Reserach Overview

Dr. Lin’s research focuses on developing alternative, non-Boolean, non-CMOS computing paradigms capable of penetrating the digital CMOS computational efficiency barrier posed by quantum-related device physics. Specifically, we investigate how to transcend deterministic com- puting by natively exploiting randomness-driven physical phenomenon, either from CMOS devices under extreme conditions or from emerging spin-torque devices, to effectively compute. In short, Dr. Lin’s creative activities concentrate on answering the single question: how to leverage the physics of devices to compute algorithms natively? To this end, Dr. Lin has worked with five Ph.D students and numerous undergraduate students in the following research tracks

Hardware-Assisted Large-Scale Neuroevolution for Multi-agent Learning

Sponsored by DAPRA, this project seeks funding to purchase key equipment to establish an FPGA-based high-performance multiagent training platform and its associated software. The re- quested acquisition of BEE4-W (Berkeley Emulation Engine) hardware platforms costs $201,500.00 in total. achieve massively parallel analog-logic computation while consuming ultra-low energy. The successful execution of this equipment acquisition has greatly enhanced ongoing DARPA and ARO funded activities that culminate in a real-world demonstration of multiagent learning for coordinated UGV operations. Specifically, our achieved hardware assistance resulting from the new equipment has 1) alleviated the computing performance bottleneck imposed by software-based simulation while conducting multiagent learning research, 2) significantly increased the multiagent HyperNEAT learning intensity to facilitate solving real-world problems, 3) enabled training teams at much larger sizes than previously possible in simulation, and 4) made training adaptive neural networks with genuine synaptic plasticity feasible. Furthermore, the proposed equipment acquisi- tion has expanded the research capability at the University of Central Florida (UCF) in the area of Robotics, Artificial Intelligence (AI), and evolvable hardware by forging a close collaboration between two research teams at UCF: one focusing on multiagent robot training (based on Hyper- NEAT technology and a hive brain) and the other on high-performance reconfigurable computing (BCM, MARC, and evolvable hardware). In addition, this equipment acquisition has enabled us for the first time to set up a practical educational platform at UCF for many of our students who are currently working (or will work) for defense-related industry and who want to learn real-world hardware design. All these outcome from the proposed equipment acquisition has benefited various tasks relevant to DoD missions in the field, such as dynamic cordoning, coordinated surveillance, and multi-robot teleoperation.

Minimum-Energy Bio-Inspired AnaLogic Computing

This project aims at leveraging the physics of field-effect devices to perform computational tasks by strategically integrating the probabilistic reasoning and learning based on discrete random field theory with the traditional logic circuit design methodology based on deterministic Boolean algebra. Specifically, the PI first models and analyzes the stochastic switching behavior in minimum- energy CMOS transistors under ultra-low V DD (≈ 50mV) both analytically and experimentally. Subsequently, the PI develops field-theoretic methodology to optimize a large-scale logic circuit built with such stochastic switching devices in order to improve its robustness. Finally, the PI exploits the stochastic switching behavior natively to design and implement AnaLogic circuits (between analog and logic circuits) that emulate a robust self-motion algorithm inspired by fly eye based on optical flow extraction. The successful execution of this proposed research has introduced a new computing circuit paradigm, which embraces and exploits randomness instead of avoiding or circumventing it. Furthermore, this proposed approach has inspired a totally unconventional design paradigm for emerging nanoscale device technology with severe device variability and switching uncertainty. Also, this proposed field-theoretic approach has offered a rich mathematical structure, therefore broadened current digital circuit design theories, which are largely based on the principle that deterministic Boolean circuits can flawlessly emulate propositional logic deduction governed by the Boolean algebra. Finally, this proposed field-theoretic methodology has enabled more accurate understanding of existing logic circuit design methods, especially their limitations when directly applied to future device technologies driven by ultra-low V DD .

Bio-Inspired Logic Design with Graph and Field Theory

Sponsored by NSF (National Science Foundation), this project proved that the stochastic na- ture of emerging nanoscale device technology calls for a new transformative approach to digital circuit design, which embraces and exploits randomness instead of avoiding or circumventing it. This project realized such a vision by strategically integrating the fundamental insights gained from studying gene regulatory networks with the traditional logic circuit design methodology based on deterministic Boolean algebra. The successful execution of this project has developed a new logic design paradigm that achieves circuit robustness for stochastically imperfect transistors and interconnects. To this end, we has reformulated the traditional Boolean-based digital design prob- lem as a probabilistic logic-labeling problem and attempt to solve it with two approaches. In a graph-theoretic approach, we have adopted concepts and techniques such as network attractors, spectral graph decomposition, and eigen-graph theory. In addition, to fully consider the stochas- tic nature of both devices and interconnects, we have heavily utilized random field theory that allows us to exploit field-theoretic techniques such as isling graph fitting and orthogonal random field decomposition. Finally, we proposed two unconventional logic design methodologies. In bio- inspired logic scaffolding, we have developed techniques to integrate stabilizing circuits into logic design via distributed graph or field fortification. In self-correcting logic design, we have applied error-correction coding theory via maximal graph or field separation.

Metaphysical and Probabilistic-Based Computing Transformation

Sponsored by NSF (National Science Foundation, this project aims at effectively tackling the upcoming “zettabytes” data explosion, which requires a huge quantum leap in our computing power and energy efficiency. This proposed work revolves around Metaphysical and Probabilis- tic Computing Transformation (iMPACT), an innovative paradigm that transcends deterministic computing by natively exploiting randomness-driven physical phenomenon, either from CMOS de- vices under extreme conditions or from emerging spin-torque devices, to effectively compute. It possesses two transformative features. First, the iMPACT embraces and exploits quantum-induced randomness as an invaluable information carrier, not the “villain of correct computation” to be suppressed. Second, the theoretical foundation underpinning the iMPACT is based on neither the Boolean algebra in digital circuit nor the nonlinear amplifying and filtering in analog circuit. In- stead, it tightly brings together the algorithmatic essence of computing and the quantum-induced metaphysical randomness through the powerful framework of stochastic-based computing transfor- mation. In short, the iMPACT framework has fundamentally deviated from all existing computing paradigms by viewing randomness as an effective information carrier for computing, not merely the opposite of determinism (i.e., the lack of order or predictability), therefore enabling intrigu- ing computing methodologies that can potentially achieve extremely high computing performance, ultra-low energy consumption, and superior error resilience. The iMPACT has employed a diverse set of theoretic foundations: the convolution theorem, Taylor expansion, or even the Wiener-Askey polynomial chaos, thus highly versatile.