Nature Magazine Cover of the Issue with IBM’s aLDPC Paper Showing the Torus Topologies Used in this Code. Credit: IBM

IBM’s Research in qLDPC Codes

The popular press is buzzing with IBM’s announcement that they have published an article about the research of qLDPC (Quantum Low Density Parity Check) codes in Nature magazine. But attentive readers of the Quantum Computing Report will note that we first reported about this last August when the preprint of this paper describing this research was first posted on the arXiv and again this last December in our article about IBM’s roadmap.

LDPC codes are a concept that originated in classical computing and the concept is now being adopted in quantum by an increasing number of companies including IBM, Photonic Inc., Alice & Bob, Infleqtion, and others. The advantage of qLDPC codes is that they provides more error protection for a fewer number of qubits than other approaches such as the surface code. To give a specific example from the paper, IBM’s qLDPC code which uses 144 physical qubits to create 12 logical qubits with a code distance of 12, notated as [[144, 12, 12]], would give slightly better performance than a surface code [[1452, 12, 11]], a savings of over 10X in the number of qubits.

The disadvantage of a qLDPC code is that it requires higher levels of qubit connectivity in order to work. A lot of IBM’s research involved figuring out how to minimize the additional qubit connectivity needed to implement this code. The result is that IBM has come up with something called a Bivariate Bicycle (BB) code. Implementing this code will require a change in topology from their previous heavy hex topology. Previously, the heavy hex had four connections (3 couplers plus 1 control) for the qubits while future processors that support qLDPC will need seven connections (6 couplers plus 1 control). The paper provides details on how they intend to do this using a thickness-2 architecture along with development of a long range coupler.

So it is important to note that right now right now, this is still in the design phase and IBM has not yet demonstrated this on real hardware. Their roadmap indicates they will be working to achieve this demonstration within the next two years with a future processor codenamed Kookaburra.

We should also note that this research goes along with other related areas they are working on. Error correction algorithms become more efficient and more effective as the physical error rate continues to decline. So IBM is working on both improving the intrinsic error rate of the qubits through materials engineering, chip fabrication, etc. and they are also researching various error suppression techniques through better controls in order to prevent the physical errors from happening in the first place. Another important effort is their research in magic state distillation. This technique is needed to implement all the necessary gate operations, including non-Clifford gates, in order to implement a universal gate set that will allow programming of any possible quantum algorithm.

For more details about IBM research in qLDPC codes, you can read a blog they have posted here as well as the full paper in Nature magazine here. For more information about IBM’s research in magic state distillation, you can read their blog posting here and another paper in Nature magazine here.

Retirement of Cloud Simulators

What was not reported in the popular press, but also quite significant, is an announcement from IBM that they are retiring their cloud simulators as of May 15, 2024. For many years, end uses were able to simulate their quantum programs on several different simulators that were hosted on the IBM cloud. There are five different simulators that could support simulations of different types simulations for programs using up to 32 to 5000 qubits depending upon the specific simulator. A web page showing these simulators can accessed here. Part of IBM’s reasoning for retiring these simulators is their focus on utility-scale computing. Their belief is that because actual quantum hardware continues to increase its capability, it may be more effective for users to work with the actual hardware rather than to use simulators, at least for the larger programs.

As a replacement for the cloud simulation, IBM is recommending users take advantage of the  Qiskit Runtime local testing mode that will allow a user to simulate a program on their own personal computer before submitting it for a run on one of IBM’s real quantum processors. The Qiskit Runtime local testing mode includes a “Fake Backend” capability that allows one to run a simulation using parameters that mimic the characteristics of a specific IBM quantum processor.

IBM’s notice of the retirement of their cloud simulators can be seen here.

March 30, 2024