Skhynix Inc.

Korea Advanced Institute of Science and Technology

Yonsei University

Junhyeok Jang

Myoungsoo Jung

Seungkwan Kang

Miryeong Kwon

Donghyun Gouk

Dongsuk Oh

Hyunkyu Choi

Provided is an apparatus for accelerating graph neural network (GNN) pre-processing, the apparatus including a set-partitioning accelerator configured to sort each edge of an original graph stored in a coordinate list (COO) format by a node number, perform radix sorting based on a vertex identification (VID) to generate a COO array of a preset length, and perform uniform random sampling on some nodes of a given node array, a merger configured to merge the COO array of the preset length to generate one sorted COO array, a re-indexer configured to assign new consecutive VIDs respectively to the nodes selected through the uniform random sampling, and a compressed sparse row (CSR) converter configured to the edges sorted by the node number into a CSR format.

Method and apparatus for accelerating GNN pre-processing

A data storage device includes one or more nonvolatile memory devices each including a plurality of unit storage spaces; and an address recommending circuit configured to recommend a unit storage space among the plurality of unit storage spaces to process a write request, wherein the address recommending circuit applies feature data to a neural network to recommend the unit storage space, and wherein the feature data is generated based on request information for the write request, a target address corresponding to the write request, an address of data stored in the plurality of unit storage spaces.

Data storage device for determining a write address using a neural network

An electronic device is provided. A memory controller, having an improved response time for an input/output request and increased capacity of Dynamic Random Access Memory (DRAM) according to the present disclosure, includes an available-time prediction component configured to perform a machine learning operation using a Recurrent Neural Network (RNN) model based on input/output request information about an input/output request input from a host, and to predict an idle time representing a time during which the input/output request is not expected to be input from the host and a data compression controller configured to generate, in response to the idle time longer than a set reference time, compressed map data by compressing map data which indicates a mapping relationship between a logical address provided by the host and a physical address indicating a physical location of a memory block included in the memory device.

Memory controller and operating method thereof

Growing community of inventors

Daejeon, South Korea

Junhyeok Jang