Quantum bits, or qubits, are the fundamental building blocks of quantum computers, which promise to revolutionize many fields by leveraging the unique properties of qubits. While researchers are investigating various qubits, such as color centers in silicon, neutral atoms, and trapped ions, superconducting qubits based on solid-state electrical circuits can be easily fabricated using existing manufacturing processes, are simple to control, and are readily interconnected—properties that make them a leading candidate for building a scalable quantum computer.

However, superconducting qubits can only hold their quantum state for a very short time (the “coherence time”), whereas current methods for accurately determining their state and then reading and transmitting this information to a computer—an essential operation for processing quantum information—are slow and prone to errors. These limitations hamper the development of reliable superconducting quantum circuits for a future quantum computer.

Now, a team of researchers from the Accelerator Technology & Applied Physics (ATAP) Division at Lawrence Berkeley National Laboratory (Berkeley Lab) has collaborated with colleagues from the University of California, Berkeley, and the University of Massachusetts Amherst to develop QubiCML. This innovative and potentially game-changing technique combines conventional electronic control circuits known as field-programmable gate arrays (FPGAs) with machine learning (ML) to accurately measure the real-time state of superconducting qubits at intermediate stages in a quantum circuit.

According to Neel Rajeshbhai Vora, a scientific engineering associate in ATAP’s Berkeley Accelerator Controls & Instrumentation (BACI) Program and the lead author of the paper on QubiCML, this is “the first time ML-powered quantum state discrimination has been successfully integrated on an FPGA platform for mid-circuit measurements of superconducting qubits.”

Vora says QubiCML could help bring quantum computing closer to becoming a reality and provide ultra-high-precision control technology for particle accelerators and laser systems.

From bits to qubits

A qubit differs from a classical bit in that it can be 0, 1, or both simultaneously. This quantum property, called superposition, enables quantum computers to solve complex problems that are beyond the capabilities of today’s digital computers.

Current methods for determining the state of a superconducting qubit have significant limitations. They involve measuring the qubit at the end of a quantum circuit and then transferring the data to a classical computer for processing. This process, however, is time-consuming; for instance, transmitting the state typically takes tens of milliseconds, several orders of magnitude longer than the qubit’s coherence time.

This delay, explains Gang Huang, a staff scientist at BACI who worked on the development of QubiCML, prevents real-time measurements of the qubit’s state and “is a major limitation for advancing superconducting qubits as a reliable platform for building scalable quantum computers.”

“Superconducting quantum processors are also highly susceptible to errors when reading and retrieving quantum information. Even state-of-the-art quantum computers experience over ten percent readout errors for some qubit systems.”

According to Huang, mid-circuit quantum state measurements, where a qubit’s state is measured at pre-defined intervals during a quantum circuit’s operation, could reduce the time qubits need to maintain coherence, enabling real-time error detection and correction. It could also help to replace some quantum operations with classical logic to simplify quantum circuits. “This approach allows direct adjustments to quantum algorithms on the quantum processor, saving time and allowing the reuse of qubits, which is especially beneficial given their scarcity.”

He adds that performing real-time, mid-circuit measurements on superconducting qubits requires a low-latency, high-accuracy technique: “Until now, such a technique didn’t exist.”

Harnessing the power of ML

Current qubit state discrimination methods are computationally intensive and costly, making them impractical for mid-circuit measurements of superconducting qubits. To overcome these limitations, the researchers took a new approach that combines ML techniques with classical electronic control circuits like FPGAs to conduct precise, real-time mid-circuit measurements of superconducting qubits.

This approach utilizes and expands upon the capabilities of QubiC, an open-source FPGA-based control system developed by Berkeley Lab. “QubiC is equipped with digital-to-analog converters for generating radio-frequency pulses for qubit manipulation, analog-to-digital converters for measuring the qubit response, and digital signal processing units to process the data on an FPGA,” explains Yilun Xu, a BACI research scientist and principal investigator for the QubiCML project.

Mid-circuit measurement with QubiCML with a readout time of 500 ns. (Credit: Berkeley Lab)

To design QubiCML, the researchers first needed to calibrate the qubits, a standard practice when working with superconducting qubits. Then, they optimized the digital local oscillator (DLO) to improve the accuracy of state discrimination and set up the superconducting qubit circuits to be in either a 0 or a 1 state. After collecting qubit data from 50,000 readout samples for each qubit state, they trained an ML model using 65 trainable parameters, which they ran until high-fidelity qubit discrimination was achieved. They then deployed their model on a radio frequency system-on-chip FPGA platform, integrating radio frequency, analog, and digital circuits with memory blocks and microprocessors on a single chip.

Xu says this approach offers significant advantages over current methods. “It has much lower latency because it can perform real-time quantum state discrimination directly on an FPGA platform and achieves excellent qubit readout fidelity. It also allows for the exploration of multi-level qubit systems.”

