IBM Extends Roadmap Up to 4,158 Qubits Using Multiprocessing and Advanced Software

IBM has extended its quantum processor roadmap and added four additional processors that will be introduced between 2023 and 2025. A key to these developments includes different types of multiprocessing resulting in a 4,158 qubit machine, codenamed Kookaburra, in 2025.

The first processor, called Heron, will be introduced in 2023 with a density of 133 qubits per chip. Although this may seem to be a step backwards form the 1,121 Condor chip schedule to also be introduced in 2023, it actually can be seen as a step forward because it pioneers several key architectural concepts that will be critical for future development. First, Heron will incorporate redesigned gates and a feature called tunable couplers. These are Josephson junctions placed between qubits to help isolate them and minimize crosstalk. The tunable coupler was first used by Google when they redesigned the 72 qubit Bristlecone chip to the 54 qubit Sycamore chip. It greatly improved the qubit quality metrics and allowed Google to proceed with their Quantum Supremacy experiment. Similarly, the tunable coupler will provide for faster gate speed and lower crosstalk than IBM’s previous processors.

The other key innovation with Heron will be the ability of multiple Heron chips to communicate classically and use common control hardware. Depending upon the ultimate cooling capabilities of the IBM System 2 enclosure that they are developing with BlueFors they may be able to place multiple Heron chips in the same dilution refrigerator. Although the classical communication between processors is not as powerful as quantum level communications, there still can be some immediate advantage to users. For example, if a user submits a problem that requires 1500 shots, they could run 500 of the shots on each processor and run the problem in one-third of the time. IBM is also developing software called entanglement forging and circuit knitting that might be able to take a quantum program that uses 300 qubits, split the program apart into three 100 qubit chunks, run each on one of the processors, and send the results to a classical processor that can combine the three results together in a process called circuit knitting.

The year 2024 will see IBM introduce quantum communication between processors with two different processors using two different methods. The first will be to configure a system consisting of three Heron chips within the same dilution refrigerator that are connected via quantum chip-to-chip couplers. IBM believes they can achieve with no degradation in qubit connectivity or qubit fidelity and it will create a 408 qubit chip called Crossbill. They indicate that as far as a user is concerned, even though the Crossbill physically consists of three individual chips, it will look no different than if all 408 qubits were on a single monolithic chip.

The second processor that IBM will develop in 2024 will be a 1386+ processor called Flamingo that will consist of three or more 462 qubit chips networked together. The architecture of this will be different from Crossbill because it will use what IBM calls long-range couplers for connecting qubit chips located in different dilution refrigerators through a cryogenic cable of about a meter long. Unlike the chip-to-chip couplers described above in Crossbill, IBM has indicated that the qubit connectivity and fidelities will not be able to maintain the same levels as on chip qubits. There will be some degradation, but it is too early for IBM to indicate how much. Nonetheless, it will require smart software to account for this degradation and include algorithms that can allocate the qubit operations to minimize this affect.

In 2025, IBM will continue extending these technologies to network together three or more of the Flamingo chips to create a 4,158+ system called Kookaburra and pave the way for continued developments based upon these technologies in 2026 and beyond.

It should be noted that this represents IBM’s development roadmap and its possible that the processors that IBM will make available to users might be different or have a later availability. In the meantime, IBM will be busy in 2022 with the release of their Osprey 433 qubit processor, a target to increase their Quantum Volume number by 4X from 256 to 1024, another target to increase their CLOPs measurement by over 3X from 2.9K to 10K, and to introduce the dynamic circuits capability to enable mid-circuit measurement and qubit reuse.

On the software side they will continue to enhance the Qiskit Runtime to make it easier and faster for users to implement various hybrid classical/quantum algorithms. They are also working on software that can help mitigate errors in the hardware so that users can achieve more accurate results and releasing introduced primitives, basic routines that make it easier for users to perform quantum computations without thinking about the intricacies of hardware and error handling. The overall goal with software is to increase accessibility, simplicity, and power of quantum computing for developers using their systems and accelerate the time when quantum advantage can be achieved with real world applications.

IBM has posted a news release and a blog article that describes additional details about their roadmap announcement that you can find here and here. A YouTube video is also available here.

May 10, 2022