Sequential writes to flash memory

[0001] This application claims priority to U.S. Provisional Patent Application Ser. No. 61/162,571 filed Mar. 23, 2009, the disclosure of which is incorporated by reference herein in its entirety. This application also claims priority to U.S. Provisional Patent Application Ser. No. 61/169,035 filed Apr. 14, 2009, the disclosure of which is incorporated by reference herein in its entirety.

[0002] Unless otherwise indicated herein, the approaches described in this section are not prior art to the claims in this application and are not admitted to be prior art by inclusion in this section.

[0003] In conventional NAND flash memory, flash cells are organized into pages and pages are organized into blocks. Conventional NAND flash memory has some design limitations, such as having to write an entire page of cells to write a single flash cell. Further, if any of the flash cells in the page have previously been written, those flash cells have to be erased prior to writing. To erase those flash cells the entire block of pages containing those flash cells also has to be erased.

[0004] In part to address these design limitations, writing to flash cells of NAND flash memory often involves copying data from every page of cells within a block to a buffer, erasing the entire block, modifying the data associated with the flash cells in the buffer, and writing every page of cells within the block. This does not permit good data throughput in many situations, especially those using random writes. A random write is a write of data to any randomly chosen location within the flash memory and often involves no more than a few bytes of data.

[0005] This summary is provided to introduce subject matter that is further described below in the Detailed Description and Drawings. Accordingly, this Summary should not be considered to describe essential features nor used to limit the scope of the claimed subject matter.

[0006] In one embodiment, a method is described that comprises storing first data associated with a first write request in a write buffer that is separate from a flash memory, the first write request being configured to request that at least part of the first data be written to the flash memory, determining that the first data and second data associated with a second write request are sufficient for a sequential write to be performed to the flash memory, and performing a sequential write of the first data and the second data to the flash memory.

[0007] In another embodiment, a method is described that comprises receiving a request to write first data to a flash memory, determining that the first data is not sufficient to be written through a sequential write, storing the first data in a write buffer that is separate from the flash memory until sufficient second data associated with one or more additional write requests is stored in the write buffer such that the sequential write is capable of being performed using the first and the second data, and performing the sequential write of the first and the second data to the flash memory in response to determining that the first and the second data are sufficient to perform the sequential write.

[0008] In still another embodiment, a system is described that comprises a flash-memory controller, a write buffer separate from a flash memory, and a write-buffer controller configured to cause the flash-memory controller to store first data associated with a first write request in the write buffer, the first write request configured to request that at least part of the first data be written to the flash memory, determine that the first data and second data associated with a second write request are sufficient for a sequential write to be performed to the flash memory, and perform a sequential write of the first data and the second data to the flash memory.

[0014] This disclosure describes apparatuses and techniques for making sequential writes to flash memory. In some cases this is accomplished using a write buffer to store random writes. The stored random writes are written to the flash memory when sufficient random writes are stored to perform a sequential write. Sequential writes are faster than random writes from a data throughput perspective. These techniques effectively increase the data throughput of flash memory by grouping multiple random writes into one sequential write.

[0015] In the discussion that follows, an operating environment is described. Methods are also described that may be employed in the operating environment as well as other environments. In the discussion below, reference will be made to the environment by way of example only and, therefore, implementations described below are not limited to the environment.

[0016] Operating Environment

[0017] FIG. 1 illustrates an operating environment 100 having a flash-controller system 102. Flash-controller system 102 is communicatively linked with computing device 104 through communication link 106 and with a flash memory 108 through communication link 110. Link 106 and link 110 is each a wired communication link and/or a wireless communication link. Flash memory 108 may include multiple blocks of flash memory, such as NAND flash memory, and may be located within flash-controller system 102.

[0018] Flash-controller system 102 receives write requests 112 from computing device 104. Each write request 112 describes data to be written to flash memory 108. After multiple write requests 112 have been received, flash-controller system 102 performs a sequential write 114 having data from these write requests 112. Sequential write 114 may be a partial or complete sequential write. A partial sequential write includes data that fills at least two bytes worth of cells in a block (but less than all of the cells in a block) of flash memory 108. A complete sequential write includes data that fills all of the cells in a block of flash memory 108.