The team tested QubiCML on three superconducting qubits and demonstrated that it took only 54 nanoseconds to perform each qubit inference (state measurement) with an average accuracy of 98.46%. Additionally, the readout time was just 500 nanoseconds, which is orders of magnitude faster than the qubit’s coherence time and is considered state-of-the-art in the quantum computing community.

According to Vora, these findings demonstrate that QubiCML has the potential “to become the standard method for real-time state discrimination and provides the quantum community with a powerful tool to explore and implement advanced quantum algorithms and applications.”

Impact of readout pulse optimization. (Credit: Berkeley Lab)

“I am excited by the team’s recent advancements in machine learning-based mid-circuit qubit discrimination in quantum computing systems,” says Qing Ji, who heads BACI. “These results are essential for enhancing qubit readout accuracy and laying the groundwork for ML-driven, real-time qubit control and manipulation.”

Qing adds that BACI’s work in qubit control also serves as a platform for developing ultra-high precision control technology for particle accelerators and lasers, where ML-driven real-time optimization can significantly improve the performance of complex systems.

“This innovative integration of machine learning techniques leveraging particle accelerator control technologies takes us another step closer to the realization of quantum computing,” says ATAP Division Director Cameron Geddes. “Our BACI team is working at the forefront of this effort, supporting the Department of Energy’s quantum initiatives while reinforcing its mission in creating technology for future accelerators.”

The team is now investigating ways to improve the readout time by utilizing ML to shorten the time series of the qubit signal. They say this will improve the technique’s accuracy and reduce the computational resources required for the readout process. They are also working on extending QubiCML to multi-stage qubit systems and have already successfully implemented it on a three-level quantum system called a quantum trit (qutrit).

“Open-source classical control systems are critical for moving towards scalable quantum computing. This impressive work by the team led by Berkeley Lab highlights the breadth and depth of scientific leadership at the Quantum Systems Accelerator (QSA),” says QSA Director Bert de Jong.

Led by Berkeley Lab, QSA is one of five National Quantum Information Science (QIS) Researcher Centers funded by the U.S. Department of Energy (DOE)’s Office of Science dedicated to cutting-edge research on science’s most challenging problems. ATAP researchers work with QSA to advance quantum advantage in scientific applications. They are also working with the DOE’s Office of High Energy Physics for the Quantum Information Science Enabled Discovery (QuantISED) program (managed by the Physics Division at Berkeley Lab), for example, by proposing a novel approach that uses quantum sensing to detect dark matter.

 

The research presented here was supported by funding from the Quantum Systems Accelerator under the U.S. Department of Energy’s Office of Science National Quantum Information Science Research Centers. Additional work referenced in this article was supported by the U.S. Department of Energy’s Office of Science, the Office of High Energy Physics, the National Science Foundation, and the Defense Advanced Research Projects Agency.

 

To learn more …
  1. Neel R. Vora, Yilun Xu, Akel Hashim, Neelay Fruitwala, Ho Nam Nguyen, Haoran Liao, Jan Balewski, Abhi Rajagopala, Kasra Nowrouzi, Qing Ji, K. Birgitta Whaley, Irfan Siddiqi, Phuc Nguyen, and Gang Huang. “ML-Powered FPGA-based Real-Time Quantum State Discrimination Enabling Mid-circuit Measurements,” June 2024. https://doi.org/10.48550/arXiv.2406.18807
  2. Yilun Xu, Gang Hang, Jan Balewski, Ravi Naik, Alexis Morvan, Bradley Mitchell, Kasra Nowrouzi, David I. Santiago, and Irfan Siddiqi. “QubiC: An Open-Source FPGA-Based Control and Measurement System for Superconducting Quantum Information Processors,” in IEEE Transactions on Quantum Engineering, vol. 2, pp. 1-11, 2021. https://doi.org/10.1109/TQE.2021.3116540
  3. Yilun Xu, Gang Hang, Neelay Fruitwala, Abhi Rajagopala, Ravi Naik, Kasra Nowrouzi, David I. Santiago, and Irfan Siddiqi. “QubiC 2.0: An Extensible Open-Source Qubit Control System Capable of Mid-Circuit Measurement and Feed-Forward,” September 2023. https://doi.org/10.48550/arXiv.2309.10333
  4. Devanshu Brahmbhatt, Yilun Xu, Neel Vora, Larry Chen, Neelay Fruitwala, Gang Huang, Qing Ji, and Phuc Nguyen.“An open-source data storage and visualization platform for collaborative qubit control,”  Rep. 14, 22703 (2024). https://doi.org/10.1038/s41598-024-72584-9

 

 

For more information on ATAP News articles, contact caw@lbl.gov.