摘要:
In a nonvolatile memory with block management system, critical data such as control data for the block management system is maintained in duplicates. Various methods are described for robustly writing and reading two copies of critical data in multi-state memory. In another aspect of the invention, a preemptive garbage collection on memory block containing control data avoids an undesirable situation where a large number of such memory blocks need be garbage collected at the same time.
摘要:
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.
摘要:
The present invention presents a number of improvements for managing erase processes in non-volatile memory. Such memory systems typically manage the memory by logically organize the basic unit of physical erase (erase block) into composite logical groupings (meta-blocks or logical group), where an erase block generally consists of a number of sectors. When an erase command is received, the specified sectors are checked against the memory system's control data. If the specified sectors span any full logical grouping, the full logical groupings can each be treated as a whole and erased according to one process (such as performing a true, physical erase), while other sectors are “logically” erased at the sector level by standard techniques.
摘要:
Methods and apparatus for writing data to non-volatile memory include maintaining one or more lists of obsolete blocks of the non-volatile memory and limiting the lists to a predetermined value. If a write operation would result in a list exceeding its predetermined value, a block erase operation is performed on one of the obsolete blocks. Valid data contained in an obsolete block selected for erasure is relocated prior to erasure.
摘要:
In a nonvolatile memory with block management system, critical data such as control data for the block management system is maintained in duplicates. Various methods are described for robustly writing and reading two copies of critical data in multi-state memory. In another aspect of the invention, a preemptive garbage collection on memory block containing control data avoids an undesirable situation where a large number of such memory blocks need be garbage collected at the same time.
摘要:
A portion of a nonvolatile memory is partitioned from a main multi-level memory array to operate as a cache. The cache memory is configured to store at less capacity per memory cell and finer granularity of write units compared to the main memory. In a block-oriented memory architecture, the cache has multiple functions, not merely to improve access speed, but is an integral part of a sequential update block system. Decisions to archive data from the cache memory to the main memory depend on the attributes of the data to be archived, the state of the blocks in the main memory portion and the state of the blocks in the cache portion.
摘要:
A portion of a nonvolatile memory is partitioned from a main multi-level memory array to operate as a cache. The cache memory is configured to store at less capacity per memory cell and finer granularity of write units compared to the main memory. In a block-oriented memory architecture, the cache has multiple functions, not merely to improve access speed, but is an integral part of a sequential update block system. The cache memory has a capacity dynamically increased by allocation of blocks from the main memory in response to a demand to increase the capacity. Preferably, a block with an endurance count higher than average is allocated. The logical addresses of data are partitioned into zones to limit the size of the indices for the cache.
摘要:
In a nonvolatile memory with block management system, critical data such as control data for the block management system is maintained in duplicates. Various methods are described for robustly writing and reading two copies of critical data in multi-state memory. In another aspect of the invention, a preemptive garbage collection on memory block containing control data avoids an undesirable situation where a large number of such memory blocks need be garbage collected at the same time.
摘要:
In a memory that is programmable page by page and each page having multiple sectors that are once-programmable, even if successive writes are sequential, the data recorded to an update block may be fragmented and non-sequential. Instead of recording update data to an update block, the data is being recorded in at least two interleaving streams. When a full page of data is available, it is recorded to the update block. Otherwise, it is temporarily recorded to the scratch pad block until a full page of data becomes available to be transferred to the update block. Preferably, a pipeline operation allows the recording to the update block to be set up as soon as the host write command indicates a full page could be written. If the actual write data is incomplete due to interruptions, the setup will be canceled and recording is made to the scratch pad block instead.
摘要:
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.