Sculpting Synapses: A Filament Strengthening Reset for Enhanced Neuromorphic Computing

Authors

  • Shehla Yasmeen Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, Virginia 24061, USA
  • Aaron DiFilippo Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, Virginia 24061, USA
  • Marius Orlowski Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, Virginia 24061, USA

DOI:

https://doi.org/10.63002/asrp.406.1835

Keywords:

Neuromorphic Computing, synaptic plasticity, filamentary switching, ReRAM, CBRAM, memristor, filament strengthening

Abstract

The implementation of learning algorithms in neuromorphic hardware demands analog synaptic elements with linear and symmetric weight updates—a longstanding challenge for memristive devices. While filamentary Resistive Random-Access Memory (ReRAM) is a leading candidate, its typical SET and RESET operations provide only coarse control. This work reports a novel constrained RESET operational regime in Cu/TaOx/Pt cells, where a compliance-current-limited RESET unexpectedly strengthens the conductive filament instead of weakening it. This non-destructive phenomenon transforms RESET from a binary erase into a precise synaptic sculpting tool, particularly effective in the high-resistance state where fine tuning is crucial. Through electrostatic simulations, we reveal a self-limiting morphological transition from a conical to a stable hourglass filament shape, explaining the resistance reduction. We demonstrate that by employing the 1T1R structure together with the incremental step program/gate voltage verify algorithms (ISPVA and IGVVA), the strengthening RESET effect can be exploited analogously to the SET regime: it enables symmetric weight updates, facilitates complex learning rules such as spike timing dependent plasticity, and stabilizes faint memories, thereby advancing the development of robust, bio inspired neural networks.

Downloads

Published

17-11-2026