Get the newsletter
IBM · AI / ML · Filed Mar 3, 2025 · USPTO published Aug 13, 2026 · Aug 13 release

IBM Patents a Machine-Learning Method for Quieting Quantum Circuit Noise

Quantum computers make errors constantly, and a big reason is that calculations have to be rerouted around hardware limitations in ways that add noise. IBM is now using machine learning to make that rerouting smarter and quieter.

Workflow showing a user submitting a quantum circuit to a transpilation service that routes it using machine learning models before execution. Drawing from patent filing US 2026/0236819 A1.
Workflow showing a user submitting a quantum circuit to a transpilation service that routes it using machine learning models before execution.All 11 drawings ↓
Publication number
US 2026/0236819 A1(US20260236819A1)
Applicant
International Business Machines Corporation
Filed · Published
Mar 3, 2025 · Aug 13, 2026
Inventors
Daniel Josef Egger, Elena Peña Tapia, Stefan Woerner, David Kremer Garcia, Juan Cruz Benito
Status when we published
Waiting for an examiner
Document
20 claims
US classification
706/62

What IBM's quantum circuit routing fix actually does

You're running a calculation on a quantum computer, and the machine scrambles part of the answer because the instructions had to be reshuffled to fit the hardware. That reshuffling, called circuit routing, is one of the main reasons quantum computers produce wrong results today.

IBM's patent describes a system where a machine-learning model learns the best way to do that reshuffling. Instead of a static rule, the model is trained to find routes that keep the number of steps low and the noise low at the same time. It treats the whole problem like a game, scoring each possible route by how much error it introduces.

The practical upside is that the same quantum hardware could produce more accurate results without changing a single physical component. The fix is entirely in software, which matters because building better quantum chips is slow and expensive.

From the filing · CLAIM 1
encoding a structure of a quantum circuit to route and noise of a built routed quantum circuit in a cost function; and performing quantum circuit routing using a trained machine learning model based on minimizing said cost function.

Translation: The system uses AI to map out quantum operations while simultaneously calculating how to reduce errors in the hardware.

How the reinforcement-learning agent routes qubits

Quantum computers run programs as sequences of quantum gates (operations on quantum bits, or qubits). The problem is that most quantum hardware only allows certain pairs of qubits to interact directly. When a program needs two qubits that aren't neighbors, the circuit has to be rerouted by inserting extra SWAP gates (operations that move quantum information around). Every extra gate adds noise and potential error.

IBM's method encodes the structure of the quantum circuit (the specific sequence and arrangement of gates) into a reinforcement-learning environment. Think of reinforcement learning like training a dog: an agent tries actions, gets scored on the results, and learns over time to prefer actions that score well.

The scoring function here, called a cost function, penalizes two things at once:

  • High gate count (more gates means more chances for error)
  • High noise, measured directly from the routed circuit's gate layers

A trained model then performs the routing by choosing moves that minimize that score. Because the model has learned from many examples, it can find good routes quickly at inference time, without re-solving the problem from scratch each time a new circuit arrives.

From the filing · THE ABSTRACT
A noise of a built routed quantum circuit is encoded in a cost function, such as by measuring the noise in the layer(s) of gates of the built routed quantum circuit.

Translation: The software tracks how much interference occurs at each step of the process to help the computer find a cleaner path.

What this means for practical quantum computing reliability

For anyone using a quantum computer, whether a pharmaceutical researcher or a financial risk analyst, more accurate results from the same hardware is a concrete, immediate payoff. Routing errors today often force teams to run calculations many times and average the results to reduce noise, which is expensive and slow. A routing step that's already noise-aware at the software level reduces how much of that repetition is needed.

This is also a signal that IBM sees software-level optimization as a near-term path to better quantum performance, at a time when physical qubit quality improvements are still slow-going. The approach connects to a broader push across the industry to squeeze more from existing hardware, and IBM's quantum-related filings are among this week's Big Tech patents tracking that trend.

Source: US 2026/0236819 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time IBM patent news breaks.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings · 11 sheets

11 drawing sheets from US 2026/0236819 A1 · click any drawing to enlarge

Quantum computer system network architecture, drawing sheet 1 from US 2026/0236819 A1
Sheet 1 Quantum computer system network architecture
Machine learning software engine components, drawing sheet 2 from US 2026/0236819 A1
Sheet 2 Machine learning software engine components
Reinforcement learning environment processing steps, drawing sheet 3 from US 2026/0236819 A1
Sheet 3 Reinforcement learning environment processing steps
Quantum circuit initial and parameterized states, drawing sheet 4 from US 2026/0236819 A1
Sheet 4 Quantum circuit initial and parameterized states
Quantum gate sequence and circuit mapping, drawing sheet 5 from US 2026/0236819 A1
Sheet 5 Quantum gate sequence and circuit mapping
Gate parameter optimization quantum circuit, drawing sheet 6 from US 2026/0236819 A1
Sheet 6 Gate parameter optimization quantum circuit
Quantum circuit routing and execution workflow, drawing sheet 7 from US 2026/0236819 A1
Sheet 7 Quantum circuit routing and execution workflow
Computer hardware network infrastructure system, drawing sheet 8 from US 2026/0236819 A1
Sheet 8 Computer hardware network infrastructure system
Show the other 3 sheets
IBM Patents a Machine-Learning Method for Quieting Quantum Circuit Noise patent drawing, sheet 9
Sheet 9
IBM Patents a Machine-Learning Method for Quieting Quantum Circuit Noise patent drawing, sheet 10
Sheet 10
IBM Patents a Machine-Learning Method for Quieting Quantum Circuit Noise patent drawing, sheet 11
Sheet 11
Patent filing page

More in AI training & infrastructure

All 316 AI training & infrastructure patents →

Watchlists this filing belongs to

All watchlists →