US 20050144516 A1
The present invention presents techniques for the linking of physical blocks of a non-volatile memory into composite logical structures or “metablocks”. After determining an initial linking of good physical blocks into metablocks, a record of the linking is maintained in the non-volatile memory where it can be readily accessed when needed. In one set of embodiments, the initially linking is deterministically formed according to an algorithm and can be optimized according to the pattern of any bad blocks in the memory. As additional bad blocks arise, the linking is updated using by replacing the bad blocks in a linking with good blocks, preferably in the same sub-array of the memory as the block that they are replacing.
1. A method of operating a memory system including a controller and a non-volatile memory, wherein the non-volatile memory is comprised of a plurality of units of erase, the method comprising:
establishing a set of metablock linkings, each comprised of a plurality of units of erase, by which the controller accesses the non-volatile memory; and
storing a record of said metablock linkings in the non-volatile memory.
2. The method of
3. The method of
4. The method of
determining that a unit of erase in a first of said metablock linkings is defective;
updating the first metablock linking so that it no longer contains said defective unit of erase; and
storing a record of the updated linking in the non-volatile memory.
5. The method of
6. The method of
7. The method of
8. The method of
subsequent to said replacing the defective unit of erase with another one of said units of erase, updating said list of unlinked units of erase.
9. The method of
10. The method of
maintaining a list of unlinked units of erase;
determining that one or more units of erase in a first of said metablock linkings is defective; and
adding the non-defective units of erase in the first metablock to the list of unlinked units of erase.
11. The method of
determining that a unit of erase in a first of said metablock linkings is defective;
determining whether an alternate unit of erase is available for the defective unit of erase; and
in response to determining that an alternate unit of erase is not available, removing the first metablock from the set of metablock linkings.
12. The method of
13. The method of
14. The method of
15. The method of
16. The method of
17. The method of
18. A method of operating a memory system including a controller and a non-volatile memory, the memory comprising at least two independently accessible subarrays, wherein the individual subarrays are divided into a plurality of non-overlapping physical structures, and wherein at least one physical structure from individual ones of said subarrays are linked to form a composite logical structure whereby the controller accesses the composite logical structure components together as a unit, the method comprising:
determining that a physical structure in said composite logical structure is defective; and
in response to said determining that a physical structure in said composite logical structure is defective, replacing said defective physical structure in said composite logical structure with an alternate physical structure while maintaining the non-defective ones of the physical structures linked to form said composite logical structure.
19. The method of
20. The method of
21. The method of
22. The method of
23. The method of
24. The method of
25. The method of
prior to said determining that a physical structure in said composite logical structure is defective, forming an initial linking of physical structures to form said composite logical structure, and wherein said replacing includes updating said linking.
26. The method of
selecting the physical structures forming the composite logical structure from a pool of available physical structures.
27. The method of
forming the composite logical structure according to an algorithm.
28. A memory system including a controller and a non-volatile memory, wherein the non-volatile memory is comprised of a plurality of units of erase, wherein the controller accesses the non-volatile memory according to a set of metablock linkings, each comprised of a plurality of units of erase, wherein the controller establishes the set of metablock linkings in a deterministic manner.
29. The memory system of
30. The memory system of
31. The memory system of
32. The memory system of
33. The memory system of
34. The memory system of
35. The memory system of
36. The memory system of
37. The memory system of
38. The memory system of
39. The memory system of
40. The memory system of
41. The memory system of
42. The memory system of
43. A method of operating a non-volatile memory system having an array of memory storage elements organized in at least two planes, wherein the individual planes are divided into a plurality of non-overlapping blocks of storage elements wherein a block contains the smallest group of memory storage elements that are erasable together, and the individual blocks are divided into a plurality of pages of storage elements wherein a page is the smallest group of memory storage elements that are programmable together, comprising:
linking at least one block from individual ones of said at least two planes to form a metablock wherein the metablock components are erased together as a unit;
redirecting a metablock component block associated with a defective physical block to a spare physical block in the same plane if a spare physical block is available in the same plane;
assigning the metablock components not associated with the defective physical block to a spare block area if no spare physical block is available in the same plane; and
updating a defect map structure, wherein the defect map structure includes an entry corresponding to the metablock.
