Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RRAM-Based Bio-Inspired Circuits for Mobile Epileptic Correlation Extraction and Seizure Prediction

Hao Wang, Lingfeng Zhang, Erjia Xiao, Xin Wang, Zhongrui Wang, Renjing Xu

TL;DR

The paper tackles energy efficiency in mobile seizure prediction by integrating EEG correlation feature extraction with non-volatile RRAM in a bio-inspired hardware circuit. It proposes an 18×18 RRAM overlap-correlation array and a two-layer ANN mapped on chip for seizure prediction, achieving a sensitivity of 91.2% and a FPR/h of 0.11 on CHB-MIT with a lead time of 29.2 minutes. The hardware delivers an area of about 0.83 mm^2, latency of 62.2 μs, and total energy of 1.515 μJ for a 3-second window, corresponding to ~81% energy savings over the best prior approaches. By leveraging a WOx memristor model, DAC-less on-chip computation, and correlation-based features, the work demonstrates a promising direction for low-power neuromorphic EEG processing and can be extended to additional EEG tasks.

Abstract

Non-invasive mobile electroencephalography (EEG) acquisition systems have been utilized for long-term monitoring of seizures, yet they suffer from limited battery life. Resistive random access memory (RRAM) is widely used in computing-in-memory(CIM) systems, which offers an ideal platform for reducing the computational energy consumption of seizure prediction algorithms, potentially solving the endurance issues of mobile EEG systems. To address this challenge, inspired by neuronal mechanisms, we propose a RRAM-based bio-inspired circuit system for correlation feature extraction and seizure prediction. This system achieves a high average sensitivity of 91.2% and a low false positive rate per hour (FPR/h) of 0.11 on the CHB-MIT seizure dataset. The chip under simulation demonstrates an area of approximately 0.83 mm2 and a latency of 62.2 μs. Power consumption is recorded at 24.4 mW during the feature extraction phase and 19.01 mW in the seizure prediction phase, with a cumulative energy consumption of 1.515 μJ for a 3-second window data processing, predicting 29.2 minutes ahead. This method exhibits an 81.3% reduction in computational energy relative to the most efficient existing seizure prediction approaches, establishing a new benchmark for energy efficiency.

RRAM-Based Bio-Inspired Circuits for Mobile Epileptic Correlation Extraction and Seizure Prediction

TL;DR

The paper tackles energy efficiency in mobile seizure prediction by integrating EEG correlation feature extraction with non-volatile RRAM in a bio-inspired hardware circuit. It proposes an 18×18 RRAM overlap-correlation array and a two-layer ANN mapped on chip for seizure prediction, achieving a sensitivity of 91.2% and a FPR/h of 0.11 on CHB-MIT with a lead time of 29.2 minutes. The hardware delivers an area of about 0.83 mm^2, latency of 62.2 μs, and total energy of 1.515 μJ for a 3-second window, corresponding to ~81% energy savings over the best prior approaches. By leveraging a WOx memristor model, DAC-less on-chip computation, and correlation-based features, the work demonstrates a promising direction for low-power neuromorphic EEG processing and can be extended to additional EEG tasks.

Abstract

Non-invasive mobile electroencephalography (EEG) acquisition systems have been utilized for long-term monitoring of seizures, yet they suffer from limited battery life. Resistive random access memory (RRAM) is widely used in computing-in-memory(CIM) systems, which offers an ideal platform for reducing the computational energy consumption of seizure prediction algorithms, potentially solving the endurance issues of mobile EEG systems. To address this challenge, inspired by neuronal mechanisms, we propose a RRAM-based bio-inspired circuit system for correlation feature extraction and seizure prediction. This system achieves a high average sensitivity of 91.2% and a low false positive rate per hour (FPR/h) of 0.11 on the CHB-MIT seizure dataset. The chip under simulation demonstrates an area of approximately 0.83 mm2 and a latency of 62.2 μs. Power consumption is recorded at 24.4 mW during the feature extraction phase and 19.01 mW in the seizure prediction phase, with a cumulative energy consumption of 1.515 μJ for a 3-second window data processing, predicting 29.2 minutes ahead. This method exhibits an 81.3% reduction in computational energy relative to the most efficient existing seizure prediction approaches, establishing a new benchmark for energy efficiency.
Paper Structure (15 sections, 4 equations, 5 figures, 1 table)

This paper contains 15 sections, 4 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Works
  3. Seizure Algorithms
  4. Memristors and Epilepsy Analysis
  5. System Implementation
  6. Memristor Model
  7. RRAM-Based overlap correlation
  8. EEG-Extracting
  9. EEG-Computing
  10. Evaluation
  11. Experimental preparation
  12. Correlation extraction
  13. Seizure Prediction
  14. Energy Estimation
  15. CONCLUSION

Figures (5)

  • Figure 1: Comparison diagram.
  • Figure 2: RRAM Model Simulation Diagram. (a) Graph of the current through the memristor versus the voltage difference across its terminals. (b) Relationship between the current through the memristor and the number of pulses, upon inputting 2000 pulses of 1.6V, 500ns, and 0.8V, 500ns. (c) Diagram of the simulation logic used for (d). (d) Graph showing the relationship between pulse overlap ($\lvert \Delta t \rvert$) and the rate of change in conductance ($\Delta G$).
  • Figure 3: EEG Circuit and System Logic. (a) EEG-Extracting Circuit and Logic. (b) Simulation partial results of the EEG-Extracting phase using RRAM ($G_{18-1}$) in LTspice. (c) Schematic of the ANN and circuit logic of the first layer network. (d) On-chip EEG-Computing system
  • Figure 4: RRAM-Based System Framework. (a) EEG Signal encoding. (b) Correlation Extraction. (c) Seizure Prediction
  • Figure 5: Signal Encoding(a) and Correlation Extraction Example. (b)(c)(d) Predicted as non-disease: (b) shows the voltage difference between Channel-2 and Channel-1 for Patient-1; (c) corresponds to the changes in memristor conductance associated with this voltage difference; (d) displays the correlation map. (e)(f)(g) Predicted as disease: (e) shows the voltage difference between Channel-2 and Channel-1 for Patient-1; (f) corresponds to the changes in memristor conductance associated with this voltage difference; (g) displays the correlation map.