Hardware for Edge AI: Embracing Analog and Physical Computing Paradigm

This presentation introduces a physics-based computing paradigm and architecture, harnessing the collective dynamics of coupled oscillators to enable massive parallelism and energy-efficient computation. This approach overcomes the inherent limitations of classical von Neumann computing, enabling the execution of highly complex functions with remarkably low power consumption. At the core of this physics-based computing lies the interactive dynamics of devices within an open system, naturally minimizing their energy by transitioning to the ground state.

The talk will cover computational theory pertaining to physical computing, as well as the materials and devices essential for its physical implementation. Physical computing leverages the intrinsic nonlinearities of devices and the memory of the physical system, enabling energy-efficient, in-memory computations with no data transfer, thus making it suitable as an Ising machines for solving NP-hard problems. Additionally, it serves as an energy-efficient hardware accelerator for AI applications. In conclusion, I will address the current challenges and advancements in energy-efficient computing.