For decades, computer technology has steadily achieved astonishing gains in performance, but it is becoming increasingly unclear how much longer this rapid progress can continue without considerable technological innovation. A computer rendering of an all-optical switch that combines a graphene layer (pattern) and a nanowaveguide (gray pointer). Courtesy of Masaaki Ono. For example, the complementary metal oxide semiconductor (CMOS) transistors that have enabled such remarkable advancements in processing power now face physical constraints that could limit future gains in performance. “It is difficult to further improve the speed without increasing the power for CMOS devices, and this is a fundamental problem,” said Masaya Notomi, a senior distinguished scientist focused on optoelectronics at NTT Basic Research Laboratories in Atsugi, Japan. “In other words, it’s the end of Moore’s law,” he said, referring to the historic trend of processor capacity roughly doubling every few years even as manufacturing costs steadily fall. Similar limitations loom in the realm of computer memory, where the rate at which magnetic memory can be written and read has largely plateaued at frequencies of several gigahertz. All-optical switching technologies could potentially help propel the next generation of electronics past these hurdles. The systems are based on the notion of using laser light to directly control the physical state of another material. Optically flipping the magnetic bits in a hard drive, or modulating other optical signals in the context of a computer processor are examples. And although most of this work is early stage and remains largely the domain of academic researchers, the past 15 years have seen some remarkable demonstrations of how devices based on all-optical switching may enable fast and ultra-energy-efficient data storage, long-range signal transmission, and computer processing. The Kimel group is exploring all-optical switching strategies that use bursts of polarized light to magnetically encode data (top). Courtesy of A. Stupakiewicz and A. Kimel. Magnetic storage based on all-optical switching could allow far faster and more energy-efficient data recording than is currently possible (bottom). Courtesy of A. Stupakiewicz and A. Kimel. Writing with light Some of the most concrete progress has been with optical writing of magnetic memory. In a conventional hard drive, data is encoded by using a magnetic read/write head to flip the magnetic states of individual bits within the storage medium. Achieving this switching quickly requires additional power and also produces excess heat, but all-optical switching systems could offer a more energy-efficient route to accelerated magnetic recording. “We see already now that if we can switch with light, we don’t dissipate too much power in the recording medium,” said Alexey Kimel, a condensed matter physicist at Radboud University in the Netherlands. “Fundamentally, nothing limits the ability of light to switch magnets without leaving any energy in the medium.” The notion of optical writing of magnetic memory has been around for decades, premised on the idea of using a focused beam to directly stimulate switching of magnetic states. The first bona fide demonstration came from Theo Rasing’s group at Radboud University in 20071. Kimel, who was part of Rasing’s team at the time, said the researchers had been making progress using polarized light to manipulate magnetic spins in semiconductor materials. However, they remained unable to completely flip these magnetic states until they began working with gadolinium iron cobalt (GdFeCo), an alloy widely used in magneto-optical recording. The alloy was supplied by collaborators in Japan. “We took those materials from our friends and studied the laser-induced dynamics, and absolutely unexpectedly found that something switches in the medium,” Kimel said, noting that the effect was sufficiently robust so that they were able to confirm it within the space of an evening. The Kimel group has devised optical elements that can shape the polarization of femtosecond laser pulses to achieve ultrafast magnetic recording with the help of light. Courtesy of A. Stupakiewicz. Gadolinium appears to be the “secret sauce” in this alloy, generating a powerful magnetic field that favors light-induced switching — and at first, it seemed like the field might both begin and end with GdFeCo. “There’s something weird about gadolinium where you can do this single-pulse, heat-induced switching,” said Eric Fullerton, who studies magnetic recording at the University of California, San Diego. “And for the first seven years, that was the only material that ever showed it.” But subsequent research by Fullerton and colleagues from 20142 demonstrated the feasibility of optical switching with a broader range of ferromagnetic materials. However, this success comes with some caveats. First, there appears to be a stringent requirement for circularly polarized light, which is not essential for GdFeCo-based systems. Additionally, multiple rapid pulses are necessary for switching in ferromagnetic materials, rather than just a single burst. This initially required the use of expensive femtosecond pulse lasers, which can cost hundreds of thousands of dollars. But Fullerton said switching can now be achieved with more manageable regimens of picosecond laser bursts. Alternately, he added, combining a layer of GdFeCo with a ferromagnetic layer can achieve single-pulse switching with a more versatile range of magnetic materials. In an experimental setting, these systems can smash the speed records set by standard magnetic hard drives. “Information gets encoded within about a picosecond,” Fullerton said of the multipulse ferromagnetic systems used by his group. “You’re going like a thousand times