Skip to main content
NASA Advanced Supercomputing (NAS) Division World-class modeling, simulation, and supercomputing in support of the NASA mission

Introduction to MemComputing and its Applications in High Performance Computing

AMS Seminar Series

Speaker: Fabio Lorenzo Traversa, MemComputing, Inc.

Presentation

Seminar Slide Deck (PDF-6.7MB)

Abstract

MemComputing attempts to formalize the general concept of computing with and in memory (or computational memory), as opposed to conventional computing paradigms where the processing unit and memory are assumed physically separated entities exchanging information [1,2]. To study this alternative computing paradigm, we have introduced the Universal Memcomputing Machine (UMM) as the abstract model describing the class of non-von Neumann architectures leveraging the computational memory as central building block [1]. Proven to be equivalent to a non-deterministic Turing machine, the UMM leads to intriguing theoretical conclusions such as efficient solution of non-deterministic polynomial (NP) problems, with particular emphasis on combinatorial optimization [1,2].

The challenge is therefore finding a practical realization of a UMM possessing enough requirements to efficiently solve problems that are combinatorial in nature. To this end, we have introduced the novel, unconventional, computational architecture based on Self-Organizing Gates (SOG), which we call Self-Organizing Circuits (SOCs) [3]. Each SOG is designed to reach an equilibrium if and only if a given relation among terminal states is satisfied. For example, if it is a boolean relation, we have a Self-Organizing Logic Gate [3]. Unlike standard logic gates, they exploit the unique properties of being input/output terminal agnostic. Once assembled to form a Self-Organizing Logic Circuit, they take advantage of the collective state of the system to perform a computation. This creates nonlocal long-range correlations [4] that enable exceptionally fast convergence for solutions of hard combinatorial problems. To illustrate, I will discuss several results using this approach to solve hard instances of MaxSAT problems [5] and Integer linear Programming [6], accelerating the training of deep learning neural networks [7] and challenging problems hoped to be solved one day by quantum computers. These problems are at the core of High-Performance Computing. For example, applications to scheduling a multitude of machines performing multiple jobs, or allocating memory, are important problems for HPC and it will be discussed how to efficiently address them with Memcomputing.

  • [1] F. L. Traversa and M. Di Ventra, Universal Memcomputing Machines, IEEE Transactions on Neural Networks and Learning Systems 26, 2702 (2015).
  • [2] M. Di Ventra, F. L. Traversa, Perspective: Memcomputing: Leveraging memory and physics to compute efficiently, Journal of Applied Physics, 123, 180901 (2018).
  • [3] F. L. Traversa and M. Di Ventra, Polynomial-time solution of prime factorization and NP-complete problems with digital memcomputing machines, Chaos: An Interdisciplinary Journal of Nonlinear Science 27, 023107 (2017).
  • [4] M. Di Ventra, F. L Traversa, I. V. Ovchinnikov, Topological field theory and computing with instantons, Annalen der Physik, 529, 1700123 (2017).
  • [5] F. L. Traversa, P. Cicotti, F. Sheldon, and M. Di Ventra, Evidence of an exponential speed-up in the solution of hard optimization problems, Complexity, 2018, Article ID 7982851 (2018).
  • [6] F. L. Traversa, and M. Di Ventra, MemComputing Integer Linear Programming, arXiv preprint arXiv:1808.09999 (2019).
  • [7] H. Manukian, F. L. Traversa, M. Di Ventra, Accelerating deep learning with memcomputing, Neural Networks, 110, 1-7 (2019).
  • [8] F. Sheldon, F. L. Traversa, M. Di Ventra, Taming a non-convex landscape with dynamical long-range order: memcomputing the Ising benchmarks, Phys. Rev. E 100, 053311 (2019).
  • [9] F. L. Traversa, Aircraft Loading Optimization: MemComputing the 5th Airbus Problem, submitted, arXiv preprint arXiv:1903.08189 (2019).

Biography

Photo of Fabio Lorenzo Traversa
Fabio Lorenzo Traversa serves as the Chief Technology Officer (CTO) of MemComputing, Inc. He co-founded the company in 2016 with fellow physicist Max Di Ventra, and noted San Diego-based tech entrepreneur John Beane. Prior to joining MemComputing, Inc., Traversa served as a research scientist at the University of California, San Diego where he specialized in unconventional computing methods with Di Ventra. Together, they invented the technology known as memcomputing. To date, they’ve received three patents related to the technology with more pending domestically and internationally. Traversa received a Laurea degree in Nuclear Engineering and a Ph.D. in Physics from Politecnico di Torino in 2004 and 2008, respectively. He was a Research Fellow with the Electronics Department at Politecnico di Torino in 2008. From 2009 to 2014, he was with the Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona (UAB), first as a Post-Doctoral Researcher and then as a research fellow. During the same period, he was also a Visiting Researcher at the University of California, San Diego, and at New York University. From 2015 to 2016 he was a Scientist at the University of California, San Diego and in 2016 a Visiting Professor at Politecnico di Torino. Internationally, he is well known for his invited talks on electronics, physics, and unconventional computing. He has authored 80+ scientific papers published in top tier referred journals. His research interests cover unconventional computing, memcomputing, high performance computing, numerical algorithms, computational complexity, circuit theory, circuit design, electronic transport in nano-devices, quantum transport, quantum computing, stability analysis of nonlinear circuits and systems, and noise analysis of nonlinear circuits.