UCSB Research Advances Probabilistic Computing in Optimization Benchmarks Beyond Quantum Systems
Probabilistic Computer
In the quest to solve challenging optimization issues, new research from the University of California, Santa Barbara (UCSB) has produced an important discovery: it shows that the probabilistic computer (p-computer) can outperform a top quantum annealer on common “spin-glass” benchmarks.
The p-computer, which is constructed from probabilistic bits (p-bits), continues to demonstrate its usefulness even though it is still unclear when a commercial quantum computer will be able to outperform classical (non-quantum) machines in speed and energy efficiency for real-world combinatorial optimization problems. Associate Professor Kerem Çamsarı of UCSB’s electrical and computer engineering (ECE) is in charge of this specialty field. His team has contributed to the development of p-computing as a viable substitute for addressing these difficult problems.
You can also read Cryogenic Quantum Computing future: from mK to Mega-Qubits
Establishing a New Classical Baseline
“Pushing the Boundary of Quantum Advantage in Hard Combinatorial Optimization with Probabilistic Computers” describes the discovery. The study directly addressed a claim that a privately built quantum computer had previously solved spin-glass problems faster and more effectively than any competing technology. It was led by postdoctoral researcher Shuvro Chowdhury from the Çamsarı lab and included 14 co-authors, including Çamsarı and ECE department chair Luke Theogarajan.
Using these common spin-glass benchmarks, the UCSB team showed that their p-computer design could outperform a state-of-the-art quantum annealer. “Power of p-bits and their future in computing” is what Professor Theogarajan said this result demonstrates.
According to the study, the p-computer offers a convincing and scalable classical approach to resolving challenging optimization problems when it is co-designed with hardware to perform potent Monte Carlo algorithms. The scientists demonstrated that the p-computer was faster and more efficient by concentrating on two important techniques used for 3D spin glasses: adaptive parallel tempering and discrete-time simulated quantum annealing.
A “new and rigorous classical baseline, clarifying the landscape for assessing a practical quantum advantage,” is established. According to Çamsarı, “The answer is no” when they ran a problem this size and asked if a quantum machine was a better solution than any other method. Nevertheless, the study also draws attention to an important comparison: it discovered that p-computers are not more effective than quantum machines in handling large-scale problems. This finding casts doubt on the idea of a useful quantum advantage and creates new research opportunities.
You can also read Centre for Commercialization of Regenerative Medicine CCRM
The Challenge of Scale: Leveraging Millions of P-bits
Building p-computers at previously unheard-of sizes was necessary to achieve this performance. To study behavior at even greater scales, the team customized existing chips and ran simulations on CPUs using millions of p-bits.
In actuality, the decision to employ so many p-bits was just coincidental. When working with a very large number of p-bits, two Ph.D. students from the University of Messina in Italy, Andrea Grimaldi and Eleonora Raimondo, saw non-intuitive favorable behavior. According to Chowdhury, it took around a year to come up with a hypothesis describing why using a large number of p-bits in parallel boosts performance in such an unexpected way.
After two years of work, Chowdhury produced a report that proved the greatly scaled p-computer’s efficacy. Çamsarı clarified that although their earlier chip had just tens of thousands of p-bits, they may now create one with millions because to advancements in semiconductor technology.
P-bits: A Scalable and Implementable Technology
The algorithms were demonstrated to be “readily implementable using currently available hardware” by the researchers. The team demonstrated that a 3 million-p-bit processor might produce better outcomes by collaborating with chip designers and running simulations. This simulation, which was carried out with Taiwanese chip manufacturer TSMC, verified that such a chip could be produced today with current technology. Massive parallelism could be used by specialized processors to significantly increase energy efficiency and speed up certain computations by orders of magnitude.
Probabilistic bits, often known as P-bits, are defined as falling in between quantum bits (qubits) and classical bits (0 or 1). Because they can be constructed and run at ambient temperature, they are intrinsically probabilistic and classical, avoiding the major cooling and error correction issues that many quantum computers encounter. P-bits are frequently referred to as a “poor man’s qubit” since they employ less complex technology to carry out a useful subset of computations that qubits can.
Magnetic Tunnel Junctions (MTJs) in conjunction with CMOS technology can be used to construct P-bits, which fluctuate between states according to an energy barrier. This method enables them to be utilized not just for combinatorial optimization but also for machine learning, imitating some elements of quantum systems and acting as stochastic neurons in neural networks.
In a separate study, Çamsarı and colleagues at Northwestern University compared the asynchronous architectures his group initially created to synchronous probabilistic computers with concurrent p-bit updates. This work accomplished two significant milestones: it showed that very efficient p-bits can be produced by controlling magnetism with voltage, and it demonstrated that a precisely synchronized architecture, in which all bits update “like dancers moving in lockstep,” may match the performance of traditional architectures.
You can also read Sparrow Quantum gets €27.5M for Photonic Quantum Computing
