Abstract: A memory device includes an array of memory cells and a controller configured to access the array of memory cells. The controller is further configured to program a first number of bits to a first memory cell of the array of memory cells and program a second number of bits to a second memory cell of the array of memory cells. The controller is further configured to following a period after programming the second number of bits to the second memory cell, merge at least a subset of the first number of bits stored in the first memory cell to the second number of bits stored in the second memory cell without erasing the second memory cell such that the second number of bits plus at least the subset of the first number of bits are stored in the second memory cell.

Abstract: A method of performing automatic tuning on a deep learning model includes: utilizing an instruction-based learned cost model to estimate a first type of operational performance metrics based on a tuned configuration of layer fusion and tensor tiling; utilizing statistical data gathered during a compilation process of the deep learning model to determine a second type of operational performance metrics based on the tuned configuration of layer fusion and tensor tiling; performing an auto-tuning process to obtain a plurality of optimal configurations based on the first type of operational performance metrics and the second type of operational performance metrics; and configure the deep learning model according to one of the plurality of optimal configurations.

Abstract: Exemplary methods, apparatuses, and systems include an adaptive pre-read manager for controlling pre-reads of the memory device. The adaptive pre-read manager receives a first set of data bits for programming to memory. The adaptive pre-read manager performing a first pass of programming including a first subset of data bits from the set of data bits. The adaptive pre-read manager compares a set of threshold operating differences to a set of differences between multiple operating conditions during the first pass of programming and current operating conditions. The adaptive pre-read manager performs an internal pre-read of the programmed first subset of data bits. The adaptive pre-read manager performs a second pass of programming using the internal pre-read and a second subset of data bits from the first set of data bits.

Abstract: Exemplary methods, apparatuses, and systems including a programming manager for controlling writing data bits to a memory device. The programming manager receives a first set of data bits for programming to memory. The programming manager writes a first subset of data bits to a first wordline during a first pass of programming. The programming manager writes a second subset of data bits of the first set of data bits to a buffer. The programming manager receives a second set of data bits for programming. The programming manager writes the second subset of data bits of the first set of data bits to the first wordline during a second pass of programming to increase a bit density of memory cells in the first wordline in response to receiving the second set of data bits.

Abstract: Methods, apparatuses and systems related to maintaining stored data are described. The apparatus may be configured to refresh the stored data according to schedule that includes different delays between successive refresh operations.

Abstract: A system related to providing multi-layer code rates for special event protection with reduced performance penalty for memories is disclosed. Based on an impending stress event, extra error correction code data is utilized to encode user data obtained from a host. The user data and first error correction code data are written to a first block and the extra error correction code data is written to a second block. Upon stress event completion, pages having user data with the extra error correction code data are scanned. If pages of the first block are unable to satisfy reliability requirements, a touch-up process is executed on each page in the first block to reinstate the first block so that the extra error correction code data is no longer needed. The extra error correction code data is deleted from the second block and the second block is made available for user data.

Abstract: A memory device includes a memory array including wordlines and at least one string of cells. Each cell of the at least one string of cells is addressable by a respective wordline. The memory device further includes control logic, operatively coupled to the memory array, to perform operations including generating gate-induced drain leakage (GIDL) with respect to the at least one string of cells, and causing a grounding voltage to be applied to a set of wordlines to ground each cell of the at least one string of cells addressable by each wordline of the set of wordlines. The grounding voltage applied to the set of wordlines enables transport of positive charge carriers generated by the GIDL. In some embodiments, the positive charge carriers neutralize a buildup of negative charge carriers generated during a seeding phase of a program refresh operation.

Abstract: A memory device includes an array of memory cells and a controller configured to access the array of memory cells. The controller is further configured to program a first number of bits to a first memory cell of the array of memory cells and program a second number of bits to a second memory cell of the array of memory cells. The controller is further configured to following a period after programming the second number of bits to the second memory cell, merge at least a subset of the first number of bits stored in the first memory cell to the second number of bits stored in the second memory cell without erasing the second memory cell such that the second number of bits plus at least the subset of the first number of bits are stored in the second memory cell.

Abstract: Exemplary methods, apparatuses, and systems include a quick charge loss (QCL) mitigation manager for controlling writing data bits to a memory device. The QCL mitigation manager receives a first set of data bits for programming to memory. The QCL mitigation manager writes a first subset of data bits of the first set of data bits to a first memory block of the memory during a first pass of programming. The QCL mitigation manager writes a second subset of data bits of the first set of data bits to the first memory block during a second pass of programming in response to determining that the threshold delay is satisfied.

Abstract: A method includes receiving first data, determining a number of programming operations performed on a plurality of flash memory cells subsequent to a most recent erase operation performed on the plurality of flash memory cells, encoding the first data to provide a first write-once memory (WOM) encoded data, and storing the first WOM encoded data, based at least in part on the determined number of programming operations, within a number of the plurality of flash memory cells.

Abstract: A method includes causing a first set of memory cells, associated with a first wordline of a memory array, to be programmed with a first set of threshold voltage distributions; causing a second set of memory cells, associated with a second wordline adjacent to the first wordline, to be programmed with a second set of threshold voltage distributions; after programming the second set of cells, causing the first set of memory cells to be coarse programmed with an intermediate third set of threshold voltage distributions that is at least twice in number compared to the first set; and causing the first set of memory cells to be fine programmed with a final third set of threshold voltage distributions. At least some threshold voltage distributions of the final third set of threshold voltage distributions have wider read window margins than those of the intermediate third set of threshold voltage distributions.

