CAREER: Scalable Physics-Inspired Ising Computing for Combinatorial Optimizations

University Of California-Santa Barbara, Santa Barbara CA

Investigators

Abstract

Ising computing is an alternative computing paradigm inspired by the natural physical phenomena of ferromagnetism among atomic spins. In this unconventional computing approach, artificial spins organized in a graph dynamically interact, propelling the system toward rapid convergence to the minimum energy state – a representation of the optimal solution. The Ising computer, leveraging such convergence behavior, exhibits exponential acceleration compared to classical counterparts, particularly excelling in solving intricate optimization problems across diverse sectors such as logistics, manufacturing, supply chain management, drug discovery, and financial portfolio optimization. Despite ongoing efforts to develop Ising computers using classical and emerging technologies, none have effectively addressed critical challenges related to scalability, reconfigurability, and connectivity – essential factors for realizing practical Ising computing solutions. This project aims to tackle these challenges by constructing mixed-signal and digital application-specific integrated circuit (ASIC) hardware accelerators. The project will integrate research and education by introducing new undergraduate and graduate level courses at the university, by providing opportunity to work on hands-on projects on integrated circuit design, thus addressing a much-needed national workforce development for the Semiconductor Industry as, e.g., articulated in the recent Chips and Science Act. The specific approaches of the project are categorized into three key areas. Firstly, the initial approach aims to tackle scalability challenges by implementing many physical spins with fewer local spin interactions. This involves integrating compact latch circuits in a mixed-signal Ising computer, providing a large-scale Ising computer without the need for off-chip random number generators, which is a crucial feature for addressing large-scale combinatorial optimization problems. Beyond the scalability, the approach also aims to substantially reduce computing latency by leveraging massive parallelism through continuous-time operation. The second approach will implement a flexible digital Ising computer to address the issue of hardware overhead, which comes from mapping complex problems to the Ising computer with simpler interconnects in a regular grid topology, such as a lattice graph. The flexible Ising computer aims to amalgamate spatial and temporal (spatio-temporal) spin connectivity to achieve maximum reconfigurability, thereby minimizing hardware overhead. The resulting Ising computer with flexible spatio-temporal interactions between spins is anticipated to significantly reduce the required number of physical spins and enhance accuracy. Lastly, this project aims to implement an in-memory Ising computer with all-to-all spin interconnects to address connectivity challenges by embedding spins in the memory array, interconnected via a massive network of switches. This approach is designed to enhance connectivity and streamline spin interactions, contributing to the overall efficiency and effectiveness of the Ising computer. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.