Abstract: Non-volatile memory program failure recovery via redundant arrays enables higher programming bandwidth and/or reduced latency in some storage subsystem implementations, e.g. a solid-state disk. Data to program N portions of a plurality of non-volatile memories is received at a non-volatile memory controller. The data includes particular data to program a particular one of the N portions. The particular data is stored in an allocated buffer associated with the non-volatile memory controller. Programming the particular data to a particular one of the non-volatile memories is begun. Redundancy information sufficient to recover from failures of M of the N portions is updated. The allocated buffer is freed. At least one of the storing, the beginning programming, the updating, and the freeing is in response to the receiving of the particular data. The freeing is prior to the particular non-volatile memory completing the programming.

Abstract: An I/O device is coupled to a computing host. In some embodiments, the device is enabled to utilize memory of the computing host not directly coupled to the device to store information such as a shadow copy of a map of the device and/or state of the device. Storage of the shadow copy of the map enables one or both of the device and the computing host to utilize the shadow copy of the map, such as to decrease read latency. Storage of the state enables the device to save volatile state that would otherwise be lost when the device enters a low-power state. In some embodiments, the device implements one or more non-standard modifiers of standard commands. The non-standard modifiers modify the execution of the standard commands, providing features not present in a host protocol having only the standard commands.

Abstract: An apparatus includes a non-volatile memory and a controller. The controller is coupled to the non-volatile memory and configured to (i) measure a rate of free space consumption in the non-volatile memory, (ii) measure a rate of free space production in the non-volatile memory, and (iii) adjust a rate of a recycling process in response to the measured rate of free space consumption and the measured rate of free space production.

Abstract: An apparatus having an interface and a circuit is shown. The interface is coupled to a memory that is nonvolatile. The circuit is configured to (i) read a plurality of codewords from a block in the memory based on a program/erase count associated with the block, (ii) count a number of iterations used to decode the codewords and (iii) decrease a code rate of an error correction coding used to program the block in response to the number of iterations exceeding a threshold.

Abstract: An apparatus having a circuit and one or more processor is disclosed. The circuit is configured to receive a codeword from a memory. The memory is nonvolatile. The codeword generally has one or more errors. The processors are configured to generate read data by decoding the codeword repeatedly. The decoding includes a soft-decision decoding that uses a plurality of parameters calculated by (i) a first procedure, (ii) a second procedure in response to a plurality of failures of the decoding to converge using the first procedure and (iii) a third procedure in response to another failure of the decoding to converge using the second procedure.

Abstract: An apparatus includes a non-volatile memory and a controller. The controller is operatively coupled to the non-volatile memory and configured to perform read and write operations on the non-volatile memory using codewords as a unit of read access. The controller includes an error correction engine configured to perform an error correction on codewords read from the non-volatile memory, and, if the error correction fails, to perform one or more retry procedures. The controller is further configured to perform one or more background procedures as a result of the error correction or one or more of the retry procedures not being successful and send an error message as a result of all of the retry procedures not being successful. The one or more background procedures are directed to determining a cause of the error correction failure.

Abstract: Described embodiments provide a media controller to read data stored in a media. The media controller determines a value for each bit of a shortened codeword from the media. The shortened codeword includes a plurality of non-shortened bits of a full codeword, where the full codeword includes the plurality of non-shortened bits and one or more shortened bits. Shortened bits correspond to bits unused in the shortened codeword. The media controller converts the determined values for each bit of the shortened codeword into a first set of log-likelihood ratio (LLR) values. The full codeword is decoded using the first set of LLR values for the shortened codeword. The media controller dampens one or more LLR values corresponding to non-shortened bits of the codeword to produce a second set of LLR values and decodes the second set of LLR values.

Abstract: A Sharing Data Fabric (SDF) causes flash memory attached to multiple compute nodes to appear to be a single large memory space that is global yet shared by many applications running on the many compute nodes. Flash objects stored in flash memory of a home node are copied to an object cache in DRAM at an action node by SDF threads executing on the nodes. The home node has a flash object map locating flash objects in the home node's flash memory, and a global cache directory that locates copies of the object in other sharing nodes. Application programs use an applications-programming interface (API) into the SDF to transparently get and put objects without regard to the object's location on any of the many compute nodes. SDF threads and tables control coherency of objects in flash and DRAM.

Abstract: Described embodiments track a read disturb limit of a solid-state media coupled to a media controller. The media controller receives a read operation from a host device. In response to the received read operation, the media controller determines one or more associated regions of the solid-state media accessed by the read operation and reads the associated regions to provide read data to the host device. Based on a probability value corresponding to each of the associated regions, the media controller selectively increments a read count of each of the associated regions. Based upon each read count, the media controller determines whether each region has reached a read disturb limit. If a given region has reached the read disturb limit, the media controller relocates data of the given region to a free region of the solid-state media. Otherwise, the media controller maintains the data in the given region.

Abstract: Adaptive ECC techniques for use with flash memory enable improvements in flash memory lifetime, reliability, performance, and/or storage capacity. The techniques include a set of ECC schemes with various code rates and/or various code lengths (providing different error correcting capabilities), and error statistic collecting/tracking (such as via a dedicated hardware logic block). The techniques further include encoding/decoding in accordance with one or more of the ECC schemes, and dynamically switching encoding/decoding amongst one or more of the ECC schemes based at least in part on information from the error statistic collecting/tracking (such as via a hardware logic adaptive codec receiving inputs from the dedicated error statistic collecting/tracking hardware logic block). The techniques further include selectively operating a portion (e.g., page, block) of the flash memory in various operating modes (e.g. as an MLC page or an SLC page) over time.

