Patent Application
20160027518
This application is based upon and claims the benefit of priority from Japanese Patent Application No. 2014-149076, filed Jul. 22, 2014, the entire contents of which are incorporated herein by reference.
Exemplary embodiments described herein relate to a memory device and a method for controlling the same.
Data stored in a NAND flash memory, which is used in solid state drives (SSD) and other devices, is erased by a block, and is programmed by a page. When data is programmed at a page of a block, the page has to be not programmed after being erased. When a write request is received from a host device and there is no block to program the data, a block have to be erased before being programmed for writing the data. It makes latency of the write request longer because of the erasure.
One or more of the embodiments provide a memory device that stabilizes latency of write requests from a host device.
According to an embodiment, there is provided a memory device connectable to a host device including a non-volatile semiconductor memory unit including a plurality of memory blocks, each of the memory blocks including a plurality of pages, and a control unit configured to carry out data writing in the non-volatile semiconductor memory unit in units of a page and data erasing in the non-volatile semiconductor memory unit. When the control unit carries out data writing of a plurality of pages, the control unit splits data erasing of one memory block into a plurality of sub erasing steps and carries out one sub erasing step between the data writing of pages.
Hereinafter, embodiments will be described with reference to the accompanying drawings.
The host I/F 10 is connected to the host device 2 through connection and transfers data between the host device 2 and the memory device 1 under a communication protocol. Example of the connection is a serial bus such as serial advanced technology attachment (SATA), an address bus, a data bus, and the like. The host I/F 10 receives a read/write command from the host device 2, the read/write command includes a logical address (logical block addressing, LBA) and a size of requested data to be read/written (for example, the number of sectors, and one sector has, for example, 512 bytes).
The memory controller 20 controls the entire operation of the memory device 1. The memory controller 20 interprets a command transmitted from the host device 2 through the host I/F 10, and writes data to the NAND memory 50 via the NAND I/F 40 or read out data from the NAND memory 50.
The data buffer 30 is a memory used to temporarily store data received from the host device 2 until the memory controller 20 stores the data in the NAND memory 50, or temporarily store data read from the NAND memory 50 until the memory controller 20 transmits the data to the host device 2. The data buffer 30 is, for example, a general-purpose memory such as a static random access memory (SRAM), and a dynamic random access memory (DRAM).
The NAND I/F 40 controls the NAND memory 50 and includes an error correction circuit.
The NAND memory 50 is a NAND type flash memory and includes a memory cell array formed of a plurality of non-volatile memory elements. It is possible to read and write data in units of a page. It is possible to erase data in units of a block, which is an integer multiple of the page. In the description below, a memory area of the NAND memory 50 in which one page data is stored is referred to as a page. A memory area of the NAND memory 50 in which one block data is stored is referred to as a block. Accordingly, one block consists of a plurality of pages. In one example, a size of one page is 4 KB and 64 pages are contained in one block. The memory cell array includes a plurality of blocks.
Here, the NAND memory is used as the non-volatile semiconductor memory, but in alternative embodiments other memory different from the NAND memory may be used.
Data writing to the NAND memory 50 will be described using
In particular, as illustrated in
The memory controller 20 includes, as a logical-to-physical translation table, a map from an LBA to a physical address indicating a memory location in which each data is written. When a new data writing request is transmitted from the host device and the LBA there in is the same as the LBA designated by a previous data writing request, the memory device 1 writes new data at a page of an erased block where data has not been written. In this case, the memory device 1 makes a page which holds the data of the previous write request invalid, and makes a page which holds the data of the write request valid.
Here, when data is invalid, it means that there is no logical address in association with the data. When data is valid, it means that there is a logical address associated with the data. When a page is invalid, it means that there is no logical address associated with the page. When a page is valid, it means that there is a logical address associated with the page.
As more data are written to the NAND memory 50, available blocks for data writing decrease and invalid pages increase. As a result, capacity of the NAND memory 50 available for data writing decreases. Accordingly, the memory device 1 performs a garbage collection at an appropriate timing, so that a block available for data writing is secured.
At least one erased block is required in order to perform this garbage collection, which means that data volume available for data writing from the host device 2 (declared capacity) is smaller than a physical capacity of the NAND memory 50 (actual capacity).
