Harvard details $12 million effort to use AI against quantum computing errors
The funded team will design error-correction methods around specific scientific calculations, then test them on hardware at Harvard, MIT and QuEra.
Loading page…
The funded team will design error-correction methods around specific scientific calculations, then test them on hardware at Harvard, MIT and QuEra.
Listen to this story
The $12 million ASQC project will train AI to identify which quantum errors matter for scientific workloads, then design tailored correction codes and decoders. The plan links those designs to hardware tests at Harvard, MIT and QuEra, a step toward checking whether application-specific methods can support reliable scientific calculations. The award was part of a DOE Phase II Genesis Mission round announced October 8; Harvard’s November 20 account added the project’s funding amount, partners and research plan.
An earlier neural-network decoder from Lukin, Yelin and colleagues reportedly cut error rates by up to 17 times and ran thousands of times faster than existing versions; those are not ASQC results.
The consortium includes Cornell, Los Alamos, Lawrence Berkeley and Sandia national laboratories, alongside MIT, QuEra, NVIDIA and Anthropic.
AI models will focus on errors relevant to quantum simulation, materials science, chemistry and nuclear science.
A Harvard-led research team will use AI to tackle quantum computing errors around the scientific calculations researchers want to perform. Harvard disclosed November 20 that the collaboration has received $12 million from the U.S. Department of Energy to develop error correction tailored to problems in fields including chemistry and materials science.
The project is called Application-Aware Error Correction Codesign for Scientific Quantum Computing, or ASQC. Harvard physicist Mikhail Lukin is its principal investigator. The award belongs to an earlier funding round: DOE listed the Harvard-led project among 12 Phase II Genesis Mission selections in its October 8 announcement. Harvard’s new account supplies the project’s funding amount, partners and research plan.
The underlying obstacle is that quantum bits, or qubits, are inherently unstable and vulnerable to environmental noise. Errors make it harder to perform larger, more complex calculations. Quantum error correction addresses that problem. Harvard says research published by Lukin and collaborators last year suppressed errors far enough that adding qubits would reduce errors rather than introduce more—a foundation the funded project will build on.
ASQC’s goal is an AI-driven, fault-tolerant system: error correction intended to support reliable quantum computation despite errors. Its application-aware approach will start with scientific problems relevant to DOE, rather than treating the intended calculation as separate from the correction methods. The team’s planned workflow connects learning, design and physical testing:
“This is really an attempt to establish scientific foundations for useful fault-tolerant quantum computers,” Lukin said, “and really accelerate this progress using AI.”
Mikhail Lukin, ASQC principal investigator, speaking to the Harvard Gazette
Members of the team have already used AI on parts of this problem. In an April preprint, Lukin, Harvard physicist Susanne Yelin and colleagues described a neural-network decoder. According to Harvard’s account, it cut error rates by up to a factor of 17 compared with existing versions and ran thousands of times faster. Those figures describe the earlier decoder work, not results from the newly detailed ASQC program.
Harvard also points to recent work with QuEra and Anthropic in which the researchers used large language models to discover quantum circuits. That work covered tasks in numerical computation, chemistry, optimization, quantum simulation and number theory. It is another strand of the team’s existing AI research, distinct from ASQC’s planned effort to tailor error correction to scientific applications.
The consortium brings Harvard together with MIT, Cornell, Los Alamos National Laboratory, Lawrence Berkeley National Laboratory and Sandia National Laboratories. QuEra, NVIDIA and Anthropic are also participating. Harvard physicists Markus Greiner and Yelin will work alongside Lukin. DOE’s Advanced Scientific Computing Research division administers the program, which combines academic researchers, national-laboratory scientists and technology companies in the same funded effort.
The planned hardware tests give the collaboration a concrete destination beyond AI-generated designs. Harvard, MIT and QuEra will provide the machines on which the team tests whether its application-specific correction methods work.
Loading discussion...
Join the conversation
Explain which approach you think deserves more public funding.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.