New ThRIve method boosts CNN inference robustness in PIM architectures

Researchers have developed ThRIve, a novel training methodology designed to enhance the thermal robustness of Convolutional Neural Network (CNN) inference on Processing-In-Memory (PIM) architectures. This approach utilizes low-rank adaptation to selectively store parameters on hardware less susceptible to thermal noise, thereby maintaining consistent inference accuracy across a wide operating temperature range. ThRIve-enabled systems have demonstrated accuracy variations within 2% of ideal noise-free performance and offer significant energy-delay product reductions compared to traditional SRAM-based PIM systems. AI

IMPACT Enhances the reliability and efficiency of CNN inference on specialized hardware, potentially enabling wider adoption of PIM for machine learning.

RANK_REASON This is a research paper detailing a new methodology for improving hardware architecture performance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• CNN
• machine learning
• PIM
• Processing in memory
• Sram
• static random-access memory
• ThRIve
• Vibhanshu Sharma

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →