Invention Description
Designing high-performance analog circuits, such as analog-to-digital converters (ADCs), is a complex and time-intensive process that typically requires significant expert knowledge and manual tuning. Traditional design methods rely heavily on iterative trial-and-error, making it difficult to efficiently meet strict performance targets like signal-to-noise and distortion ratio (SNDR) and spurious-free dynamic range (SFDR). As circuit complexity increases, the gap between manual design capabilities and the need for faster, automated solutions continues to grow. This creates a need for intelligent frameworks that can streamline analog design while maintaining expert-level accuracy.
Researchers at Arizona State University have developed an AI-driven framework to automate and optimize ADC circuit design. The technique is composed of two primary steps. In the first step, a deep-learning neural network model is used to predict key performance metrics, guiding the optimization process. In the second step, the circuit parameters are iteratively adjusted to meet target specifications with the framework ensuring the generated designs remain realistic and aligned with training data. This framework was demonstrated on sample-and-hold blocks for SAR ADCs, with the approach achieving over 12-bit SNDR performance with high predictive accuracy.
This advanced AI-driven framework effectively bridges the gap between expert manual design and scalable workflows to automate and optimize ADC schematic design.
Potential Applications
- Automated design of ADC front-end analog blocks for high-performance mixed-signal ICs
- Accelerated analog and RF IC development in consumer electronics, communications, and automotive sectors
- Design automation tools integrating AI/ML for semiconductor companies and EDA vendors
- Custom analog circuit design assistance to reduce time-to-market and improve yield
- Telecommunications infrastructure requiring precise data conversion
Benefits and Advantages
- Incorporates expert circuit intuition into AI training for improved design outcomes
- Enables efficient multi-metric optimization
- Uses novel techniques to enhance realism and convergence in parameter tuning
- Achieves high accuracy predictions validated through SPICE simulations
- Automates complex analog design tasks traditionally reliant on experienced designers
- Optimizes critical performance parameters such as SNDR and SFDR
For more information about this opportunity, please see
Guo et al – Balancing Speed and Accuracy for Robust Analog-Mixed Signal Circuit Design using Closed-Loop Reinforcement Learning with Ensemble Neural Network Surrogates – ISCAS 2026