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

2025-11-03

AI Summary

To address the lack of open-source support and non-ideality compensation capabilities for memristor-based compute-in-memory hardware in neuromorphic edge AI, this work designs and implements the first fully open-source memristor interface and computing board. The hardware employs a mixed-signal read-write-verify loop, while the software stack integrates high-level APIs and a chip-in-the-loop fine-tuning framework. We propose a novel voltage-incremental proportional-integral (VIPI) weight mapping method and a closed-loop fine-tuning strategy to enable high-precision analog conductance programming and dynamic compensation of device non-idealities. Experimental validation on MNIST handwritten digit recognition and real-world robot obstacle avoidance demonstrates successful model deployment, on-chip weight loading, and online adaptive learning—comprehensively verifying an end-to-end workflow from offline training to continuous edge-side optimization.

Technology Category

• Machine Learning: Hardware-aware ML
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Learning & Optimization for CV

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Research challenges in human and human-AI computation
• Responsible Web: Machine-in-the-loop, human agency and autonomy

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.

Problem

• Developing an open-source hardware interface for memristor crossbar systems
• Creating software APIs for neuromorphic inference and on-device learning
• Programming analog conductances and mitigating device non-idealities for edge AI

Innovation

• Open-source memristor interfacing system for edge AI
• Hardware interface with mixed-signal read-program-verify loops
• VIPI method programs weights with chip-in-loop fine-tuning