Publications by authors named "Vinayak Bharat Naik"
Article Synopsis
- - Memristors are promising for efficient multiply-accumulate (MAC) operations in analog in-memory computing (AiMC), but variations can impact accuracy, typically addressed with error-prone on-chip training that can wear out memristors.
- - This study introduces a hardware-software codesign using magnetic tunnel junctions (MTJs) that allows off-chip calibration, maintaining software accuracy without the drawbacks of on-chip training while demonstrating low variations across mass-produced devices.
- - The proposed off-chip training method adjusts deep neural network parameters for accurate AiMC inference, showing improved performance in MAC operations and achieving results comparable to software baselines, demonstrating MTJs' potential for AI applications.
