This episode of the Physics World Weekly podcast features a lively chat about some of the best physics done this year as we unveil our Top 10 Breakthroughs of 2020. Our choices run the gamut from medical physics to particle astrophysics – and we even have two “Holy Grails” in the fields of superconductivity and semiconductor physics.
The Top 10 serves as the shortlist for the Physics World Breakthrough of the Year award, which will be announced on 17 December. Links to all the nominees, more about their research and the criteria for the awards can be found here.
One of the highlights in the Physics World calendar is the announcement of our Breakthrough of the Year, which will be made this year on Thursday 17 December.
Today, we are revealing the 10 finalists for 2020, which serves as a shortlist from which we will pick the Breakthrough of the Year.
This year’s Top 10 Breakthroughs were selected by a crack team of five Physics World editors, who have sifted through hundreds of research updates published on the website this year. In addition to having been reported in Physics World in 2020, our selections must meet the following criteria:
Significant advance in knowledge or understanding
Importance of work for scientific progress and/or development of real-world applications
Of general interest to Physics World readers
Here are the Physics World Top 10 Breakthroughs for 2020, in no particular order. Come back next week to find out which one has bagged the Breakthrough of the Year award – and in the meantime you can listen to four of the judges talk about the Top 10 in the Physics World Weekly podcast.
Ideal system: A strontium ion trapped in an electric field. The measurement on the ion lasts only a millionth of a second. (Courtesy: F Pokorny et al., Stockholm University)
To Markus Hennrich and colleagues at Stockholm University, Sweden, together with researchers at the universities of Siegen in Germany and the Basque Country and Seville in Spain, for using a series of “weak” measurements (the subject of Physics World’s 2011 Breakthrough of the Year) to probe the nature of superposition collapse in quantum mechanics. While the act of measurement usually forces quantum systems into definite classical states, the work of Hennrich and colleagues showed that some measurements do not destroy all quantum information. By taking a series of “snapshots” during experiments on a single ion of strontium, the team revealed that measurements are not instantaneous, but instead gradually convert superposition states into classical ones. Because weak measurements could in principle allow errors to be detected in quantum states without destroying those states in the process, the work might be used to improve error correction in quantum computers.
To Haocun Yu and Lee McCuller of the Massachusetts Institute of Technology and their colleagues on the LIGO Scientific Collaboration for showing that quantum-scale correlations can leave their mark on macroscopic objects weighing tens of kilograms. The researchers explored the exquisite interplay between the laser beam of a LIGO interferometer and its mirrors – each of which weighs 40 kg. They observed that radiation noise contributes to the motion of the mirrors, which is a result of Heisenberg’s uncertainty principle. When using squeezed vacuum states of laser light they showed that the quantum noise drops below the standard quantum limit, which demonstrates quantum correlations between the laser beam and the mirrors. The research could lead to the improved detection of gravitational waves by LIGO, Virgo and future observatories.
To the Borexino collaboration for observing neutrinos from the carbon–nitrogen–oxygen (CNO) cycle in the Sun. To do so the team had to first painstakingly minimize the effects of background radiation in the Borexino detector, which comprises 278 tonne of ultrapure liquid scintillator located deep inside a mountain at Italy’s Gran Sasso Infn Laboratories. The observation confirms a theory of stellar nucleosynthesis first proposed over 80 years ago. It should also encourage physicists to use the next generation of neutrino detectors to try to resolve the “metallicity puzzle” of the Sun – a mystery regarding the abundance of carbon, nitrogen and oxygen in the star.
To Noel Clark and colleagues at the University of Colorado Boulder and the University of Utah in the US, for observing a ferroelectric nematic phase of matter in liquid crystals more than 100 years after it was predicted to exist. In this phase, all the molecules within specific patches, or domains, of the liquid crystal point in roughly the same direction – a phenomenon known as polar ordering that was first hypothesized by Peter Debye and Max Born back in the 1910s. Clark and colleagues found that when they applied a weak electric field to an organic molecule known as RM734, a striking palette of colours developed towards the edges of the cell containing the liquid crystal. In this phase, RM734 proved far more responsive to electric fields than traditional nematic liquid crystals. Although further work is required to identify materials that display the phenomenon at room temperatures, ferroelectric nematics could find applications in areas from new types of display screens to reimagined computer memory.
