A semiconductor device includes a substrate, a plurality of word line structures disposed over the substrate to be spaced apart from each other in a first direction perpendicular to a surface of the substrate. Each of the plurality of word line structures extends in a second direction parallel to the surface of the substrate. In addition, the semiconductor device includes a switching layer disposed over the substrate to contact side surfaces of the plurality of word line structures, and bit line structures disposed over the substrate to extend in the first direction and to contact a surface of the switching layer. The switching layer is configured to perform a threshold switching operation, and has a variable programmable threshold voltage.

Techniques for burst memory write operations are disclosed. In the illustrative embodiment, a memory die is limited in how quickly it can perform memory write operations that it receives from a microcontroller due to thermal constraints. The memory die can mitigate the need for the microcontroller to perform a costly rank switch to send an operation to another die by buffering memory write operations. The microcontroller can then send several consecutive memory write operations to a first memory die before switching to a second memory die. The first memory die can then perform the memory write operations while the microcontroller has moved on to other memory operations.

A method for training an artificial neural network circuit is provided. The artificial neural network circuit includes a crossbar circuit that has a plurality of input bars, a plurality of output bars crossing the plurality of input bars, and memristors each of which includes a variable conductance element provided at corresponding one of intersections of the input bars and the output bars.

An output, representing synaptic weights of a neural network can be received from first memory elements. The output can be compared to a known correct output. A random number can be generated with a tuned bias via second memory elements. The weights can be updated based on the random number and a difference between the output and the known correct output.

Methods, systems, and devices for adjustable programming pulses for a multi-level cell are described. A memory device may modify a characteristic of a programming pulse for an intermediate logic state based on a metric of reliability of associated memory cells. The modified characteristic may increase a read window and reverse a movement of a shifted threshold voltage distribution (e.g., by moving the threshold voltage distribution farther from one or more other voltage distributions). The metric of reliability may be determined by performing test writes may be a quantity of cycles of use for the memory cells, a bit error rate, and/or a quantity of reads of the first state. The information associated with the modified second pulse may be stored in fuses or memory cells, or may be implemented by a memory device controller or circuitry of the memory device.

An electronic device includes analog-to-digital converters each configured to receive an analog input signal and output a digital output signal corresponding to the analog input signal, an analog input signal generator configured to generate analog input signals provided to each analog-to-digital converter based on input voltages and weight data, an input signal distribution information generator configured to generate input signal distribution information indicating a distribution of the analog input signals for each of the analog-to-digital converters, an analog-to-digital converter group classifier configured to classify the analog-to-digital converters into a plurality of first analog-to-digital converter groups based on the input signal distribution information, and an analog-to-digital converter input range optimizer configured to determine an input range of each first analog-to-digital converter group based on the input signal distribution information, and each analog-to-digital converter is configured to operate according to an input range of a corresponding first analog-to-digital converter groups.

A neural network circuit is described that includes a first sample-and-hold circuit, a reference voltage generation circuit, a first comparator circuit, and a first output circuit. The first sample-and-hold circuit generates a first analog voltage based on a first output current output by a first neural network computation array. The reference voltage generation circuit generates a reference voltage based on a first control signal. The first comparator circuit is connected to the first sample-and-hold circuit and the reference voltage generation circuit, and outputs a first level signal based on the first analog voltage and the reference voltage. The first output circuit samples the first level signal based on a second control signal, and outputs a first computation result that meets the first computation precision.

A method for neural network computation using adaptive data representation, adapted for a processor to perform multiply-and-accumulate operations on a memory having a crossbar architecture, is provided. The memory comprises multiple input and output lines crossing each other, multiple cells respectively disposed at intersections of the input and output lines, and multiple sense amplifiers respectively connected to the output lines. In the method, an input cycle of kth bits respectively in an input data is adaptively divided into multiple sub-cycles, wherein a number of the divided sub-cycles is determined according to a value of k. The kth bits of the input data are inputted to the input lines with the sub-cycles and computation results of the output lines are sensed by the sense amplifiers. The computation results sensed in each sub-cycle are combined to obtain the output data corresponding to the kth bits of the input data.

A method is presented for mapping weights for kernels of a neural network onto a crossbar array. In one example, the crossbar array is comprised of an array of non-volatile memory cells arranged in columns and rows, such that memory cells in each row of the array is interconnected by a respective drive line and each column of the array is interconnected by a respective bit line; and wherein each memory cell is configured to receive an input signal indicative of a multiplier and operates to output a product of the multiplier and a weight of the given memory cell onto the corresponding bit line of the given memory cell, where the value of the multiplier is encoded in the input signal and the weight of the given memory cell is stored by the given memory cell.

The apparatuses and methods described herein may operate to measure a voltage difference between a selected access line and a selected sense line associated with a selected cell of a plurality of memory cells of a memory array. The voltage difference may be compared with a reference voltage specified for a memory operation. A selection voltage(s) applied to the selected cell for the memory operation may be adjusted responsive to the comparison, such as to dynamically compensate for parasitic voltage drop.