LIMCA leverages Large Language Models to automate the complex, knowledge-intensive design of analog In-Memory Computing (IMC) architectures for neural network acceleration.

• Transforms manual circuit design into an automated process using LLMs to generate SPICE netlists and simulate complex circuits
• Achieves 90% accuracy in netlist generation without requiring any fine-tuning of the LLM
• Enables rapid design space exploration with 300x reduction in development time compared to manual methods
• Democratizes IMC design by lowering the expertise barrier, potentially accelerating hardware innovation

This breakthrough matters for Engineering by transforming hardware design workflows, reducing time-to-market, and enabling more efficient exploration of emerging computing architectures that can deliver substantial energy efficiency improvements for AI systems.

LIMCA: LLM for Automating Analog In-Memory Computing Architecture Design Exploration