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: An SSD controller dynamically adjust read thresholds in a NVM to reduce errors due to device threshold voltage distribution shifts, thus improving performance, reliability, and/or cost of a storage sub-system, such as an SSD. In a first aspect, the controller periodically performs offline tracking on a portion of the NVM. The controller reads a representative sub-portion with current read thresholds. If the read meets a condition, then the controller reads the sub-portion with sample read thresholds, estimates the device threshold voltage distributions, and adjusts the current read thresholds of the portion to calculated new operating read thresholds of the sub-portion. In a second aspect, the portion includes data with a known statistical average number of zero and/or one bits.

Abstract: Storage address space to NVM address, span, and length mapping/converting is performed by a controller for a solid-state storage system that includes a mapping function to convert a logical block address from a host to an address of a smallest read unit of the NVM. The mapping function provides span and length information corresponding to the logical block address. The span information specifies a number of contiguous smallest read units to read to provide data (corresponding to the logical block address) to the host. The length information specifies how much of the contiguous smallest read units relate to the data provided to the host. The converted address and the length information are usable to improve recycling of no longer needed (e.g. released) portions of the NVM, and usable to facilitate recovery from outages and/or unintended interruptions of service.

Abstract: An apparatus having mapping and interface circuits. The mapping circuit (i) generates a coded item by mapping write unit bits using a modulation or recursion of past-seen bits, and (ii) calculates a particular state to program into a nonvolatile memory cell. The interface circuit programs the cell at the particular state. Two normal cell states are treated as at least four refined states. The particular state is one of the refined states. A mapping to the refined states mitigates programming write misplacement that shifts an analog voltage of the cell from the particular state to an erroneous state. The erroneous state corresponds to a readily observable illegal or atypical write sequence, and results in a modified soft decision from that calculated based on the normal states only. A voltage swing between the particular state and the erroneous state is less than between the normal states.

Abstract: A paired queue apparatus and method comprising request and response queues wherein queue head and tail pointer update values are communicated through an enhanced pointer word data format providing pointer indicator information and optional auxiliary information in a single transfer, wherein auxiliary information provides additional system communication without consuming additional bandwidth. Auxiliary information is optionally contained in a response data entry written to a response queue or in a request entry written to a request queue.

Abstract: A method for using a variable-size flash translation layer. The method includes reading an entry in a map based on a read logical block address in a read request to obtain both a physical address of a particular page in a memory and information regarding compressed data with a variable size; converting the information to both an address in the particular page and a number of read units in the memory that contain the compressed data; and reading the compressed data from at least the particular page in the memory based on the address and the number of read units.

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: A method for recovery after a power failure. The method generally includes a step of searching at least some of a plurality of pages of a memory to find a first erased page in response to an unsafe power down. A step may move stored data located between a particular word line in the memory that contains the first erased page and a previous word line that is at least two word lines before the particular word line. Another step may write new data starting in a subsequent word line that is the at least two word lines after the particular word line that contains the first erased page.

Abstract: An apparatus includes a memory and a controller. The memory may be configured to store data. The controller may be configured to process a plurality of read/write operations to/from the memory, receive a codeword from the memory, generate a plurality of syndromes of the codeword at a plurality of possible code rates, generate a plurality of count values by counting a number of unsatisfied parity checks in each of the plurality of syndromes, generate a plurality of normalized values by dividing the plurality of count values by a plurality of lengths of the plurality of possible code rates respectively, and determine a bit error rate value of the memory based on a lowest value among the plurality of normalized values.

Abstract: An apparatus having a memory and a controller is disclosed. The memory is configured to store a codeword. The controller is configured to (i) determine one or more least-reliable bit positions in a soft-decision version of the codeword in response to failing to decode a hard-decision version of the codeword, (ii) generate a trial codeword by selecting at random a respective value in one or more trial positions among the least-reliable bit positions in the hard-decision codeword and (iii) perform a hard-decision decoding of the trial codeword.

Abstract: An apparatus includes a non-volatile memory and a controller. The non-volatile memory includes a plurality of R-blocks. The controller is coupled to the non-volatile memory. The controller is configured to (i) write data using the R-blocks as a unit of allocation and (ii) perform recycling operations selectively on either an entire one of the R-blocks or a portion less than all of one of the R-blocks.