This application is related to U.S. patent application entitled “NON-VOLATILE MEMORY AND METHOD WITH BLOCK MANAGEMENT SYSTEM”, by Peter Smith, Alan Bennett, Alan Bryce, Sergei Gorobets, and Alan Sinclair with which it is filed concurrently and which is hereby incorporated by reference:
This invention relates generally to semiconductor non-volatile data storage systems, and more specifically, to a system and method for forming physical blocks into larger logical structures that accommodate defects in non-volatile data storage systems.
Nonvolatile memory devices such as flash memories are commonly used as mass data storage subsystems. Such nonvolatile memory devices are typically packaged in an enclosed card that is removably connected with a host system, and can also be packaged as the non-removable embedded storage within a host system. In a typical implementation, the subsystem includes one or more flash devices and often a subsystem controller.
Current commercial memory card formats include that of the Personal Computer Memory Card International Association (PCMCIA), CompactFlash (CF), MultiMediaCard (MMC), Secure Digital (SD), MemoryStick, and MemoryStick-Pro. One supplier of these cards is SanDisk Corporation, assignee of this application. Host systems with which such cards are used include digital cameras, cellular phones, personal computers, notebook computers, hand held computing devices, audio reproducing devices, and the like.
The nonvolatile memory devices themselves are composed of one or more arrays of nonvolatile storage elements. Each storage element is capable of storing one or more bits of data. One important characteristic of the nonvolatile memory array is that it retains the data programmed therein, even when power is no longer applied to the memory array. In contrast, a volatile memory device requires that the power to the array be refreshed periodically in order to preserve the data contained in the volatile memory array. Another characteristic of nonvolatile memory is that once a cell contained within a nonvolatile memory array is programmed, that cell must be erased before it can be reprogrammed with a new data value.
The physical means for storing the charge in the memory cell can be implemented by using a floating gate transistor, such as an electrically erasable programmable read only memory (EEPROM). One known problem with floating gate devices such as EEPROMs is that the floating gate eventually wears out and breaks down after a very large number of write, program and erase cycles. When this happens, the cell is no longer usable and must be taken out of the list of available memory cells in the array. This sort of defect is called a “grown” defect. In one commercially available implementation, these defects are dealt with by mapping out the defective cells and substituting the physical addresses of good memory cells for the newly detected defective memory cells. Examples of implementations where defective cells or sectors are mapped out and replaced are described in U.S. Pat. No. 5,659,550 issued on Aug. 19, 1997 by Mehrotra et al.; U.S. Pat. No. 5,671,229 issued on Sep. 23, 1997 by Harari et al.; and in U.S. Pat. No. 5,862,080 issued on Jan. 19, 1999 by Harari et al., which applications are expressly incorporated herein in their entirety by this reference.
Arrays of nonvolatile memory cells typically are partitioned into groups to provide for efficient implementation of read, program and erase functions. For example, in many nonvolatile memory architectures, the memory cells are arranged into a larger group called a unit of erase. This unit of erase is the smallest number of memory cells that are erasable at one time.
The size of the unit of erase depends on the memory architecture that is being implemented. In earlier nonvolatile memories, the unit of erase was a block that was the same size as a standard 512-byte disk drive sector. In one commercial form, each block contained enough cells to store one sector of user data plus some overhead data related to the user data and/or to the block in which it was stored. In order to ensure that the blocks of cells were individually erasable, the blocks had to be sufficiently isolated from one another.
Because this isolation took up valuable space on the integrated circuit chip, another memory architecture was developed in which the unit of erase was made significantly larger so there would be less space required for such isolation. An example of this large block system architecture is described in U.S. Pat. No. 6,580,638 issued on Jun. 17, 2003 by Conley et al., which is a continuation of U.S. Pat. No. 6,426,893 issued on Jul. 30, 2002. Both of these patents are expressly incorporated herein in their entirety by this reference. In a large block system, the unit of erase is often further partitioned into individually addressable pages that are the basic unit for reading and programming user data (unit of programming and/or reading). In one commercial implementation, the unit of erase is a metablock. A metablock is a virtual unit of erase that is composed of multiple physical units of erase. These multiple physical units of erase can be used in parallel program and erase operations, but are addressed as a single logical block.