The perovskite thin-film X-ray detector is 100 times more sensitive than conventional detectors and requires no outside power source. (Courtesy: Los Alamos National Laboratory)
To Wanyi Nie and colleagues at Los Alamos National Laboratory for using thin-film perovskites to create an extremely sensitive X-ray detector. Using a synchrotron beamline to characterize their thin-film perovskite detectors, the researchers found that the X-ray absorption coefficients of the perovskite materials were on average 10 to 40 times higher than that of silicon for higher-energy X-rays. They also demonstrated that the new X-ray detectors are 100 times more sensitive than conventional silicon-based devices. This new type of solid-state X-ray detector could enable medical and dental imaging at extremely low radiation dose, enabling the same quality image to be generated using a much-reduced X-ray dose, making scans safer for patients. Nie also notes that it should be possible to fabricate large-scale detector arrays at far lower cost than for semiconductor detectors.
To Kostya Trachenko of Queen Mary University of London, Bartomeu Monserrat and Chris Pickard of the University of Cambridge and Vadim Brazhkin of the Russian Academy of Sciences for calculations showing that the upper limit on the speed of sound in solids and liquids depends on just two dimensionless quantities – the fine structure constant and the proton-to-electron mass ratio. The team’s theoretical prediction is backed up by experimental data of the speed of sound in a range of solid materials and a calculation of the speed of sound in metallic hydrogen – a material that is yet to be created in the lab but should have the fastest speed of sound. The research provides insight into how fundamental constants impose bounds on physical properties.
To Andrea Alù, Qiaoliang Bao, Cheng-Wei Qiu and an international team of collaborators at the City University of New York, National University of Singapore, Monash University, China University of Geosciences and the University of Texas at Austin, for showing that dispersion- and diffraction-free propagation of light is possible, with a resolution that beats the diffraction limit by more than an order of magnitude, in twisted layers of 2D molybdenum trioxide. Their work builds on the discovery of “magic-angle” graphene – Physics World’s Breakthrough of the Year in 2018 – by using twisted layers of 2D materials to change the behaviour of propagating photons, rather than electrons. Just as the electron version of twistronics has led to a flurry of research on superconductivity and electron states, the new photonics variant has important implications for nano-imaging, quantum optics, computing and low-energy optical signal processing.
Silicon sees the light: Elham Fadaly (left) and Alain Dijkstra in their Eindhoven lab. (Courtesy: Sicco van Grieken/SURF)
To Elham Fadaly, Alain Dijkstra and Erik Bakkers at Eindhoven University of Technology in the Netherlands, Jens Renè Suckert at Friedrich-Schiller-Universität Jena in Germany and an international team for creating a silicon-based material with a direct band gap that emits light at wavelengths used for optical telecommunications. Normally, silicon has an indirect electronic band gap, which means that it is a poor emitter of light and must be integrated with other semiconductor materials to create optoelectronic devices. To create a direct band gap, the researchers had to grow crystals of silicon-germanium alloy with a hexagonal crystal structure, rather than the usual diamond-like structure. They did this by creating nanowires of the alloy, which emitted infrared light. As well as having applications in optical telecoms and optical computing, the new silicon-based material could be used to create chemical sensors.
To a team headed up by Joao Seco at the German Cancer Research Centre in Heidelberg and Simon Jolly at University College London, for demonstrating how a mixed particle beam could enable simultaneous cancer therapy and treatment monitoring. The idea is to use a beam containing both carbon ions, which provide therapeutic irradiation of the target tumour, and helium ions, which travel straight through the patient and can therefore be used for imaging. In experiments at the Heidelberg Ion Beam Therapy Center using pelvis phantoms, the researchers showed that even small inflations of an air balloon inside the phantom caused an observable change in helium range. They also demonstrated that small phantom rotations changed the measured signal. The experiments reveal the potential of using a mixed beam to monitor intra-fractional anatomy changes, enabling more accurate delivery of particle therapy and, ultimately, providing better outcomes for cancer patients.
This research by Ranga Dias and colleagues is described in a paper in Nature. The paper has since been retracted by the journal.
To Ranga Dias and colleagues at the University of Rochester and the University of Nevada Las Vegas in the US for observing superconductivity at temperatures up to 15 °C in a hydrogen-rich material under immense pressure. Superconductors carry electrical current with no electrical resistance and have a range of applications from the high-field magnets used in MRI scanners to particle accelerators. Practical devices based on superconductors must be chilled to very cold temperatures, which is costly and can involve the use of helium, so a long-standing goal of condensed-matter physicists has been to develop a material that is a superconductor at room temperature. The carbonaceous sulphur hydride material made by Dias and colleagues shattered the previous high-temperature record by about 35 degrees and is the first to claim room-temperature superconductivity. While a pressure of 2.6 million atmospheres was required to achieve room-temperature superconductivity, the researchers think it may be possible to reduce the pressure by changing the chemistry of the material.
