Sandia Labs FY22 Laboratory Directed Research & Development Annual Report

AI-ENHANCED CODESIGN FOR NEXT-GENERATION NEUROMORPHIC CIRCUITS AND SYSTEMS. Circuit design involves an assembly of

useful for mission-relevant applications in nuclear deterrence, code generation, and supply chain management. The PI from this one-year project was invited to give two presentations, and the team filed a Technical Advance, with plans for a patent application. (PI: Suma George Cardwell)

interconnected circuit elements that perform a specific objective function. Neuromorphic computing is a prime candidate to employ AI enhanced methods in developing circuits, systems, and architectures due to plethora of novel devices, diverse methods of encoding/decoding of inputs/ outputs, and neuro-inspired algorithms. The Sandia research focused on answering two key questions. One, can AI-enhanced codesign tools accelerate

circuit design? Two, can AI-enhanced tools be leveraged for novel device

discovery? The team, in collaboration with Sandia Alliance partner Georgia Tech, focused on design rather than just circuit optimization solutions and leveraged reinforcement learning and evolutionary algorithms. The result was a new AI-enhanced framework for automated circuit design and approaches for novel device discovery. The complete framework can be extended beyond circuit design and will be

New neuromorphic computing framework employs AI-enhancements to facilitate circuit design but has many opportunities for use across the DOE mission.

63

LABORATORY DIRECTED RESEARCH & DEVELOPMENT