Stochastic Magnetic Tunnel Junctions to Solve Real-World Optimization Problems
Abstract
The objective of this project is to develop the necessary fundamental and algorithmic knowledge to create approximate classical sampling (ACS) machines that can solve optimization problems relevant to the Department of Defense (DoD) with higher throughput and lower power and area consumption compared to their stochastic and quantum counterparts. We propose to achieve this goal by innovating in algorithms, as well as in stochastic magnetic tunnel junction materials and devices, which, when combined, will lead to significant advances in the hardware implementations of large-scale stochastic binary networks.Current state-of-the-art approaches to solving classically difficult problems include coherent Ising machines based on lasers, D-Wave#s quantum Ising machine, and other Ising machines that use CMOS and different kinds of nano-oscillators. However, the scalability of such approaches and their ability to computewith lower Size, Weight, and Power (SWaP) consumption remains anopen question. To go beyond the state-of-the-art, we propose to implement ACS machines that will tackle factorization-type digital processing problems, as well as Ising-type optimization problems that are of high priority for DoD Tactical Mission systems, such as security, cryptography, realtime sensing, transportation, routing, and many search algorithms. We will use quantum-inspired #reversible# gates (e.g., Toffoli gates) to realize the digital platform. We believe that Toffoli gates at the hardware level will provide building-block functionalities to support a scalable, low-power, andhigh-speed ACS engine. Moreover, the proposed quantum-inspired approaches to ACS will allow us to modularize circuit design, which can simplify the design of ACS hardware.On the materials and devices front, we will focus on Stochastic Magnetic Actuated Random Transducers (SMART), a concept pioneered by the PI. These CMOS-compatible SMART devices will serve as stochastic fluctuating bits (p-bits) for realizing Toffoli gates. SMART p-bits have an active write scheme, well-controlled stochastic properties even in the presence of environmental variations, and can operate in the GHz regime, surpassing the performance of their counterparts. SMART p-bits will thus enable the PIs to more efficiently design the architectures and algorithms and progressively scale the framework to solve more complex problems.The long-term vision of this research is to enable efficient, portable hardware that can solve large-scale optimization problems at room temperature, thereby enhancing and expanding the DoD#s mission capabilities in the future.Approved for Public Release.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 11, 2023
- Source ID
- N000142312771
Entities
People
- Andrew Kent
Organizations
- New York University
- Office of Naval Research
- United States Navy