Accuracy and energetic efficiency of computation in the post-Moore-law era: a Stochastic Thermodynamics approach


CALL: 2017

DOMAIN: MS - New Functional and Intelligent Materials and Surfaces


LAST NAME: Polettini



HOST INSTITUTION: University of Luxembourg

KEYWORDS: Finite-Time ComputationInexact ComputationStochastic ThermodynamicsEnergetic Efficiency

START: 2018-06-01



Submitted Abstract

The process of miniaturization of the microprocessor, which has sustained the tremendous spreading of digital technologies, is slowing down and might eventually come to a halt as it meets its fundamental thermodynamic limits, both in the process of computation and in the process of transportation of information. At the macroscopic level, keeping within an energetic budget is crucial to avoid over-heating at room temperature of personal devices; furthermore, the energy expenditure to maintain major data centers and High Performance Computing facilities cool should remain a small share of the world’s energy consumption. At the microscopic level, as the electronic components become smaller, the operating voltages become comparable to the random voltage generated by thermal noise (the environment), which produces false bit flips and makes computation inaccurate. Therefore, keeping within a fixed energy budget is at the same time a formidable constraint and the occasion to venture into new research directions that demand a better integration between all levels involved in the process, from the algorithmic one, to the technological, to the architecture of the network. In this respect, many recent lines of research propose a slow-down of the computational task and to trade energetic feasibility with accuracy (whenever the tasks need not be too precise). As infinitely-fast computation is accurate but expensive, and infinitely-slow computation is cheap but completely unreliable, there exists a (class of) optimums in between. The main objective of this project is to find and characterize this class. The physical theory that allows to study the trade-off between velocity, thermodynamic efficiency, and accuracy is Stochastic Thermodynamics, which was recently formalized into a complete theory but has among its milestones the work by Johnson and Nyquist on the analysis of electrical noise in circuits. The project’s goal is to develop a set of tools that will allow to make claims about optimal conditions for the “next switch” (any technology that might replace the transistor with a slower technology), or for the next integration step. The project will be based on the mathematics of Langevin systems applied to the electrical elements and circuits that are fairly representative of IT technologies.

This site uses cookies. By continuing to use this site, you agree to the use of cookies for analytics purposes. Find out more in our Privacy Statement