Ease of scaling could make probabilistic computing competitive with current-day quantum computers, though limitations of the design prompt researchers to dub it the “poor man’s qubit.”
Researchers at Purdue University and Tohoku University have built quasi-quantum “probabilistic computer” using a modified type of magnetoresistive random-access memory (MRAM) to approximate the behavior of a qubit, the building block of a quantum computer.
In classical computing, a bit can either hold a value of 0 or 1, while a qubit can hold values of 0 and 1 simultaneously. The Purdue/Tohoku probabilistic computer uses a p-bit, which “rapidly fluctuate” between 0 or 1. In a whitepaper published in Nature on Wednesday detailing their proof-of-concept, researchers were able to factor 945 and 35,161 into primes using an 8 p-bit machine. Quantum computers that can factor long integers in trivial amounts of time could be used to break widely-used RSA encryption schemes, prompting research into post-quantum cryptography.
While quantum computers have been capable of this scale of integer factorization for years, the MRAM-derived p-bit design can be operated at room temperature, while quantum computers rely on aggressive helium cooling to operate. Different types of qubits—superconducting, topological, trapped ion, etc.—are being actively evaluated by researchers globally, though the relative novelty of these designs, combined with their inherent physical properties, presents difficulties in scaling up for many-qubit machines, an encumbrance not applicable to p-bits.
Combined with the modest cooling requirements, the prospect of scalability for p-bit systems in the near-term appears quite possible—the p-bit design was demonstrated to provide ten times higher energy efficiency, with a 300x area advantage compared to conventional computers.
innoThe capabilities of p-bits are not identical to true qubits, however, leading Purdue’s Supriyo Datta to call p-bits “a poor man’s qubit.” According to the whitepaper, “for a subclass of quantum systems, quantum annealing can be approximated with replicated p-bit networks.” D-Wave is the largest commercial purveyor of quantum annealer systems, which are oriented toward solving quadratic unconstrained binary optimization (QUBO) problems.
Current commercial use of quantum annealers includes path optimization and training for machine learning, as these tasks can easily be expressed as QUBO problems.