Programming Systems for Reconfigurable Analog Hardware

Speaker: Sara Achour, Assistant Professor, Stanford University
Date: March 2, 2022

Analog computing platforms, implemented on a range of electronic and optical substrates, are becoming pervasive and crucial for satisfying the computational needs of different domains. Delivering the potential of such hardware platforms to domain specialists is a challenge as these devices exhibit a host of analog behaviors that must be considered when mapping computation to the hardware. This talk will first overview some compilation techniques for automatically configuring this class of analog hardware to execute dynamical system computations. This class of hardware leverages the physical behavior of transistors to emulate the target dynamical system. Under this paradigm, voltages and currents within the device implement state variables in the dynamical system. The presented techniques include (1) a circuit synthesis technique that automatically programs the analog hardware to implement a given dynamical system computation (2) a circuit transformation technique that modifies a given circuit to respect the operating range limitations of the hardware and reduces the effect of analog noise, quantization error, and manufacturing variations on the computation. If time permits, we will also discuss how the lessons learned developing these compilation techniques could potentially be leveraged to effectively target other analog computing platforms.