Abstract
Neural networks span a wide range of applications of industrial and commercial significance. Binary neural networks (BNN) are particularly effective in trading accuracy for performance, energy efficiency, or hardware/ software complexity. Here, we introduce a spintronic, re-configurable in-memory BNN accelerator, PIMBALL: Processing In Memory BNN AcceL(L)erator, which allows for massively parallel and energy efficient computation. PIMBALL is capable of being used as a standard spintronic memory (STT-MRAM) array and a computational substrate simultaneously. We evaluate PIMBALL using multiple image classifiers and a genomics kernel. Our simulation results show that PIMBALL is more energy efficient than alternative CPU-, GPU-, and FPGA-based implementations while delivering higher throughput.
| Original language | English (US) |
|---|---|
| Article number | A41 |
| Journal | ACM Transactions on Architecture and Code Optimization |
| Volume | 16 |
| Issue number | 4 |
| DOIs | |
| State | Published - Oct 2019 |
Fingerprint
Keywords
- Binary neural networks
- Computational random access memory
- Non-volatile memory
- Processing in memory
Cite this
PIMBALL : Binary neural networks in spintronic memory. / Resch, Salonik; Khatamifard, S. Karen; Chowdhury, Zamshed Iqbal; Zabihi, Masoud; Zhao, Zhengyang; Wang, Jian Ping; Sapatnekar, Sachin S.; Karpuzcu, Ulya R.
In: ACM Transactions on Architecture and Code Optimization, Vol. 16, No. 4, A41, 10.2019.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - PIMBALL
T2 - Binary neural networks in spintronic memory
AU - Resch, Salonik
AU - Khatamifard, S. Karen
AU - Chowdhury, Zamshed Iqbal
AU - Zabihi, Masoud
AU - Zhao, Zhengyang
AU - Wang, Jian Ping
AU - Sapatnekar, Sachin S.
AU - Karpuzcu, Ulya R.
PY - 2019/10
Y1 - 2019/10
N2 - Neural networks span a wide range of applications of industrial and commercial significance. Binary neural networks (BNN) are particularly effective in trading accuracy for performance, energy efficiency, or hardware/ software complexity. Here, we introduce a spintronic, re-configurable in-memory BNN accelerator, PIMBALL: Processing In Memory BNN AcceL(L)erator, which allows for massively parallel and energy efficient computation. PIMBALL is capable of being used as a standard spintronic memory (STT-MRAM) array and a computational substrate simultaneously. We evaluate PIMBALL using multiple image classifiers and a genomics kernel. Our simulation results show that PIMBALL is more energy efficient than alternative CPU-, GPU-, and FPGA-based implementations while delivering higher throughput.
AB - Neural networks span a wide range of applications of industrial and commercial significance. Binary neural networks (BNN) are particularly effective in trading accuracy for performance, energy efficiency, or hardware/ software complexity. Here, we introduce a spintronic, re-configurable in-memory BNN accelerator, PIMBALL: Processing In Memory BNN AcceL(L)erator, which allows for massively parallel and energy efficient computation. PIMBALL is capable of being used as a standard spintronic memory (STT-MRAM) array and a computational substrate simultaneously. We evaluate PIMBALL using multiple image classifiers and a genomics kernel. Our simulation results show that PIMBALL is more energy efficient than alternative CPU-, GPU-, and FPGA-based implementations while delivering higher throughput.
KW - Binary neural networks
KW - Computational random access memory
KW - Non-volatile memory
KW - Processing in memory
UR - http://www.scopus.com/inward/record.url?scp=85073722481&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85073722481&partnerID=8YFLogxK
U2 - 10.1145/3357250
DO - 10.1145/3357250
M3 - Article
AN - SCOPUS:85073722481
VL - 16
JO - ACM Transactions on Architecture and Code Optimization
JF - ACM Transactions on Architecture and Code Optimization
SN - 1544-3566
IS - 4
M1 - A41
ER -