Refining Qubit Calibration Models with Supervised Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing qubit calibration models are often too simple and fail to accurately capture the complex parameters and imperfections in qubit calibration data, leading to unreliable extraction of qubit parameters.

Innovation Solution

The use of supervised machine learning algorithms to refine qubit calibration models by learning perturbations from calibration data, allowing for the capture of features not accounted for in the original models, such as higher-order effects and unforeseen physics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If simple qubit calibration models are used, then the calibration process is easier and faster, but the accuracy and reliability of parameter extraction deteriorates

Engineering Contradiction:
Improvecalibration speedVSAvoidparameter extraction accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent combines simple parametric models with machine learning models to create a hybrid calibration model. The parametric model provides physical interpretability and fast computation, while the machine learning component captures complex nonlinear relationships and imperfections in the data, achieving both speed and accuracy in parameter extraction

Inventive Principle:
Principle #40Composite materials

Solution Approach 2:

The calibration model is divided into multiple components: a base parametric model for capturing primary physical relationships and separate machine learning models for correcting systematic errors and capturing higher-order effects. This segmentation allows each component to specialize in specific aspects of the calibration problem

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If complex qubit calibration models are used, then the accuracy of parameter extraction improves, but the device complexity and computational requirements increase

Engineering Contradiction:
Improveparameter extraction accuracyVSAvoidmodel complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

Rather than using a fully complex model for all calibration tasks, the patent applies machine learning corrections selectively to address specific deficiencies in the parametric model, such as systematic errors and higher-order effects, maintaining simplicity where it suffices while adding complexity only where needed

Inventive Principle:
Principle #16Partial or excessive action

3Ease of manufacture

If traditional calibration methods are used, then the process is simpler to implement, but the ability to capture higher-order effects and unforeseen physics is lost

Engineering Contradiction:
Improveimplementation simplicityVSAvoidmodel adaptability to complex phenomena
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent introduces machine learning models as intermediary components that bridge the gap between simple parametric models and complex physical phenomena. These intermediaries learn to capture higher-order effects and unforeseen physics from calibration data without requiring explicit physical models for each effect

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3711004B1Refining qubit calibration models using supervised learning
Publication Date: 2025.02.19 GOOGLE LLC
  • EP3711004B1 patent drawingFigure 1
  • EP3711004B1 patent drawingFigure 2
  • EP3711004B1 patent drawingFigure 3

AI summary

A computer-implemented method for refining a qubit calibration model is described. The method comprises receiving, at a learning module, training data, wherein the training data comprises a plurality of calibration data sets, wherein each calibration data set is derived from a system comprising one or more qubits, and a plurality of parameter sets, each parameter set comprising extracted parameters obtained using a corresponding calibration data set, wherein extracting the parameters includes fitting a qubit calibration model to the corresponding calibration data set using a fitter algorithm. The method further comprises executing, at the learning module, a supervised machine learning algorithm which processes the training data to learn a perturbation to the qubit calibration model that captures one or more features in the plurality of calibration data sets that are not captured by the qubit calibration model, thereby to provide a refined qubit calibration model.