Reconfigurable reservoir computing platform (IMAGE)
Caption
a, Chaotic dynamic trajectory diagram of the Lorenz system, demonstrating the nonlinear characteristics and butterfly effect. b, Schematic of the μ value computation process using 16 kb SRM reservoir computing platform. The precise μ value is calculated by inputting x, y, z time-series data from the Lorenz system using the SRM array. When the calculated μ value deviates from the expected value beyond the preset threshold δ, the system will readjust weights through ridge regression, iteratively optimizing the accuracy of μ value estimation. c, Edge-end prediction application scenarios for complex natural systems, including animal migration, cloud variation, and ocean current disturbance. d, Stand-by units operating in rectification region with ultra-low current output, enabling significant power consumption reduction. e, Processing units showing continuous conductance modulation under pulse signals, achieving 128 discrete resistance states with autonomous conductance recovery at low voltage. These units enable sequence recognition and implement conductance interactions through routing units to establish the connections. f, Routing units exhibiting reversible resistance states: high voltage induces low-resistance state while negative voltage restores initial state.
Credit
©Science China Press
Usage Restrictions
Use with credit.
License
Original content