A memory module includes at least two memory devices. Each of the memory devices perform verify operations after attempted writes to their respective memory cores. When a write is unsuccessful, each memory device stores information about the unsuccessful write in an internal write retry buffer. The write operations may have only been unsuccessful for one memory device and not any other memory devices on the memory module. When the memory module is instructed, both memory devices on the memory module can retry the unsuccessful memory write operations concurrently. Both devices can retry these write operations concurrently even though the unsuccessful memory write operations were to different addresses.

This disclosure describes various examples of a system which uses a multi-bank, multi-port shared memory system that may be implemented as part of a system on a chip. The shared memory system may have particular applicability in the context of an artificial reality system, and may be designed to have distributed or varied latency for one or more memory banks and/or one or more components or subsystems within the system on a chip. The described shared memory system may be logically a single entity, but physically may have multiple memory banks, each accessible by any of a number of components or subsystems. In some examples, the memory system may enable concurrent, common, and/or shared access to memory without requiring, in some situations, full locking or arbitration.

Methods, systems, and devices for reduced pin status register are described. An apparatus may include a first memory die and a second memory die each coupled with a data bus. The apparatus may further include a controller coupled with the first memory die and the second memory die via the data bus that is configured to transmit a first command associated with a first operation to the first memory die and a second command associated with a second operation to the second memory die. The controller may further transmit a third command concurrently to the first memory die and the second memory die, the third command requesting a first status of the first operation and a second status of the second operation. The controller may receive the first status and the second status concurrently via the data bus from the first memory die and the second memory die.

A memory controller for accessing a memory, comprises a holding circuit which holds a plurality of read or write access requests from a bus master, a read/write control circuit which selects one of the access requests in the holding circuit and issues a read command or a write command; and an active control circuit which selects the access request held in the holding circuit and issues an active command, wherein the active control circuit includes a generation circuit that generates number of activated read commands and number of activated write commands, and a selection circuit that, when the number of activated read commands is not less a threshold, issues the active command of an read access, and when the number of activated write commands is not less than the threshold, issues the active command of a write access.

A memory control circuit is configured to access a memory including a plurality of banks. The memory control circuit comprises: a holding unit configured to hold an access request from an external circuit; a management unit configured to manage states of the plurality of banks; a determination unit configured to determine, based on an access type of an access request held in the holding unit and the states of the plurality of banks, which access type of command issuance that is read or write is to be prioritized; and an issuance unit configured to issue a command of an access request corresponding to the access type determined to be prioritized by the determination unit, among the access requests held in the holding unit.

A memory controller includes an arbiter for selecting memory requests from a command queue for transmission to a DRAM memory. The arbiter includes a bank group tracking circuit that tracks bank group numbers of three or more prior write requests selected by the arbiter. The arbiter also includes a selection circuit that selects requests to be issued from the command queue, and prevents selection of write requests and associated activate commands to the tracked bank group numbers unless no other write request is eligible in the command queue. The bank group tracking circuit indicates that a prior write request and associated activate commands are eligible to be issued after a number of clock cycles has passed corresponding to a minimum write-to-write timing period for the bank group of the prior write request.

A memory controller includes a command queue and an arbiter for selecting entries from the command queue for transmission to a DRAM. The arbiter transacts streaks of consecutive read commands and streaks of consecutive write commands. The arbiter transacts a streak for at least a minimum burst length based on a number of commands of a designated type available to be selected by the arbiter. Following the minimum burst length, the arbiter decides to start a new streak of commands of a different type based on a first set of one or more conditions indicating intra-burst efficiency.

A computing device, including a processor configured to perform data transfer scheduling for a hardware accelerator including a plurality of processing areas. Performing data transfer scheduling may include receiving a plurality of data transfer instructions that encode requests to transfer data to respective processing areas. Performing data transfer scheduling may further include identifying a plurality of transfer path conflicts between the data transfer instructions. Performing data transfer scheduling may further include sorting the data transfer instructions into a plurality of transfer instruction subsets. Within each transfer instruction subset, none of the data transfer instructions have transfer path conflicts. For each transfer instruction subset, performing data transfer scheduling may further include conveying the data transfer instructions included in that transfer instruction subset to the hardware accelerator. The data transfer instructions may be conveyed in a plurality of sequential data transfer phases that correspond to the transfer instruction subsets.

Apparatuses and methods related to command selection policy for electronic memory or storage are described. Commands to a memory controller may be prioritized based on a type of command, a timing of when one command was received relative to another command, a timing of when one command is ready to be issued to a memory device, or some combination of such factors. For instance, a memory controller may employ a first-ready, first-come, first-served (FRFCFS) policy in which certain types of commands (e.g., read commands) are prioritized over other types of commands (e.g., write commands). The policy may employ exceptions to such an FRFCFS policy based on dependencies or relationships among or between commands. An example can include inserting a command into a priority queue based on a category corresponding to respective commands, and iterating through a plurality of priority queues in order of priority to select a command to issue.

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.