Fig. 9: Schematic comparison of calculation step between vN and NvN architecture.
From: Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture
Equation (7) is used as an example to indicate the difference between traditional vN architecture (a) and proposed NvN architecture (b) based on the processing-in-memory (PIM) technology. Here xl, wl, bl, and ξl are input, weight, bias, and nonlinear activation function of the lth layer, respectively; pl = xl × wl; sl = bl + pl.
