Quantum Payment Optimization Chapter 2 -Faceoff Xanadu. Qiskit. DWave.
The principles of quantum mechanics, mainly tunneling, superposition, and entanglement, can be exploited to perform computations in one of two types of devices: a quantum annealer or a universal quantum computer. As the name implies, universal quantum computers can be used for any calculation or algorithm defined by the Church-Turing thesis. However, they are currently in their infancy in the noisy intermediate quantum state (NIQS), computing shallow algorithms with few parameters. Quantum annealers are used exclusively for optimization problems and finding a minimum, they are tested and proven to outperform classical computations in certain situations.
Universal Quantum Computers
Unlike classical computers, quantum computers rely on the quantum bit or qubit to represent data. The qubit differs from the bit in that it is in a superposition of two states and can take on a continuous range of values between them. Operators called gates are used to evolve the state of the qubit. They are unitary and reversible (irreversibility would cause a loss of information). Two common operations are placing the qubits in superposition (the Hadamard operator) and entangling qubits (the CNOT operator). The results of the computation are obtained when the states are measured. Thus, quantum computations are a prepare-evolve-measure cycle as opposed to the load-run-read cycle of classical computers.
There are many different architectures and designs for building a gate-based quantum computer. Superconducting and photonics-based QCs are gaining lots of attention and funding. To program these quantum circuits, several development libraries are being developed. Circuits are often first tested on simulators and then run on real quantum hardware. Two popular ones are Qiskit by IBM and Pennylane by Xanadu.
Qiskit stands for Quantum Information Science Kit and is designed to allow users to quickly program circuits that can be compiled on quantum computers, primarily IBMs superconducting transmon quantum computers. It consists of three modules for accessing the quantum computing stack:
Terra provides elements for programming circuits and optimizing them.
Aer provides a C++ simulator and performing noisy realistic simulations.
Ignis provides a framework for understanding and mitigating noise and error correction.
Qiskit is a well-developed SDK with the most extensive community supporting it.
Some of the challenges of quantum computing are their need to operate at cryogenic temperatures, fidelity (the ability for a qubit to remain incoherent), scalability, and a small number of qubits and circuit depth (how many operations can be performed). Some of these challenges can be overcome using photons and have been the focus of Canadian-based company Xanadu. Using light instead of particles introduces photonic modes or Qmodes instead of qubits. Using these Qmodes to make calculations is a different paradigm and results in continuous-variable quantum computing similar to analog computers. The change from qubits to qmodes requires a different set of gates and operations. The SDK strawberry fields have been designed to program photonic circuits. Xanadu has also developed a higher-level program, Pennylane, designed to work on any quantum hardware.
Quantum Annealers
quantum annealers are not universal and only run adiabatic quantum computing algorithms in which the system's energy is lowered until the lowest energy state, the global minimum, is found. They are mainly used to solve NP-hard optimization, formulated as quadratic unconstrained binary optimizations (QUBOs). These QUBOs are easily converted into an Ising model - a graph of magnetic dipoles that can interact with their Neighbours. The Ising model is perfectly mapped to the quantum processing unit (QPU) in annealers, a lattice of interconnected qubits. The number of configurable states in the Ising model represents possible solutions. The number of states is quickly simulated by exploiting the superposition of qubits. Since the states are entangled, adjusting parameters changes the possible solutions and the one with the lowest energy level can be found. Dwave produces the only truly quantum commercial annealer currently with over low-noise qubits. Its Ocean software provides a suite of tools for solving NP-hard problems that can be optimized and submitted to the QPU.
There are several types of simulated quantum annealers. For example, Fujitsu has a digital annealer that emulates quantum annealing using C-mos circuits to calculate the possible parallel states. Another notable annealer is Toshiba's simulated bifurcation machine. Bifurcation is the sudden branching into two parts representing qualitatively different solutions for a nonlinear system. This is used to model the superposition of |1> and |0> states of qubits. These methods are slower than quantum but provide stability and offer more qubits than quantum can currently deliver.
The three quantum computing architectures discussed here all have their unique advantages and challenges. For example, universal quantum computers have been able to differentiate themselves from annealers. However, annealers have developed and are in the position of strengthening their niche of solving np-hard problems. We will likely see further differentiation between the different universal quantum computing hardware and software, but the future is unknown, and all paths are in a superposition.