Editorial

Novel computing paradigms: Quo vadis?

Unconventional computation (also non-classical, novel, or emerging computation) [4], [6], [11], [20] is a broad and interdisciplinary research area with the main goal to go beyond the standard models and practical implementations of computers, such as the von Neumann computer architecture and the abstract Turing machine, which have dominated computer science for more then half a century. This quest, in both theoretical and practical dimensions, is motivated by a number of trends. First, it is expected that, without disruptive new technologies, the ever-increasing computing performance and storage capacity achieved with existing technologies will eventually reach a plateau. The main reason for this is fundamental physical limits on the miniaturization of today’s silicon-based electronics (see, e.g. [14]). Second, novel ways to synthetically fabricate chemical and biological assemblies, for example through self-assembly, self-replication (e.g. [9]), or bio-engineering (e.g. [5], [21]) allow one to create systems of unimagined complexity. However, we currently lack the methodologies and the tools to design and program such massively parallel and spatially extended unconventional “machines.” Third, many of today’s most important computational challenges, such as for example understanding complex biological and physical systems by simulations or identifying significant features in large, heterogeneous, and unstructured data sets, may not be well suited for classical computing machines. That is, while a classical Turing-universal computer, at least from a theoretical perspective, can in principle solve all of these challenging problems (as any other algorithmic problem), the general hope is that unconventional computers might solve them much more efficiently, i.e. orders of magnitude faster and using much less resources.

Not everything that looks like a computation in a physical system is useful and not everything that does some information processing can be used to solve problems. A common, if slightly abused, example is that of a falling ball, which can be interpreted as an “unconventional” computer that solves the second order differential equation of Newton’s second law. As a matter of fact, a significant research effort has been spent on similar examples, with the goal to characterize the types of computations, i.e. the laws governing the underlying dynamics behind various physical phenomena. However, a falling ball is a pretty useless computer that can only solve one particular equation with different initial conditions. Interpreting the solution, storing and recalling it, and interfacing the computing unit with other units to perform further computations, is virtually impossible. Thus, while most physical systems solve some equations and most biological organisms process information in some way, we should refrain from calling these systems “computers” until we can harness and interpret the underlying processes with a specific computation in mind. Given a physical, a biological, or a chemical system that is supposed to act as a computer, the question is not only what, if anything, this system computes, but also, and more importantly, What are the characteristics of such a computation? (in terms of speed, size, integration density, or power consumption)? What are the limitations? What kind of problems can be solved and how efficiently? How can we “program” the system to perform a specific computation? and How can we interface the result of the computation with traditional computers to post-process, analyze, and store it?