TY - GEN
T1 - Optimizing Deep Neural Network Precision for Processing-in-Memory
T2 - 2025 International Conference on Electronics, Information, and Communication, ICEIC 2025
AU - Hwang, Inseong
AU - Jang, Jihoon
AU - Kim, Hyun
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - This paper analyzes the detailed measurement of memory bottlenecks in processing-in-memory (PIM) systems on deep neural networks (DNNs) with two precisions (INT8/FP32) by utilizing the memory bottleneck metrics. The impact of INT8, a helpful data movement efficiency improvement on DNN, was examined to determine which precision is more optimal for a PIM system. The results demonstrate that INT8 alleviates the overall memory bottleneck, while LLC MPKI of Softmax with high computational complexity increases from 3.459 to 16.725 and LFMR of the FC layer decreases from 99.795% to 99.483%, but it is hard to expect considerable improvement. For this reason, processing the Softmax and FC layers in PIM when targeting INT8 DNN models is anticipated to enhance performance significantly.
AB - This paper analyzes the detailed measurement of memory bottlenecks in processing-in-memory (PIM) systems on deep neural networks (DNNs) with two precisions (INT8/FP32) by utilizing the memory bottleneck metrics. The impact of INT8, a helpful data movement efficiency improvement on DNN, was examined to determine which precision is more optimal for a PIM system. The results demonstrate that INT8 alleviates the overall memory bottleneck, while LLC MPKI of Softmax with high computational complexity increases from 3.459 to 16.725 and LFMR of the FC layer decreases from 99.795% to 99.483%, but it is hard to expect considerable improvement. For this reason, processing the Softmax and FC layers in PIM when targeting INT8 DNN models is anticipated to enhance performance significantly.
KW - Deep Neural Network
KW - Memory Bottleneck Analysis
KW - Multi-Precision
KW - Processing-In-Memory
UR - http://www.scopus.com/inward/record.url?scp=86000008653&partnerID=8YFLogxK
U2 - 10.1109/ICEIC64972.2025.10879758
DO - 10.1109/ICEIC64972.2025.10879758
M3 - Conference contribution
AN - SCOPUS:86000008653
T3 - 2025 International Conference on Electronics, Information, and Communication, ICEIC 2025
BT - 2025 International Conference on Electronics, Information, and Communication, ICEIC 2025
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 19 January 2025 through 22 January 2025
ER -