Abstract: A semiconductor device is disclosed including a memory module formed from a pair of semiconductor dies mounted face to face to each other at the wafer level. These die pairs are formed using wafer-to-wafer bonding technology, where the wafers may be bonded to each other when they are of full thickness. Once bonded, respective inactive surfaces of the wafers may be thinned and then the die pairs diced from the wafers to form a completed memory module. When the wafers are bonded face to face, they compensate each other, mechanically resulting in the die pair having a minimum warpage.

Abstract: A memory system is provided that includes a first memory array including a first memory cell, a second memory array including a second memory cell, and a memory controller configured to determine a threshold voltage of the second memory cell to compensate a drift of a threshold voltage of the first memory cell and/or determine an offset voltage of the second memory cell to compensate an offset voltage of the first memory cell.

Abstract: Apparatuses and techniques are described for reliably storing bad column data in a memory device. Units of bad column data and related units of error detection data are stored in non-adjacent groups of memory cells connected to a word line in a ROM block. A unit of bad column data and a related unit of error detection data can be stored in respective groups of memory cells which are relatively far apart from one another along the word line. This helps ensure that a defect in some NAND strings will not affect both the unit of bad column data and a related unit of error detection data. In another aspect, a unit of bad column data and a related unit of error detection data can be stored using different input/output circuits to further increase robustness.

Abstract: A method reading memory using bi-directional sensing, including programming first memory cells coupled to a first word-line using a normal programming order; programming second memory cells coupled to a second word-line using a normal programming order; reading data from the first memory cells by applying a normal sensing operation to the first word-line; and reading data from the second memory cells by applying a reverse sensing operation to the second word-line. Methods also include receiving an error associated with reading data from the first memory cells; and then reading the data from the first memory cells by applying a reverse sensing operation to the first word-line. Method also include receiving an error associated with reading the data from the second memory cells; and then reading the data from the second memory cells by applying a normal sensing operation to the second word-line.

Abstract: A method of concurrently programming a memory. Various methods include: applying a non-negative voltage on a first bit line coupled to a first memory cell; applying a negative voltage on a second bit line coupled to a second memory cell, where the negative voltage is generated using triple-well technology; then applying a programming pulse to the first and second memory cells concurrently; and in response, programming the first and second memory cells to different states. The methods also include applying a quick pass write operation to the first and second memory cells, by: applying a quick pass write voltage to the first bit line coupled to the first memory cell, where the quick pass write voltage is higher than the non-negative voltage; applying a negative quick pass write voltage to the second bit line coupled to the first memory cell, where the negative quick pass write voltage is generated using triple-well technology.

Abstract: A method of concurrently programming a memory. Various methods include: applying a non-negative voltage on a first bit line coupled to a first memory cell; applying a negative voltage on a second bit line coupled to a second memory cell, where the negative voltage is generated using triple-well technology; then applying a programming pulse to the first and second memory cells concurrently; and in response, programming the first and second memory cells to different states. The methods also include applying a quick pass write operation to the first and second memory cells, by: applying a quick pass write voltage to the first bit line coupled to the fist memory cell, where the quick pass write voltage is higher than the non-negative voltage; applying a negative quick pass write voltage to the second bit line coupled to the first memory cell, where the negative quick pass write voltage is generated using triple-well technology.

Abstract: A non-volatile memory device comprising a memory cell region having a plurality of co-planar memory cell planes arranged in a plane parallel to a semiconductor substrate, with each memory cell plane comprising a plurality of sub-planes disposed adjacent one another along an axis that is parallel to the substrate. Further, each memory cell plane comprises a plurality of sense amplifier regions arranged along the axis in an alternating pattern with the sub-planes such that adjacent to each sub-plane is a sense amplifier region and each sense amplifier region is operable with respect to at least a fraction of the bit lines of the two sub-planes immediately adjacent the sense amplifier region.

Abstract: A non-volatile memory apparatus is provided and includes a substrate having a major surface extending longitudinally. A stack of first and second sets of word lines and insulating layers extends along and over the major surface longitudinally and alternating with and overlying one another vertically to define a device region. The first and second sets of word lines each respectively extends longitudinally beyond a first and second side of the device region a decreasing longitudinal distance from the device region as a vertical distance from the major surface increases to define first and second stepped contact regions. Word line contacts extend vertically in the first and second stepped contact regions. The second set of word lines in the first stepped contact region do not contact the word line contacts and the first set of word lines in the second stepped contact region do not contact the word line contacts.

Abstract: Techniques and apparatuses are provided for determining the efficiency of a charge pump. A charge pump is driven during a measurement period without limiting its input current. The driving can include ramping up the output of the charge pump from an initial level to a final level and maintaining the output at the final level. A counter counts a number of clock pulses provided to the charge pump during the measurement period. Using a current mirror which limits the input current, a ratio of the current mirror is determined which results in a similar number of clock pulses during the measurement period. The ratio indicates an efficiency of the charge pump and can be used to set control parameters such as ramp up rate and clock frequency.

Abstract: Methods for reducing manufacturing cost and improving the reliability of non-volatile memories using NAND strings with polysilicon channels and p-type doped source lines are described. A NAND string may include a polysilicon channel that is orthogonal to a substrate and connects to a boron doped source line at a source-side end of the NAND string. To reduce the likelihood of the polysilicon channel being cut-off or pinched near the source-side end of the NAND string, a thicker polysilicon channel may be formed near the source-side end of the NAND string while a thinner polysilicon channel may be formed for the remainder of the NAND string by diffusing boron into a first portion of the polysilicon channel corresponding with the thicker polysilicon channel and then etching the polysilicon channel with etchants that exhibit a reduction in their etch rate at a boron concentration above a threshold concentration.