Abstract: An apparatus comprising a non-volatile memory and a controller. The controller is coupled to the non-volatile memory and configured to (i) accumulate a read disturb count for a first region of the non-volatile memory, (ii) accumulate error statistics for a second region of the non-volatile memory, (iii) determine, based upon both the read disturb count and the error statistics, whether the first region has reached a read disturb limit, and (iv) in response to determining that the first region has reached the read disturb limit, relocate at least some data of the first region.

Abstract: An apparatus includes a first circuit and a second circuit. The first circuit may be configured to generate a reduced representation of an input sequence of characters by replacing a repetition of a sequence of one or more characters by a code representing the repetition of the sequence of one or more characters. The second circuit may be configured to generate a compressed representation of the input sequence of characters in response to the reduced representation of the input sequence of characters. The second circuit is generally configured to recognize the code representing the repetition of the sequence of one or more characters and take into account the repetition of the sequence of one or more characters during a compression operation.

Abstract: An apparatus includes a first circuit and a second circuit. The first circuit may be configured to generate (i) a compressed representation of an input sequence of characters and (ii) statistics regarding one or more types of information in the compressed representation of the input sequence of characters. The second circuit may be configured to generate a compressed bitstream representation of the input sequence of characters in response to the compressed representation of the input sequence of characters and one or more of the statistics regarding the one or more types of information in the compressed representation of the input sequence of characters. The bitstream values encoding the one or more types of information are generally determined based upon a frequency of occurrence of the one or more types of information in the compressed representation of the input sequence of characters.

Abstract: Scalable control/management data structures enable optimizing performance and/or attempting to achieve a particular performance target of an SSD in accordance with host interfacing, number of NVM devices, NVM characteristics and size, and NVM aging and performance decline. Pre-scaled data structures are included in SSD controller firmware loadable at system initialization. Static data structure configurations enable load-once-operate-for-product-lifetime operation for consumer applications. Dynamic configurations provide sequences of data structures pre-scaled to optimize operation as NVM ages and performance declines. Pre-configured adjustments in data structure size included in consecutive configurations periodically replace earlier configurations at least one time during product lifetime, producing a periodic rescaling of data structure size to track changes in aging NVM.

Abstract: A non-volatile storage system having Non-Volatile Memory (NVM) provides self-journaling and hierarchical consistency, enabling low-latency recovery and force unit access handshake. Mappings between host addresses and addresses in the NVM are maintained via one or more map entries, enabling locating of host data written to the NVM. Objects stored in the NVM include sufficient information to recover the object solely within the object itself. The NVM is managed as one or more data streams, a map stream, and a checkpoint stream. Host data is written to the data streams, map entries are written to the map stream, and checkpoints of map entries and other data structures are written to the checkpoint stream. Time markers embedded in the streams enable determination, during recovery, that selected portions of the streams are inconsistent with each other and are to be discarded.

Abstract: An apparatus having a first circuit and a second circuit is disclosed. The first circuit may be configured to (i) read data from a region of a memory circuit during a read scrub of the region and (ii) generate a plurality of statistics based on (a) the data and (b) one or more bit flips performed during an error correction of the data. The memory circuit is generally configured to store the data in a nonvolatile condition. One or more reference voltages may be used to read the data. The second circuit may be configured to (i) update a plurality of parameters of the region based on the statistics and (ii) compute updated values of the reference voltages based on the parameters.

Abstract: Described embodiments provide a lookup engine that receives lookup requests including a requested key and a speculative add requestor. Iteratively, for each one of the lookup requests, the lookup engine searches each entry of a lookup table for an entry having a key matching the requested key of the lookup request. If the lookup table does not include an entry having a key matching the requested key, the lookup engine sends a miss indication corresponding to the lookup request to the control processor. If the speculative add requestor is set, the lookup engine speculatively adds the requested key to a free entry in the lookup table. Speculatively added keys are searchable in the lookup table for subsequent lookup requests to maintain coherency of the lookup table without creating duplicate key entries, comparing missed keys with each other or stalling the lookup engine to insert missed keys.

Abstract: The disclosure is directed to protecting data of a scalable storage system. A scalable storage system includes a plurality of nodes, each of the nodes having directly-attached storage (DAS), such as one or more hard-disk drives and/or solid-state disk drives. The nodes are coupled via an inter-node communication network, and a substantial entirety of the DAS is globally accessible by each of the nodes. The DAS is protected utilizing intra-node protection to keep data stored in the DAS reliable and globally accessible in presence of a failure within one of the nodes. The DAS is further protected utilizing inter-node protection to keep data stored in the DAS reliable and globally accessible if at least one of the nodes fails.

Abstract: A data processing system includes one or more nodes, each node including a memory sub-system. The sub-system includes a fine-grained, memory, and a less-fine-grained (e.g., page-based) memory. The fine-grained memory optionally serves as a cache and/or as a write buffer for the page-based memory. Software executing on the system uses a node address space which enables access to the page-based memories of all nodes. Each node optionally provides ACID memory properties for at least a portion of the space. In at least a portion of the space, memory elements are mapped to locations in the page-based memory. In various embodiments, some of the elements are compressed, the compressed elements are packed into pages, the pages are written into available locations in the page-based memory, and a map maintains an association between the some of the elements and the locations.

Abstract: A distributed database system has multiple compute nodes each running an instance of a database management system (DBMS) program that accesses database records in a local buffer cache. Records are persistently stored in distributed flash memory on multiple storage nodes. A Sharing Data Fabric (SDF) is a middleware layer between the DBMS programs and the storage nodes and has API functions called by the DBMS programs when a requested record is not present in the local buffer cache. The SDF fetches the requested record from flash memory and loads a copy into the local buffer cache. The SDF has threads on a home storage node that locate database records using a node map. A global cache directory locks and pins records to local buffer caches for updating by a node's DBMS program. DBMS operations are grouped into transactions that are committed or aborted together as a unit.