Vangelis Dikopoulos “Physics-Inspired Computing Architectures for Optimization and High-Impact Applications”

Abstract:
Modern computing faces a fundamental challenge. Digital computing sacrifices computational potential by imposing deterministic abstractions on inherently analog silicon, and is further limited by restricted parallelism, memory bottlenecks, and synchronous operation, making hard combinatorial optimization problems intractable at scale. Quantum computing, though promising, remains limited by sparse qubit connectivity, high error rates, and kilowatt-scale power budgets. Meanwhile, the demand for fast, energy-efficient computing continues to grow across AI, cryptography, computational biology, logistics, and finance. Physics-inspired analog computing offers a compelling new paradigm for combinatorial optimization and beyond: hardware whose natural dynamics directly implement computation in silicon free of digital abstractions.

In this talk, I will present three silicon-validated physics-inspired analog accelerator chips that tackle hard combinatorial optimization problems by exploiting the energy landscape and natural dynamics of coupled oscillator systems in 28nm CMOS technology. (1) I will present Daedalus, a mixed-signal 3-SAT solver that natively maps Boolean satisfiability onto a coupled oscillator fabric, achieving state-of-the-art solution time and energy efficiency. (2) I will introduce MEDUSA, a 200-variable analog k-SAT solver that scales the solver architecture and advances the dynamical system algorithm, surpassing D-Wave’s quantum annealer in solution time and energy efficiency. (3) I will also introduce RXO-LDPC, the first physics-inspired oscillator-based LDPC decoder, which recasts error-correction decoding as a combinatorial optimization problem and achieves more than three orders of magnitude lower bit error rate than conventional Belief Propagation decoders. 

Beyond the three systems, I will outline a research vision that extends physics-inspired computing to high-impact applications such as MIMO communications, protein folding, probabilistic AI, and quantum error-correction decoding – paving the way for high-performance, energy-efficient computing on some of the hardest problems in science and engineering.

Bio:
Vangelis (Evangelos) Dikopoulos is a PhD candidate in Electrical and Computer Engineering at the University of Michigan, Ann Arbor. His research focuses on physics-inspired mixed-signal accelerators for combinatorial optimization, resulting in three silicon-validated prototypes with publications at IEEE ISSCC, IEEE JSSC, IEEE ESSERC, and IEEE IEDM. Prior to Michigan, he received the diploma in Electrical and Computer Engineering from the University of Patras, Greece. He received the 2024 Qualcomm Innovation Fellowship for North America.

Details

• Date: Tuesday, April 7
• Time: 10:00 am - 11:00 am
• Category: Special Seminar
• Location: Grier B (34-401B)

Host

• Luqiao Liu
• Email: chadcoll@mit.edu