Multibit neural inference in a N-ary crossbar architecture

2026-04-28

AI Summary

This work addresses the challenges of quantization error, system non-idealities, and stochastic noise in multi-bit neural network inference using N×N crossbar arrays by proposing an analog in-memory computing framework based on magnetic tunnel junctions (MTJs). The framework supports matrix-vector multiplication and integrates principal component analysis (PCA) for dimensionality reduction, weight quantization, and a comprehensive noise model. The study reveals that systematic errors dominate overall performance degradation, whereas device-specific random noise has a comparatively minor impact. Leveraging this insight, the authors determine an optimal number of states per cell that balances quantization error against resistance resolution to minimize total inference error. Evaluated on a 4×4 MTJ array, the approach achieves 94.48% accuracy on MNIST classification, approaching the software baseline of 97.56% and significantly narrowing the hardware-software performance gap.

Abstract

In-memory computing (IMC) enables energy-efficient neural network inference by computing analog matrix-vector multiplications (MVM) in memory crossbar arrays. In this work we present a simulation framework for N-ary crossbar architectures that retrieves MVM results with minimal implementation assumptions. The XOR and MNIST classification tasks were successfully inferred using a simulated crossbar array of (4x4) 4-states magnetic tunnel junctions (MTJ). MNIST accuracy reached 94.48% (vs. 97.56% software baseline). The software-hardware performance gap was further reduced using PCA dimensionality reduction. We identified weight quantization as the primary error source, and studied its impact alongside systematic nonidealities and random noise. We find that cell-specific random noise is less detrimental than systematic errors due to averaging across the array. Finally, we demonstrate an optimal number of states per cell that balances quantization error against resistance state resolution to minimize total MVM error.

Problem

• in-memory computing
• neural inference
• crossbar array
• weight quantization
• matrix-vector multiplication

Innovation

• in-memory computing
• N-ary crossbar
• multibit neural inference
• magnetic tunnel junction
• weight quantization