One method of forming metablocks, or “super” blocks is described in U.S. Pat. No. 6,034,897, which is expressly incorporated herein in its entirety by this reference. As described therein, for a memory having a number of devices, the same physical block in each of the devices is grouped into a metablock. Although this allows for the formation of metablocks and all of the blocks in a given metablock to be addressed by the same address, namely the address of the block in the first of the devices, it has a number of limitations. For example, as the linking of blocks into metablocks is pre-determined in this fixed configuration, when a block goes bad, the metablock to which it belongs becomes bad despite the other blocks within it still being functional. Thus, it is desirable goal to provide a system and method for an adaptable metablock arrangement without the operational overhead of establishing a new linking every time one is needed.
The various aspects of the present invention overcome these and other limitations found in the prior art for the linking of physical memory structures into composite logical structures. In the primary embodiment of the present invention, techniques are presented for the formation of metablock structure in flash memories. According to a first aspect of the present invention, the linking of block into metablocks is updated to accommodate defective portions of the memory. A system for maintaining a defect map includes a defect map structure for tracking the existence of factory and grown defects in a nonvolatile memory system that is organized into metablocks. A method for maintaining a defect map makes use of the defect map structure and the characteristics of a metablock. In one embodiment, the size of the defect map structure is equal to the parallelism of the nonvolatile memory system. In another embodiment, the remaining units of erase that cannot be assigned to a metablock will be kept in a list of spare units of erase for later use.
According to another aspect of the present invention, a record of the linking of blocks into metablocks is maintained in the non-volatile memory. The controller can then read out the linking into its volatile memory as needed. The record is updated in response to re-linkings resulting from defects. A complete linking record may be maintained in the non-volatile memory, for example in a specified location, or only a partial linking record may be maintained. In another aspect of the present invention, an initial linking of blocks into metablocks is deterministically formed by an algorithm using, for example, a firmware based implementation. In this case, the linking record stored in the non-volatile memory need only contain the deviations from the standard algorithm.
Additional aspects, features and advantages of the present invention are included in the following description of exemplary embodiments, which description should be read in conjunction with the accompanying drawings.
Generally, the memory system of the present invention maintains in its main non-volatile memory a record of the linking of blocks into multi-block metablock structures. This record contains a defect map of factory and grown defects in a map table that can be read into non-volatile memory on demand. This map is updated as grown defects are encountered and blocks are assigned to new locations. Metablocks are preferable formed by one block in each of several subarrays, or planes, and in this case, the grouping is done preferably with same-numbered blocks in each plane, with the exception of blocks that are marked bad and placed in the map. In one embodiment, the blocks are directly mapped to alternate blocks. Metablocks that would contain the defective block are then reformed with the alternate block in the corresponding plane. Rather than maintain a complete record of the linking of blocks into metablocks within the non-volatile memory, a “standard” linking can be based on an algorithm implanted, for example, in the system's firmware, with only the deviations due to defects from this algorithm needing to be stored. Those standard metablocks with no defects present will be termed usable standard metablocks. The remaining Metablocks are termed unusable standard metablocks, and the component physical blocks will be termed spare blocks.
Typically, the nonvolatile memory array includes memory cells that include one or more conductive floating gates as storage elements. The memory arrays could include other long-term electron charge storage elements instead, for example a charge trapping dielectric. A variety of nonvolatile storage elements can be used, including nonvolatile multi-state storage elements, storage elements that are Flash/EEPROM/floating gate, nonvolatile storage elements such as NROM, MNOS, MRAM/FRAM, NAND, NOR, and non-volatile storage elements that are organized into units of erasure. Non-volatile storage elements that store digital information, analog information or combinations of both can be used.
The memory cell array can be operated with two levels of charge defined for each storage element to therefore store one bit of data with each element. Alternatively, more than two storage states can be defined for each storage element, in which case more than one bit of data is stored in each storage element.
