Ph.D. Public Defense
CMOS Compatible Ising Machines with Bistable Nodes and Resistive Coupling
Supervise by Zeljko Ignjatovic
Thursday, July 21, 2022
Join via Zoom: https://rochester.zoom.us/j/8578352667
In recent years, special-purpose designs have been increasingly adopted for their efficacy in certain type of tasks such as encryption and network operations. In a related but different track of work, researchers are trying to leverage the evolution of dynamical systems to effectively carry out an entire algorithm. In many recent examples, this evolution has the effect of optimizing a particular formula called the Ising model. By configuring the dynamical system, a user is setting the machine to optimize a particular Ising formula. Reading out the state of such a system at the end of the evolution thus has the effect of obtaining a solution to the problem mapped.
The mechanism that allows these systems to seek optimal states differs by design. But they share the distinction from a von Neumann machine in that there is no explicit program to follow. Instead, these systems can be thought of as physics-driven special-purpose computing machines optimizing one particular objective function (in the form of the Ising formula). Hence, they are generally referred to as Ising machines. Ising machines have been implemented in a variety of ways with very different physics principles involved. It is unclear as yet whether some of the more complex constructions will demonstrate some fundamental advantage at a very large scale. In the near term, though, we believe an electronic version offers a much more compelling value proposition.
In this dissertation, we discuss the details of one such design which utilizes bistable, resistively-coupled networks and leverages mature CMOS technology to achieve ultimate system density and efficiency in an electronic Ising machine. We show that such an Ising machine out-performs the room-sized quantum and desktop-size optical approaches from many aspects: speed, area, energy, and quality of solution.