[0019] FIG. 2 illustrates an example of a detailed embodiment of flash-controller system 102. Flash-controller system 102 includes a write buffer 200 to which flash-controller system 102 stores write requests 112. After sufficient write requests 112 have been stored to enable sequential write 114, flash-controller system 102 communicates sequential write 114 to flash memory 108 over communication link 110.

[0020] Flash-controller system 102 also includes flash-memory controller 202, system connector 204, storage connector(s) 206, backup power-source 208, backup flash-memory 210, and, in this embodiment, flash memory 108. Flash-memory controller 202 includes write-buffer controller 212 and power-failure-backup handler 214. Write buffer 200 is a computer-readable-medium such as dynamic random access memory (DRAM) with a capacity that is a substantial percentage of the total space of flash memory 108, such as twenty-five percent. Write-buffer controller 212 stores write requests 112 and data associated with the write requests 216 in write buffer 200. This data associated with the write requests 216 is the data for which writing is being requested with the write requests 112.

[0021] Write-buffer controller 212 also stores metadata 218 in write buffer 200. Metadata 218 includes information about data 216. Metadata 218 can be used to track or organize data 216 using a data structure, such as a radix-tree data structure. Metadata 218 is updated to keep track of blocks of data in flash memory 108 that have been updated by write requests but which have not yet been written to flash memory 108. When computing device 104 requests to read data, metadata 218 is checked to see if the latest version (a most recently modified version) of the data is in write buffer 200. If the data is in a block that is in data 216, then the latest version of the data is in write buffer 200 (otherwise, the latest version of the data is not in write buffer 200). If the latest version of the data is in write buffer 200, the data is read from write buffer 200 instead of from flash memory 108. After data 216 is written to flash memory 108, metadata 218 is updated to remove entries regarding the written data.

[0022] Metadata 218 may also include another radix-tree type of data structure to arrange multiple write requests 112 into one sequential write 114. Write-buffer controller 212 uses metadata 218 to translate logical-block-addressing (LBA) to physical-block-addressing (PBA) and write changed blocks to flash memory 108 in a log-structured manner. The same radix-tree data structure can be used for this address translation to track data changes. By using the radix-tree to write back the data in a log-structured manner, wear leveling and associated garbage collection can be handled in write buffer 200, which is faster and allows flash memory 108 to be available for writing data. Furthermore, flash memory 108 may forgo space reserved for wear leveling and garbage collection.

[0023] Power-failure-backup handler 214 detects when power to flash-controller system 102 is lost, and in response to detecting that power is lost backs up data 216 and metadata 218 to backup flash-memory 210. Backup power-source 208 keeps flash-memory controller 202, backup flash-memory 210, and write buffer 200 powered while handler 214 performs a backup of data 216 and metadata 216. Backup power-source 208 is an auxiliary battery or a temporary power-source, such as super capacitors. Super capacitors, also known as electric double-layer capacitors, are electrochemical capacitors that have an unusually high energy density compared to that of common capacitors and usually operate at a lower working voltage. Backup flash-memory 210 is an area of flash memory reserved for backup of write buffer 200. This reserved area of flash memory is located within flash memory 108 or in a separate flash memory mounted on or external to flash-controller system 102. Power-failure-backup handler 214 also detects when power to flash-controller system 102 is subsequently restored, and in response to detecting that power has been restored restores data 216 and metadata 218 to write buffer 200.

[0024] System connector 204 connects to communication link 106 (e.g., is a Peripheral Component Interconnect Express “PCIe” connector). Storage connector 206 connects to communication link 110.

[0025] Note that one or more of the entities shown in FIGS. 1 and/or 2 may be further divided, combined, and so on. Each of these entities can be hardware, software, firmware, or a combination thereof, and/or stored on computer-readable-media and executed by one or more processors (not shown). For example, write-buffer controller 212 and power-failure-backup handler 214 can be software and/or firmware executed by flash-memory controller 202 or as driver software executed by computing device 104.