In the memory device 1, a ratio of data volume of data requested to be written from the host device 2 (writing-requested data) with respect to data volume of the garbage collection target data is within a predetermined range. For simple description, this ratio is set to a constant value C(GC(Garbage Collection)/H (Host)) (C(GC/H)) and the embodiments will be described below.
When the writing-requested data and the garbage collection target data are written to the NAND memory 50, it is necessary to also write a predetermined volume of logical-to-physical translation table data. This ratio, that is, a value which is obtained by dividing a sum of data volume of the writing-requested data and the garbage collection target data (user data) by data volume of the logical-to-physical translation table data to be written to the NAND memory 50, is within a predetermined range. For simple description, the minimum value of this ratio is set to C(UD(User Data)/T(Translation Table Data)) (C(UD/T)), and the embodiments will be described below.
The memory device 1 splits data erasing operation of one block of the NAND memory 50 (referred to below simply as “block erasing”) into a plurality of data erasing operations and performs the spit data erasing operations (referred to below simply as “split erasing”) accordingly. That is, split erasing is performed each time a predetermined volume of user data is written to the NAND memory 50.
Each time writing-requested data of one block is written, garbage collection target data of C(GC/H) block is written. In this case, it is necessary to erase data from (1+C(GC/H)) blocks of the NAND memory 50 per one block of the writing-requested data, in order to secure sufficient number of blocks for the writing-requested data and the garbage collection target data.
C(GC/H) is a parameter that is configurable by the firmware and may be determined based on, for example, the declared capacity or the actual capacity. When data stored in a block of the NAND memory 50 is transferred and then data in the block is erased during a previous garbage collection, respective quantity of valid pages and invalid pages in a block to be erased may be measured, and then C(GC/H) may be determined based on the practical measurement result. The LBA of writing-requested data may be stored, and then C(GC/H) may be determined based on an access pattern indicating that these LBAs are continuous or sparse. The more continuous LBAs become, the higher probability becomes that a block may be detected where data movement is completed in small amount during garbage collection, and the smaller C(GC/H) may become.
One block of the NAND memory 50 corresponds to C(P (Page)/B(Block)) pages in size. In other words, one block consists of C(P (Page)/B(Block)) pages. C(P (Page)/B(Block)) is a constant determined by characteristics of the NAND memory 50. Data writing unit in the NAND memory 50 is one page, and accordingly, data of one page is written during one data writing process. Thus, it is necessary to perform data erasing of one block each time writing-requested data of (C(P/B)/(1+C(GC/H)) pages is written by in the NAND memory 50. In other words, it is necessary to perform data erasing of one block per (C(P/B)/(1+C(GC/H)) times of data writing of the writing-requested data. For convenience of explanation, (C(P/B)/(1+C(GC/H)) is referred to as C(Span) below.
A deadline to complete erasing of one block (erasing completion deadline) is a timing when data of a last page, among C(Span) pages, is written to the NAND memory 50.
When block erasing of the NAND memory 50 is performed through the split erasing, there is a minimum time required for each of the split erasing process. Accordingly, when block erasing is performed for one block, an available splitting number has an upper limit. The upper limit is an integer determined by characteristics of the NAND memory 50, and the like. For convenience of description, the upper limit is referred to as C(MaxSplit) hereinafter. Thus, the number of split erasing of erasing in one block should be a smaller one of C(MaxSplit) and C(Span).
When data writing request on data of C(Span) pages is received from the host device 2, a time period necessary for writing to the NAND memory 50 is the sum of a time period for writing the writing-requested data to the NAND memory 50, a time period for writing the garbage collection target data to the NAND memory 50, and a time period for erasing one block of the NAND memory 50. A value obtained by dividing this value by a smaller value of C(MaxSplit) and C(Span), which is equal to a total time period necessary for writing one page of the writing-requested data to the NAND memory 50, is the maximum latency of a write request from the host device 2 to the memory device 1.
A block to be erased is determined by the time the data erasing starts. That is, a block to be erased is determined at a timing early enough so that data writing on a number of pages equal to a smaller value of C(MaxSplit) and C(Span) can be completed prior to the erasing completion deadline.
