Structure Design and Characteristics of 3T Sense-Switch Pflash for Computing-Inmemory

In this paper, we propose a novel structure for a 3T senseswitch pFlash (SSpF) that is specially designed for computing-in-memory. By leveraging the effect of programming pulse width modulation, the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3 ~\mathrm{T}-\text{SSpF}$</tex> can be adjusted to more than a hundred conductance states. Furthermore, the conductance level and conductance variation of a 256k 3T-SSpF array are evaluated, and a typical neural network is implemented and applied to the MNIST and CIFAR-10 datasets. The network shows reasonable performance, with a loss of accuracy less than 1 %; this approach will therefore enable the realization of more accurate NVM-based neural networks for edge inference.