US 12,169,636 B2
Computational SSD accelerating deep learning service on large-scale graphs
Myoungsoo Jung, Daejeon (KR); Miryeong Kwon, Daejeon (KR); and Donghyun Gouk, Daejeon (KR)
Assigned to KOREA ADVANCED INSTITUTE OF SCIENCE AND TECHNOLOGY, Daejeon (KR)
Filed by KOREA ADVANCED INSTITUTE OF SCIENCE AND TECHNOLOGY, Daejeon (KR)
Filed on Jan. 9, 2023, as Appl. No. 18/151,645.
Claims priority of application No. 10-2022-0003627 (KR), filed on Jan. 10, 2022; and application No. 10-2022-0133576 (KR), filed on Oct. 17, 2022.
Prior Publication US 2023/0221876 A1, Jul. 13, 2023
Int. Cl. G06F 3/06 (2006.01)
CPC G06F 3/0644 (2013.01) [G06F 3/061 (2013.01); G06F 3/0679 (2013.01)] 15 Claims
OG exemplary drawing
 
1. A computational storage supporting graph machine learning acceleration, the computational storage comprising:
an operation unit disposed near a storage;
a graph storage unit configured to store a graph data set in the storage or provide an interface for accessing the graph data set and output the graph data set and metadata for managing the graph data set;
a graph execution unit configured to convert a graph machine learning model programmed by a host in a form of a data flow graph into a data flow graph having a preset format, download the data flow graph having the preset format to a memory of the operation unit, and execute the downloaded data flow graph to perform graph machine learning preprocessing and graph machine learning inference; and
an accelerator generation unit configured to download a bit file of the host, set a setting memory value based on the bit file to design a hardware logic of the operation unit, and generate a graph machine learning inference accelerator.