ScholarWorks@성균관대학교

초록

Floating gate memristors (FGMEMs) made of 2-dimensional (2D) materials, operating as two- or multi-terminals to charge and discharge a graphene floating gate (FG), can mimic the functions of synapses and neurons for neuromorphic computing. A large-area chemical vapor deposition for transition metal dichalcogenides (semiconductors) and h-BN (insulators) enables wafer-scale integration. However, it faces challenges with uniformity, thickness control, and oxidation. Here, we demonstrate reliable 4-inch wafer-scale integration of FGMEM arrays using well-established materials of 2D graphene FG, 3D Al2O3 tunneling insulator and 3D ZnO channel. Among 200 random devices in the 4-inch wafer memristor array, 92.5 % exhibit on/off ratios exceeding 103, averaging 106. The bottom graphene FGMEM (B-FGMEM) forms a uniform Al2O3/graphene interface, whereas the step height from patterned ZnO in the top graphene FGMEM (T-FGMEM) results in a rough, incomplete interface. B-FGMEMs demonstrate a retention of 4 × 104 s and endurance of 104 cycles, surpassing the stability of T-FGMEMs by over 10 times. Furthermore, B-FGMEMs exhibit high stability against electrical fatigue of 4000 cycles. In pattern recognition simulations, B-FGMEM achieves a better accuracy of 88.2 % with excellent non-linearity (βp = 1.8, βd = 1.7) compared to the 66.9 % accuracy of T-FGMEM with poor non-linearity (βp = 2.8, βd = 4.6). © 2025 Elsevier B.V.

키워드