China has successfully launched a space telescope to study some of the most energetic events in the universe. Lifting off today from the Xichang Satellite Launch Center at 4.14 a.m. local time, the Gravitational wave high-energy Electromagnetic Counterpart All-sky Monitor (GECAM) represents one of the first new “all-sky” devices that will monitor fast-radio bursts, high-energy neutrinos and magnetars. The craft is set to begin operations as soon as it enters orbit, doing so for three years.
We’re really excited to see GECAM fly on time, as we overcame numerous technical difficulties and made it through the pandemic
Shaolin Xiong
To study these events, GECAM consists of two satellites – each weighing 160 kg – that will orbit on opposite sides of the Earth at an altitude of about 600 km. Each GECAM satellite features a dome-shaped array of 25 gamma-ray detectors and eight charged particle detectors. They will search for cosmic events happening in the energy range of 6 keV – 5 MeV and, within a couple of minutes of detection, GECAM will send out alerts to telescopes around the world for follow-up observations.
The idea of GECAM emerged following the announcement in February 2016 of the first detection of gravitational waves by the US-based Laser Interferometer Gravitational-Wave Observatory (LIGO). A year later and LIGO, working together with the VIRGO gravitational-wave detector in Italy, spotted the first gravitational wave produced by the merger of two neutron stars – an observation that was followed up by the Fermi Gamma-ray Space Telescope and other observatories around the world, kick-starting the era of multimessenger astronomy. Likewise, GECAM spot the gamma-rays bursts that are related to the production of gravitational waves by such comic mergers.
“We’re really excited to see GECAM fly on time, as we overcame numerous technical difficulties and made it through the pandemic,” says GECAM’s principal investigator Shaolin Xiong from the Institute of High Energy Physics, Chinese Academy of Sciences (CAS) in Beijing.
GECAM is the first of a line-up of space science missions to be launched by China in the coming five years. Other missions include the Advanced Space-based Solar Observatory, the Einstein Probe as well as the Solar wind Magnetosphere Ionosphere Link Explorer – a joint mission between CAS and the European Space Agency.
Identifying anatomic changes shown on MRI scans during a course of pencil-beam proton therapy, and adapting treatment plans accordingly, could improve the treatment quality, effectiveness and safety. In a retrospective study, researchers at St. Jude Children’s Research Hospital determined that 27% of paediatric patients, the majority with brain tumours, would have benefited from mid-treatment plan changes.
Pencil-beam proton therapy delivers radiation doses to a target with high accuracy, but if the target shifts, the dosimetric advantage afforded by the Bragg peak may be compromised and the risk of toxicities to adjacent tissue increased. Prior clinical trials comparing outcomes of paediatric patients treated with either proton or photon therapy have produced mixed results with regard to toxicity.
The researchers hypothesized that adherence to the treatment plan without adjusting for anatomic changes may have compromised the ability of proton therapy to maximally spare normal tissue while adequately covering the target volume. As such, they investigated whether altering treatment plans to compensate for changes in tumours or adjacent organs-at-risk (OARs) could improve target dose delivery and reduce normal tissue dose. They published their findings in the International Journal of Radiation Oncology, Biology, Physics.
The study included 73 children with a variety of cancers who received proton therapy at St. Jude during a 30-month period starting January 2017. The patients had between one and seven MRI exams, most of which were performed between weeks two and four of their treatments. Children with low-grade gliomas or rhabdomyosarcomas underwent weekly MRI exams.
The researchers evaluated 230 MRI scans acquired in the treatment position during the patients’ proton therapy courses, recording patterns of temporal and spatial anatomic changes. To identify scenarios that would benefit significantly from re-optimization, they compared the calculated radiation dose for the anatomy-of-the-day without treatment adaptation with the dose that would have been delivered with an adapted plan.
