Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing data on a memory controller. One of the methods comprises obtaining a first request and a second request to access respective data corresponding to the first and second requests at a first memory device of the plurality of memory devices; and initiating interleaved processing of the respective data; receiving an indication to stop processing requests to access data at the first memory device and to initiate processing requests to access data at a second memory device, determining that the respective data corresponding to the first and second requests have not yet been fully processed at the time of receiving the indication, and in response, storing, in memory accessible to the memory controller, data corresponding to the requests which have not yet been fully processed.

Techniques for scheduling memory operations are disclosed in which alternate read/write commands within a multi-bank memory operation are delayed beyond a minimum timing parameter in order to increase memory data bus utilization. The remaining read/write commands are not delayed beyond the minimum timing parameter. Every other clock cycle (e.g., even clock cycles) within the memory operation is reserved for activate commands, while other commands such as sync and read/write are scheduled on the intervening clock cycles (e.g., odd clock cycles). For memory devices for which a sync command (which causes a clock of the memory data bus to start) is to precede a corresponding read/write command by a number of clock cycles that would place it in a cycle reserved for activate commands, a particular operation mode is disclosed in which the memory device internally delays a received sync command.

A computing device includes a host processor to execute a host driver to create a host-side interface, the host-side interface emulating a first Ethernet interface, assign the host-side interface a first medium access control (MAC) address and a first Internet Protocol (IP) address. Memory components are disposed on a substrate. A memory channel network (MCN) processor is disposed on the substrate and coupled between the memory components and the host processor. The MCN processor is to execute an MCN driver to create a MCN-side interface, the MCN-side interface emulating a second Ethernet interface. The MCN processor is to assign the MCN-side interface a second MAC address and a second IP address, which identify the MCN processor as a MCN network node to the host processor.

Techniques including a memory controller with a set of memory channel queues, wherein memory channel queues of the set of memory channel queues correspond to memory channels to access a set of memory modules, a first arbitration module, and a second arbitration module. The memory controller is configured to receive a first memory request from the peripheral and place one or more portions of the first memory request in the memory channel queues of the set of memory channel queues. The first arbitration module is configured to determine an arbitration algorithm, select a first memory channel queue based on the arbitration algorithm, present the one or more portions of the first memory request in the selected first memory channel queue to the second arbitration module, and output the presented one or more portions of the first memory request based on a selection by the second arbitration module.

A fine-grained dynamic random-access memory (DRAM) includes a first memory bank, a second memory bank, and a dual mode I/O circuit. The first memory bank includes a memory array divided into a plurality of grains, each grain including a row buffer and input/output (I/O) circuitry. The dual-mode I/O circuit is coupled to the I/O circuitry of each grain in the first memory bank, and operates in a first mode in which commands having a first data width are routed to and fulfilled individually at each grain, and a second mode in which commands having a second data width different from the first data width are fulfilled by at least two of the grains in parallel.

Methods, systems, and apparatus, including computer-readable media, are described for interleaving memory requests to accelerate memory accesses at a hardware circuit configured to implement a neural network model. A system generates multiple requests that are processed against a memory of the system. Each request is used to retrieve data from the memory. For each request, the system generates multiple sub-requests based on a respective size of the data to be retrieved using the request. The system generates a sequence of interleaved sub-requests that includes respective sub-requests of a first request interleaved among respective sub-requests of a second request. Based on the sequence of interleaved sub-requests, a module of the system receives respective portions of data accessed from different address locations of the memory. The system processes each of the respective portions of data to generate a neural network inference using the neural network model implemented at the hardware circuit.

A memory access unit for handling transfers of samples in a d-dimensional array between a one of m data buses, where m≧1, and k*m memories, where k≧2, is disclosed. The memory access unit comprises k address calculators, each address calculator configured to receive a bus address to add a respective offset to generate a sample bus address and to generate, from the sample bus address according to an addressing scheme, a respective address in each of the d dimensions for access along one of the dimensions from the bus address according to an addressing scheme, for accessing a sample. The memory access unit comprises k sample collectors, each sample collector operable to generate a memory select for a one of the k*m memories so as to transfer the sample between a predetermined position in a bus data word and the respective one of the k*m memories. Each sample collector is configured to calculate a respective memory select in dependence upon the address in each of the d dimensions such that each sample collector selects a different one of the k*m memories so as to allow the sample collectors to access k of the k*m memories concurrently. A memory controller may comprise m memory access units for handling transfers of samples in a d-dimensional array between m data buses and k*m memories.

A memory system may include a plurality of memory dies configured to store data therein, and a controller coupled to the plurality of memory dies through a plurality of channels, and configured to correlate at least some of a plurality of read requests and transferring the plurality of read requests to the plurality of channels, such that the plurality of read requests are processed in an interleaving way through the plurality of channels, when controlling the plurality of memory dies for the plurality of read requests. The controller may determine whether to perform the correlation operation in response to the number of the plurality of read requests, wherein the plurality of read requests include a read request for an internal operation of the controller and a read request received from a host.

An interface circuit of a memory device including a plurality of memory dies including a plurality of registers corresponding to the plurality of memory dies, respectively, the plurality of registers each configured to store command information related to a data operation command, a demultiplexer circuit configured to provide input command information to a selected register from among the plurality of registers according to at least one of a first address or a first chip selection signal, the input command information being received from outside the interface circuit, and a multiplexer circuit configured to receive output command information from the selected register from among the plurality of registers and output the output command information according to at least one of a second address or a second chip selection signal may be provided.

An image signal processor includes a line interleaving controller and an image signal processor core. The line interleaving controller receives a plurality of image data lines included in an image frame, generates one or more virtual data lines corresponding to the image frame, and outputs the plurality of image data lines and the virtual data lines sequentially line by line. The image signal processor core includes at least one pipeline circuit. The pipe line circuit includes a plurality of processing modules serially connected to sequentially process data lines received from the line interleaving controller. The line interleaving controller processes one or more end image data lines included in an end portion of the image frame based on the virtual data lines. Interference or collision between channels is reduced or prevented by processing the end image data lines in synchronization with the virtual data lines.