Use of Magnetoresistive Random-Access Memory as Approximate Memory for Training Neural Networks

2018/10/25 by Nicolas Locatelli, Locatelli, Nicolas, Adrien F. Vincent +3
Computer Science · Engineering · #Advanced Memory and Neural Computing #Applied Physics (physics.app-ph) #Emerging Technologies (cs.ET) #FOS: Computer and information sciences #FOS: Physical sciences #Ferroelectric and Negative Capacitance Devices #Neural Networks and Applications

paper · doi:10.48550/arxiv.1810.10836

openalex publication_date 2018/10/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Hardware neural networks that implement synaptic weights with embedded non-volatile memory, such as spin torque memory (ST-MRAM), are a major lead for low energy artificial intelligence. In this work, we propose an approximate storage approach for their memory. We show that this strategy grants effective control of the bit error rate by modulating the programming pulse amplitude or duration. Accounting for the devices variability issue, we evaluate energy savings, and show how they translate when training a hardware neural network. On an image recognition example, 74% of programming energy can be saved by losing only 1% on the recognition performance.

Related

• Implementing Binarized Neural Networks with Magnetoresistive RAM without Error Correction
• A New MRAM-based Process In-Memory Accelerator for Efficient Neural Network Training with Floating Point Precision
• Evaluation of STT-MRAM as a Scratchpad for Training in ML Accelerators
• Processing-In-Memory Acceleration of Convolutional Neural Networks for Energy-Efficiency, and Power-Intermittency Resilience
• SOT-MRAM-Enabled Probabilistic Binary Neural Networks for Noise-Tolerant and Fast Training