The majority of patients had brain tumours (79%) or tumours located in the head-and-neck (12%). They were prescribed a median dose of 54 Gy(RBE) in 1.8 Gy(RBE) per fraction. MRI analysis revealed that 20 patients had anatomic changes in their gross tumour volume (GTV) and/or changes in the tissue density within the beam path. This included all eight patients with rhabdomyosarcomas and 24% of patients with low-grade gliomas.
Led by radiation oncologist Sahaja Acharya, the researchers determined that 11 of the 20 re-optimized treatment plans had significantly changed plan quality, defined as a 5% or greater decrease in CTV V95 (clinical target volume receiving at least 95% of the prescription dose) or a 5% or greater increase in the dose–volume parameter used as an OAR constraint.
Because of anatomic changes, seven patients experienced a significant reduction in CTV coverage with the delivered plan, ranging from 5–16% decrease in CTV V95. Four patients experienced a significant increase in dose to the brain stem, hippocampus and/or optic apparatus, which could have been prevented if the delivered plan had been adapted. The team note that none of the patients who had re-optimized plans experienced local or distant failures. Additionally, none experienced grade 3 or higher toxicities or developed radiation necrosis.
“Of the 27% of patients who demonstrated anatomic change during proton therapy, more than 50% had suboptimal delivered plans,” write the authors. “Re-optimization of these plans resulted in an improvement of CTV coverage up to 16% and a decrease in OAR dose up to 46%.” Because most changes were identified during the second or third week of a six-week course of treatment, the re-optimization effect would have been significant.
“Certain tumour histologies and locations might derive more benefit than others from this type of approach. Specific patterns of anatomic change, unrelated to tumour or patient characteristics, might help in selecting patients for MRI-based adaptive therapy,” the researchers write.
“We are planning to conduct a prospective pilot study that will specify timing and frequency of on-treatment MRI for different disease categories,” Acharya tells Physics World. “Through this study, we seek to understand the number of patients we need to image in order to detect one plan deviation.”
Acharya is also the principal investigator of an accruing phase II study (NCT04065776) investigating the feasibility of reducing radiation doses to the hippocampi by using proton therapy for midline or suprasellar low-grade gliomas. Children enrolled in this study also undergo a weekly MRI scan, and treatment plans are adapted to shrinking tumour volumes if there is more than a 20% change in GTV. All children are also followed with longitudinal neurocognitive testing.
“This study might provide some insight on whether adapting to a smaller target volume can spare critical regions of the brain, such as the hippocampus, for children with midline or suprasellar low-grade gliomas, and whether sparing such regions will preserve neurocognition,” explains Acharya.
You know what the scientific method is until you try to define it: it’s a set of rules that scientists adopt to obtain a special kind of knowledge. The list is orderly, teachable and straightforward, at least in principle. But once you start spelling out the rules, you realize that they really don’t capture how scientists work, which is a lot messier. In fact, the rules exclude much of what you’d call science, and includes even more of what you don’t. You even begin to wonder why anyone thought it necessary to specify a “scientific method” at all.
In his new book The Scientific Method: an Evolution of Thinking from Darwin to Dewey, the University of Michigan historian Henry Cowles explains why some people thought it necessary to define “scientific method” in the first place. Once upon a time, he writes, science meant something like knowledge itself – the facts we discover about the world rather than the sometimes unruly way we got them. Over time, however, science came to mean a particular stepwise way that we obtain those facts independent of the humans who follow the method, and independent of the facts themselves.
Cowles’s story of this transformation begins with the naturalist Charles Darwin, who worried that professional scientists would regard his mixture of observing, recording and analysing things like barnacles, birds and worms as not rigorous enough to count as legitimate science. Darwin’s account of scientific method, in short, was largely motivated by his own professional anxiety. That anxiety motivated him to reflect on, as Cowles says, “how you thought what you thought”. Darwin found the answer in what he was studying – nature itself.
A natural view
Just as nature takes alternative forms of life and selects among them, Darwin argued, so scientists take hypotheses and choose the most robust. Nature has its own “method”, and humans acquire knowledge in an analogous way. Darwin’s scientific work on living creatures is indeed rigorous, as I think contemporary readers will agree, but in the lens of our notions of scientific method it was hopelessly anecdotal, psychological and disorganized. He was, after all, less focused on justifying his beliefs than on understanding nature.
