Scalable Probabilistic Computers for Optimization and Quantum Simulation
Abstract
Approved for Public Release Scalable Probabilistic Computers for Optimization and Quantum Simulation (Camsari)PI: Kerem Camsari, El,ectrical and Computer Engineering, University of California, Santa Barbara Award Number #: N00014-21-S-F008Program Officer: Dr. Ian, Appelbaum The worldwide effort to build quantum computers (QC) is centered around the following challenge: Can QCs solve a computat,ional problem faster, better or more efficiently than all existing classical methods? After decades of research from hardware to qua,ntum algorithms, present-day noisy intermediate scale quantum (NISQ) has now achieved quantum supremacy, albeit in contrived computa,tional tasks. This project aims to develop probabilistic computers (PCs) to tackle a similar question: Can PC so to perform all exis,ting methods to solve some practical problem better, faster or more efficiently than any other method? Unlike all known candidates f,or QC, PCs can operate at room temperature and they can be built in many different hardware platforms including nanoscale devices wi,th demonstrated scalability in present-day technology. Even though PCs do not exhibit the key quantum properties of phase-coherence, and many-body entanglement, correlated p-bits built out of scalable substrates have been shown applicable to training neural networ,ks solving combinatorial optimization problems with industrial relevance, performing fast probabilistic inference, simulating the ph,ysics of quantum systems, all of which are applications envisioned for near-term QCs. From a hardware, architecture and algorithms-l,ayers perspective, our ambitious goal in this project is to investigate whether PCs can outperform all known classical methods for p,ractical problems in combinatorial optimization and quantum simulation. To meet this ambitious goal, we outline three specific objec,tives:(1) Develop a complete hardware architecture for stochastic magnetic tunnel junction (MTJ)-based PCs with physics-based models,, accounting for variations. Due to their dense integrability, CMOS compatibility and unique noise amplification, stochastic MTJs ar,e an ideal choice for nanodevice based PCs.(2) Develop architecture-hardware co-design techniques to drastically improve sampling ra,tes of scalable PCs, achieving ideal parallelism that increases linearly with system size.(3) Demonstrate and benchmark computationa,l advantages of PCs for practical combinatorial optimization and quantum simulation tasks using CMOS emulations of PCs and projectio,ns for MTJ-based PCs. This project builds on the rapidly developing field of domain-specific computation with for computationally ha,rd problems, leveraging emerging nanodevices and existing technology. Preliminary results based on small experimental prototypes and, emulations using CMOS technology suggest performance projections for scaled PCs with orders of magnitude improvements in energy eff,iciency and performance compared to classical computers. Key to these advantages are the room temperature operation, scalability of, magnetic nanodevices and feasible CMOS integration. Froman applied perspective, real world applications of scaled PCs include probl,ems critical for DoD, including logistics / vehicle routing problems, tactical communications and probabilistic decision making (in,ference) under uncertainty. Abstract Scalable Probabilistic Computers for Optimization and Quantum Simulation (Camsari)1
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 08, 2022
- Source ID
- N000142212471
Entities
People
- Kerem Camsari
Organizations
- Office of Naval Research
- United States Navy
- University of California, Santa Barbara