In some implementations, the present disclosure relates to a method. The method includes obtaining a set of weights for a neural network comprising a plurality of nodes and a plurality of connections between the plurality of nodes. The method also includes identifying a first subset of weights and a second subset of weights based on the set of weights. The first subset of weights comprises weights that used by the neural network. The second subset of weights comprises weights that are prunable. The method further includes storing the first subset of weights in a first portion of a memory. A first error correction code is used for the first portion of the memory. The method further includes storing the second subset of weights in a second portion of the memory. A second error correction code is used for the second portion of the memory. The second error correction code is weaker than the first error correction code.

Methods, systems, and devices for full multi-plane operation enablement are described. A flash controller can determine that a first plane of a set of planes of a memory die is an invalid plane. The flash controller can issue a single descriptor associated with a multi-plane operation for the set of planes of the memory die. The single descriptor can include a plurality of commands for the multi-plane operation in which the first command of the plurality of commands can be a duplicate of a second command of the plurality of commands based on the first plane being the invalid plane. In some cases, a negative-and (NAND) controller can receive the single descriptor associated with the multi-plane operation for the set of planes of a memory die. The NAND controller can issue a plurality of commands for the multi-plane operation based on receiving the single descriptor.

A three-dimensional stacked integrated circuit (3D SIC) that can have at least a first 3D XPoint (3DXP) die and, in some examples, can have at least a second 3DXP die too. In such examples, the first 3DXP die and the second 3DXP die can be stacked. The 3D SIC can be partitioned into a plurality of columns that are perpendicular to each of the stacked dies. In such examples, when a first column of the plurality of columns is determined as failing, data stored in the first column can be replicated to a second column of the plurality of columns. Also, for example, when a part of a first column of the plurality of columns is determined as failing, data stored in the part of the first column can be replicated to a corresponding part of a second column of the plurality of columns.

A solid state drive having a drive aggregator and a plurality of component solid state drive, including a first component solid state drive and a second component solid state drive. The drive aggregator has at least one host interface, and a plurality of drive interfaces connected to the plurality of component solid state drives. The drive aggregator is configured to generate, in the second solid state drive, a copy of a dataset that is stored in the first component solid state drive. In response to a failure of the first component solid state drive, the drive aggregator is configured to substitute a function of the first component solid state drive with respect to the dataset with a corresponding function of the second component solid state drive, based on the copy of the dataset generated in the second component solid state drive.

A method for accessing a memory module includes; encoding first data of a first partial burst length to generate first parities and first cyclic redundancy codes, encoding second data of a second partial burst length to generate second parities and second cyclic redundancy codes, writing the first data and the second data to first memory devices, and writing the first parities, the first cyclic redundancy codes, the second parities, and the second cyclic redundancy codes to a second memory device and a third memory device.

Storage devices include a memory array which stores host data received from a host computing device. During normal operations, the storage device may encounter a fatal error which can halt functionality. To restore functionality without system disconnection or third-party interventions, the storage device can store recovery data within a host memory buffer prior to encountering a fatal error. The recovery data can be replay protected memory data and/or firmware recovery data that can be written to the host memory buffer upon power on or during a firmware update. When a fatal error occurs, the recovery data can be accessed to try and rebuild file and mapping systems to restore full operation of the storage device. When full operational restoration is not possible, host data can at least be copied from the storage device prior to utilizing firmware recovery data to restore the storage device to an erased but functional state.

A system can respond to detection or prediction of an uncorrectable error (UE) in memory based on fault-aware analysis. The fault-aware analysis enables the system to generate a determination of a specific hardware element of the memory that is faulty. In response to detection of an error, the system can correlate a hardware configuration of the memory device with historical data indicating memory faults for hardware elements of the hardware configuration. Based on a determination of the specific component that likely caused the UE, the system can identify a region of memory associated with the detected UE and mirror the faulty region to a reserved memory space of the memory device for access to data of the faulty region.

A workgroup-computing-entity-based fail-safe/evolvable hardware core structure is disclosed which includes a 3-hierarchical-level 6-workgroup-Basic-Building-Block (6-wBBB) created to supplant the node-computing-entity-based non-fail-safe/limited evolvable von-Neumann core structure of 3-hierarchical-level 3-node-BBB, (i.e., base-level IO-devices/mid-level main memory/top-level CPU) and all the first-time fail-safe workgroup systems can be subsequently generated in the second period along the workgroup-computing evolutionary timeline. Furthermore, based on the first 6-wBBB evolvable architecture, the workgroup evolutionary processes can go up to 7 generations in creating all the necessary workgroup-computing entity-based hardware core structures, so that all the real-time intelligent workgroup-computing systems can be generated in the third period along the workgroup-computing evolutionary timeline.

Workloads, e.g., synthetic workloads, on one or more storage systems in an dynamic, automated manner, for example, to load test the one or more storage systems. A distributed system may be employed in which a workload information server (WIS) serves one or more clients referred to herein as workload control components (WCCs) that analyze workload information of the one or more storage systems, and control the modification of workloads thereon based on this analysis, through the WIS. The WIS also may serve one or more clients referred to herein as workload generation controllers (WGCs) that monitor workloads on the one or more storage systems, report workload information to the WIS and generate, modify or remove workloads on the one or more storage systems according to instructions received from the WIS in response to requests (e.g., hints) from the one or more WGCs.

Runtime memory cell row defect detection and replacement includes detecting in a memory of a computer system operating in a runtime operating system mode, a defective row of memory cells having at least one defective cell. In response to the detection of the defective row, interrupting the operating system of the computer system and, in a runtime system maintenance mode, replacing the defective row of memory cells with a spare row of memory cells as a replacement row of memory cells. Execution of the operating system is then resumed in the runtime operating system mode Other aspects and advantages are described.