In the lens of our notions of scientific method, Charles Darwin’s scientific work was hopelessly anecdotal, psychological and disorganized
Robert P Crease
Following Darwin, the American “pragmatists” – 19th-century philosophers such as Charles Peirce and William James – developed more refined accounts of the scientific method that meshed with their philosophical concerns. For Peirce and James, beliefs were not mental judgements or acts of faith, but habits that individuals develop through long experience. Beliefs are principles of action that are constantly tested against the world, reshaped and tested again, in an endless process. The scientific method is simply a careful characterization of this process.
For Cowles, though, the turning point into the modern conception of scientific method occurred in 1910 when fellow pragmatist John Dewey published a book called How We Think. In it, Dewey specified five steps that epitomize what we now think of as the scientific method, which was no longer intimately connected with the experience of the individual scientist but instead became a set of rules that turns those who adopt them into scientists. These rules were: “(i) a felt difficulty; (ii) its location and definition; (iii) suggestion of possible solution; (iv) development by reasoning of the bearings of the suggestion; (v) further observation and experiment leading to its acceptance or rejection; that is, the conclusion of belief or disbelief”.
While Dewey intended these steps to describe organic human thinking and how they can function in education, it was all too easy to interpret them as a special way to think, and to calcify them into what Cowles calls “a manual for technical practice”. Authors of textbooks seized on the idea as a way of explaining what science was and why it worked. Dewey’s steps, Cowles says, were quickly turned into “a symbolism of the separation of science from everyday thinking, a talisman of scientific exceptionalism”. The result, he argues, was “a shared method that seemed less and less natural as the 20th century wore on” though seemingly confirmed by episodes carefully selected from the history of science.
While the ruminations of Darwin and the early pragmatists on method had assumed that it was a way of thinking, in the end professional anxieties and the desire to elevate and insulate scientific knowledge above other kinds of knowledge transformed it into something apart from thinking, something potentially programmable into a computer. In this way “the scientific method” – an abstract set of rules purporting to be about scientific practice – had been turned into a logo; an identifier and attestation.
The critical point
Reviewing Cowles’s book, the Stanford University historian Jessica Riskin recently argued that the “scientific method” originated not within science itself, but “in the popular, professional, industrial, and commercial exploitation of its authority” (New York Review of Books 2 July 2020). Integral to this idea, she writes, was the claim that “science held an exclusive monopoly on truth, knowledge, and authority, a monopoly for which ‘the scientific method’ was a guarantee”. Yet it is possible, Cowles argues, to reject such a view of scientific method and to think of science “as the flawed, fallible activity of some imperfect, evolving creatures and as a worthy, even noble pursuit”.
To me, the scientific method is a good example of how philosophers and others create a problem by overstressing the orderly and formal character of scientific practice, and then set themselves the task of solving the problem they have created. As I argued in my own book The Workshop and the World, scientists working in a laboratory should be compared to a jury who place themselves for a short time in isolation to evaluate evidence. The jury members bring their experience to the deliberations, which is essential to being able to judge fairly. Those who think that hard-and-fast rules can be drawn up to guarantee justice have never been on a jury.
3D printing techniques are transforming many areas of manufacturing. One such technique, laser powder bed fusion (LPBF), is particularly attractive because it can be used to make complex metal parts that would be difficult or impossible to manufacture conventionally. However, it suffers from a major drawback in the form of tiny voids that weaken and degrade the metal. Researchers in the US and China have now identified how these voids are generated, and how they become trapped as the metal solidifies – findings that could help manufacturers find ways to control them, and thereby improve 3D metal-printing processes.
In LPBF, a high-power laser, guided by a digital computer-aided design and drafting model, is scanned across a thin layer of metal powder. The heat from the laser melts the metal powder in localized regions and fuses it to the layer of metal directly underneath. While this process is highly versatile and fairly speedy, the objects it creates are often riddled with microscopic, vapour-filled pores that drastically limit the material’s toughness and fatigue resistance.
“Keyhole” structures
When a high-power laser boils molten metal, it often generates deep and narrow gaps. These gaps are known as keyholes, and it had long been suspected that they were related to defects in the finished 3D-printed part. The exact relationship between keyholes and porosity was not, however, fully understood.
If this bubble never reconnects to the main keyhole, Rollett explains that it can instead collapse, generating an acoustic shock wave as it does so. This shock wave pushes the remaining pores away from the keyhole, thereby ensuring that they survive long enough to become trapped in the re-solidifying metal.
