Neuromorphic Circuit Simulation with Memristors: Design and Evaluation Using MemTorch for MNIST and CIFAR

TL;DR

The paper investigates the feasibility of memristor-based in-memory neuromorphic computing for CNN inference on MNIST, CIFAR10, and CIFAR100 using Memtorch. The authors train lightweight PyTorch CNNs, patch them to memristive crossbars with a VTEAM memristor model, and systematically vary max input voltage, tile size, and ADC resolution while simulating non-idealities such as stochasticity, device faults, endurance, and retention. They report near-digital inference accuracy under ideal/near-ideal conditions (losses < ~1%), and provide a detailed analysis of how hardware non-idealities influence performance, identifying practical design choices (e.g., 64×64 tiles, 8-bit ADC, MIV ≈ 6–9V) that balance accuracy and speed. The results demonstrate Memtorch’s utility for evaluating memristive neuromorphic architectures and quantify the practical impact of device non-idealities, informing design trade-offs for energy-efficient in-memory CNN accelerators.

Abstract

Memristors offer significant advantages as in-memory computing devices due to their non-volatility, low power consumption, and history-dependent conductivity. These attributes are particularly valuable in the realm of neuromorphic circuits for neural networks, which currently face limitations imposed by the Von Neumann architecture and high energy demands. This study evaluates the feasibility of using memristors for in-memory processing by constructing and training three digital convolutional neural networks with the datasets MNIST, CIFAR10 and CIFAR100. Subsequent conversion of these networks into memristive systems was performed using Memtorch. The simulations, conducted under ideal conditions, revealed minimal precision losses of nearly 1% during inference. Additionally, the study analyzed the impact of tile size and memristor-specific non-idealities on performance, highlighting the practical implications of integrating memristors in neuromorphic computing systems. This exploration into memristive neural network applications underscores the potential of Memtorch in advancing neuromorphic architectures.

Figures (16)

• Figure 1: Basic Layout of Von Neumann Architecture.
• Figure 2: Architecture of a Multi-Layer Perceptron (MLP).
• Figure 3: Architecture of a VGG-style CNN.
• Figure 4: Memristor Circuit Symbol.
• Figure 5: Memristor states and switching mechanism.