TY - JOUR
T1 - Array-Integrated Memristor with an Interference-Suppressed Pulse Scheme for Multibit Neuromorphic and Edge Computing
AU - Noh, Minseo
AU - Byun, Yongjin
AU - Kim, Gimun
AU - Park, Junhyeok
AU - Kim, Sungjoon
AU - Kim, Sungjun
N1 - Publisher Copyright:
© 2025 American Chemical Society
PY - 2025/9/9
Y1 - 2025/9/9
N2 - In this study, we developed a Pt/Al/TiOy/TiOx/HfO2/Pt memristor device featuring an optimized annealing process and an integrated TiOyovershoot layer to mitigate current overshoot during electroforming, achieving current-compliant-free and forming-free features. Extensive characterization demonstrated stable resistive switching properties, including a high on/off ratio (∼10), reliable retention, and endurance across a 24 × 24 crossbar array. Multilevel cell operation enabled precise programming, achieving up to 6-bit levels through the Incremental Step Pulse with Verify Algorithm (ISPVA) method. The device’s synaptic potential was further evaluated using the Extended Modified National Institute of Standards and Technology (EMNIST) data set. ISPVA-based training achieved superior classification accuracy of 92.6% for a subset (N = 6) and 83.34% for the full alphabet (N = 26), outperforming conventional incremental pulse methods. Furthermore, resistive switching voltage range-based program sequencing makes weight transfer accurate. These findings highlight the Pt/Al/TiOy/TiOx/HfO2/Pt memristor as a core synaptic element for scalable, high-density, and energy-efficient neuromorphic computing systems.
AB - In this study, we developed a Pt/Al/TiOy/TiOx/HfO2/Pt memristor device featuring an optimized annealing process and an integrated TiOyovershoot layer to mitigate current overshoot during electroforming, achieving current-compliant-free and forming-free features. Extensive characterization demonstrated stable resistive switching properties, including a high on/off ratio (∼10), reliable retention, and endurance across a 24 × 24 crossbar array. Multilevel cell operation enabled precise programming, achieving up to 6-bit levels through the Incremental Step Pulse with Verify Algorithm (ISPVA) method. The device’s synaptic potential was further evaluated using the Extended Modified National Institute of Standards and Technology (EMNIST) data set. ISPVA-based training achieved superior classification accuracy of 92.6% for a subset (N = 6) and 83.34% for the full alphabet (N = 26), outperforming conventional incremental pulse methods. Furthermore, resistive switching voltage range-based program sequencing makes weight transfer accurate. These findings highlight the Pt/Al/TiOy/TiOx/HfO2/Pt memristor as a core synaptic element for scalable, high-density, and energy-efficient neuromorphic computing systems.
KW - crossbar array
KW - EMNIST
KW - neuromorphic computing
KW - overshoot layer
KW - synaptic behaviors
UR - https://www.scopus.com/pages/publications/105015475045
U2 - 10.1021/acsaelm.5c01300
DO - 10.1021/acsaelm.5c01300
M3 - Article
AN - SCOPUS:105015475045
SN - 2637-6113
VL - 7
SP - 8211
EP - 8226
JO - ACS Applied Electronic Materials
JF - ACS Applied Electronic Materials
IS - 17
ER -