Expectations for Digital Annealer, a Technology That Leads the New AI Era
Digital Annealer is a digital circuit inspired from quantum mechanics. This new technology can rapidly solve combinatorial optimization problems, which are difficult to solve with today's general-purpose computers.
It is predicted to take several decades before we can put to practical use the highly anticipated quantum computers of the next generation. However, society already has myriad problems that need to be solved, so to develop solutions to these problems using AI technology, we must first solve combinatorial optimization problems.
Digital Annealer is attracting attention worldwide as a computer that can perform combinatorial optimization at extraordinary speeds. This technology can be used immediately and is now the technology closest to being put into practical use. We spoke to Hirotaka Tamura, a research fellow at Fujitsu Laboratories and an innovator who pioneers technology that yields new possibilities, about how he engages in innovation and what ideas he can offer.
Toward Solving Combinatorial Optimization Problems at High Speed
"Many things in this world can vary significantly in quality depending on how well their components are combined. When combining one thing with another, it is important to maximize value, minimize cost, and enhance synergy. For example, when figuring out the shortest route to visit all major cities in the world and then return, mapping out optimal routes for self-driving cars, or optimizing investment portfolios, there must be a system that can constantly determine the optimal combination of actions when the task of decision-making is left up to AI or robots.
"Adding elements results in an exponential, unmanageable increase in combinations, a number too vast for today's computers to determine which combinations are good and which are not. Though computers and their features have evolved significantly, we now face the problem of calculations outpacing technology."
Reconsidering the Optimal Solution at This Moment
How did you arrive at this reverse innovation concept, which uses today's digital circuits to realize a technology that the quantum computing of the future aims to achieve?
"Compared to the rate of technological advancements preceding the invention of digital integrated circuits, improvements in performance have more or less stagnated. However, the realities facing society continue to demand better computer performance. Everyone is saying, 'AI is amazing. Look at what we can do!' However, if hardware advancement stops, the technologies we dream about will never come to fruition.
"As for hardware that specializes in solving combinatorial optimization problems, quantum computing has a technology called quantum annealing, which many researchers now study. However, many unknowns remain regarding what kinds of problems quantum annealers can solve. I often doubted whether the results obtained were in fact the optimal solution, and I even started to doubt whether a quantum computer is needed. This caused me to think of how it might be possible to achieve the same objective at high speed using digital circuits.
"Even with a digital computer, if the hardware is tailored to this purpose, performance can be improved substantially. Despite it being dedicated hardware, if it can be employed in various applications to solve combinatorial optimization problems, there is a big market for it, and many people can make use of it. Thus, Digital Annealer was developed by attempting to create hardware that is both highly functional and versatile."
Digitally Transcending the Limits of Digital Technology
Research labs create various experimental devices and test hypotheses. * This photo shows a past study.
What are some of the obstacles and challenges you face when digitally developing hardware tailored to solve combinatorial optimization problems?
"Some obstacles remain, but when developing the Digital Annealer, the first challenge was to make it run as fast as possible. How to achieve the speed? That was our challenge.
In addition to today's technological approaches, we are evaluating various solutions to successfully develop a high-speed digital annealer. One major challenge is to figure out how to implement this in an appropriate order to maximize competitiveness.
"The second challenge is to ensure that it is sufficiently versatile as a piece of hardware for solving combinatorial optimization problems. In other words, it must have capacity to facilitate use in various settings and a large market that demands it. To achieve this, we devised a fully connected architecture that can combine every bit.
In any case, the Digital Annealer is a technology that is as yet incomplete. We will continue to do research to refine it."
Even If It Seems Unlikely to Work, We Learn by Attempting
What methods do you use to find the optimal solutions to problems?
"Methods for solving combinatorial optimization problems using digital circuits were intensively studied during the second neural net boom, which spanned from the late 1980s to the 1990s. However, in those days, focus was placed on processor performance, which was improving visibly, so there were few efforts to build dedicated hardware besides processors.
"Developing dedicated hardware was something that had not been previously attempted. When we started designing the Digital Annealer, it was not some sort of great revelation, but rather a series of attempts that we made without any expectation to succeed. Some results were surprisingly hopeful. If we had not conducted such simulations then, I do not think we would have made these discoveries.
"If logical thinking is all it took to create something like this, it would be no hassle at all. Once you have racked your brain as much as you can, the next step is to explore various directions by trial and error. You must try all kinds of things without expecting success. Even ideas that do not seem promising at first can lead to unexpected revelations when you try them. I think it is crucial to attempt to pursue an idea and then use what you learned in that attempt to take the next step. This is one of the most enjoyable parts of research."
Conceptualizing a Larger Flow While Facing Immediate Problems
Massive piles of research notes can be found in the wake of Tamura's research.
"The way I approach my research is to think about the field I am studying, the problems of current technology, what methods can solve such problems, and what advancements will occur in the future. About 70% of this thought process consists of thinking about hardware implementation and structure, as well as the technical challenges that occur daily. The remaining 30% is to consider the larger technological flow.
"When I was a university student, the age of vacuum tubes was coming to an end. Transistors were developed, which later became integrated circuits. I observed how this technology's scale steadily expanded, and I watched dynamic changes taking place. That is when I naturally began to wonder what lay ahead. The development of the Digital Annealer was the result of asking, 'How should the future of hardware be?'
"People tend to think we can simply rely on great visionaries to come up with great ideas, but in actuality, it is people like myself who are solving individual problems while following the flow of technology who must think of solutions themselves. In fact, I think that the people who are struggling with the challenges immediately in front of them are the ones who have a better view of what lies ahead."
Those familiar with Tamura's career as a researcher surmise that he found the path to the Digital Annealer by his knowledge of both current computer architecture and quantum mechanics as well as by approaching the problem from his background in both digital and quantum science. Instead of following the path of quantum science because it is a current trend, the root of his innovation may stem from the attempt to reverse this concept, and to consider how the desired results could be achieved faster through digital technology.
In Part II, we speak with Tamura about the collaborative research he conducted with the University of Toronto to materialize his idea as well as his thoughts on the Digital Annealer.
(Continue to Part II.)
Hirotaka Tamura, research fellow at Fujitsu Laboratories
Hirotaka Tamura graduated from the Department of Electrical and Electronic Engineering of the University of Tokyo in 1977. He completed his doctoral course at the same university in 1982, whereupon he joined Fujitsu Laboratories. After studying the Josephson device, the quantum effect device, and other topics, he began researching CMOS circuits.
He is a research fellow at Fujitsu, a Doctor of Engineering, a recipient of the 51st Okochi Memorial Prize, and an IEEE Fellow.