Crossbarred-Memristor Architecture Search
Memristor Crossbar Arrays (MCAs) are customarily implemented in In-Memory Computing (IMC) presently, which enable inference to swift and power-efficient for specialized neural networks and display the potential of memory wall assuagement. However, the co-optimization in Neural Architecture Search (NAS) of NNs and devices restraints demonstrate arduous search space and demanding hardware constraints. The search space traverses kernel size, depth, width and resolution of input; and the hardware constraints extends to circuit topology and computing variation, which result in accuracy degeneration.
In order to tackle these challenges, we proposed a fast customizable method of crossbarred-memristor architecture search, named \textbf{Xmas}. Xmas collectively delves into both aforementioned search space with a special-crafted weighted sampling function and hardware requirements to scout the most robust and suitable neural networks for customizable MCAs. Additionally, device variations are added in the search strategy rather than typically studied in evaluation.