Abstract: A neural network accelerator includes a control circuit configured to control a learning operation for a neural network by performing a plurality of learning steps; an operation processor configured to perform the learning operation under the control of the control circuit; and an operation memory storing the embedding table that has a plurality of embedding entries and coupled to the operation processor, wherein the operation processor performs a first embedding operation using an embedding entry required for a current learning step, and performs a second embedding operation using an embedding entry that is required for a next learning step and is not affected by the current learning step.

Abstract: A device transmits or receives a packet in a memory network including one or more processors and/or one or more memory devices. The device includes a key storage unit configured to store a one-time password (OTP) key that is shared with a target node, an encryption unit configured to encrypt a transmission packet with the OTP key stored in the key storage unit and to transmit the encrypted transmission packet to the target node, and a decryption unit configured to decrypt a receiving packet from the target node with the OTP key stored in the key storage unit. The device is a processor or a memory device in the memory network.

Abstract: An apparatus and a job scheduling method are provided. For example, the apparatus is a multi-core processing apparatus. The apparatus and method minimize performance degradation of a core caused by sharing resources by dynamically managing a maximum number of jobs assigned to each core of the apparatus. The apparatus includes at least one core including an active cycle counting unit configured to store a number of active cycles and a stall cycle counting unit configured to store a number of stall cycles and a job scheduler configured to assign at least one job to each of the at least one core, based on the number of active cycles and the number of stall cycles. When the ratio of the number of stall cycles to a number of active cycles for a core is too great, the job scheduler assigns fewer jobs to that core to improve performance.

Abstract: A memory network includes a first local memory group, a second local memory group, and multiple first channels. The first local memory group includes multiple first memory devices that are not directly connected to each other. The second local memory group includes multiple second memory devices that are not directly connected to each other. The first channels are configured to connect the first memory devices to the second memory devices in a one to one relationship.

Abstract: A memory device includes a memory cell region including a plurality of memory cells; a memory cell controller configured to control read and write operation for the memory cell region; one or more NDP engines configured to perform a near data processing (NDP) operation for the memory cell region; a command buffer configured to store an NDP command transmitted from a host; and an engine scheduler configured to schedule the NDP operation for the one or more NDP engines according to the NDP command.

Abstract: A device transmits or receives a packet in a memory network including one or more processors and/or one or more memory devices. The device includes a key storage unit configured to store a one-time password (OTP) key that is shared with a target node, an encryption unit configured to encrypt a transmission packet with the OTP key stored in the key storage unit and to transmit the encrypted transmission packet to the target node, and a decryption unit configured to decrypt a receiving packet from the target node with the OTP key stored in the key storage unit. The device is a processor or a memory device in the memory network.

Abstract: A memory device includes a memory cell region including a plurality of memory cells; a memory cell controller configured to control read and write operation for the memory cell region; one or more NDP engines configured to perform a near data processing (NDP) operation for the memory cell region; a command buffer configured to store an NDP command transmitted from a host; and an engine scheduler configured to schedule the NDP operation for the one or more NDP engines according to the NDP command.

Abstract: A processor and corresponding method are described including cores having a thread set allocated based on a pre-set implementation order, and a controller configured to receive scheduling information determined based on an implementation pattern regarding the allocated thread set from one of the cores and transmit the scheduling information to another of the cores. The one of cores determines the scheduling information according to characteristics of an application when implementation of the thread set is completed. Each of the cores re-determines an implementation order regarding the allocated thread set based on the determined scheduling information.

Abstract: An apparatus and a job scheduling method are provided. For example, the apparatus is a multi-core processing apparatus. The apparatus and method minimize performance degradation of a core caused by sharing resources by dynamically managing a maximum number of jobs assigned to each core of the apparatus. The apparatus includes at least one core including an active cycle counting unit configured to store a number of active cycles and a stall cycle counting unit configured to store a number of stall cycles and a job scheduler configured to assign at least one job to each of the at least one core, based on the number of active cycles and the number of stall cycles. When the ratio of the number of stall cycles to a number of active cycles for a core is too great, the job scheduler assigns fewer jobs to that core to improve performance.

