Abstract: A storage device includes a storage array with multiple planes organized as plane groups, where the planes of a plane group receive and process commands in parallel. The storage device includes a storage controller that receives a command from a host controller. In response to receipt of the command the storage controller provides ready information for all planes to the host controller. The multiple planes can optionally have independent multiplane read operation (IMPRO). Each plane group can have a first plane and a second plane, and the storage controller can optionally read data from the first plane of a plane group in response to a virtual ready signal for the first plane, before the second plane of the plane group is ready.

Abstract: The disclosed technology may comprise a mechanism that provides address range based memory hints, based on software memory access patterns. The mechanism includes an interface that can be in the form of a new instruction (“mem_range_hint”) that includes as operands (i) the start of the virtual address (“start_vaddr”), (ii) memory region size (“size”), and (iii) a hint (“hint”) associated with the memory access pattern.

Abstract: The disclosure includes a system and methods provide for optimizing performance of disaggregated memory architectures in terms of time and energy. Examples of the systems and methods disclosed herein provide for a near memory compute proximate to a disaggregated memory that can be implemented to receive, from a compute node, one or more requests to perform computation functions on data stored at the disaggregated memory and collect telemetry data for the disaggregated memory, a near memory compute proximate to the disaggregated memory, and the compute node. The systems and methods disclosed herein can also model a plurality of configurations for executing the one or more requests based on the telemetry data, select a modeled configuration of the plurality of modeled configurations for executing the one or more requests, and assign one or more of a plurality of data operators of the near memory compute according to the selected modeled configuration.

Abstract: A volume management server may regularly check the status of persistent volumes to determine if they have been released. A database with information about the persistent volumes may be updated to indicate a current date as the release date when a persistent volume is found to be released without a release date. This enables the volume management server to determine, in the future, how long the persistent volume has been released. When the difference between the current date and the release date for a persistent volume reaches a first predetermined age threshold, the volume management server notifies a set of users. When the difference between the current date and the release date for the persistent volume reaches a second predetermined age threshold, the volume management server causes deletion of the persistent volume.

Abstract: The present disclosure provides methods and apparatuses for programming operating system (OS) data before a surface mount technology (SMT) process. In some embodiments, a method includes erasing a plurality of memory cells in a memory block, classifying word lines coupled to the memory block into first word lines to be programmed with OS data and second word lines to be programmed in a state pattern, programming, with a multi-bit program, the OS data into first memory cells of the plurality of memory cells coupled to the first word lines, and programming second memory cells of the plurality of memory cells coupled to the second word lines to have the state pattern.

Abstract: Examples described herein relate to federated management of data operators across multiple near-memory compute (NMC) nodes attached to memory devices in a network-attached memory system. Federated management includes loading, executing, and scaling data operators across the multiple NMC nodes together as a group. Examples include receiving a data access request from a client application and loading data operators in the multiple NMC nodes based on a data access pattern associated with the data access request. Examples include scaling the data operators based on performance metrics for the data operators or the multiple NMC nodes in correlation with client application performance. The multiple NMC nodes may dynamically scale the data operators based on request-load, execution frequency of data operators, resource availability, or other scaling strategies. Examples also include loading and scaling the data operators based on one or more of request characteristics or data operator characteristics.

Abstract: A storage system, a file storage and reading method, and a terminal device are provided. A storage system including a UFS is used as an example. The storage system configures different medium forms of the UFS, for example, configures persistent storage space of the UFS including SLC and TLC medium forms, to implement hybrid medium multi-disk storage. Based on the storage system, a hot file related to an application running rate is preferentially selected and stored in an SLC storage area of the UFS, and files on a mapped multi-disk are mounted on a same directory of a file system, so that files can be migrated between the SLC storage area and a TLC storage area without user perception, to improve performance of the storage system.

Abstract: Disclosed is a storage device which includes a nonvolatile memory device, and a storage controller that controls the nonvolatile memory device and updates metadata based on an operation of the nonvolatile memory device, and the storage controller includes a journal data generator that generates a plurality of journal data associated with the update of the metadata, and a journal data replayer that replays the plurality of journal data in parallel to restore the metadata.

Abstract: A method for controlling a data allocatable neural processing unit is provided. The method may comprise: determining a dependency determination between data blocks of a specific layer of at least one neural network model; allocating memory banks to store the data blocks based on the dependency determination; and storing each data block in the allocated memory bank.

Abstract: A method, a system and an accelerator for managing memory access among one or more computing entities may continuously monitor data access to at least one memory module, associated with a respective source computing entity. Based on the monitoring, embodiments may identify a memory area of the source memory module that contains a predetermined quantity of data that is expected to be used by a sink computing entity, and transmit a first version of content of the identified memory area to the sink computing entity. Embodiments may then identify an explicit request, from the sink computing entity to the source computing entity, for accessing data of the identified memory area. Embodiments may subsequently calculate a difference between the first version of content of the identified memory area and a current content of the identified memory area, and transmit the calculated difference to the requesting sink computing entity.