There maybe situations where a data writing request from the host device 2 is not received at all, such as when all the data in the NAND memory 50 are requested to be erased. In this case, in order to increase a response speed with respect to an erasing request from the host device 2, when the request of erasing only is continuously received from the host device 2 predetermined times, erasing of the NAND memory 50 is not split but performed collectively. In another example, when a battery is built in the memory device 1 and the power supply is cut off while the memory device 1 is operated, whether or not it is possible to continue a process after the termination of the power supply is determined based on a battery capacity when the power supply is cut off. Then, how to perform erasing after the termination is determined according to the determination result. When it is determined that battery capacity is not enough to continue the erasing, erasing is not restarted. When it is determined that battery capacity is enough to continue the data erasing, the data erasing is not split, but continuously performed to complete the data erasing.
When a write request is not received from the host device 2, it seems that it is not necessary to perform data writing to the NAND memory 50. However, if data is not written at a page in an erased block for a predetermined time after data has been erased from the block, an error is likely to occur when data is written and then read at the pages of the block not yet written to. To avoid this issue, when a write request is not received from the host device 2 for a predetermined time period, dummy data is written to the NAND memory 50, and data writing is completed at the all pages of the block within the predetermined time. In this manner, the dummy data writing is scheduled as if a write request was received from the host device 2, and a timing of erasing new writing destination is determined.
In
Since C(Span) is 16, one block is erased each time data of 16 pages is written to the NAND memory 50. Since C(MaxSplit) is 16, it is possible to split the data erasing of one block into 16 steps. Accordingly, it is possible to perform the split erasing, each time data of one page is written to the NAND memory 50. A target block on which the split erasing is performed is determined before data is written at the first page, the 17th page, the 33th page, and the 49th page.
For example, when 10 msec is used for one split erasing each time the data writing is performed on the writing-requested data from the host device 2, the maximum latency of a write request from the host device 2 is 20 msec, which is a summation of a time taken to perform data writing of the writing-requested data from the host device 2 and a time used to perform one split erasing. Accordingly, it is possible to guarantee that the maximum latency of a write request from the host device 2 is 20 msec.
When the maximum latency of a write request from the host device 2 is designated, the memory controller 20 determines time of one split erasing to meet the maximum latency. For example, when the maximum latency of a write request from the host device 2 is designated as 15 msec, it is necessary that the time used to perform one split erasing is 5 msec. In this case, a summation of processing time taken to erase one block is 80 msec.
Here, for comparison, it is assumed that single block erasing (without split) is performed whenever a blank block for data writing becomes necessary. In this case, the data writing of the writing-requested data from the host device 2 to the NAND memory 50 may have to be carried out after the block erasing of a writing destination block is performed. As the processes have to be performed in the order of the block erasing and then the data writing, the latency is increased by a time taken to perform the block erasing.
On the other hand, according to the first embodiment described above, it is possible to equalize a latency of a write request from the host device 2 to the sum of a time taken for the data writing of the writing-requested data to the NAND memory 50 and a time taken to perform one split erasing from the NAND memory 50. Further, it is possible to guarantee the maximum latency of a write request from the host device 2.
This second embodiment is different from the first embodiment in that when a write request from the host device 2 is not received, the garbage collection target data is written, instead of the dummy data. In this case, it is possible to proceed with the garbage collection and to decrease C(GC/H) in the long term.
Alternatively, when a write request from the host device 2 is not received, no data maybe written and new block erasing may not be performed. Even when there is a timing when the data writing of the writing-requested data can be performed, a write request from the host device 2 may not be received. In such as case, an erased block may be already prepared for when a next write request from the host device 2, and it is not necessary to perform new block erasing. That is, block erasing may be skipped as with writing of dummy data to the NAND memory 50.
According to the second embodiment described above, as in the first embodiment, it is possible to guarantee the maximum latency of a write request from the host device 2. Further, it is possible to proceed with the garbage collection and to decrease C(GC/H) in the long term.
