My dissertation research was in applications of analog electronic computing in a modern setting. Analog computing is promising in solving problems in continuous mathematics. Emerging computing domains in robotics and machine learning rely on data and algorithms that are more numerical, which are solved in an unnatural way using digital, discrete-time numerical methods. Existing numerical techniques, running on digital computers, may be less efficient than analog approaches, especially as digital computing faces limitations in transistor scaling.
I have worked with a team of PhD students to build prototype analog computers, fabricated in integrated silicon technology. Our chips feature electronic integrators and multipliers, and operate on continuous-time, analog signals. I have dedicated my efforts to building the interface between the analog computer and a conventional digital computer. The digital hardware, software routines, and programing language I developed help port modern applications to analog hardware.
Thus far, we have demonstrated using analog computing to solve nonlinear ordinary differential equations used in mobile robotics control. We have used analog computing, in conjunction with a multigrid numerical method, to solve elliptic partial differential equations that arise in physics simulations. Our analog computer architecture holds promise in solving vexing nonlinear systems of equations, and in optimization problems in non-convex spaces.
As a computer architecture student, I have also published work answering questions such as “what are the time consuming algorithms and computation in mobile robotics?” and “how do chip designers use simple encryption in hardware to protect against malicious computer hardware?”
As integrated circuit designers create specialized chips, they rely on components licensed from other design teams. These designs may conceal backdoor circuitry that delivers secrets to an unauthorized party, obliterating all the software security built on top of the hardware. As chip design becomes a global effort, secure design techniques are direly needed to mitigate these backdoors. I contributed a security design pattern in my research project titled Security Engineering for Backdoor-Free Crypto Hardware, where we masked secret data to untrusted components.
Improved hardware security would help teams to collaborate in production of cheaper specialized computers. As pervasive computing allows more information to be attached to physical objects, better networks are required to aggregate and present that information.