IBM Demonstrates First ‘Multi-Processor’ for Quantum Processing

IBM divided and parsed the quantum solution into two connected quantum processing units, demonstrating greater processing capability than either individual unit.

IBM recently published research in Nature on a world's first: the team demonstrated a pair of quantum processing units (QPUs) working simultaneously on the same algorithm. In the project, IBM connected two of its Eagle 127-qubit QPUs with a classical, real-time electronics link to demonstrate practical multi-processor quantum computing.

IBM has previously demonstrated QPUs with more than 1,000 qubits with its Condor superconducting qubit quantum processor. The Condor is 50% more dense than the previous record holder. Despite the increase in qubit density, IBM still believes that the future of quantum computing, like conventional high-performance computing (HPC), will involve multiple processors operating together. The newly demonstrated dual-QPU setup is the first step in that direction. The test circuit involved in the IBM demonstration required 142 qubits, greater than the 127-qubit Eagle QPUs.

The Quantum “Chandelier”

The figure below shows an IBM quantum computer with its cryogenic casing removed.

IBM quantum computer based on the Eagle processor with 127 qubits. Image used courtesy of Maeil Business Newspaper

Each round plate is a temperature separation stage with the temperature dropping with each layer. The wires or tubes are microwave waveguides that communicate with the individual qubits. The QPU is the small square near the center of the bottom layer, cooled to near absolute zero.

QPUs are not built as fixed quantum logic gates. Rather, the qubits are configured into specific algorithmic circuits while running. These circuits can be changed within the same quantum chip. In concept, this has some similarities to field programmable gate arrays (FPGAs). FPGAs configure a logic circuit at power-up time and then operate that circuit as though it were hard-etched silicon. The circuit can be different at each powerup or chip reset.

When a problem is assigned to a QPU, the quantum circuit design is compiled with conventional digital logic and transferred to the chip. The result is read via microwaves and converted and interpreted with digital logic. Once the result is accessed, the quantum chip can repeat the assignment with different data or be given a new configuration.

The Fundamentals of Qubits and Quantum Chip Architecture

Quantum computers operate on different principles than conventional digital computers. A digital computer is based on the relatively simple MOSFET switch. MOSFETs are combined into logic gates and flip flops etched in silicon, which perform computations and run code. The results, correct or not, are absolute. This is called deterministic computing.

Structure of 127-qubit IBM Eagle QPU. Image used courtesy of IBM

The equivalent fundamental component of a quantum computer is the qubit. A qubit, or quantum bit, can exist in multiple states simultaneously, giving it greater information and logic capability than an on/off MOSFET. Unlike deterministic logic, qubit logic operates based on probability. The qubits are continually in multiple states and reading them will collapse the multi-state qubit into a final, most probable result. It isn’t guaranteed correct but is at a high enough state of probability to effectively be called correct. 

Individual qubits are energized and accessed with microwaves through a waveguide at a specific frequency. Qubits are fabricated in a grid pattern to create a multi-qubit chip. The chip is cooled to near absolute zero during operation. The need to supercool the quantum device and the waveguide architecture are two of the challenges in expanding the qubit grid size of a quantum computer.

Doubling Up With Circuit Cutting

With a conventional CPU-based computer, more complex problems require more clock cycles. With a quantum computer, if the algorithm circuit is too complex for the number of available qubits, it must be broken into pieces. First, part of the problem is solved, then the quantum circuit is reconfigured, and the second part of the problem is solved. This is called circuit cutting, where the fabric of a complex quantum circuit is cut in two. The two halves are processed sequentially, and the sub-solutions are combined with conventional digital logic. Since the computation occurs within one QPU, sequential processing is referred to as local operations (LO).

IBM Quantum System Two with two IBM Quantum Eagle processors connected by real-time link. Image used courtesy of IBM

Sequential quantum computing can be used to solve many problems. However, it more than doubles the amount of time needed and increases the overall number of qubits required due to sampling overhead. The alternate solution, as demonstrated in the IBM experiment, involves splitting a quantum circuit across multiple QPUs, called local operations and classical communication (LOCC). In doing so, the effective circuit can be larger than either QPU could hold on its own. This allows the system to simultaneously compute cut circuits with the chips joined by conventional digital circuitry. 

In IBM's recent research, each QPU was connected to a conventional processing unit. These processors joined the two QPUs, even simulating some of the quantum phenomena that quantum computers depend upon. The net result was a larger effective qubit grid.

Quantum computers operate based on probability, which is inherently more error-prone than deterministic computing. Error rates are decreased by repeating the computation multiple times. For each cycle, the algorithm has to be compiled and injected into the QPU. With LO circuit cutting, the circuitry is less complex, but the software run, including the compile and injection cycle, must be repeated more times than with LOCC. LOCC only needs one compile cycle because the total circuit isn’t changed. The end result is faster operations with LOCC. However, LO yields higher-quality results because it doesn't require the external joining circuitry.

Supercomputing of the Future

IBM’s quantum-centric supercomputing (QCSC) roadmap envisions systems combining multiple QPUs with multiple classic processors like CPUs and GPUs with work divided based on which processor type is best suited to a given algorithm. However, this demonstration isn’t the last word. As QPUs are built with larger grids and with improvements in software stacks and conventional hardware, it may turn out that either LO or LOCC, perhaps both have a future in the quantum world.