Daniel Saunders survey was easy to read, despite its 22 pages. It is really more of a survey than a review, just like its title says. It was initially released on 23 Nov 2016, but published at arXiv on 18 Apr 2017 (and on 27 Apr 2017: "Withdrawing since this submission was in bad taste").

It starts the introduction with: "The memcomputing model recently introduced by di Ventra, et al. [1] provides a novel way of harnessing nonlinear circuits to perform massively parallel computation." After some details, this point is reinforced again: "A memcomputing device therefore has no need for a separate CPU, and may save on computation time (due to overcoming the von Neumann bottleneck [8]) and energy expenditure."

It then quickly comes to the heart of the matter: "It is claimed in [3] that the digital memcomputing machine is able to solve general NP problems with only polynomial resources (i.e., polynomial time, space, and energy use), ... It is further argued that there exists a simulation of the digital memcomputing machine (...) implementing an NP algorithm which has only a polynomial blowup in running time, and may therefore resolve the question of P vs. NP in favor of equality."

So the introduction concludes with: "This claim is surprising, as the P vs. NP problem has been open for decades, withstanding a great deal of complexity theory research. This paper therefore serves to provide skepticism on this point, and to consider the relation of the memcomputing concept to classical computation models in general." before outlining the organization of the paper.