Despite the role that keyholes play in creating voids, Rollett and colleagues emphasize that keyholes themselves are not flaws. In fact, they have a positive role to play in LPBF, as they increase the efficiency of the laser: once inside a keyhole, the beam undergoes multiple reflections that enhance laser light absorption. It is only under certain conditions that the keyhole changes shape and becomes unstable, generating unwanted pores in the process.
Stay out of the “danger zone”
To find out what these conditions are, the researchers studied how the speed at which the laser is scanned over the metal power relates to keyhole instability. They found that keyholes become more unstable if the laser is scanned too slowly across the metal powder. Such slow scanning speeds increase the laser’s power in a localized region, allowing the metal there to overheat.
The team found that there is a well-defined boundary between stable and unstable keyholes. “You can think of the boundary as a speed limit, except it is the opposite of driving a car,” Rollett says. “In this case, it gets more dangerous as you go slower. If you’re below the speed limit, then you are almost certainly generating a defect.” However, he continues, “As long as you stay out the ‘danger zone’ (that is, too hot, too slow), the risk of leaving defects behind is quite small.”
The researchers, who report their work in Science, say they now plan to investigate ultrafast keyhole dynamics as well as other modes of instability in the laser melting process. “We also plan to study the development of microstructure from the rapid solidification and cooling inherent in LPBF additive manufacturing,” Rollett tells Physics World.
While world events are often difficult to predict, true randomness is surprisingly hard to find. In recent years, physicists have turned to quantum mechanics for a solution, using the inherently unpredictable behavior of photons to generate the truly random numbers that underpin many modern cryptographic protocols. Now, a new study promises to make this process of quantum random number generation more accessible, by showing that it is possible to produce certifiably random numbers quickly using a system built with off-the-shelf components.
When numbers are used to securely encode information, the randomness of those numbers is crucial: a string of truly random numbers is one that a hacker can never guess. In classical physics, however, all processes – even chaotic ones – are deterministic, making true randomness impossible. To illustrate this, study lead author David Drahi, a physicist at the University of Oxford, UK, notes that classically, a simple coin flip is about as random as it gets. However, he continues, “if you know the mass of the coin, if you can see the coin, if you can look at the wind, you can predict where it is going to land”. Classical randomness is therefore limited by the existence of information about the physical process meant to produce it.
In the quantum world, in contrast, “there are these fundamentally non-deterministic processes,” says Nathan Walk, a physicist at Freie Universitat Berlin, Germany and a co-author on the study. The results of quantum measurements, he adds, are inherently unpredictable, because their outcome does not exist in any meaningful way until the measurement has been made and the wavefunction of the system has, famously, collapsed.
Certifiably random
In developing their random number generator (RNG), the study’s authors focused not only on producing randomness, but also on confirming that this randomness originates from a non-deterministic quantum process rather than some incidental classical noise in the experiment. “There is not a test you can do on a string of numbers to tell if it’s random,” Walk notes. “You can’t certify strings. But you can certify processes.”
Drahi, Walk and collaborators built such a quantum certification process – essentially an additional measurement – into their protocol for generating random numbers. They also developed rigorous theoretical proofs of its effectiveness and demonstrated that it can be implemented in practice by performing experiments using quantum light. According to Renato Renner, a physicist at ETH Zurich in Switzerland who was not affiliated with the study, such steps are important for creating a practical system. “You really want to have a device that produces some certificate, otherwise you don’t really profit from any quantum advantages,” he says.
In their experiments, the researchers send laser light (a photonic state) into one input of a beam splitter while the other input is kept void, resulting in a zero signal (a vacuum state). The consequent pair of output beams is then measured using two separate detectors. Because each photon that arrives at the beam splitter has an equal (50%) chance of being reflected or transmitted, the difference between the numbers of photons recorded by each detector is unpredictable. It is bound to be a random number, Drahi explains.
To confirm that randomness generated in this way is reliable and useful, the researchers performed another measurement on the photonic state before it reaches the beam splitter. In this certification measurement, the light signal is discarded if it would not generate the desired amount of randomness at the end of the experiment. This can happen if the laser signal contains either too few photons or too many. Too few, and the number of possible unpredictable events will be too low for the measurement to be sufficiently random. Too many, and the detectors will hit their maximum value, making the measurement fully predictable.
The inclusion of this certification measurement means that the researchers have theorized and built a device that not only produces randomness at a fast rate of 8.05 gigabits per second, but also ensures the quality of that randomness in real time. According to Feihu Xu, a physicist at the University of Science and Technology of China who was not involved in the work, this “development of a formal framework to monitor and certify the randomness” stands out even though some other ideas in the study have been explored before.
