Advances in Analog Architectures and Randomized Algorithms: Computational and Energy Efficiencies

Efficient numerical computation is foundational to many disciplines, whose computing demands continue to grow dramatically, yet their continued scaling is jeopardized by the slowdown in Moore’s law. Multi-function multi-way analog (MFMWA) architectures represent a promising computing framework that supports in-memory computation of matrix operations and offers the potential for tremendous improvements in computation time and energy consumption, but at the expense of inherent unpredictability and noise which prevents such architectures from supporting reliable numerical computation in general. In this paper, we summarize and review a recent novel combination of MFMWA architectures and randomized algorithms tailored to these architectures that seek to mitigate the detrimental impact of imperfect analog computations with the goal of realizing the promising benefits of such advances in analog architectures for numerical speedups and energy reductions across various disciplines. Through a mixture of analysis, physical measurements and simulations, we organize and review recent research results that demonstrate the potential to achieve orders of magnitude reduction in both computation time and energy consumption for broadly applicable core numerical methods with algorithmic solution accuracy similar to digital computers. Together, these results provide understanding and insights into the computational and energy efficiencies of such a combination of MFMWA architectures and tailored randomized algorithms.