Abstract: A memory network includes a first local memory group, a second local memory group, and multiple first channels. The first local memory group includes multiple first memory devices that are not directly connected to each other. The second local memory group includes multiple second memory devices that are not directly connected to each other. The first channels are configured to connect the first memory devices to the second memory devices in a one to one relationship.

Abstract: A processor and corresponding method are described including cores having a thread set allocated based on a pre-set implementation order, and a controller configured to receive scheduling information determined based on an implementation pattern regarding the allocated thread set from one of the cores and transmit the scheduling information to another of the cores. The one of cores determines the scheduling information according to characteristics of an application when implementation of the thread set is completed. Each of the cores re-determines an implementation order regarding the allocated thread set based on the determined scheduling information.

Abstract: A multi-core apparatus includes cores each including an active cycle counting unit configured to store an active cycle count, and a stall cycle counting unit configured to store a stall cycle count. The multi-core apparatus further includes a job scheduler configured to determine an optimal number of cores in an active state based on state information received from each of the cores, and adjust power to maintain the optimal number of cores.

Abstract: An apparatus and a job scheduling method are provided. For example, the apparatus is a multi-core processing apparatus. The apparatus and method minimize performance degradation of a core caused by sharing resources by dynamically managing a maximum number of jobs assigned to each core of the apparatus. The apparatus includes at least one core including an active cycle counting unit configured to store a number of active cycles and a stall cycle counting unit configured to store a number of stall cycles and a job scheduler configured to assign at least one job to each of the at least one core, based on the number of active cycles and the number of stall cycles. When the ratio of the number of stall cycles to a number of active cycles for a core is too great, the job scheduler assigns fewer jobs to that core to improve performance.

Abstract: An apparatus and a job scheduling method are provided. For example, the apparatus is a multi-core processing apparatus. The apparatus and method minimize performance degradation of a core caused by sharing resources by dynamically managing a maximum number of jobs assigned to each core of the apparatus. The apparatus includes at least one core including an active cycle counting unit configured to store a number of active cycles and a stall cycle counting unit configured to store a number of stall cycles and a job scheduler configured to assign at least one job to each of the at least one core, based on the number of active cycles and the number of stall cycles. When the ratio of the number of stall cycles to a number of active cycles for a core is too great, the job scheduler assigns fewer jobs to that core to improve performance.

Abstract: An apparatus and a job scheduling method are provided. For example, the apparatus is a multi-core processing apparatus. The apparatus and method minimize performance degradation of a core caused by sharing resources by dynamically managing a maximum number of jobs assigned to each core of the apparatus. The apparatus includes at least one core including an active cycle counting unit configured to store a number of active cycles and a stall cycle counting unit configured to store a number of stall cycles and a job scheduler configured to assign at least one job to each of the at least one core, based on the number of active cycles and the number of stall cycles. When the ratio of the number of stall cycles to a number of active cycles for a core is too great, the job scheduler assigns fewer jobs to that core to improve performance.

Abstract: A multi-core apparatus includes cores each including an active cycle counting unit configured to store an active cycle count, and a stall cycle counting unit configured to store a stall cycle count. The multi-core apparatus further includes a job scheduler configured to determine an optimal number of cores in an active state based on state information received from each of the cores, and adjust power to maintain the optimal number of cores.

Abstract: An apparatus and a job scheduling method are provided. For example, the apparatus is a multi-core processing apparatus. The apparatus and method minimize performance degradation of a core caused by sharing resources by dynamically managing a maximum number of jobs assigned to each core of the apparatus. The apparatus includes at least one core including an active cycle counting unit configured to store a number of active cycles and a stall cycle counting unit configured to store a number of stall cycles and a job scheduler configured to assign at least one job to each of the at least one core, based on the number of active cycles and the number of stall cycles. When the ratio of the number of stall cycles to a number of active cycles for a core is too great, the job scheduler assigns fewer jobs to that core to improve performance.