[Paper Review] Metaplastic and Energy-Efficient Biocompatible Graphene Artificial Synaptic Transistors for Enhanced Accuracy Neuromorphic Computing

Dmitry Kireev, Samuel Liu+5 more|arXiv (Cornell University)|Mar 8, 2022

Advanced Memory and Neural Computing63 references106 citations

TL;DR

This paper introduces biocompatible bilayer graphene-based artificial synaptic transistors (BLAST) that mimic synaptic plasticity using a dry Nafion ion-selective membrane, achieving ultra-low energy switching at ~50 aJ/µm²—over ten times more efficient than prior 2D material synapses. The devices exhibit metaplasticity, enabling superior performance in neuromorphic image classification tasks compared to ideal linear synapses, making them strong candidates for energy-efficient, bio-integrated online learning systems.

ABSTRACT

CMOS-based computing systems that employ the von Neumann architecture are relatively limited when it comes to parallel data storage and processing. In contrast, the human brain is a living computational signal processing unit that operates with extreme parallelism and energy efficiency. Although numerous neuromorphic electronic devices have emerged in the last decade, most of them are rigid or contain materials that are toxic to biological systems. In this work, we report on biocompatible bilayer graphene-based artificial synaptic transistors (BLAST) capable of mimicking synaptic behavior. The BLAST devices leverage a dry ion-selective membrane, enabling long-term potentiation, with ~50 aJ/m^2 switching energy efficiency, at least an order of magnitude lower than previous reports on two-dimensional material-based artificial synapses. The devices show unique metaplasticity, a useful feature for generalizable deep neural networks, and we demonstrate that metaplastic BLASTs outperform ideal linear synapses in classic image classification tasks. With switching energy well below the 1 fJ energy estimated per biological synapse, the proposed devices are powerful candidates for bio-interfaced online learning, bridging the gap between artificial and biological neural networks.

Motivation & Objective

• To develop energy-efficient, biocompatible artificial synapses for neuromorphic computing that can interface directly with biological systems.
• To overcome limitations of rigid, toxic materials in existing neuromorphic devices by using flexible, biocompatible 2D materials like graphene and Nafion.
• To achieve sub-femtojoule energy efficiency per synaptic event, approaching biological synapse levels.
• To demonstrate metaplasticity in artificial synapses, enabling adaptive learning superior to linear synaptic models.
• To validate device performance through neuromorphic simulations on standard image classification benchmarks.

Proposed method

• Fabricated macroscale (mBLAST, 10–100 mm²) and microscale (µBLAST, ~400 µm²) BLAST devices using bilayer graphene electronic tattoos transferred onto Nafion-117 membranes.
• Utilized a dry Nafion membrane as a solid polymeric electrolyte with mobile proton clusters to enable protonic gating of graphene channel conductance.
• Applied current pulses through the Nafion to drive movement of positively charged proton clusters, modulating the electrical double layer at the graphene interface and tuning channel conductance.
• Measured conductance changes via source-measure units with 0.1 V drain-source bias and pulsed gate current to emulate synaptic weight updates.
• Conducted pulse trains, relax periods, ramp/level tests, and temperature dependence studies to assess retention, linearity, and stability.
• Performed neuromorphic simulations using CrossSim with stochastic lookup tables to model device nonlinearity, noise, and process variation, training two-layer MLPs on MNIST, Fashion-MNIST, and UCI-HAR datasets.

Experimental results

Research questions

• RQ1 Can biocompatible, flexible 2D material-based synaptic transistors achieve energy efficiency comparable to biological synapses?
• RQ2 Does the use of a dry Nafion membrane enable stable, repeatable, and reversible conductance modulation in graphene-based synaptic devices?
• RQ3 Can metaplasticity in artificial synapses outperform ideal linear synaptic models in deep learning tasks?
• RQ4 How does device performance, including conductance retention and temperature stability, scale across different device sizes and operating conditions?
• RQ5 To what extent can experimental device characteristics be accurately modeled in neuromorphic simulations for real-world classification tasks?

Key findings

• The BLAST devices achieved a switching energy of ~50 aJ/µm², which is at least an order of magnitude lower than previous reports on 2D material-based artificial synapses.
• The devices demonstrated long-term conductance retention with minimal decay over 150–200 seconds after pulse trains, indicating stable synaptic states.
• Conductance changes were highly repeatable and scalable across multiple pulse amplitudes and widths, with symmetric and linear responses across a wide range of conductance levels.
• Neuromorphic simulations showed that metaplastic BLASTs outperformed ideal linear synapses in image classification tasks on MNIST and Fashion-MNIST, demonstrating the advantage of nonlinearity in learning.
• The devices exhibited robust performance across temperatures from 0°C to 80°C, with only minor conductance variations, indicating thermal stability suitable for practical deployment.
• The use of paired BLAST devices to encode synaptic weights via conductance difference reduced dynamic range requirements and minimized noise coupling, improving simulation fidelity.