公开(公告)号：US12277073B2

公开(公告)日：2025-04-15

申请号：US18472137

申请日：2023-09-21

Applicant: Kwangwoon University Industry-Academic Collaboration Foundation

Inventor： Young Ho Gong ,  Woo Hyuck Park ,  Ye Bin Kwon ,  Donggyu Sim

IPC: G06F13/00 ,  G06F9/50 ,  G06F13/16

Abstract: According to a quantization interconnect apparatus and an operating method thereof according to the exemplary embodiment of the present disclosure, in a quantization artificial neural network accelerator system, the quantization is performed in the interconnect bus according to a precision without separate processing of the CPU/GPU so that as compared with the quantization by a host processor and an accelerator according to a quantization method of the related art, a number of instructions is reduced to improve the performance/memory efficiency. Further, a computational burden of the host process is reduced to reduce the power and improve the performance.