Data from the host is typically just provided to the memory system identified as logical sectors. For example, the host will send data described in terms of a starting logical sector and the total number of sectors. Within the host, these logical sectors may be structured into larger units, but this information is typically not passed on to the memory's controller. (Knowledge of these host structures can be used, though, to have physical structures in the memory or controller constructs that reflect these host structures as they will reflect how the host transfers data: for example, memories are typically structured to store user data based on a physical sector that is the same size as a logical sector.) The controller organizes the host structures into logical constructs that mimic the physical structure for more efficient storage of them in the physical memory. For example, in a common arrangement the memory system controller groups logical sectors (as defined by the host) into logical blocks that correspond in size to the physical structure of a block, which is the physical unit of erase for a flash type memory. The controller can then maintain the relation of the physical structures to the logical constructs, for instance in the form of a Sector Address Table (SAT), and update this correspondence as the logical to physical relation changes.
For greater efficiency, memory systems often increase parallelism by introducing larger structures. For instance, in a memory system having a number of semi-autonomous arrays that can be written, read, or both, concurrently, blocks from different arrays are grouped into “metablock” structures and the controller will form logical blocks of data into corresponding logical metablocks, allowing for the multiple blocks to be operated upon at the same time. A metablock can be formed from planes with a single memory chip or one or more planes distributed across several memory chips. In this way, by allowing a group of (logically) sequential sectors spans over multiple physical blocks that can be read/programmed/erased in parallel, then this group can be read/programmed/erased concurrently with a minimal number of non-concurrent read, program, or erase operations. This arrangement is beneficial for sequential read and writes of logically contiguous sectors of relatively large amounts of data.
The structure of
Each plane typically has its own data registers and programming circuits. This allows for simultaneous programming into the blocks of memory cells of each of the planes. Each plane is operable semi-independently of the other planes. For convenience of illustration, the planes depicted in
In systems where there are a large number of blocks, the planes can be logically divided into zones as shown in
The unit of operation used by the memory system as applied to the scenarios in
A metablock extends across multiple planes and typically includes one block from each of these planes, as shown by the arrows in each of the sets of connected arrows shown in
The more general situation for the linking of blocks into metablocks is shown in
Several metablock linkings are shown, where each metablock MB is numbered according to the first row of the first block in the linking; for example, the metablock starting with block 0 is labeled MB0, and so on. Although
As shown in
The first metablock as shown in
Alternatively, a metablock could include only a subset of the number of planes contained within a single zone. This example is not shown but can be useful if variable degrees of parallelism are desired. A system having variable degrees of parallelism could allow for blocks to move between areas having different parallelism. For example, a metablock that contained two blocks instead of four could move from an area where the parallelism is two (two planes) to an area where the parallelism is four (as in our four-plane example). If two metablocks each containing two members exist in such a system, they could be combined into one four-member metablock. One caveat to using such a mixed-parallelism system is that the defect map would have to be reconciled to account for the varying parallelism.
In general, using metablocks increases the parallelism of the memory system by allowing for more blocks to be processed at one time in parallel. This increased parallelism significantly increases the performance of the memory system. The memory system maintains the identity of the individual blocks within the various metablocks. For example, the identity of the individual blocks within the various metablocks can be maintained as a linked list in any convenient location within the system.
Usually, overhead data stored along with user data of each page will include an address, logical and/or physical, sufficient to identify the plane, zone and block in which the pages reside, as well as an offset of the page within that block. An address map is then created within a memory of the controller by the controller reading these address fields of the page overhead data. This is usually done for part of the memory at a time, preceding a programming, reading or erasing operation directed to that part of the memory.
The first figure,
In one embodiment of the invention, a feature of the metablock is that the first virtual block address (associated with the first physical block address) is the only information required to determine the locations of the remaining physical blocks associated with that particular metablock, the rest of the linked metablocks being determined by a linking algorithm. This reduces the size of the data required for storing the metablocks. If a defect is encountered in the first physical block, then the entire metablock becomes invalid and gets mapped out of the address space because there is no longer any mechanism for pointing to the remaining members of the metablock located across each of the available planes or planes. An example of this will be shown in the discussion of
(Under the naming convention being used, there is no metablock MB1 at the point of
Note that in
The mapping table 1101 is one example of a record of the linking of blocks to form metablocks. One technique that allows for adaptability of the metablock composition is, whenever a metablock is needed for a write process, to dynamically form a metablock from a pool of available good blocks. Although this technique allows for adaptability for the metablock structure, it then requires that the linking be re-established whenever a block is needed; further, it requires that each plane be accessed in order to determine this pool. According to one of the primary aspects of the present invention, this record is stored at a location in the non-volatile memory. This record can then be read from the flash memory into volatile memory (or an alternate non-volatile memory) on the controller on demand for use in address translation.
