Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenMENA: An Open-Source Memristor Interfacing and Compute Board for Neuromorphic Edge-AI Applications

Ali Safa, Farida Mohsen, Zainab Ali, Bo Wang, Amine Bermak

TL;DR

OpenMENA addresses the need for an open, reproducible memristor-based edge AI platform by delivering a complete hardware-software stack, including an 8×8 memristor crossbar interface, a firmware-software stack, and a VIPI-based weight programming method. It combines constrained optimization-based training to learn positive weights with a voltage-incremental programming scheme and chip-in-the-loop fine-tuning to mitigate hardware non-idealities, and validates end-to-end on digit recognition and robot control tasks. The results demonstrate the feasibility of end-to-end training and adaptation on memristor crossbars at the edge, highlighting the platform’s potential to democratize memristor-enabled continual learning and local plasticity-enabled edge AI.

Abstract

Memristive crossbars enable in-memory multiply-accumulate and local plasticity learning, offering a path to energy-efficient edge AI. To this end, we present Open-MENA (Open Memristor-in-Memory Accelerator), which, to our knowledge, is the first fully open memristor interfacing system integrating (i) a reproducible hardware interface for memristor crossbars with mixed-signal read-program-verify loops; (ii) a firmware-software stack with high-level APIs for inference and on-device learning; and (iii) a Voltage-Incremental Proportional-Integral (VIPI) method to program pre-trained weights into analog conductances, followed by chip-in-the-loop fine-tuning to mitigate device non-idealities. OpenMENA is validated on digit recognition, demonstrating the flow from weight transfer to on-device adaptation, and on a real-world robot obstacle-avoidance task, where the memristor-based model learns to map localization inputs to motor commands. OpenMENA is released as open source to democratize memristor-enabled edge-AI research.

OpenMENA: An Open-Source Memristor Interfacing and Compute Board for Neuromorphic Edge-AI Applications

TL;DR

OpenMENA addresses the need for an open, reproducible memristor-based edge AI platform by delivering a complete hardware-software stack, including an 8×8 memristor crossbar interface, a firmware-software stack, and a VIPI-based weight programming method. It combines constrained optimization-based training to learn positive weights with a voltage-incremental programming scheme and chip-in-the-loop fine-tuning to mitigate hardware non-idealities, and validates end-to-end on digit recognition and robot control tasks. The results demonstrate the feasibility of end-to-end training and adaptation on memristor crossbars at the edge, highlighting the platform’s potential to democratize memristor-enabled continual learning and local plasticity-enabled edge AI.

Abstract

Memristive crossbars enable in-memory multiply-accumulate and local plasticity learning, offering a path to energy-efficient edge AI. To this end, we present Open-MENA (Open Memristor-in-Memory Accelerator), which, to our knowledge, is the first fully open memristor interfacing system integrating (i) a reproducible hardware interface for memristor crossbars with mixed-signal read-program-verify loops; (ii) a firmware-software stack with high-level APIs for inference and on-device learning; and (iii) a Voltage-Incremental Proportional-Integral (VIPI) method to program pre-trained weights into analog conductances, followed by chip-in-the-loop fine-tuning to mitigate device non-idealities. OpenMENA is validated on digit recognition, demonstrating the flow from weight transfer to on-device adaptation, and on a real-world robot obstacle-avoidance task, where the memristor-based model learns to map localization inputs to motor commands. OpenMENA is released as open source to democratize memristor-enabled edge-AI research.

Paper Structure

This paper contains 12 sections, 3 equations, 7 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. System Design
  3. Memristor control and interfacing circuit
  4. Memristor conductance control and readout signals
  5. Training and Weight Setting Algorithms
  6. Software model training via constrained optimization
  7. Writing weights to the OpenMENA board
  8. Chip-in-the-loop fine-tuning
  9. Experimental Results
  10. Digit classification
  11. Robot obstacle avoidance
  12. Conclusion

Figures (7)

  • Figure 1: The proposed OpenMENA system for memristor crossbar control and interfacing. The OpenMENA PCB is mounted on an Arduino Due board for general purpose digital control. An 8-by-8 knowm memristor crossbar is mounted on the OpenMENA socket. The complete system is interface via a companion python library featuring both inference and weight setting functionalities. All code and design files are released as open-source to help democratize research in memristor-based neuromorphic AI.
  • Figure 2: Block diagram of the proposed OpenMENA system.
  • Figure 3: Memristor crossbar control and readout circuit.
  • Figure 4: Bipolar memristor conductance control strategy.
  • Figure 5: Digit classification test accuracy in function of the model's output decision threshold. Test accuracy is reported both for our proposed VIPI method and for conventional PI control.
  • ...and 2 more figures