Fig. 3: System architecture and coarse solver for solving the exponential function.
From: High-concurrency tri-mode memristor-based ordinary differential equation solver
a Our memristor-based ODE solving architecture, which can be reconfigured into coarse solver mode and fine solver mode. b System flowchart for using fixed-point iteration to solve Runge-Kutta methods. c Array mapping diagram of the implicit Gauss-Legendre method of order 6. We use 6 different step sizes to ensure efficient solving. d The result of solving the exponential function ten times using fixed step sizes of 0.1 and 0.5 (with error bars representing the standard deviation), and our developed adaptive stepsize. e Comparison of MSE (mean squared error) and normalized operations solved with three different step sizes. f Comparison of latency and energy consumption of CPU/GPU and MCS at the same accuracy.