In one set of embodiments, the record of the linking information can be maintained in a special non-volatile memory location, such as somewhere outside the normal physical blocks where user data is stored. For instance, this record can be maintained in one of the reserved areas in plane 0 of
By use of such a table, the linking of blocks into metablocks can be updated to accommodate defects and, by keeping a record in the non-volatile memory, it is readily accessible when needed, thereby overcoming the limitation of the prior art discussed in the Background.
The defect map aspect of the present invention is illustrated in
As noted above, one of the principle aspects of the present invention is that the controller maintains a record of the linking table in the non-volatile memory. This record may be a complete listing of the blocks forming each metablock corresponding to a given physical address, such as logical-to-physical mapping table 1101 of
The described mechanisms employed for linking physical blocks into metablocks are designed to maximize the number of metablocks for any distribution of bad physical blocks, and to allow new metablocks to be created when physical blocks fail during the life of the system. As above, metablock number is defined from the address of the physical block in the first plane incorporated in the metablock.
A “standard” metablock comprises physical blocks whose physical block addresses are a deterministic function of the metablock number. For example, this could be the arrangement of metablock MB0 in
In another example, as in the arrangement of metablock MB2 in
The physical blocks are linked into metablocks according to rules that maximize the possible number of standard metablocks. When a metablock is being linked, a standard block is created if the required physical blocks are available. If any physical block required for formation of a standard metablock is bad, the metablock number is classified as unused. If any physical block required for formation of a standard metablock is bad, the metablock number is classified as unused. In this case, other usable physical blocks which are consequently left unlinked are classified as spare, as shown for metablock n+1 in
Another aspect of the present invention is the optimization of the standard linking pattern based on the pattern of defects on the card or to otherwise maximize the number of usable standard metablocks. When a card is being formatted (or re-formatted), the pattern of bad blocks can be scanned and the decision about the ‘standard’ linking pattern can be made so that they match each other as much as possible. In many cases, the matching of the algorithm for the standard linking to the bad block pattern will reduce the needed re-linked metablock and meta- to physical address translation overhead, thereby optimizing (at least initially) the linking. An example of this is shown in
The standard block pattern can be unique for every card, for every re-formatting, or both. This can provide an extra level of data security, where every card has a unique method of data scrambling across different blocks. Consequently, standard metablocks need not necessarily have the same configuration, although they will have been formed using the same rule. In another embodiment, the host may want to provide the rule and code for the metablock linking rule, so that the algorithm is not system specific.
When metablocks are being linked during an initial formatting process, some physical blocks near the beginning of block address space, or near chip or die boundaries, cannot be linked into standard metablocks, as shown in
Metablocks which cannot exist as standard blocks may be structured as re-linked metablocks. Re-linked metablocks may be formed during the initial formatting process of the memory system, or in response to failure of a physical block within an existing metablock. Metablock n+4 in
In a preferred embodiment, it is preferable to re-link blocks only within the same die or chip, as a re-linking running across to another die or chip may trigger additional die/chip reads or programs. For example, in an exemplary case of a 4-plane meta-block, where the meta-block spans over 2 chips with 2 planes each, the data can be programmed to (or read from) all 4 planes, which requires 2 chip read/program in parallel. (This may not be 100% parallel but pipelined, so that the page read, block program and erase can be done in parallel, but the data bus being shared so that data transfers are not parallel). In this case, if a metablock is re-linked so that a physical block is taken from another, third chip, than it may be not possible to access it concurrently with the other two and the operation time can double.
If a bad physical block in plane 0 is substituted by a spare block from plane 0, the metablock number is defined by that of the spare block, as shown for metablock n+1 in
As noted above with respect to
In the embodiment where only the deviations from the standard linking are maintained in a record in the non-volatile memory, the management data structures for metablock linking can be contained within two types of sectors in one or more dedicated block linkage management blocks in flash memory. This arrangement has the advantage, with respect to saving the complete linking, of requiring the storage of less information. Metablock numbers of every re-linked metablock in the memory system are contained in a link table in a set of link table sectors. No entries for standard metablocks need be stored in the link table, because their physical block addresses are generated deterministically from the metablock number. An RF (re-linked) flag can be used as an attribute of a meta-block, indicating whether the block is standard, and all the blocks are linked in a standard fashion, or whether it is re-linked and requires additional block address translation through the link table (LT). The RF flag can be determined by analyzing the LT and then stored in SRAM, or else stored in a control data structure in the main non-volatile memory as an attribute for the meta-block. Alternately, a list of re-linked sectors can be keep in a list that can be searched through to determine which meta-blocks are re-linked. Entries in the link table can be ordered by metablock number, and each entry has fields for metablock number and block address of each linked physical block.
