Physical analogy computing
Please login to view abstract download link
Current computational demand, driven by the integration of artificial intelligence (AI) at large scale, robotics & automation, complex systems research, and edge-computing, is hindered by the power constraints of modern infrastructure and the physical limitations for digital compute architectures [1, 2]. This necessitates a shift toward unconventional, heterogeneous computing paradigms, e.g., hybrid analog-digital computing, which promise superior energy efficiency and real-time processing [3]. Integrating these architectures remains challenging, particularly regarding challenges such as frequent digital-analog conversions in hybrid systems [4]. We introduce the REconfigurable Discrete Analog Computer (REDAC), the world’s largest hybrid analog-digital platform. Its physical configuration can be programmed from within existing digital infrastructure. REDAC serves as a versatile technology platform for the fast-track development of hybrid architectures; by leveraging modularity, scalability, and a full software stack, it enables rapid prototyping and validation of complex problem sets. While currently optimized for modern integrated circuit technologies in industrial and edge-computing applications, the REDAC-architecture remains technology-agnostic. Crucially, REDAC bridges the gap between conceptual design and hardware realization, dramatically reducing the time for chip development for next-generation hybrid computing solutions. Here, we demonstrate how anabrid and its partners in academia and industry utilize REDAC to apply analog computing to diverse fields, such as sustainable AI. We address the hardware- and software-related challenges setting up hybrid architectures where digitally stored information is fed into analog compute circuits. Furthermore, we examine energy-constrained environments, such as edge infrastructure, where collaborators at the Deutsche Telekom Chair for Communication Networks (TU Dresden) are exploring the hybrid analog-digital computing capabilities of the REDAC platform.
