 104 writes the user data from the host 106 to the active block A0, and then returns to step S206. In the above, the controller 104 executes step S206 first, and then executes steps S208 and S212, which means that the controller 104 will preferentially move the user data of the data transfer process. In another embodiment, the controller 104 may first perform steps S208 and S212, and then perform step S206. Under this setting, the controller 104 will preferentially write the user data of the host 106 to the active block A0.

Step S214: The controller 104 turns off the active block A0. If the active block A0 has no free space to store data, the controller 104 closes the active block A0 and writes EOB information to the last page of the active block A0.

Step S210: The controller 104 determines the data movement amount of the data movement process, wherein the data movement amount may be a fixed value, determined by the ratio of the amount of data in the source block to the amount of user data from the host 106, or determined by the source area The amount of data in the block and the proportion of the free space of the active block A0 are determined. The source block may be a source block of a garbage collection process or a non-garbage collection process. The amount of data moved is preferably the amount of valid data.

Step S216: The controller 104 sets the flag cleanflag to "START", where "START" represents the active block A0 as the destination block of the data transfer, and is ready to perform the data transfer.

Step S218: The controller 104 transfers the user data of the data transfer amount from the source block to the active block A0. The controller 104 can move the user data of the source block to the active block A0 at once or in sections. Assume that the source block has x user data, and the amount of user data from the host 106 is n*x, the ratio of the two is 1:n, x for example=16, n for example=10. If the data is moved for a segment, the controller 104 can move M pieces of user data to the active block A0 in each segment, M for example=4, and the number of each segment moved=4. Figure 3 illustrates an example of segmented movement. After the user data Data#0...Data#3 is moved to the active block A0, the host 106 can be inserted to write user data to the active block A0, and then the user data Data#4...Data#7 is moved to the active block A0; the remaining eight user data are also moved from the source block to the active block A0 in the same way. If the data transfer is performed at one time, the user data is not inserted into the host 106 and the controller 104 moves all 16 user data to the active block A0. In addition, when moving data, M is preferably a multiple of the section, and can cover integer multiple pages. For example, if the controller 104 writes only one page of data at a time, M=4, indicating that four sections of a page are moved together. In one embodiment, the controller 104 writes the information of one super page at a time, M=16.

Step S220: The controller 104 sets the flag cleanflag to "FINISH", where "FINISH" indicates that the one-time or segmented data transfer has been completed.

Step S222: The controller 104 writes the user data from the host 106 to the active block A0. After the user data of the data transfer amount has been transferred at this stage, the controller 104 changes the source of the data to the host 106, that is, writes the user data from the host 106 to the active block A0. The controller 104 can write a fixed number of user data, or 40 (n*M, n=10 and M=4) pieces of user data calculated from the above ratio to the active block A0.

Step S224: The controller 104 determines whether the data transfer procedure is completed? If yes, go to step S226, otherwise go to step S216. If the controller 104 moves the user data of the source block to the active block A0 in stages, the controller 104 will repeatedly execute steps S216 to S224, for example, the number of times of repetition = 4 (a total of 16 pens and 4 pens per segment) , Then repeat 4 times and move in four stages).

Step S226: The controller 104 sets the flag cleanflag to "DISABLE".

Step S228: The controller 104 changes the source block to an idle block. At this time, the number of idle block pools is increased by one. Step S228 may be irrelevant whether the active block A0 is closed. The number of idle blocks can be filled in time.

In one of step S212, step S214, step S218, and step S222, the controller 104 further includes updating the logical-physical address mapping table according to the user data stored in the active block A0. Among them, the controller 104 preferably first After generating the physical-logical address mapping table of the user data of the active block A0, the physical-logical address mapping table is used to update the logical-physical address mapping table.

In one embodiment, step S222 may include time limit judgment. If the time limit is exceeded, the flow proceeds to step S224.

In one embodiment, the amount of data for the segmented move (S218) and the segmented write (S222) depends on the effective data amount of the source block and the ratio of the free space of the active block A0. If there is only a small amount of free space in the active block A0, but the amount of effective data in the source block is large, the controller 104 increases the effective data moved by each segment (S218), or/and the host 106 user who restricts the writing of each segment Information (S222). In this way, before the host block A0 is closed, the source block can be completely moved.

The user's operating habits may cause the device to repeatedly power off and power on (called power cycling). For example, mobile phone users open the cover to view the information and then cover it. Large amounts of idle blocks are consumed and garbage collection needs occur. Certain blocks may also be read too frequently, leading to ECC failed movement, early move, wear leveling movement, or other valid data movement requests. This case allowed the excessively low number of idle blocks to be replenished in time.

The above operation of the controller 104 on the flash memory 102 may also be implemented by other structures. Anyone who does not configure the active block A1 for data transfer between blocks belongs to the scope of protection in this case. In this case, the non-volatile memory control method can be realized by the aforementioned concept.

Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. Anyone who is familiar with this skill can do some modifications and retouching without departing from the spirit and scope of the present invention. Therefore, the present invention The scope of protection shall be as defined in the scope of the attached patent application.

100: data storage device; 102: flash memory; 104: controller; 106: host; 108: idle block pool; 110: data block pool; A0: active block; Data#0...Data#7: User data; GC: garbage collection path; Non_GC: non-garbage collection of other effective data movement path; S202...S228: step.

Figure 1 illustrates a data storage device 100 according to an embodiment of the case; and Figures 2A-2B illustrate a high-performance data storage method implemented according to an embodiment of the case with a flowchart, and uses a flag cleanflag to control the movement of data; and Figure 3 illustrates an example of segmented movement.

100: data storage device

102: Flash memory

104: controller

106: Host

108: idle block pool

110: data block pool

A0: Active block

GC: garbage collection path

Non_GC: Non-garbage collection of other effective data movement path

Claims (22)

A data storage device includes: a non-volatile memory; and a controller to operate the non-volatile memory according to the requirements of a host, wherein: the controller configures one from a plurality of idle blocks of the non-volatile memory The active block fills in the write data requested by the host; when the number of idle blocks is less than a critical number, the controller uses the active block as a transfer destination of valid data of a first source block; and The controller uses the active block as a destination for the transfer of valid data of the second source block when a second source block meets an endangered condition. The data storage device as described in item 1 of the patent application scope, in which: the controller allows valid data of one source block to be completely moved to the active block before allowing valid data of another source block to be moved to the active Block. The data storage device as described in item 2 of the patent application scope, wherein: the controller moves the valid data of a source block to the active block in a plurality of batches, and allows the host to send Write data to fill in the active block. The data storage device as described in item 3 of the patent application scope, wherein: the controller estimates a ratio between the effective data amount of a source block and the free space of the active block, and sets each batch according to the ratio The amount of effective data moved. The data storage device as described in item 4 of the patent application scope, wherein: according to the ratio, the controller sets the amount of written data that allows the host to fill the active block between two batches. The data storage device as described in item 3 of the patent application scope, wherein: the controller estimates a ratio between the effective data amount of a source block and the amount of written data requested by the host, and sets each batch according to the ratio The amount of effective data moved at one time. The data storage device as described in item 1 of the patent application scope, wherein: the second source block that meets the endangered condition is an error correction failure. The data storage device as described in item 1 of the patent application scope, wherein: the second source block that meets the endangered condition is for preventive removal. The data storage device as described in item 1 of the scope of the patent application, wherein: the second source block that meets the endangered condition is an average loss move. The data storage device as described in item 1 of the patent application scope, wherein: after the controller transfers all the valid data of a source block to the active block, the controller releases the data before the active block finishes writing The source block is an idle block. The data storage device as described in item 2 of the patent application scope, wherein: the controller operates a flag; and the controller only uses the flag after all valid data of a source block is moved to the active block Marking the active block allows saving valid data from other source blocks. A non-volatile memory control method, including: operating a non-volatile memory according to the requirements of a host; configuring an active block from a plurality of idle blocks of the non-volatile memory to fill in write data requested by the host; When the number of these idle blocks is less than a critical number, the active block is used as the transfer destination of valid data of a first source block; and when a second source block meets an endangered condition, The active block is the transfer destination of the valid data of the second source block. The non-volatile memory control method as described in item 12 of the patent application scope further includes: after the valid data of one source block is completely moved to the active block, the valid data of another source block is allowed to be moved to The active block. The non-volatile memory control method described in item 13 of the patent application scope further includes: moving the valid data of a source block to the active block in multiple batches, and allowing the host between different batches The written data sent is filled in the active block. The non-volatile memory control method as described in item 14 of the patent application scope further includes: estimating a ratio between the effective data amount of a source block and the free space of the active block, and setting each according to the ratio The amount of effective data for batch movement. The non-volatile memory control method as described in item 15 of the patent application scope further includes: according to the ratio, setting the amount of written data that allows the host to fill the active block between two batches. The non-volatile memory control method described in item 14 of the patent application scope further includes: estimating a ratio between the effective data amount of a source block and the amount of written data required by the host, and setting according to the ratio The amount of effective data moved by each batch. The non-volatile memory control method as described in item 12 of the patent application scope, wherein: the second source block satisfying the endangered condition is an error correction failure. The non-volatile memory control method as described in item 12 of the patent application scope, wherein: the second source block that meets the endangered condition is for preventive removal. The non-volatile memory control method as described in item 12 of the patent application scope, wherein: the second source block that satisfies the endangered condition is an average loss move. The non-volatile memory control method as described in item 12 of the patent application scope further includes: after moving all the valid data of a source block to the active block, it is released before the end of the active block is written The source block is an idle block. The non-volatile memory control method described in item 13 of the patent application scope further includes: operating a flag; and after all the valid data in a source block is moved to the active block, it is marked with the flag The active block allows the storage of valid data from other source blocks.

Images