In an extension of the link table structure, the deviations from the basic algorithm establishing the standard linking can themselves be based on a set of rules; for example, the non-standard linking could be limited to a single plane or limiting how far from its original row a block can be re-linked. Only when the re-linking is an exception to all of the algorithms need it be explicitly kept in the non-volatile memory.
A particular example of a link table and how the information it contains can be compressed is described with respect to
Of the metablocks shown in
The compression algorithm used for block re-linking shown in
The full set of entries in the link table may be spread across multiple link table sectors. A link table sector is created or modified by writing it to the next available sector position in the block linkage management block. Multiple copies of a link table sector may therefore exist, with only the latest version being valid. An index to the positions of valid link table sectors in the block linkage management block can be contained in a field in the spare physical block list sector in the block linkage management block, which is described below. The metablock number relating to the first entry in each valid link table sector can be held in a list in controller RAM, to allow easy identification of the relevant link table sector for any metablock number. If an additional entry must be added to a link table sector which already contains the maximum number of entries, the sector is split into two new sectors, each with half the entries, and the new entry is inserted in the appropriate new sector.
The block addresses of all spare physical blocks are held as entries in an unordered list in the spare physical block list sector. In some embodiments, it may be convenient to limit the maximum number of spare blocks per plane per chip. In an exemplary embodiment, this is limited to 16, as in most cases this should be large enough. If there are more than 16 spare blocks in a plane, they will be “lost”.
Blocks that become spare, because either a standard metablock cannot be created during initial configuration of the flash memory into metablocks or because a physical block linked into a standard or re-linked metablock has failed, are added as new entries to the end of the spare physical block list. When a spare block is required for creation of a re-linked metablock, the first block in the relevant plane closest to the beginning of the spare physical block list can be used. The spare physical block list sector also contains a field with an index to valid link table sector locations.
When a metablock re-link operation takes place and link table sectors are created or modified, the spare physical block list and link table sector index fields can both be modified in the spare physical block list sector. A valid spare physical block list sector is therefore the last written sector in the block linkage management block. When the block linkage management block is full, it can be compacted during a control write operation by rewriting all valid sectors to a new block location, and the full block can be subsequently erased.
To simplify the saving of spare physical block sectors and limit the amount of searching that is required when looking for a replacement block the spare physical block list can be stored by chip, die, and plane. The actual number of sectors required depends on the size of memory device, number of devices, and the interleaving method used. As noted above, it may be convenient to store only a certain number of physical blocks per plane. The number held will depend upon the expected failure rate of blocks and the number of blocks per plane. This means that each spare physical block list sector will hold the spare physical blocks for several planes. The firmware will calculate the correct sector to load and then index into the sector to load the correct values.
When the number of spare physical block lists changes due to re-linking or another bad block being found, a new spare physical block list sector will be written. This may mean two spare physical block list sectors covering the same physical area. The Boot Sector holds the spare physical block list Indices containing the index of the valid sector. If no spare physical blocks exist in an area covered by a spare physical block list, then the sector will not exist in the Boot Block.
Referring back to
Although the various aspects of the present invention have been described in terms of a firmware based implementation, they may also be implemented in a hardware based implementation. A hardware engine on the controller or, preferably, on a memory chip itself can do the re-linking. The re-linking information can be stored on the chip itself at a special memory location. The controller can then treat all the blocks as standard, and the engine will translate the block addresses to the true internal addresses. The memory will appear to the host (or controller) as a memory with only standard blocks, or as a memory with large physical blocks, without having to know about either the smaller block structures or the details of any re-linking.
Although specific examples of various aspects of the present invention have been described, it is understood that the present invention is entitled to protection within the scope of the appended claims.