Systems, apparatuses, and methods for dynamically coalescing multi-bank memory commands to improve command throughput are disclosed. A system includes a processor coupled to a memory via a memory controller. The memory also includes processing-in-memory (PIM) elements which are able to perform computations within the memory. The processor generates memory requests targeting the memory which are sent to the memory controller. The memory controller stores commands received from the processor in a queue, and the memory controller determines whether opportunities exist for coalescing multiple commands together into a single multi-bank command. After coalescing multiple commands into a single combined multi-bank command, the memory controller conveys, across the memory bus to multiple separate banks, the single multi-bank command and a multi-bank code specifying which banks are targeted. The memory banks process the command in parallel, and the PIM elements process the data next to each respective bank.

A memory device includes command logic allowing for a command protocol allowing interruption of a first command sequence, such as a page write sequence, and then to proceed directly to receive and decode a second command sequence, such as a read sequence, without latency associated, completing the first command sequence. Also, the command logic is configured to be responsive to a third command sequence after the second command sequence and its associated embedded operation have been completed, which completes the interrupted first command sequence and enables execution of an embedded operation identified by the first command sequence. A memory controller supporting such protocols is described.

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.

An apparatus includes a scheduler module operatively coupled to each memory block from a set of memory blocks via a shared address bus. The scheduler module is configured to receive a group of memory commands from a set of memory controllers. Each memory controller from the set of memory controllers is uniquely associated with a different memory block from the set of memory blocks. The scheduler module is configured to classify each memory command from the group of memory commands into a category based at least in part on memory commands previously sent to the set of memory blocks via the shared address bus. The scheduler module is configured to select an order in which to send each memory command from the group of memory commands to the set of memory blocks via the shared address bus based at least in part on the category of each memory command.

Methods, apparatus, systems, and articles of manufacture are disclosed to determine memory access integrity based on feedback from memory. An example apparatus includes an access reconstruction controller including an output, a first input configured to be coupled to memory, and a second input configured to be coupled to a memory signal generator; a comparator including a first input coupled to the output of the access reconstruction controller, a second input configured to be coupled to an arbiter, and an output configured to be coupled to the arbiter; and a data integrity monitor including an input coupled to the second input of the comparator and configured to be coupled to the arbiter and an output coupled to the output of the comparator and configured to be coupled to the arbiter.

An apparatus and method are provided for controlling access to a memory device. The apparatus has a pending access requests storage that is used to store access requests waiting to be issued to the memory device, and memory access control circuitry is then used to issue to the memory device access requests selected from the pending access requests storage. Access requests are received at an interface of the apparatus from at least one requesting device, and access request evaluation circuitry within the apparatus is arranged to apply criteria to determine, for a current access request, whether to accept that current access request or reject that current access request. The criteria applied takes account of at least one access timing characteristic of the memory device. The access request evaluation circuitry is responsive to determining that the current access request is to be accepted, to cause that current access request to be stored in the pending access requests storage. However, if instead the access request is rejected, it is prevented from being added to the pending access requests storage at that time, and instead a rejection indication is issued to the requesting device that issued that current access request. This provides a mechanism for significantly improving the performance of the memory device by providing more selectivity as to what access requests are accepted into the pending access requests storage.

Methods, systems, and apparatus, including computer-readable media, are described for an integrated circuit that accelerates machine-learning computations. The circuit includes processor cores that each include: multiple channel controllers; an interface controller for coupling each channel controller to any memory channel of a system memory; and a fetch unit in each channel controller. Each fetch is configured to: receive channel data that encodes addressing information; obtain, based on the addressing information, data from any memory channel of the system memory using the interface controller; and write the obtained data to a vector memory of the processor core via the corresponding channel controller that includes the respective fetch unit.

Apparatus identifies a set of M output memory addresses from a larger set of N input memory addresses containing at least one non-unique memory address. A comparator block performs comparisons of memory addresses from a set of N input memory addresses to generate a binary classification dataset that identifies a subset of addresses from the set of input addresses, where each address in the subset identified by the binary classification dataset is unique within that subset. Combination logic units receive a predetermined selection of bits of the binary classification dataset and sort its received predetermined selection of bits into an intermediary binary string in which the bits are ordered into a first group identifying addresses belonging to the identified subset, and a second group identifying addresses not belonging to the identified subset. Output generating logic selects between bits belonging to different intermediary binary strings to generate a binary output identifying a set of output memory addresses containing at least one address in the identified subset.

Systems, apparatuses, and methods for dynamically coalescing multi-bank memory commands to improve command throughput are disclosed. A system includes a processor coupled to a memory via a memory controller. The memory also includes processing-in-memory (PIM) elements which are able to perform computations within the memory. The processor generates memory requests targeting the memory which are sent to the memory controller. The memory controller stores commands received from the processor in a queue, and the memory controller determines whether opportunities exist for coalescing multiple commands together into a single multi-bank command. After coalescing multiple commands into a single combined multi-bank command, the memory controller conveys, across the memory bus to multiple separate banks, the single multi-bank command and a multi-bank code specifying which banks are targeted. The memory banks process the command in parallel, and the PIM elements process the data next to each respective bank.

An embodiment of an apparatus may comprise one or more substrates, and a controller coupled to the one or more substrates, the controller including circuitry to control access to a memory, convert an address for a transaction for the memory from a first address in a first address space to a second address in a second address space, determine a bandwidth bypass condition for the transaction based on a bandwidth of memory transactions for the memory, and provide the second address for the transaction to a scheduler at a time based at least in part on the determined bandwidth bypass condition. Other embodiments are disclosed and claimed.