2024 International Conference on Solid State Devices and Materials

2024 International Conference on Solid State Devices and Materials

Sep 1 - Sep 4, 2024Arcrea HIMEJI
International Conference on Solid State Devices and Materials
2024 International Conference on Solid State Devices and Materials

2024 International Conference on Solid State Devices and Materials

Sep 1 - Sep 4, 2024Arcrea HIMEJI

[B-2-02]1S1R Multi-Level-Cell for Dense Quantized Recurrent Spiking Neural Network Inference Computing

〇Joel Minguet Lopez1, Manon Dampfhoffer2, Gabriele Navarro1, Mathieu Bernard1, Catherine Carabasse1, Niccolo Castellani1, Thomas Magis1, Chiara Sabbione1, Gabriel Molas1, François Andrieu1(1. Univ. Grenoble Alpes, CEA, Leti (France), 2. Univ. Grenoble Alpes, CEA, List (France))
https://doi.org/10.7567/SSDM.2024.B-2-02
We experimentally validated a Multi-Level-Cell (MLC) programming strategy in OxRAM+OTS 1S1R devices. Up to 5-level experimental encoding capabilities per device are demonstrated on 80nm cells, thanks to both 1S1R RESET voltages and SET currents engineering. Moreover, the ability to satisfactory perform a 1S1R MLC read operation is validated, promising a stable read current margin of ~2 decades during 10^9 cycles. Furthermore, the 1S1R MLC pertinence for dense hardware implementation of the synaptic weights of Quantized Recurrent Spiking Neural Networks (QRSNN) for Keyword Spotting (KWS) speech recognition task is demonstrated.