[0026] Also note that flash-controller system 102 can be implemented in different manners. Flash-controller system 102 can be a circuit board including various integrated circuits (ICs), a system-on-chip (SoC), or a device including multiple circuit boards. Furthermore, flash memory 108 can be included within flash-controller system 102 or as part of a separate flash-memory device. Flash-controller system 102 can be included as part of various different devices, such as part of a computing device, a portable flash-memory device, a video processing and/or rendering device, a television set-top box, or a gaming device.

[0027] Techniques for Sequential Writes

[0028] This disclosure describes techniques for improving flash memory write performance, such as by storing multiple writes into a sequential write. Two methods are illustrated below, aspects of which may be implemented in hardware, firmware, software, or a combination thereof. These methods are shown as a set of acts that specify operations performed by one or more entities and are not necessarily limited to the order shown.

[0029] FIG. 3 illustrates a method 300 for improving flash-memory write performance. At 302, data associated with a write request is stored in a write buffer that is separate from a flash memory. Consider method 300 in the context of FIGS. 1 and 2. At 302, write-buffer controller 212 stores data 216 in write buffer 200. Write buffer 200 also contains additional data from other write requests, such as from prior iterations of operation 302.

[0030] At 304, the data and additional data associated with one or more additional write requests are determined to be sufficient for a sequential write to be performed on the flash memory. Continuing the ongoing example, at 304, write-buffer controller 212 determines, using metadata 218, that enough data 216 has been stored to enable a sequential write 114 to be performed on flash memory 108. Data associated with two writes that affect cells in a same block but not all cells in the block is sufficient to perform a partial sequential write, which enables better performance than performing the writes separately. Data associated with two or more writes that affect all cells in a block is sufficient to perform a complete sequential write.

[0031] At 306, a sequential write of the first data and the second data is performed on the flash memory. At 306, write-buffer controller 212 performs a sequential write 114 of at least a portion of data 216 to flash memory 108. This portion of data 216 is then removed from write buffer 200 and metadata 218 is updated. The sequential write may be a partial or complete sequential write.

[0032] In some embodiments, the sequential write at 306 is performed in response to there being sufficient data for a sequential write (for a partial sequential or for a complete sequential write). In other embodiments, the sequential write at 306 is performed more or less often and based on other factors. For example, write-buffer controller 212 may be configured to determine when write buffer 200 is full or nearly full (e.g., in danger of becoming full such as by at least a threshold amount of buffer 200 being filled). In response to this determination write-buffer controller 212 performs a sequential write. A block of data in flash memory 108 may have pending write requests 112 stored within write buffer 200, for instance. While these pending write requests do not contain enough data to fill an entire block of cells (a complete sequential write), it is still more efficient to write them sequentially than separately. Write buffer controller 212 combines these write requests 112 into a partial sequential write request. This partial sequential write writes to a block of cells but does not contain data sufficient to write to every cell in the block. The corresponding area of write buffer 200 is freed up for further use. If no combination of write requests is found, a single write request may be performed.

[0033] Write-buffer controller 212 may use metadata 218 to select which portions of data 216 to be written through partial sequential writes. For example, a block of data in flash memory 108 having more pending write requests 112 stored within write buffer 200 is selected for a partial sequential write before a block of data in flash memory 108 having fewer pending write requests 112 stored within write buffer 200 is selected for a partial sequential write.

[0034] Further, write buffer controller 212 is configured, in some embodiments, to monitor the rate at which data is incoming via write requests 112. If the incoming data rate is low, write buffer controller 212 performs complete or partial sequential writes or single writes because they do not impact (or impact very little) overall data throughput. During this period of low incoming data rate, write buffer controller 212 may also pre-erase blocks of data in flash memory 108 for later use.

[0035] By storing random-write requests in write buffer 200, flash-memory controller 202 becomes immediately available to handle the next incoming write requests. This allows the incoming write requests to continue at a higher rate than if flash-memory controller 202 waits for each write request to be performed prior to handling the next request. This in turn improves overall data throughput.

