Resistor-Capacitor Elements for Neural Calculations
摘要
Implementation of complex psychical functions of an artificial intelligence calls for components with nontrivial time dynamics. At the same time, these components must be compact, energy-efficient and convenient for production of large arrays of elements. A resistor-capacitor cell looks very promising for this sort of elements. The cell consists of a variable resistor and two (serial and parallel) capacitors. The presence of the two capacitors means that there are two additional characteristic times related to two RC circuits. The inclusion of the serial capacitor prevents constant or slowly varying voltages from changing the state of the information carrier – the resistance of the variable resistor. However, our research demonstrates that it can be done with the help of electrical transient processes resulting from voltage changes, or high-frequency harmonic signals. The paper considers this sort of processes and determines the conditions for successful data writing/reading applicable to both single elements and crossbar-type structures. Special attention is given to different roles of the two capacitors. The serial capacitor is a differentiator or high-pass filter that prevents the effect of slow voltage drifts. The parallel capacitor, though shunting the variable resistor, often acts as a short-time storage element affecting the variable resistor – a long-time storage element. The transfer function of a resistor-capacitor cell (at a given resistance of the resistor) is a complex function, which allows us to control not only signal amplitudes, but also their phases during matrix-vector multiplications. Moreover, holographic principles start working, which reveals new opportunities of application of resistor-capacitor structures.