Artificial neural networks (ANNs) have enabled major advances in tasks such as image recognition, speech processing, and natural language understanding. However, their hardware implementation is often dominated by the cost of data movement between memory and processing units. In conventional von Neumann architectures, repeated transfers of weights and activations between memory and compute engines result in substantial energy consumption and latency, limiting the efficiency of AI systems, particularly for edge devices with strict power and resource constraints.

Analog and mixed-signal (AMS) computing provides an alternative approach by performing computation directly in the physical domain using device and circuit dynamics. By leveraging the intrinsic properties of electronic devices, AMS architectures can execute operations such as multiply-and-accumulate with significantly improved energy efficiency compared to purely digital implementations.