Memristive Computing for NP-Hard AI Problems

Abstract

The memristor has definitely shown abilities that could revolutionize computing in the coming decades [1][2][3][4]. Its unique adaptive properties are ideal for computational purposes and, so far, they have motivated the exploration of novel computing paradigms [5][6][7]. The pinched current-voltage hysteresis feature indicates the potential of using it in a continuous operational mode as part of an analog computational paradigm [8][9][10][11]. For example, reported properties of network configurations of memristors, as presented in Chap. 7, showed that composite memristive systems significantly improve the efficiency of logic operations via massive analog parallelism, where calculation consists in the evolution of the memristance of all the involved devices. Massive parallelism is commonly found in nature, hence massively parallel computing systems and architectures constitute a great technological engineering challenge [12][13][14].However, in Chap. 7 we extensively discussed the major disadvantages and the computing inefficiencies of memristive networks which are mostly attributed to the dependence of the computing medium behavior on the symmetry of both the underlying circuit geometry and the employed devices. Such inefficiencies were properly treated with the inclusion of asymmetries in the computing structure through a variety of composite memristive components [15][16][17]. Nevertheless, this necessary range of electronic components available so that the requirements of different application are met, reduces significantly the overall structural homogeneity and simplicity of the hardware (HW). Moreover, among the mentioned drawbacks we distinguish: (i) the need to access all the memristors sequentially, both for their initialization and for the read-out of the network state; (ii) the power consumption (the minimum necessary applied input voltage normally varies). It was mentioned, though, that the sparse nature of such memristor network-based computations resembles certain operational features and computing capabilities of

Memristors in Cellular-Automata-Based Computing:A Review

“…Inspired by the theoretical work of Itoh and Chua described above, the authors in [92] described in detail the Memristive Cellular Automata architecture where works like the ones presented in [93][94][95] were analyzed. In these works, the authors have presented several fields where MCAs have been successfully applied, such as shortest-path computing, sorting, and bin-packing problems.…”