“Untrusted” light source
The researchers’ inclusion of real-time randomness certification in their experiment also has consequences for its possible future applications, because the study’s theoretical framework proves that the RNG protocol is partly independent of the devices used to implement it. For instance, the light source used can be “untrusted” – the randomness analysis is independent of any information about it. As long as the light signal passes the certification measurement, the properties of the device that produced it do not affect the quality of final randomness. This flexibility means that the group’s quantum RNG could in principle be used by “a person or a computer that knows nothing about quantum physics”, Walk notes. The numbers obtained in this fashion would be reliably random regardless of operator’s expertise, he adds, since the protocol automatically ensures its own performance.
As a bonus, the research team managed to construct their experiment using affordable off-the-shelf components – a feature that allowed Drahi to ship the experimental setup from Oxford to some of his co-authors in Moscow. “It got there and it worked,” he says, noting that it would have been virtually impossible to ship all of the components for a more exotic quantum device across a continent, have a different researcher assemble them upon arrival, and then successfully generate random numbers at the same high speed. This level of practicality, combined with the rigorous approach to confirming the randomness of their random numbers as well as generating them, sets up this study as a promising starting point for the development of real-world quantum devices providing reliably true randomness. “It could have very broad applications,” Renner concludes.
Natural acoustic metamaterials found on the wings of some moths could help the insects avoid being eaten by bats – according to Marc Holderied and colleagues at the UK’s University of Bristol. By doing a combination of simulations and experiments, the team found that coupled vibrations of wing scales enable the moths to absorb ultrasound over a broad range of frequencies. The discovery could lead to the development of bio-inspired sound proofing materials with the potential to perform far better than current designs.
As a key source of prey for echolocating bats, moths are under evolutionary pressure to evade capture. While many moth species are very good at hearing the ultrasound used by bats, many other species lack this ability. Instead, these insects use acoustic camouflage to avoid detection. This involves sound waves being scattered or absorbed by sub-wavelength size structures on the moth’s wings.
Scientists have studied similar metamaterial structures on butterfly wings, which can have special optical properties that produce vibrant iridescent colours. However, these scales are not suitable for dampening sound, leading Holderied’s team to ask what is different about moths’ wings? In their study, the researchers investigated sound dampening mechanisms in two earless species of moth – Antheraea pernyi and Dactyloceras lucina.
Microscope and tomography
They first used a combination of scanning electron microscopy and micro-computed tomography to study the shapes and arrangements of the wing scales of both species. The images revealed overlapping tile arrangements of paddle-shaped scales, with stalks attached to stiff, lightweight membranes. These layers were typically less than 0.3 mm thick, which is far shorter than the ultrasound wavelengths used by bats.
In subsequent experiments, Holderied and colleagues showed that these layers could significantly dampen sound waves within a broad range of ultrasound frequencies. Sound at frequencies as low as 20 kHz is absorbed and the layers display a maximum absorption of 72% at 78 kHz.
Using computer models, the researchers worked out the mechanisms responsible for these advanced absorption properties. The scales vary significantly in terms of their sizes and shapes – depending on their location on the wing. The team found that this leads to variations in the natural resonant frequencies of the scales. Furthermore, their models revealed a strong coupling between the vibrations of neighbouring scales on shared membranes. This means that collectively, the scales act to dampen acoustic waves over a broad range of frequencies. This is wholly unlike the acoustic behaviour of butterfly wings, whose uniform scales can only resonate in small, localized clusters.
The wings of the two moth species are the first confirmed examples of natural acoustic metamaterials. Their design could be mimicked to create lightweight, ultrathin soundproofing panels, with the potential to out-perform the thick, porous absorbers currently used for sound insulation.
Proton therapy holds the promise of delivering highly conformal dose distributions to a target tumour volume, due to the finite range of the proton beam. However, the accuracy of this delivery can be affected by factors such as target motion, which may lead to overdosing regions of healthy tissue and underdosage of the tumour volume.
“Proton therapy is very sensitive to changes in the position and shape of the patient as a whole and the patient’s organs, both between fractions and within one fraction,” explained Sonja Schellhammer, recipient of the Donal Hollywood Award for the best abstract presented at the recent ESTRO 2020 congress. “It would be ideal if we could monitor the proton range simultaneously with the patient anatomy in real time.”
