Agile Application-Centric Design of Reconfigurable Analog Circuits

Speaker: Yu-Neng Wang, PhD Student, Stanford University
Date: March 8, 2023

Recently, there has been significant interest in domain-specific reconfigurable analog architectures that implement unconventional computational paradigms. Designing efficient reconfigurable analog architectures requires coordination between domain specialists and analog designers, as the optimization metrics are domain-specific and holistic (e.g., classification accuracy) but are nontrivial to translate to traditional circuit performance metrics (e.g., gain, bandwidth) The domain specialist-provided functional specification is fluid and can be modified to improve the performance of the analog implementation. However, enabling domain-specialists to contribute to the early design-space exploration of analog architectures remains a challenge, as current analog circuit modeling approaches and design productivity tools require extensive circuit expertise to use and modify.

We introduce Ark, a design system that enables non-circuit experts to rapidly design and simulate physically plausible reconfigurable analog architectures. With Ark, the analog designer constructs a specification that captures the architectural design space and provides design rules to ensure the constructed designs are physically plausible. Ark both verifies that domain specialist-provided architectures are physically plausible, and automatically generates dynamical systems for transient simulation. We evaluate Ark by crafting a specification for ladder filter network designs, and confirm that Ark produces simulations that match the circuit dynamics and preserves the intended functionality.