In a noisy and imprecise world, the definitive 0s and 1s of today’s computers can get in the way of accurate answers to messy real-world problems. So says an emerging field of research pioneering a kind of computing called probabilistic computing. And now a team of researchers at MIT have pioneered a new way of generating probabilistic bits (p-bits) at much higher rates—using photonics to harness random quantum oscillations in empty space.
The deterministic way in which conventional computers operate is not well-suited to dealing with the uncertainty and randomness found in many physical processes and complex systems. Probabilistic computing promises to provide a more natural way to solve these kinds of problems by building processors out of components that behave randomly themselves.
The approach is particularly well-suited to complicated optimization problems with many possible solutions or to doing machine learning on very large and incomplete datasets where uncertainty is an issue. Probabilistic computing could unlock new insights and findings in meteorology and climate simulations, for instance, or spam detection and counterterrorism software, or next-generation AI.
The team can now generate 10,000 p-bits per second. Is the p-circuit next?
The fundamental building blocks of a probabilistic computer are known as p-bits and are equivalent to the bits found in classical computers, except they fluctuate between 0 and 1 based on a probability distribution. So far, p-bits have been built out of electronic components that exploit random fluctuations in certain physical characteristics.
But in a new paperpublished in the latest issue of the journal Science, the MIT team have created the first ever photonic p-bit. The attraction of using photonic components is that they operate much faster and are considerably more energy efficient, says Charles Roques-Carmes, a science fellow at Stanford University and visiting scientist at MIT, who worked on the project while he was a postdoc at MIT. “The main advantage is that you could generate, in principle, very many random numbers per second,” he adds.
At the heart of the team’s p-bit is a component called an optical parametric oscillator (OPO), which is essentially a pair of mirrors that bounce light back and forth between them.
The light does not travel in a physical vacuum, however, in the same sense that outer space is a vacuum. “We do not actually pump a vacuum,” Roques-Carmes says. “In principle...it’s in the dark. We’re not sending in any light. And so that’s what we call the vacuum state in optics. There’s just no photon, on average, in the cavity.”
When a laser is pumped into the cavity, the light oscillates at a specific frequency. But each time the device is powered up, the phase of the oscillation can take on one of two states.
Which state it settles on depends on quantum phenomena known as vacuum fluctuations, which are inherently random. This quantum effect is behind such well-observed phenomena as the Lamb shift of atomic spectra and the Casimir and van der Waals forces found in nanosystems and molecules, respectively.
“We can keep the random aspect that just comes from using quantum physics, but in a way that we can control.”
—Charles Roques-Carmes, Stanford University
OPOs have previously been used to generate random numbers, but for the first time the MIT team showed they could exert some control over the randomness of the output. By injecting the oscillator with incredibly weak laser pulses—so weak there is less than a single photon per pulse on average—they could alter the probability with which it takes a particular phase state.
This ability to influence, but not deterministically set, the phase state of the OPO makes it a promising way to generate p-bits, say the researchers. “We can keep the random aspect that just comes from using quantum physics, but in a way that we can control the probability distribution that is generated by those quantum variables,” says Roques-Carmes.
The team says they were able to generate 10,000 p-bits per second of signal obeying a given probability distribution. In other words, they can make 10 kilo-p-bits per second that—at the present level of probabilistic computing technology at least—seem to behave in the ways required to build a probabilistic computer.
The team built their device using a large tabletop set of optical components, so building a practical probabilistic computer using these principles will require considerable work. But Yannick Salamin, a postdoc at MIT’s Research Laboratory of Electronics, says there are no fundamental barriers. “We wanted to show the physics of it, so we built this large system,” he says. “But if you are interested in scaling up and miniaturizing and so on, there are experts in this area that can do it.”
Kerem Çamsari, assistant professor of electrical engineering and computer science at the University of California, Santa Barbara, says the MIT group’s work is “very exciting,” but he’d like to see this proof of concept built out to a wider scale than just individual p-bits. “It would be great to see follow-up work beyond single p-bits to correlated photonic p-circuits,” he says.
Morgan Mitchell, professor of atomic quantum optics at the Institute of Photonic Sciences (ICFO) at the Technical University of Catalonia, in Barcelona, says the new work “is interesting in the context of classical optical computing,” but he cautions against reading too much into the initial results. “It will be interesting to see if the authors can quantify” the extent to which the p-bits’ state is due to vacuum-generated randomness rather than other sources of apparent randomness such as environmental noise or device imperfections.
- Intel Starts R&D Effort in Probabilistic Computing for AI ›
- Waiting for Quantum Computing? Try Probabilistic Computing ›
Edd Gent is a freelance science and technology writer based in Bengaluru, India. His writing focuses on emerging technologies across computing, engineering, energy and bioscience. He's on Twitter at @EddytheGent and email at edd dot gent at outlook dot com. His PGP fingerprint is ABB8 6BB3 3E69 C4A7 EC91 611B 5C12 193D 5DFC C01B. His public key is here. DM for Signal info.
Margo Anderson is the news manager at IEEE Spectrum. She has a bachelor’s degree in physics and a master’s degree in astrophysics.