A quantum processing unit, or QPU for short, is the brain of a quantum computer. It is where the qubits reside, and where computation takes place. Like the central processing unit, or CPU, of a classical computer, a QPU needs quite a bit of supporting infrastructure. Also like a CPU, that supporting infrastructure can vary quite a bit. In fact, quantum computer hardware varies considerably more than classical computer hardware at this point.
Analogies are often made, but they sometimes fall short. For example, a Microcontroller Tips article titled “What’s a quantum processing unit?” compares QPUs to microcontrollers, but QPUs don’t actually control anything; in fact, they need to be controlled themselves. Furthermore, challenges cited in regard to control, temperature, and fabrication apply to superconducting processors and silicon spin processors, but not to all modalities. Neutral atoms, specifically, have none of the cited limitations, which makes an important point about the variety of quantum computers that are in development. In fact, a Medium article titled “Quantum Processing Units (QPUs)” by QuAIL Technologies makes that point.
A Huawei article titled “What is Quantum Processing Unit(QPU)” compares QPUs to CPUs, which is a little bit closer. The claim about faster calculations needs to be clarified, however, because QPUs actually calculate much slower than CPUs. They calculate with greater efficiency, though, so computation time can be less for certain classes of problems. Like the Microcontroller Tips article, the noise, scalability, and connectivity issues cited in this article apply to some modalities, but not to neutral atoms and other modalities.
The closest analogy to a QPU, at least in regard to how QPUs will be used in the future, is the graphics processing unit, or GPU. Most computation will continue to be performed by CPUs, as it is now, and select calculations will continue to be performed by GPUs and other specialized processors, as they are now. However, one of these specialized processors included in this mix will be QPUs. Although QPUs and GPUs will perform different calculations, their use as a sort of roleplayers will be the same in this sense.
It is worth noting that quantum processing unit power has no standardized metric. The first comparison that is always made is the number of qubits available. However, the capabilities and efficiency of QPUs are influenced by coherence times, connectivity, gate speeds, error rates, and other factors. Manufacturers publish metrics that make their own processors look capable, and so the industry could use an impartial standards organization.
The quantum technology that goes into a quantum processor can vary significantly. The computing process can vary quite a bit, as well. Just a few examples of qubit modalities and their supporting technologies to showcase the extent of this variety include:
· Neutral atoms in vacuum chambers, cooled and controlled by lasers
· Electronic circuits, cooled by dilution refrigerators and controlled by microwaves
· Individual electrons, trapped in vacuum chambers like ionized atoms, but cooled by dilution refrigerators and controlled by microwaves
· Photonic integrated circuits at room temperature, but with cryogenically-cooled detectors and controlled by physical hardware
There are quite a few more modalities in development than are listed above, but these four alone already demonstrate quite a bit of variation. Some of the other modalities use similar technologies, but they differ in what actually constitutes a qubit.
The potential applications of quantum computation are classically intractable problems. Because QPUs are much slower than CPUs, CPUs will continue to solve most real-world problems. However, despite this speed disadvantage, QPUs are much more efficient at solving certain really hard problems, including problems that are infeasible for high performance computing (HPC). Some of the challenging problems for which quantum computation is being researched include:
· Combinatorial optimization problems, which are a broad classification of problems that become harder to solve as problem sizes scale upward
· Quantum chemistry, which will enable the precise simulation of molecules that can only be approximate with classical computation, if they can be simulated at all
· Machine learning, which involves voluminous classical data that can be exponentially compressed as it is mapped to qubits
· Integer factorization, which, unfortunately, has applications in breaking public cryptosystems and making much of the world’s communication vulnerable to decryption
· Random number generation (RNG), which has applications in cybersecurity and artificial intelligence (AI)
· Science, in general, ranging from the physics governing the qubits to the engineering challenges of controlling them, and much more
RNG for cryptographic purposes is already being commercialized, and scientific advancements are forthcoming continuously. The next of these to become reality will likely be combinatorial optimization problems. The Rydberg states of neutral atoms have been proven to naturally solve several classifications of these problems, and the largest publicly-available quantum computer is currently the 256-qubit Aquila.
For more-specific examples of some of these applications, be sure to check out “Top Applications Of Quantum Computing for Machine Learning” and “Quantum Computing Use Cases for Financial Services.”
.svg){kind=small}
