"Race Logic" Machines Could Be Key to Solving Computationally-Challenging Problems Efficiently
By using when a bit flips, rather than what it flips to or from, to encode information, race logic could dramatically change computation.
A team at the National Institute of Standards and Technology (NIST) has developed a novel electronic platform which, they say, can solve a range of computationally-challenging problems while drawing a fraction of the energy of current systems: a race logic system.
"Race logic, an arrival-time-coded logic family, has demonstrated energy and performance improvements for applications ranging from dynamic programming to machine learning," the researchers explain in the abstract to their paper. "However, the various ad hoc mappings of algorithms into hardware rely on researcher ingenuity and result in custom architectures that are difficult to systematize."
"We propose to associate race logic with the mathematical field of tropical algebra, enabling a more methodical approach toward building temporal circuits. This association between the mathematical primitives of tropical algebra and generalized race logic computations guides the design of temporally coded tropical circuits. It also serves as a framework for expressing high-level timing-based algorithms."
The race logic system differs from a binary computer, which encodes bits as either 0 or 1, by representing information as time signals - caring more about when a bit is flipped than what it was flipped from or two. The result: A system which requires much less power than its binary equivalent while addressing computationally-challenging work.
"In this work we attempt to make the first steps at generalizability of temporal computing," the researchers write. "We provide a generalizeable data-path and a mathematical algebra, expanding the logical framework of race logic. This leads to novel circuit designs that are informed by higher level algorithmic requirements."
"The properties of abstraction and composability offered by the mathematical framework coupled with native storage from the temporal memory lend themselves to generalization. We design a state machine that can carry out both specialized and general graph algorithms, such as Needleman-Wunsch and Dijkstra’s algorithm, respectively. The potential for general purpose graph accelerators built on temporal computing motivates further exploration of temporal state machines."
The team's work, which has been proven in simulation but not yet instantiated as a physical prototype, has been published under open-access terms on arXiv.org.