Abstract: An SSD controller maintains a zero count and a one count, and/or in some embodiments a zero/one disparity count, for each read unit read from an SLC NVM (or the lower pages of an MLC). In an event that the read unit is uncorrectable in part due to a shift in the threshold voltage distributions away from their nominal distributions, the maintained counts enable a determination of a direction and/or a magnitude to adjust a read threshold to trade the threshold voltage shift and restore the read data zero/one balance. In various embodiments, the adjusted read threshold is determined in a variety of described ways (counts, percentages) that are based on a number of described factors (determined threshold voltage distributions, known stored values, past NVM operating events). Extensions of the forgoing techniques are described for MLC memories.

Abstract: A method for using a variable-size flash translation layer is disclosed. Step (A) receives a read request to read data corresponding to a logical block address from a nonvolatile memory. Step (B) reads a particular entry of a map to obtain (i) a physical address of a particular page of the nonvolatile memory, (ii) an offset in the particular page to compressed data previously stored and (iii) a length of the compressed data. The particular entry is associated with the logical block address. Step (C) converts the offset and the length to (i) an address of a given read unit in the particular page and (ii) a number of the read units to be read. Step (D) reads from the particular page at most the number of the read units starting from the given read unit. An offset and length granularity are finer than one read unit.

Abstract: An apparatus includes one or more error-correction decoders, a buffer, and at least one processor. The buffer may be configured to store data to be decoded by the one or more error-correction decoders. The at least one processor is generally enabled to send messages to the one or more error-correction decoders. The messages may contain datapath control information corresponding to data in the buffer to be decoded by the one or more error-correction decoders. The one or more error-correction decoders are generally enabled to decode the data read from the buffer according to the corresponding datapath control information.

Abstract: An apparatus having a device and a circuit is disclosed. The device is configured to convey a codeword. The circuit is configured to (i) receive the codeword from the device, (ii) generate a syndrome by performing a portion less than all of an iterative decoding procedure on the codeword, (iii) generate a value by counting a number of unsatisfied parity checks in the syndrome and (iv) generate a quality metric of the device according to the value.

Abstract: In an SSD controller reading from flash memory, subsequent to failure of an initial soft-decision decoding attempt based on a nominal LLR, soft-decision re-decoding attempts are made using compensated LLR soft-decision information sets, pre-calculated at respective read-equilibrium points corresponding to mean shifts and variance change in the actual charge-state distributions of the flash memory channel. According to embodiment, soft-decision re-decoding attempts are performed without a retry read, or overlapped with one or more retry reads. By overlapping re-decoding with one or more retry reads, the probability of successful decoding increases, the need for further retry reads diminishes, and throughput is improved. The LLR compensation becomes very effective over a large number of retry reads, improving decoding reliability and achieving close to optimal bit error rates, even in the presence of large channel variation.

Abstract: A method for data recovery after a power failure is disclosed. The method may include steps (A) to (D). Step (A) may determine that a last power-down of a solid-state drive was an unsafe power-down. Step (B) may search at least some of a plurality of pages of a nonvolatile memory of the solid-state drive to define an unsafe zone in response to the determining that the last power-down of the solid-state drive was the unsafe power-down. Step (C) may define a pad zone comprising one or more of the pages subsequent to the unsafe zone. Step (D) may resume operation of the solid-state drive by writing new data subsequent to the pad zone.

Abstract: An apparatus comprising a memory and a controller. The memory is configured to process a plurality of read/write operations. The memory comprises a plurality of memory unit granularities each having a size less than a total size of the memory. The controller is configured to process a plurality of I/O requests to the memory units of the memory that are not marked as bad on a memory unit list. The controller is configured to track a plurality of bad blocks of the memory. The controller is configured to determine which of the memory units to mark as bad based on a test of whether a unit of memory larger than a block of the memory has failed. The test is based on a threshold of the bad blocks in the unit of memory.

Abstract: Described embodiments detect an impending out-of-space (OOS) condition of a media. On startup, a media controller determines whether an impending OOS indicator is set from a previous startup. If the impending OOS indicator is not set, it is determined whether a free pool size has reached a threshold. The free pool is blocks of the solid-state media available to be written with data. If the free pool size has not reached the first threshold, while the startup time is less than a maximum startup time, garbage collection is performed on the solid-state media to accumulate blocks to the free pool. If the startup time reaches the maximum startup time and the free pool size has not reached the threshold, the impending OOS indicator is set and the media is operated in impending OOS mode. Otherwise, if the free pool size reaches the threshold, the media is operated in normal mode.

Abstract: A paired queue apparatus and method comprising request and response queues wherein queue head and tail pointer update values are communicated through an enhanced pointer word data format providing pointer indicator information and optional auxiliary information in a single transfer, wherein auxiliary information provides additional system communication without consuming additional bandwidth. Auxiliary information is optionally contained in a response data entry written to a response queue or in a request entry written to a request queue.