This third embodiment is different from the first embodiment in that, when a block to be erased is not present at a timing even though the block erasing can be performed in the NAND memory 50, data writing of the garbage collection is continued instead of the block erasing (as illustrated with data writing on 33th page through 64th page), because there is already sufficient amount of erased blocks, which is equal to or more than a predetermined value, or the like. In this case, it is possible to proceed with the garbage collection and to decrease C(GC/H) in the long term.
According to the third embodiment described above, as in the first embodiment, it is possible to guarantee the maximum latency of a write request from the host device 2. As in the second embodiment, it is possible to proceed with the garbage collection and to decrease C(GC/H) in the long term.
The fourth embodiment is different from the first embodiment in that execution timing of the block erasing is changed in case time to write a page depends on its page number in the NAND memory 50.
The NAND memory 50 may be a SLC(Single Level Cell) type NAND memory in which one bit is stored in one memory cell, and may be a MLC(Multi Level Cell) type NAND memory in which two or more bits are stored in one memory cell. When the NAND memory 50 is a MLC type NAND memory, data of two pages may be stored in each memory cell of the NAND memory 50. A page where writing has been previously performed is a lower page, and a page where writing is performed subsequent to the lower page is an upper page. In a MLC type NAND memory, in order to reduce a program disturbance due to coupling capacitance between adjacent memory cells, the order of writing at the lower page and the upper page is determined. The data writing is repeatedly performed plural times, which includes a writing operation of writing data at a memory cell and a verifying operation of verifying data written in the memory cell. However, the data writing is performed to write data to cells more precisely for the last writing, and takes longer time. As the last writing, which takes time, is performed for the last page of a block, time to write data to the last page becomes longer than time to write data to other pages. Thus, time to write data to a page depends on its page number in the NAND memory 50.
According to the fourth embodiment, when a page on which the writing-requested data from the host device 2 is to be written is a page that requires a long writing time, split erasing is not performed. In contrast, when the page is a page that needs only a short writing time, split erasing is performed.
In an example shown in
According to the fourth embodiment, it is possible to stabilize a latency or a write request from the host device 2, and to guarantee the maximum latency of a write request from the host device 2. When the garbage collection target data is written, if there is a difference of writing times depending on a page of a data writing destination, whether or not to perform the block erasing is determined based on total writing time obtained including time period to write the writing-requested data from the host device 2. Further, when the data writing on a first page takes a long period of time, the block erasing may not be performed as with the last page.
The fifth embodiment is different from the first embodiment in that when the number of pages of garbage collection target data to be written to the NAND memory 50 is different, execution timing of the block erasing is changed.
When C(GC/H) is not an integer, the number of writing pages of garbage collection target data is different depending on a timing at which a write request is received from the host device 2. In this case, the block erasing is performed when the number of writing pages of the garbage collection target data is small, and the block erasing is not performed when the number of writing pages of the garbage collection target data is large. With this operation, it is possible to stabilize a latency of a write request from the host device 2.
According to the fifth embodiment described above, as in the first embodiment, it is possible to guarantee the maximum latency of a write request from the host device 2.
In the sixth embodiment, the split erasing is performed so as to erase data of one block each time data of seven pages is written to the NAND memory 50 according to a write request from the host device 2.
The number of pages of the NAND memory 50, on which the logical-to-physical translation table is written, is ((1+C(GC/H))/C(UD/T) with respect to writing of one-page data requested to be written from the host device 2. As ((1+C(GC/H))/C(UD/T) includes division, the number of pages of the NAND memory 50, on which user data is written in accordance with a write request from the host device 2, is not necessary an integer multiple of the number of the pages on which data of the logical-to-physical translation table is written. Accordingly, when user data is written to the NAND memory 50, in accordance with a write request from the host device 2, the number of pages of the NAND memory 50, on which the logical-to-physical translation table is written, is different depending on the timing of writing the user data to the NAND memory 50 in accordance with a write request from the host device 2.
In the sixth embodiment, when number of pages on which the logical-to-physical translation table is written, is small, the block erasing is performed, and when the number of pages of the NAND memory 50, on which the logical-to-physical translation table is written, is large, the block erasing is not performed.
With this operation, it is possible to stabilize a latency of a write request from the host device 2. Determining whether or not to perform block erasing maybe performed based on total writing time obtained by adding writing time of user data in the NAND memory 50 and writing time of the logical-to-physical translation table in the NAND memory 50.