Abstract: A read is initiated with respect to a target cell. A pair of adjacent cells includes a first cell and a second cell each adjacent to the target cell. First cell state information is obtained for the first cell and second cell state information is obtained for the second cell. A state information bin is determined by applying a pre-defined operation to the first cell state information and the second cell state information of the respective pair of adjacent cells. The target cell is assigned to the state information bin. Each state information bin defines a read level offset for reading the target cell.

Abstract: A method includes causing a read operation to be initiated with respect to a set of target cells. For each target cell, a respective group of adjacent cells is adjacent to the target cell. The method further includes obtaining, for each group of adjacent cells, respective cell state information, assigning, based on the cell state information, each target cell of the set of target cells to a respective state information bin, and determining a set of calibrated read level offsets. Each state information bin is associated with a respective group of target cells of the set of target cells, and each calibrated read level offset of the set of calibrated read level offsets is associated with a respective state information bin of the set of state information bins.

Abstract: The storage device includes a non-volatile memory with control circuitry and an array of memory cells that are arranged in a plurality of word lines. The control circuitry is configured to program the memory cells in a plurality of programming loops which include applying a programming pulse to a selected word line to program at least one memory cell of the selected word line to a programmed data state. The programming loops also include simultaneously applying a verify pulse to the selected word line to verify a data state being programmed, applying a first voltage to at least one unselected word line that has not been programmed, and applying a second voltage to at least one unselected word line that has already been programmed. The first voltage is determined as a function of the programmed data state to reduce a voltage threshold distribution across the memory cells.

Abstract: A three-dimensional memory device includes an alternating stack of insulating layers and electrically conductive layers, first memory opening fill structures extending through the alternating stack and including a respective first vertical semiconductor channel having a tubular section and a semi-tubular section, second memory opening fill structures, first bit lines electrically connected to a respective subset of the first drain regions, second bit lines electrically connected to a respective subset of the second drain regions, and an erase voltage application circuit configured to electrically bias the first bit lines at a first bit line erase voltage and the second bit lines at a second bit line erase voltage during an erase operation. The first bit line erase voltage is greater than the second bit line erase voltage.

Abstract: A method for programming a memory block of a non-volatile memory structure, wherein the method provides, during a program verify operation, selecting only a partial segment of memory cells of a memory block for bit scan mode, applying a sensing bias voltage to one or more bit lines of the memory block associated with the selected memory cells, and initiating a bit scan mode of the selected memory cells.

Abstract: A memory apparatus and method of operation is provided. The apparatus includes a block having memory cells connected to word lines and arranged in strings and is divided into a first sub-block and a second sub-block each configured to be erased as a whole in an erase operation. The apparatus has a temperature measuring circuit configured to detect an ambient temperature of the apparatus. A control circuit is configured to determine a word line inhibit voltage based on the ambient temperature. The control circuit applies an erase voltage to each of the strings while simultaneously applying a word line erase voltage to the word lines associated with a selected one of the first and second sub-blocks to encourage erasing and the word line inhibit voltage to the word lines associated with an unselected one of the first and second sub-blocks to discourage erasing in the erase operation.

Abstract: Technology is disclosed for an efficient read NAND memory cells while mitigating read disturb. In an aspect, a read sequence includes a read spike that removes residual electrons from the NAND channels, followed by reading multiple different groups of memory cells, followed by a channel clean operation. The read spike and channel clean mitigate read disturb. The read spike and channel clean each take a significant amount of time to perform. However, since multiple groups of memory cells are read between the read spike and channel clean this time is essentially spread over the reading of multiple groups, thereby improving the average time to read a single group of memory cells. In one aspect, reading the multiple different groups of memory cells includes reading one or more pages from each of the groups of memory cells. In one aspect, each group is in a different sub-block.

Abstract: A three-dimensional memory device includes an alternating stack of insulating layers and electrically conductive layers, first memory opening fill structures extending through the alternating stack and including a respective first vertical semiconductor channel having a tubular section and a semi-tubular section, second memory opening fill structures, first bit lines electrically connected to a respective subset of the first drain regions, second bit lines electrically connected to a respective subset of the second drain regions, and an erase voltage application circuit configured to electrically bias the first bit lines at a first bit line erase voltage and the second bit lines at a second bit line erase voltage during an erase operation. The first bit line erase voltage is greater than the second bit line erase voltage.

Abstract: The storage device includes a non-volatile memory with control circuitry and an array of memory cells that are arranged in a plurality of word lines. The control circuitry is configured to program the memory cells in a plurality of programming loops which include applying a programming pulse to a selected word line to program at least one memory cell of the selected word line to a programmed data state. The programming loops also include simultaneously applying a verify pulse to the selected word line to verify a data state being programmed, applying a first voltage to at least one unselected word line that has not been programmed, and applying a second voltage to at least one unselected word line that has already been programmed. The first voltage is determined as a function of the programmed data state to reduce a voltage threshold distribution across the memory cells.