[0036] Furthermore, conventional NAND flash memory employs wear leveling techniques that shift data being written to blocks with lower cycle counts (the number of times the block of cells has been written). For random writes this wear-leveling leaves behind invalid data that is later cleaned up through a process known as garbage collection. Garbage collection copies valid data from a block and then erases the block freeing it for later use. If the NAND flash memory is performing garbage collection it is not immediately available for responding to write requests. This results in additional write performance degradation. Wear leveling also oftentimes requires a particular amount (e.g., approximately twenty-five percent) of the storage space of the flash memory to be reserved for garbage collection purposes.

[0037] In some embodiments, write buffer controller 212 is configured to translate logical-block-addressing (LBA) to physical-block-addressing (PBA) and write changed blocks to flash memory 108 in a log-structured manner. This allows write buffer controller 212 to maintain a record of how often blocks have been written (explicitly or relative to other bocks in flash memory 108). By writing back the data in a log-structured manner, wear leveling and associated garbage collection can be handled in write buffer 200, which is both faster and allows flash memory 108 to be available for writing data. For example, based on this record, write buffer controller 212 can determine which blocks of flash memory 108 to write to and/or for which blocks of flash memory 108 garbage collection is to be performed. Furthermore, flash memory 108 may forgo space reserved for wear leveling and garbage collection so that a particular amount of the storage space of the flash memory need not be reserved for garbage collection purposes.

[0038] FIG. 4 illustrates an example of a method 400 for improving flash-memory write performance. At 402, a request to write data to a flash memory is received. In the context of FIGS. 1 and 2, write-buffer controller 212 receives a write request 112 to write data to flash memory 108.

[0039] At 404, it is determined that the data is not sufficient to be written through a complete sequential write, such as by write-buffer controller 212. The data requested to be written may not be sufficient to fill an entire block of cells in flash memory 108.

[0040] At 406, the data is stored in a write buffer that is separate from the flash memory. The data is stored until sufficient additional data associated with one or more other write requests is stored in the write buffer such that a sequential write is capable of being performed using the data received at 402 and the additional data associated with one or more other write requests. For example, at 406, write-buffer controller 212 stores the data in write buffer 200 as part of data 216. Metadata 218 is updated to account for this change to data 216.

[0041] At 408, it is determined that the data received at 402 and the additional data associated with one or more other write requests are sufficient to perform a sequential write. This may be performed in various manners as described above. Continuing the ongoing example, write-buffer controller 212 determines that the data received at 402 and the additional data associated with one or more other write requests is stored in write buffer 200.

[0042] At 410, a sequential write of the data and additional data is performed on the flash memory. The sequential write is performed in response to determining that sufficient data and additional data are stored in the write buffer such that a sequential write is capable of being performed. For example, at 410, once enough data 216 is stored sufficient to fill a new block of cells or to modify an existing block of cells, write-buffer controller 212 performs sequential write 114 on flash memory 108. The sequential write may be partial or complete.

[0043] In some embodiments, method 400 uses a partial sequential write rather than a complete sequential write. Operation 404, in this case, may determine that the data is not sufficient to be written until some threshold percentage of an entire block of cells has been written to. This percentage can be modified depending on throughput requirements or a current rate of incoming write requests. If a current rate of incoming write requests goes up, the percentage increases. If the current rate goes down, the percentage decreases.

[0044] One or more of the techniques described above can be performed by one or more programmable processors executing a computer program to perform functions by operating on input data and generating output. Generally, the techniques can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment containing both hardware and software components. In one implementation, the methods are implemented in software, which includes but is not limited to firmware, resident software, microcode, etc. Furthermore, the methods can take the form of a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system.

[0045] For the purposes of this description, a computer-usable or computer-readable medium can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. Examples of a computer-readable medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk. Current examples of optical disks include compact disk-read only memory (CD-ROM), compact disk-read/write (CD-R/W) and DVD.

[0046] Although the subject matter has been described in language specific to structural features and/or methodological techniques and/or acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features, techniques, or acts described above, including orders in which they are performed.