PANDA: A 3.178 TOPS/W Reconfigurable Seizure Prediction ANd Detection Neural Network Accelerator for Epilepsy Monitoring
PANDA, a reconfigurable seizure prediction and detection neural network accelerator, is presented. A lightweight twostage seizure monitoring framework with temporal neural network splitting is proposed to be deployed on PANDA. Channel first-output stationary dataflow with zero activation skipping and weight cache with statistical information are employed for higher energy efficiency. A flexible instruction set is defined to make PANDA highly configurable. For seizure monitoring, PANDA achieves up to 99% sensitivity, 0.43/h false alarm rate (FAR), and 3.178 TOPS/W energy efficiency.
Authors
- Xiaoyan Song (ORCID: https://orcid.org/0000-0002-3968-8450)
- Siyuan Qiu (ORCID: https://orcid.org/0000-0003-3631-1313)
- Wenjin Wang (ORCID: https://orcid.org/0009-0005-0826-4355)
- Hailong Jiao (ORCID: https://orcid.org/0000-0002-2815-6168)
- Yuchao Yang (ORCID: https://orcid.org/0000-0003-4674-4059)
- Changchun Zhou (ORCID: https://orcid.org/0009-0005-3968-5048)
- Xinyi Song (ORCID: https://orcid.org/0000-0003-1741-4747)
- Jie Yang (ORCID: https://orcid.org/0009-0008-0873-2560)
- Xiaoyi Song
Institutions
- Peking University (CN)
- Southern University of Science and Technology (CN)
- Nanfang Hospital (CN)
- Peking University Shenzhen Hospital (CN)
- Southern Medical University (CN)
Publication Details
- Published
- 2025-06-08
- DOI
- https://doi.org/10.23919/vlsitechnologyandcir65189.2025.11075036
- Citations
- 1
- Primary Topic
- EEG and Brain-Computer Interfaces
- Type
- article
- Field-Weighted Citation Impact
- 1.43
Funders
- National Natural Science Foundation of China