US 11,921,628 B2
Data storage device for determining a write address using a neural network
Junhyeok Jang, Daejeon (KR); Seungkwan Kang, Daejeon (KR); Dongsuk Oh, Daejeon (KR); and Myoungsoo Jung, Daejeon (KR)
Assigned to SK hynix Inc., Icheon (KR); and Korea Advanced Institute of Science and Technology, Daejeon (KR)
Filed by SK hynix Inc., Icheon (KR); and Korea Advanced Institute of Science and Technology, Daejeon (KR)
Filed on Aug. 16, 2022, as Appl. No. 17/889,297.
Claims priority of application No. 10-2022-0033095 (KR), filed on Mar. 17, 2022.
Prior Publication US 2023/0297500 A1, Sep. 21, 2023
Int. Cl. G06F 12/02 (2006.01); G06F 3/06 (2006.01); G06F 12/0831 (2016.01); G06F 12/0882 (2016.01); G06F 12/0891 (2016.01)
CPC G06F 12/0246 (2013.01) [G06F 3/0611 (2013.01); G06F 3/0631 (2013.01); G06F 3/064 (2013.01); G06F 3/0679 (2013.01); G06F 12/0833 (2013.01); G06F 12/0882 (2013.01); G06F 12/0891 (2013.01); G06F 2212/7202 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A data storage device, comprising:
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 a plurality of feature data corresponding to the plurality of unit storage spaces to a neural network to recommend the unit storage space,
wherein the plurality of feature data are generated based on request information for the write request, a target address corresponding to the write request, and an address of data stored in the plurality of unit storage spaces,
wherein the address recommending circuit includes:
an embedding table storing an embedding vector corresponding to the address of the data stored in the plurality of unit storage spaces and generating a target vector corresponding to the target address;
a comparison vector generating circuit configured to generate a plurality of comparison vectors by using information of the plurality of unit storage spaces and the embedding vector stored in the embedding table, the plurality of comparison vectors respectively corresponding to the plurality of unit storage spaces; and
a multilayer neural network circuit configured to generate scores for the plurality of unit storage spaces using the plurality of feature data each being generated using a request vector corresponding to the request information, the target vector, and one of the plurality of comparison vectors.