Presentation
Cryogenic In-Memory Matrix-Vector Multiplication using Ferroelectric Superconducting Quantum Interference Device (FE-SQUID)
SessionBeyond Moore is Cool
DescriptionCryogenic in-memory implementation of machine learning (ML) algorithms can be a game-changer for quantum computing systems. The Ferroelectric Superconducting Quantum Interference Device (FE-SQUID) is promising as a memory element, thanks to its non-volatile nature, voltage-controlled switching, separate read-write paths, and scalability. Here, we demonstrate matrix-vector multiplication (MVM) using FE-SQUID-based cryogenic memory for energy-efficient implementation of ML algorithms. FE-SQUID-based memory allows optimizing the read operation for in-memory computing without imposing any limits on memory performance. We utilize this FE-SQUID-based in-memory MVM to perform a handwritten digit recognition task and achieve an average accuracy of 95% with the MNIST handwritten digit dataset.
Event Type
Research Manuscript
TimeWednesday, July 12th3:55pm - 4:10pm PDT
Location3006, 3rd Floor
Design
Emerging Device Technologies