Abstract: A method for detecting faults in a memory system includes performing an operation on at least one memory cell of the memory system. The method also includes receiving, during performance of the operation, a first clock cycle count for a first pulse of a charge pump associated with the at least one memory cell. The method also includes receiving, during performance of the operation, a second clock cycle count for a second pulse of the charge pump. The method also includes determining whether a fault will occur based on a difference between the first clock cycle count and the second clock cycle count.

Abstract: A non-volatile memory system and corresponding method of operation are provided. The system includes memory cells arranged in sectors. Each of the memory cells includes a control gate in communication with one of a plurality of word lines and a drain coupled to one of a plurality of bit lines and is configured to retain a threshold voltage. A control circuit is in communication with the memory cells and is configured to read the threshold voltage of each of the memory cells using default read parameters. The control circuit determines whether reading the non-volatile memory cells using the default read parameters is successful. The control circuit dynamically tests and adjusts read parameters based on whether reading the memory cells using the read parameters is successful in response to determining that reading the memory cells using the default read parameters is not successful.

Abstract: Systems and methods for reducing the impact of defects within a crossbar memory array when performing multiplication operations in which multiple control lines are concurrently selected are described. A group of memory cells within the crossbar memory array may be controlled by a local word line that is controlled by a local word line gating unit that may be configured to prevent the local word line from being biased to a selected word line voltage during an operation; the local word line may instead be set to a disabling voltage during the operation such that the memory cell currents through the group of memory cells are eliminated. If a defect has caused a short within one of the memory cells of the group of memory cells, then the local word line gating unit may be programmed to hold the local word line at the disabling voltage during multiplication operations.

Abstract: Systems and methods for reducing the impact of defects within a crossbar memory array when performing multiplication operations in which multiple control lines are concurrently selected are described. A group of memory cells within the crossbar memory array may be controlled by a local word line that is controlled by a local word line gating unit that may be configured to prevent the local word line from being biased to a selected word line voltage during an operation; the local word line may instead be set to a disabling voltage during the operation such that the memory cell currents through the group of memory cells are eliminated. If a defect has caused a short within one of the memory cells of the group of memory cells, then the local word line gating unit may be programmed to hold the local word line at the disabling voltage during multiplication operations.

Abstract: Consecutive logical block addresses (LBAs) are mapped to consecutive good blocks in a sequence of blocks in a memory device. For each bad block, a mapping process substitutes a next available good block. For a selected LBA, the mapping process determines a number X>1 of bad blocks before, and including, a corresponding block in the sequence, a number Y of bad blocks in the X blocks after the corresponding block in the sequence, and maps the LBA to a block which is X+Y blocks after the corresponding block, or, if the block which is X+Y blocks after the corresponding block is a bad block, to a next good block. The mapping technique can be used for a sequence of blocks in a trimmed die, where a bad block register stores physical block addresses of the trimmed away blocks.

Abstract: A random access memory is provided including a plane structure comprising a plurality of sense amplifiers, each including a local data latch, a pair of local busses connected to each of the data latches, a differential data bus, and a pair of redrivers connected between the pair of local busses and the differential data bus.

Abstract: Techniques are presented for accelerating in-memory matrix multiplication operations for a convolution neural network (CNN) inference in which the weights of a filter are stored in the memory of a storage class memory device, such as a ReRAM or phase change memory based device. To improve performance for inference operations when filters exhibit sparsity, a zero column index and a zero row index are introduced to account for columns and rows having all zero weight values. These indices can be saved in a register on the memory device and when performing a column/row oriented matrix multiplication, if the zero row/column index indicates that the column/row contains all zero weights, the access of the corresponding bit/word line is skipped as the result will be zero regardless of the input.

Abstract: Use of a NAND array architecture to realize a binary neural network (BNN) allows for matrix multiplication and accumulation to be performed within the memory array. A unit synapse for storing a weight of a BNN is stored in a pair of series connected memory cells. A binary input is applied on a pair of word lines connected to the unit synapse to perform the multiplication of the input with the weight. The results of such multiplications are determined by a sense amplifier, with the results accumulated by a counter. The arrangement extends to ternary inputs to realize a ternary-binary network (TBN) by adding a circuit to detect 0 input values and adjust the accumulated count accordingly. The arrangement further extends to a ternary-ternary network (TTN) by allowing 0 weight values in a unit synapse, maintaining the number of 0 weights in a register, and adjusting the count accordingly.

Abstract: Apparatuses, systems, and methods are disclosed for concurrently programming non-volatile storage cells, such as those of an SLC NAND array. The non-volatile storage cells may be arranged into a first block comprising a first string of storage cells that intersects with a first word line at a first storage cell, a second block comprising a second string of storage cells that intersects with a second word line at a second storage cell, a bit line electrically connectable to the first string and the second string, and controller configured to apply a programming pulse, at an elevated voltage, to the first word line and second word line to concurrently program the first and second storage cells.

Abstract: Example techniques that mitigate against memory hole shorts during an erase operation for memory cells in a string include an example method in which, during an erase operation, erase pulses are applied to the word lines of the memory string and terminated at different times based. In some instances, the erase pulses applied to the word lines of the memory string are terminated based on the temperature of the memory cells of the memory string. In further implementations, the erase pulses applied to the word lines of the memory string are boosted for different times depending on the location of the word line along the memory string during the erase operation.