Abstract: In an aspect, the disclosure is directed to a cell erase method which includes not limited to: initiating an erase operation of a block of memory cells by applying an erase condition to the block of memory cells; performing a first erase verification procedure for at least a portion of the block of memory cells in comparison with a first erase verify voltage; adjusting the first erase verify voltage in response the portion of the block of memory cells having passed the first erase verification procedure; performing a second erase verification procedure for the block of memory cells in comparison with the adjusted erase verify voltage; performing a post program verification procedure for the block of memory cells in comparison with a post program verify voltage to detect leakage current of the block of memory cells; determining whether the adjusted erase verify voltage reaches a final erase verify voltage.

Abstract: A memory system is connectable to a host having a host memory and includes a non-volatile memory that stores management data, a memory controller configured to manage caching of parts of the management data in cache lines of the host memory, and a first memory configured to store a bitmap that includes a bit indicating whether the memory controller has accessed first data stored in the host memory after power was last supplied to the memory system. The first data indicates whether or not a part of the management data corresponding thereto is stored in one of the cache lines, and the memory controller is configured to perform either a first operation of reading the first data from the host memory or a second operation of reading an initial value of the first data managed by the memory controller, based on the bitmap.

Abstract: The present disclosure discloses a virtual memory management method and apparatus supporting physical addresses larger than virtual addresses. The method comprises steps of: determining a target virtual address corresponding to an instruction fetch address or a load storage address in any one of a user mode, a supervisor mode, or a machine mode; determining a target physical address corresponding to the target virtual address by accessing a virtual memory management unit, the virtual memory management unit being internally provided with page table entries that map virtual addresses to physical addresses, the bit width of the target virtual address being possibly less than or equal to that of the target physical address, particularly in a many-core application field; and finally, returning the target physical address to a corresponding instruction fetch unit or load memory unit, thereby ensuring the correctness and validity.

Abstract: A first data entry is written to an address location of a memory resource that is neither a first physical address nor a last physical address. In response to a determination that a second data entry has a value that is greater than a value associated with the first data entry, the second data entry is written to an address location that is physically located between the address location of the memory resource to which the first data entry is written and the last physical address. In response to a determination that the second data entry has the value that is less than the value associated with the first data entry, the second data entry is written to an address location that is physically located between the address location of the memory resource to which the first data entry is written and the first physical address.

Abstract: Systems and methods for implementing maintenance operations on storage devices in place of drive-based maintenance operations are disclosed. According to an aspect, a system includes a storage controller configured to receive a plurality of media scan configurations for maintenance from a plurality of storage devices. The data maintenance algorithms implemented across storage device and storage controller is also configured to tune and/or disable drive-based maintenance routines on one or more of the plurality of storage devices. Further, the BMC and/or storage controller is configured to perform controller-based maintenance operations in replacement of the drive-based maintenance routines of the one or more of the plurality of storage devices based on the received plurality of media scan configurations.

Abstract: One example method includes assigning, at a production site, a priority to a portion of a dataset to be backed up, checking to determine if the priority meets or exceeds a threshold priority; and, when the priority meets or exceeds the threshold priority, and when an air gap between the production site and a storage vault is closed, backing up, by way of the closed air gap, the portion of the dataset to the storage vault.

Abstract: A method of operating a storage device including a nonvolatile memory, the method including: generating virtual domains each of which including page mapping and block mapping tables; receiving a data input/output (I/O) request; performing a data I/O operation corresponding to the data I/O request using the virtual domains; transmitting a data I/O response to a host device in response to the data I/O request and the data I/O operation; and changing at least one of the virtual domains based on a direct request from the host device or a change in a first parameter associated with the data I/O request, and wherein, in response to the first parameter being changed, a second parameter associated with the data I/O response is changed by changing at least one of the virtual domains and by performing the data I/O operation using the changed virtual domain.

Abstract: In an embodiment, an apparatus includes a memory access controller to be coupled to a memory and a memory management unit (MMU) coupled to the memory access controller. The MMU is to receive a memory transaction comprising an original transaction security attribute from a first device; responsive to the memory transaction comprising a first physical address of the memory, transmit the memory transaction to the memory access controller; and responsive to the memory transaction comprising a virtual address, generate a translated memory transaction comprising a translated physical address of the memory based on the virtual address and a translated transaction security attribute and transmit the translated memory transaction to the memory access controller, the translated physical address and the translated transaction security attribute associated with an operating system (OS) memory region of the memory associated with an OS. Other embodiments are described and claimed.

Abstract: Aspects of a storage device including a memory and a controller are provided. The controller may collect, by an association rule mining (ARM) model, file system data from a host file system, the file system data defining at least one attribute of a file. The controller may receive, from the host, a memory command associated with the file. The controller can associate, by the ARM model, the at least one attribute with the file. The controller may perform the memory command based on the association of the at least one attribute with the file.

Abstract: Methods, systems, and devices for suspend operations are described. A memory device may perform a write operation including one or more programming phases and one or more verify phases. The memory device may receive a read command while performing the write operation and determine whether the verify phase of the write operation is complete. The memory device may suspend a performance of the write operation in response to determining that the verify phase of the write operation is complete. The memory device may transmit first information for the write operation from a first latch to a volatile memory device in response to suspending the performance of the write operation. The memory device may perform a read operation associated with the read command in response to suspending the performance of the write operation and transferring the first information.