faster.” These devices take longer to cool between rounds of writing, however, which means repeated cycles of magnetic switching will still take longer than desired. Kimel’s group has been able to accomplish such ultrarapid switching with minimum heat loss in iron cobalt-doped garnet. Here, they achieved switching within 20 ps and were able to repeatedly flip these states at a rate of roughly 60 GHz, which is an order of magnitude faster than conventional hard drives. However, this exotic material is not well suited for routine use in commercial products. A scanning electron microscopy image of the Notomi group’s all-optical switching device. Courtesy of Masaaki Ono. Kimel said the major commercial players in the hard drive space, such as Western Digital and Seagate, have generally not made prominent research investments in the space. However, these companies have moved to license a number of patents related to all-optical switching, including an unsuccessful effort to buy up the Radboud University team’s intellectual property portfolio. Kimel believes further progress in academia may spark additional interest from industry. “The profit for them is not obvious at the moment. Maybe in the future, but not tomorrow,” he said. “But I’m still quite optimistic.” Photon-powered processing There is also a broader ecosystem of computing applications for which all-optical switching could eventually prove transformative. From the perspective of FTT’s Notomi, this technology could offer a powerful complement for existing CMOS devices that greatly boosts computing power without requiring substantially more energy. “We’re not talking about an all-optical computer,” he said. “We want to deliver very fast and energy-efficient photonic processors inside the CMOS processors.” The notion of optical writing of magnetic memory has been around for decades, premised on the idea of using a focused beam to directly stimulate switching of magnetic states. Such devices employ materials with nonlinear optical properties, where an input light beam is able to interact with matter in a manner that alters the behavior of an output light beam. Thus, one light beam acts as the switch controlling another, in a manner reminiscent of electron-based switching in conventional electronics. Notomi said early efforts at such photonic switches were extremely inefficient, requiring excessively high energy levels to achieve a switching event. “One of the reasons is that the extent of light-matter interaction is too small,” he said. One way to improve this is to physically confine the input beam, and in 2010, Notomi and colleagues showed they could use photonic crystals with tiny nanoscale cavities to achieve such confinement3. “We achieved very small energy consumption of just 400 aJ per bit,” Notomi said. “That’s hundreds of times smaller than the conventional all-optical switching devices at that time.” However, he was less impressed with the speed of switching, which was relatively comparable to the other devices — on the order of 20 to 40 ps. Because of the inherent limitations of the photonic crystal approach, the group switched gears to focus on nano- waveguides, tiny devices that can effectively control the flow of light or energy. To maximize the light response, the team coupled its waveguides with graphene. This material exhibits excellent nonlinear optical properties and a rapid response time to illumination, but it is too thin to use on its own, comprising just a single-atom-thick layer of carbon. But the nanoscale waveguide and graphene proved to be highly effective in combination. In work published this past November, Notomi’s team showed that this design can achieve the fastest and most energy-efficient all-optical switching described to date, requiring just 35 fJ of energy to achieve a switching time as low as 100 fs4. Even Notomi was surprised by this. “We had expected the response time should be around 1 ps,” he said. “We are currently working on trying to find out why we can achieve such a fast speed.” The work is still very preliminary, and Notomi estimated that it would be at least a decade before photonic integrated circuits would be developed. But he also said this area of technology is in active development at his company, NTT, and other computing giants such as IBM and Intel are investigating photonic processor technologies as well. Even if all-optical processing remains over the horizon, advancements in ultrafast, low-energy switching may soon find their way into other aspects of the computing pipeline, such as the interconnects that transfer information between various components of the computer, or even for rapidly relaying data over long-distance broadband networks. Notomi said many such systems use a communications process known as time division multiplexing, where current data transfer speeds top out at 40 Gb/s. “But our graphene switch will be able to achieve 1 Tb/s, which is 25× faster,” he said. This would also require parallel advancements in laser pulse generation and detection technologies, but the resulting systems could ultimately provide a significant turbo boost to global communications networks. References 1. C.D. Stanciu et al. (2007). All-optical magnetic recording with circularly polarized light. Phys Rev Lett, Vol. 99, Issue 4, p. 047601. 2. C-H. Lambert et al. (2014). All-optical control of ferromagnetic thin films and nanostructures. Science, Vol. 345, Issue 6202, pp. 1337-1340, https://science.sciencemag.org/content/345/6202/1337. 3. K. Nozaki et al. (2010). Sub-femtojoule all-optical switching using a photonic- crystal nanocavity. Nat Photonics, Vol. 4, pp. 477-483. 4. M. Ono et al. (2020). Ultrafast and energy-efficient all-optical switching with graphene-loaded deep-subwavelength plasmonic waveguides. Nat Photonics, Vol. 14, pp. 37-43.