講演情報
[8a-PB3-3]Temperature Dependence on Pattern Recognition of Y-Doped AlN Memristors for the Applications in Neuromorphic Computing
〇JingEn Lin1, Yi Fu1, JerChyi Wang1,2,3 (1.Chang Gung Univ. for Chang Gung University, 2.Chang Gung Memorial Hospital, 3.Ming Chi Univ. of Tech for Ming Chi University of Technology)
キーワード：
Resistive RAM、Thermal、Artificial neural network system
Heat accumulation can degrade the reliability of high-density neuromorphic devices under high-frequency operation. In this work, Y-doped AlN memristors fabricated on n+-Si substrates with a 10-nm Y-doped AlN resistive-switching layer deposited in an Ar/N2 gas mixture were characterized at different temperature. At elevated temperature, the set/reset voltages and HRS/LRS resistances decreased, indicating thermally assisted conductive filament formation. The spike-timing-dependent plasticity (STDP) response also became less stable at 55 °C, with larger long-term potentiation/depression variations. When the extracted A+/A− and τ+/τ− parameters were used in an unsupervised spiking neural network, the average pattern-recognition accuracy decreased to 72.77% with a 3.69% standard deviation. These results show that self-heating affects both synaptic behavior and learning stability, highlighting the need for thermal-aware memristor design.