According to the sixth embodiment described above, as in the first embodiment, it is possible to guarantee the maximum latency of a write request from the host device 2.
This seventh embodiment is different from the first embodiment in that a block erasing destination is selected based on the number of pages for which the data writing has not been performed among pages included in a writing destination block of data.
The maximum latency of a write request from the host device 2 is reduced when block erasing is performed in order to secure separately a writing destination block of data requested to be written from the host device 2, and a writing destination block of garbage collection target data. Accordingly, it is preferable that block erasing timings of these writing destination blocks is not overlapped.
In the seventh embodiment, an erasing target block is determined among a block that is a writing destination of the data requested to be written from the host device 2, a block that is a write destination of the garbage collection target data, and a block that is a write destination of the logical-to-physical translation table, based on capacity of available area for data writing to the NAND memory 50. The maximum latency of a write request from the host device 2 is reduced when the block erasing is performed in order to secure a writing destination of data requested to be written from the host device 2, a write destination of the garbage collection target data, and a write destination of the logical-to-physical translation table. Accordingly, block erasing timings of these writing destination blocks is not likely to be overlapped.
Since it is necessary to perform the block erasing of one block when a write request of C(Span) pages is received from the host device 2, if the number of pages prepared as a writing destination of data requested to be written from the host device 2 is equal to or less than C(Span) (Yes in Step 141), the block is erased prior to write the destination block for the data requested to write from the host device 2 (Step 142).
When the block erasing is performed on the three blocks of a writing block for the data requested to be written from the host device 2, a writing block for the garbage collection target data, and a writing block for the logical-to-physical translation table, if the number of pages secured for the logical-to-physical translation table is smaller than the number of pages needed to write the logical-to-physical translation table generated by data writing to the NAND memory 50, the block erasing is performed on the writing block for the logical-to-physical translation table prior to the writing block for the garbage collection target data. If the number of pages secured for the logical-to-physical translation table is greater than the number of pages needed to write the logical-to-physical translation table generated by data writing to the NAND memory 50, the block erasing is performed on the writing block for the garbage collection target data prior to the writing block for the logical-to-physical translation table. That is, when the block erasing is performed on the three blocks for the writing-requested data from the host device 2, the garbage collection target data, and the logical-to-physical translation table, if the number of pages secured for the logical-to-physical translation table is less than (3*C(Span)*(1+C(GC/H))/C(UD/T)), which is the number of pages for the logical-to-physical translation table generated by the data writing to the NAND memory 50 (Yes in Step 143), the block erasing is preferentially performed on the writing block for the logical-to-physical translation table (Step 144).
In order not to overlap a block erasing timing of a writing block for the garbage collection target data with a block erasing timing of a writing block for the writing-requested data from the host device 2, when the number of pages prepared as a writing for the garbage collection data is smaller than the number of writing pages for the garbage collection target data, which is (2*C(Span)*C(GC/H)) (Yes in Step 145), which are occurred while the block erasing is performed on two blocks of the writing block for the writing-requested data from the host device 2 and the writing block for the garbage collection target data block erasing is preferentially performed on the writing block for the garbage collection target data (Step 146).
When the number of pages prepared as a data writing destination is enough (No in Step 141, No in Step 143, and No in Step 145), the block erasing is randomly performed (Step 147).
According to the seventh embodiment, as in the first embodiment, it is possible to guarantee the maximum latency of a write request from the host device 2. It is possible to avoid a situation where erased block is not present when writing of the data requested to be written from the host device 2, writing of the garbage collection target data, and writing of the logical-to-physical translation table are performed.
The embodiments is not limited thereto, various changes maybe added in a range not departing from a gist of the present disclosure.
While certain embodiments have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the inventions. Indeed, the novel embodiments described herein may be embodied in a variety of other forms; furthermore, various omissions, substitutions and changes in the form of the embodiments described herein may be made without departing from the spirit of the inventions. The accompanying claims and their equivalents are intended to cover such forms or modifications as would fall within the scope and spirit of the inventions.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2014-149076 | Jul 2014 | JP | national |
