Analog in-memory calculation-oriented input range expansion method

Abstract

The invention discloses an input range expansion method oriented to simulation in-memory computing, and belongs to the technical field of in-memory computing. According to the method, an original large-range input signal is split, the original signal is split into small signals which can be mapped to a voltage range allowed by the memristor by setting different mapping multiplying powers, when the amplitude of the input signal is large, precision loss caused by single scaling is avoided through a hierarchical mapping mode, and the accuracy of the memristor is improved. Meanwhile, the mapped signal is ensured not to cause the conductance state drift of the device; according to the method, the input range can be effectively expanded under the condition that the stability of the device is not damaged, and relatively high calculation precision and energy efficiency ratio can be kept. The method is suitable for all memristive devices which are limited by an input voltage range in analog calculation; the method is suitable for various analog calculation and storage calculation integrated accelerators based on memristors, and has high universality and application value.

Images