Paul Uriarte Vicandi | Searcharxiv

arXiv2026

20 ps Non-Destructive Read and 1 ns Write Operations at <5 V in Ferroelectric HfO2/ZrO2 Non-Volatile Memories

Achieving low-voltage, nanosecond multi-level programming and non-destructive read-out of ferroelectric non-volatile memories (NVM) is critical for analog in-memory computing architectures relying on ferroelectric capacitive devices (FeCap). We integrate HfO2/ZrO2 ferroelectric nanolayers concurrently in the BEOL of CMOS and on SiO2/Si, achieving nanosecond multilevel switching with programming voltages below 5 V. Partial ferroelectric switching enhances FeCap endurance above 1011 cycles, leading to MemCapacitance (MC) states with non-destructive read-out and 10-year retention. However, experiments reveal the collapse of the MC window for read frequencies above 1 MHz. To overcome this speed limit, we introduce a novel, non-destructive readout methodology. Using electrical pulses with widths down to 20 ps, below the RC time constant of the FeCaps, we enable measurement of the polarization-dependent leakage current, providing ultrafast and non-destructive read operations at only 14 fJ.

physics.app-ph↗

arXiv2025

Energy-convergence trade off for the training of neural networks on bio-inspired hardware

The increasing deployment of wearable sensors and implantable devices is shifting AI processing demands to the extreme edge, necessitating ultra-low power for continuous operation. Inspired by the brain, emerging memristive devices promise to accelerate neural network training by eliminating costly data transfers between compute and memory. Though, balancing performance and energy efficiency remains a challenge. We investigate ferroelectric synaptic devices based on HfO2/ZrO2 superlattices and feed their experimentally measured weight updates into hardware-aware neural network simulations. Across pulse widths from 20 ns to 0.2 ms, shorter pulses lower per-update energy but require more training epochs while still reducing total energy without sacrificing accuracy. Classification accuracy using plain stochastic gradient descent (SGD) is diminished compared to mixed-precision SGD. We analyze the causes and propose a ``symmetry point shifting'' technique, addressing asymmetric updates and restoring accuracy. These results highlight a trade-off among accuracy, convergence speed, and energy use, showing that short-pulse programming with tailored training significantly enhances on-chip learning efficiency.

cs.ET↗