IBM Research Automates QEC Decoder Chip Design Using Open AI Agent Factories at 11.5x Lower Inference Cost

IBM Achieves 52x Design Speedup With AI Agents Tackling Quantum Error Correction

Using an agent factory to optimize hardware for quantum error correction, IBM Research achieved a 52-times speedup over the baseline design for approximately $26 in inference cost. A configuration using open frontier models cost $26, while a proprietary frontier model configuration cost 11.5 times more—$299—under the same search budget.

IBM Research Automates QEC Decoder Chip Design Using Open AI Agent Factories at 11.5x Lower Inference Cost

IBM Research has developed an AI agent factory to automate the design of QEC decoder chips, achieving up to 204x hardware speedup and 11.5x lower AI inference costs. This system utilizes autonomous multi-agent search and open-weight models, making QEC chip design more economically practical for fault-tolerant quantum systems. The agent factory explores architectural transformations to optimize decoding algorithms for high-rate qLDPC codes.