System-Technology Co-Optimization for 3D Monolithic Memory-Centric Computing

Low-dimensional nanostructures for monolithic 3D-integrated flexible and stretchable electronics

“…Additionally, the principal-component analysis which is performed using the off-chip classifier illustrates the good capability of correctly classifying nitrogen and six vapors. Wu et al [276][277][278] proposed the M3D scheme that incorporates Si-based CMOS logic, analog RRAM-based computing-inmemory, and ternary content-addressable memory layers, in which the thermal effect on the accuracy of deep neural networks is evaluated. 279 In the one-/few-shot learning task on the Omniglot dataset, they achieved a high classification accuracy equivalent to that of a graphics processing unit (GPU), reaching up to 97.8%, while consuming 162 times less energy.…”