Quantum technologies promise to revolutionize computation and, with it, an enormous list of sectors and society’s challenges. Still, unfortunately, they don’t seem to be functional any time soon. Halfway between them and traditional computers, the so-called physics-inspired computing like the one developed by Quside takes the maximum possible advantage of current computing technologies, saving both time and money.
Originally published in MIT Tech Review (ES)
Even if Artificial Intelligence catches most of the attention within the current technological paradigm, Quantum Technologies is another field with also the highest potential to revolutionize the world as we know it.
This field has risen so much interest that, in the last years, the EU has invested more than 1.000 million euros in quantum-related projects, both public and private. This number is only expected to grow in the next decade, thanks to different initiatives such as Quantum Flagship and the European Quantum Communication Infrastructure (EuroQCI). In the same way, the number of spin-off companies that commercialize products based on these technologies is also growing, with some universities even offering quantum engineering degrees.
Together, this financial push along with interest from the private side fosters an industrial quantum ecosystem that, doubtlessly, has come to stay and to transform our ways of living, thanks to quantum physics.
Nevertheless, this “quantum revolution” is not the first one we experience: we already have technologies that take advantage of some quantum effects. Still, there are some critical differences between this current revolution and the previous ones:
These technologies are already an industrial reality, such as quantum random number generators (QRNG). These devices, commercialized by Quside and other companies, can yield safer connectivity and communications, thanks to the genuine unpredictability that quantum physical principles ensure.
But this second quantum revolution has its own caveats, too. Even if it is possible to create the conditions to study, design, and fabricate such systems and control them with high enough precision, scaling them up is still a challenging scientific and commercial problem.
This is the case with, for example, quantum computers. Even if companies already exist that develop and sell this kind of computer, they consist of several qubits too low (and with an insufficient precision) for them to be considered a practical and accessible solution in the short- or mid-term.
Suppose we put our hopes on quantum technologies to solve the most critical social, economic, and industrial challenges we expect to face during the following decades, which demand a computational power that current methods cannot handle. In that case, it is mandatory to overcome this problem or, at least, to find a way to avoid it. That’s the reason why a lot of companies and research groups have started to focus on new ideas that can yield advantages over classical computation without having to wait for engineers to solve these enormous (though exciting) challenges that quantum computing poses. It is the case of the so-called physics-inspired computing, on which technological leaders such as Toshiba, Fujitsu, Microsoft, and Amazon are already working.
This new type of computation, which can be helpful in fields as relevant as machine learning, takes advantage of the mathematical apparatus that was initially developed for quantum computers to solve problems in platforms available today. This translates into a substantial improvement in both cost and time investments.
To understand how physics-inspired computing works, imagine lying on the beach with a glass of soda. Even if it looks flat when observed far away, the sand’s surface is a chaotic and irregular set of innumerable valleys and peaks. If you want to break such an idyllic situation, you could ask yourself: Which is the lowest valley? The classical solution consists of measuring and writing down all of the valley’s heights (which will probably differ by indistinguishable fractions of our measurement’s apparatus precision). However, this approach is by no means viable if we want to consider a sand extension more extensive than the towel we are lying on.
So if you want to solve this problem within a reasonable amount of time, a new strategy is mandatory. And here is where physics plays its role: if, by the thermodynamic principles, every system tends to evolve towards a steady state in which its energy is minimized, could we pour our soda over the sand in a careful enough way such that, by gravity’s action, the liquid gets stored in the deepest valley, in which its potential energy is minimum? In this case, the physical evolution of the system would do the work for us, and we could keep enjoying our holiday. Obviously, the answer to this question is yes, even though this “carefully enough” (adiabatically, in physics jargon) might be a bit difficult to execute.
There exist a lot of optimization techniques inspired by physical processes. Still, generally, all of them make use of two components: first, the use of acceleration platforms such as GPU or FPGA, which allow the programmer to take advantage of every single watt the computer can yield; and second, a physics simulation algorithm able to find those valleys on the beach within the shortest time possible.
Further, most of these physics-inspired computing algorithms are stochastic. To find the minimum energy state, they propose a series of actualizations of the variables conforming to the system and accept or reject them with a certain probability that depends on its energy. This procedure demands a massive amount of random numbers, especially for complex problems.
Actually, given the efficiency and velocity of currently available computing platforms, along with the high demand for randomness, the random number generation speed is the limiting factor in most applications. It is also known that a bad quality in the random streams (manifested by correlations between numbers at long ranges) could influence the algorithms’ capability to find satisfactory solutions. In such cases, both quality and velocity on the generation of such random numbers are essential.
With this philosophy in mind, Quside uses its experience as a QRNG manufacturer to provide random numbers with the highest quality and speeds possible. Furthermore, its generators can directly interact with hardware acceleration platforms, such as GPU and FPGA, which increases the efficiency of transmitting such random numbers and facilitates its usage.
This idea of yielding both maximum efficiency and quality, along with its user-friendly interface, has allowed Quside to demonstrate accelerations of up to 10x in some use cases computations associated with Monte Carlo simulations within fields apparently as different as finance, biochemistry, and logistics.
On top of that, Quside also researches new physics-inspired algorithms to find energy valleys more efficiently. On the one hand, we study how to translate such problems so that the computer can work with them. On the other hand, we research novel physical models that optimize better and faster. To this end, we merge ideas taken from parallel genetic algorithms and ingredients inspired by quantum computing that allow the system to travel from one valley to another with less effort.
Even though it may look like your problems have nothing in common with sand and soda, physics-inspired computation’s potential is enormous. Many problems of industrial interest can be translated into the task of finding the minimum-energy state of a physical system, from medical research, like the solution of proteins’ structure for the development of new drugs, to logistics problems like the minimization of the cost of distribution, and bank-related like the optimization of financial portfolios. All of them are problems that, without any doubt, can transform people’s lives. With the help of this technology, they could have more optimal and much faster solutions than the current ones, in a horizon much closer than functional quantum computers.