Back to AI Research

AI Research

A Unified Physics-Aware Quantum Machine Learning Fr... | AI Research

Key Takeaways

  • This research introduces a unified framework designed to improve device modeling in scenarios where data is scarce.
  • We present a unified reinforcement-learning (RL) framework that discovers compact parametrized quantum circuits (PQCs) for data-scarce device modeling.
  • By combining reinforcement learning (RL) with quantum-inspired computing, the authors have developed a method to create c...
  • The core of this approach is a reinforcement learning agent that discovers the best architecture for parametrized quantum circuits (PQCs).
  • The system uses a graph neural network (GNN) policy, which is optimized through proximal policy optimization (PPO).
Paper AbstractExpand

We present a unified reinforcement-learning (RL) framework that discovers compact parametrized quantum circuits (PQCs) for data-scarce device modeling. A graph neural network (GNN) policy optimized by proximal policy optimization (PPO) searches circuit architectures using leave-one-group-out cross-validation (LOGOCV) error on held-out process or geometry groups as the reward. The framework achieves the lowest mean absolute error (MAE) on all 11 targets versus six classical baselines, with 59% lower error (Ioff) and 81% tighter fold variability (VTH) for HEMTs and 84% lower error (VTH, SS, Ioff) and 82% tighter fold variability (Ioff) for NWFETs. These results demonstrate the potential of RL-selected, classically simulated PQCs as compact surrogates with low OOD error and improved physical consistency, despite imposing no explicit physical constraints, penalty terms, or device-specific equations, on the two evaluated device datasets.

This research introduces a unified framework designed to improve device modeling in scenarios where data is scarce. By combining reinforcement learning (RL) with quantum-inspired computing, the authors have developed a method to create compact, highly accurate models for complex semiconductor devices, specifically Power GaN HEMTs and logic Nanowire FETs.

How the Framework Works

The core of this approach is a reinforcement learning agent that discovers the best architecture for parametrized quantum circuits (PQCs). The system uses a graph neural network (GNN) policy, which is optimized through proximal policy optimization (PPO). To ensure the models are robust and can handle new, unseen data, the framework uses a "leave-one-group-out" cross-validation strategy. This means the system is rewarded for how well it predicts device behavior on groups of data it has never seen before, such as different manufacturing process splits or geometry combinations. The same ai systems question is explored in Measure Before You Manage, which adds a research perspective.

Achieving High Accuracy

The framework functions as a compact surrogate model that effectively mimics the behavior of physical devices. Despite not being programmed with explicit physical equations or penalty terms, the models demonstrate a high degree of physical consistency. When compared against six classical machine learning baselines, this quantum-inspired approach achieved the lowest mean absolute error across all 11 tested targets.

Key Performance Results

The framework showed significant improvements in both accuracy and stability across two distinct types of semiconductor technology: The same ai systems question is explored in DRACO, which adds a research perspective.

  • Power GaN HEMTs: The model achieved a 59% reduction in error for off-state current (Ioff) and 81% tighter variability for threshold voltage (VTH).

  • Logic Nanowire FETs: The model delivered an 84% reduction in error across threshold voltage, subthreshold swing, and off-state current, alongside an 82% improvement in the consistency of off-state current predictions.

Why This Matters

These results highlight the potential of using RL-selected, classically simulated quantum circuits to model complex hardware. By successfully predicting outcomes for unseen process and geometry configurations, this framework offers a powerful tool for device modeling that maintains high precision even when training data is limited. The ability to achieve these results without relying on hard-coded physical constraints suggests that the framework is highly adaptable to different types of electronic devices. The same reasoning question is explored in TransMeme, which adds a research perspective. as detailed in the full paper on Arxiv

More from Franklin

ThomsonresearchTransMemeresearchSelective Agent Guidance via EntropyresearchMeasure Before You ManageresearchDRACOresearchFeature Transformation Enhanced Jacobi Polynomial Graph...research

Comments (0)

No comments yet

Be the first to share your thoughts!