01 / AI · Analysis
ReLay cuts model call costs in analogue circuit layout tasks
The researchers recovered code from existing circuit layouts and added recurring design procedures to a library of shared functions.

Researchers reported that ReLay lowered model call costs by 64–77% per completed circuit layout task for language model agents, compared with generating layout shapes directly.
The paper identified a barrier to reusing existing layouts. Much of the design knowledge is embedded in their shapes, where it is hard to extract and use again.
ReLay recovers code from those layouts. The researchers used the recovered code to build a library of shared functions for procedures that recur across designs.
Reusing recovered design code
The paper explains that the recovered code can help move an existing design to a different process design kit, which provides the rules and tools for designing circuits for a manufacturing process.
The paper describes parametric resizing, which lets designers change dimensions by adjusting variables in the recovered code. The researchers added procedures that recurred across designs to the library as shared functions.
The paper lists 1,050 hand-drawn layouts from 27 circuit families in its evaluation. The researchers measured the share of calls to shared functions when building layouts held out from training.
Costs compared with direct shape generation
The researchers gave language model agents new circuit layout tasks. For the cost comparison, they used direct generation of layout shapes as the baseline.
The researchers reported similar task completion rates and electrical quality for the two approaches. The paper’s abstract did not give the completion rate for either approach.