Home / AI-Enabled Analog IC Design
AI-Enabled Analog IC Design
Agentic AI methods that assist analog integrated circuit design.
Agentic AI for Analog IC Design
Analog and mixed-signal design still depends heavily on expert intuition and many manual iterations between schematic, simulation and layout. The AMPIC Lab is developing AI-enabled methods, including agentic AI workflows, to assist each step of the analog IC design process while keeping the designer in control.
1. Specification to sizing
From a specification (gain, bandwidth, phase margin, power) to transistor sizes, using physics-based models such as gm/ID design with machine-learned surrogates to search quickly.
2. Compensation and stability
Choosing compensation networks and checking loop stability. This builds on our long-standing work on opamp compensation (split-length and indirect compensation of multi-stage opamps).
3. Layout and verification
Agentic flows that generate layout, run design-rule and layout-versus-schematic checks, and iterate on failures, using open PDKs such as SkyWater sky130.
Try it: AI-assisted opamp design
Our demo notebook sizes a two-stage Miller opamp from a specification. It trains a small neural network on a physics-based opamp model, searches the design space with it, and verifies the result against the full model. It runs in your browser on Google Colab, with no installation. The circuit model is a compact illustrative one, so the results show the method rather than a fabricated design.
Download the notebook Preview the results


To run it: open colab.research.google.com, choose File, then Upload notebook, and select the downloaded file.
Foundations
- Book chapters on high-performance CMOS operational amplifiers and fully-differential opamp design and simulation (Springer, 2016).
- Systematic design of three-stage opamps using split-length compensation, and indirect compensation of three-stage opamps (IEEE MWSCAS).
Related work: Data Converters and all research areas. Interested in working on this topic? See Opportunities.