A memory device includes a memory cell array including a plurality of memory cells; a voltage generator configured to generate voltages used for a program operation and a verify operation for the memory cells; and control logic configured to perform a plurality of program loops while writing data to the memory cell array, such that first to N-th (e.g., N>=1) program loops including a program operation and a verify operation are performed and at least two program loops in which the verify operation is skipped are performed when a pass/fail determination of the program operation in the N-th program loop indicates a pass.

A nonvolatile semiconductor memory device includes a control circuit configured to control a soft program operation of setting nonvolatile memory cells to a first threshold voltage distribution state of the nonvolatile memory cells. When a characteristic of the nonvolatile memory cells is in a first state, the control circuit executes the soft program operation by applying a first voltage for setting the nonvolatile memory cells to the first threshold voltage distribution state to first word lines, and applying a second voltage higher than the first voltage to a second word line. When the characteristic of the nonvolatile memory cells is in a second state, the control circuit executes the soft program operation by applying a third voltage equal to or lower than the first voltage to the first word lines and applying a fourth voltage lower than the second voltage to the second word line.

A method for programming a memory block of a non-volatile memory structure, comprising determining whether a number of programming/erase cycles previously applied to the block exceeds a first programming/erase cycle threshold and, if the first threshold is exceeded, determining whether the number of programming/erase cycles previously applied to the block exceeds an extended programming/erase cycle threshold. Further, if the determination is made that the extended threshold is not exceeded, the method comprises applying a two-pulse per programming loop scheme to each of the outermost strings of the block and applying a single-pulse per programming loop scheme to all other strings of the block. Alternatively, or in addition thereto, relative to a programming/erase cycle threshold, one or more outermost strings of the block may be unpermitted to be further programmed, and a “sub-block” comprised of all valid strings of the block may be defined and permitted for further programming.

A method for programming a memory block of a non-volatile memory structure, comprising determining whether a number of programming/erase cycles previously applied to the block exceeds a first programming/erase cycle threshold and, if the first threshold is exceeded, determining whether the number of programming/erase cycles previously applied to the block exceeds an extended programming/erase cycle threshold. Further, if the determination is made that the extended threshold is not exceeded, the method comprises applying a two-pulse per programming loop scheme to each of the outermost strings of the block and applying a single-pulse per programming loop scheme to all other strings of the block. Alternatively, or in addition thereto, relative to a programming/erase cycle threshold, one or more outermost strings of the block may be unpermitted to be further programmed, and a “sub-block” comprised of all valid strings of the block may be defined and permitted for further programming.

Numerous examples of a precision programming apparatus are disclosed for precisely and quickly depositing the correct amount of charge on the floating gate of a non-volatile memory cell within a vector-by-matrix multiplication (VMM) array in an artificial neural network. In one example, a neuron output circuit for providing a current to program as a weight value in a selected memory cell in a vector-by-matrix multiplication array is disclosed, the neuron output circuit comprising a first adjustable current source to generate a scaled current in response to a neuron current to implement a positive weight, and a second adjustable current source to generate a scaled current in response to a neuron current to implement a negative weight.

Numerous embodiments of a precision programming algorithm and apparatus are disclosed for precisely and quickly depositing the correct amount of charge on the floating gate of a non-volatile memory cell within a vector-by-matrix multiplication (VMM) array in an artificial neural network. Selected cells thereby can be programmed with extreme precision to hold one of N different values.

A method of operating a memory system includes programming, in a memory device, K logical pages stored in a page buffer circuit into a memory cell array, reading, from the memory device, the K logical pages programmed into the memory cell array into the page buffer circuit after a first delay time elapses, transmitting, in a memory controller, N−K logical pages to the memory device, and programming, in the memory device, N logical pages into the memory cell array based on the read K logical pages and the N−K logical pages, wherein K is a positive integer and N is a positive integer greater than K.

Memory devices might include a controller configured to cause the memory device to apply a first plurality of incrementally increasing programming pulses to control gates of a particular plurality of memory cells selected for programming to respective intended data states, determine a first occurrence of a criterion being met, store a representation of a voltage level corresponding to a particular programming pulse in response to the first occurrence of the criterion being met, set a starting programming voltage for a second plurality of incrementally increasing programming pulses in response to the stored representation of the voltage level corresponding to the particular programming pulse, and apply the second plurality of incrementally increasing programming pulses to control gates of a different plurality of memory cells selected for programming to respective intended data states.

A memory device may include: a control circuit comprising a first verification component suitable for counting the number of memory cells in the selected word line having an excessively high threshold voltage as excessive memory cells, after a program operation is completed; and a second verification component suitable for counting the number of failed bits when the number of excessive memory cells counted is greater than or equal to an excess threshold value, and suitable for outputting a pass or fail signal for the program operation according to the count of at least one of the first verification component and the second verification component.

A nonvolatile semiconductor memory device includes a control circuit configured to control a soft program operation of setting nonvolatile memory cells to a first threshold voltage distribution state of the nonvolatile memory cells. When a characteristic of the nonvolatile memory cells is in a first state, the control circuit executes the soft program operation by applying a first voltage for setting the nonvolatile memory cells to the first threshold voltage distribution state to first word lines, and applying a second voltage higher than the first voltage to a second word line. When the characteristic of the nonvolatile memory cells is in a second state, the control circuit executes the soft program operation by applying a third voltage equal to or lower than the first voltage to the first word lines and applying a fourth voltage lower than the second voltage to the second word line.