Currently, verification of proton range is mostly based on detection of secondary radiation such as prompt gamma rays or acoustic waves. Another indirect approach involves using MRI to visualize long-term radiation-induced biological effects. MRI benefits from excellent soft-tissue contrast, offers real-time imaging, delivers no radiation dose and could prove ideal for integration with proton therapy.
With this in mind, scientists at OncoRay/Helmholtz-Zentrum Dresden-Rossendorf (HZDR) are working to integrate proton therapy with MRI. Previous studies by others have shown that MRI can determine proton beam range using an off-line approach, with contrast-enhanced images recorded weeks or months after irradiation. “The question we asked in this study, is whether it would be possible to visualize the proton beam with online MRI during irradiation,” Schellhammer explained in her ESTRO presentation.
To test this idea, the researchers designed a phantom experiment using the Dresden proton therapy facility, where OncoRay, in collaboration with IBA and ASG Superconductors – Paramed MRI Unit, had integrated an open 0.22T MRI scanner with a fixed proton research beamline. The proton beam was incident upon a plastic bottle filled with deionized water and placed in the scanner’s magnetic isocentre. The team used a PMMA range shifter to tailor the proton energy such that the beam stopped within the bottle, and recorded MR image slices through the centre of the beam.
The researchers performed one scan during proton irradiation, and three immediately after irradiation, using a variety of MR sequences. For two of the six MR sequences – the proton density-weighted gradient echo (GE) sequence and the inversion recovery gradient echo (IRGE) sequence – they saw signatures in the MR image that were likely induced by the proton beam. They also observed that these signals didn’t disappear immediately after irradiation, but faded over tens of seconds.
“With this experiment, we confirmed that there is a measurable proton beam-induced signal on the MR images,” said Schellhammer. “We then asked whether this could be useful for range verification.”
Next, the team irradiated the water phantom using proton beams with energies of between 190 and 225 MeV (corresponding to different beam ranges). The observed MR signal changed in depth with increasing range, and the measured residual ranges agreed with the calculated values to within 2 mm.
Repeating the experiment with four different beam currents (1, 3, 9 and 27 nA, corresponding to dose rates of 1.7, 5, 15 and 45 Gy/s) demonstrated that the signal intensity increased with increasing beam current. MR images (recorded for 20 s) were only visible using beam currents of 3 nA or more.
To investigate whether this approach may work in a patient, the team used the same experimental set-up to irradiate other liquid phantoms (ethanol and petroleum), highly viscose materials (sugar syrup, mayonnaise and gelatine) and a tissue-mimicking material (a pork chop). While images of ethanol and petroleum showed similar beam-induced signals as seen with water, for the more viscose materials and the pork chop, no signal was seen.
“It appears that the signal is only present in liquids, so likely may not be transferable to patients,” explained Schellhammer. “But this approach may well prove useful for quality assurance of proton range in future hybrid MR-proton therapy systems.”
The mechanisms underlying the observed effects are still to be unravelled. The most probable hypothesis, Schellhammer suggested, is that irradiating a liquid increases its temperature locally, creating a density difference in which the lighter heated water rises out of the imaging volume and creates a signal void. Further investigations are needed to test this hypothesis, she noted.
“I have demonstrated that proton beam range can be accurately verified with online MRI,” Schellhammer concluded. “However, today only at high doses and in fluid-filled phantoms. The method holds potential for dosimeter-free online quality assurance of MR-integrated proton therapy. Further research towards MR-based range proton beam verification is clearly justified.”
“Alexa, play some Christmas music.” “OK Google, turn on the fairy lights.” “Hey Siri, how long do you need to cook a turkey?”
This festive season, we’ll undoubtedly be chatting to our smart speakers like they’re another member of the family, and every time, the disembodied response will be almost instantaneous.
These devices – which include Amazon’s Echo, Google’s Nest and Apple’s Homepod – have already become an extra presence in more than a fifth of UK households. Indeed, in 2019 almost 147 million units were sold globally and sales for 2020 are expected to be 10% higher still. Quite simply, smart speakers have reached an astonishing level of capability for recognizing what we say. Although speculation remains as to exactly how much they are listening to and what the collected data are used for, there is no doubt that the voice-recognition technology is amazing in its accuracy. This comes down to ultrasensitive acoustic sensors and sophisticated machine-learning algorithms interpreting speech (see box “From speech to text”).
While good enough to have made it into our homes, the development of sensors for voice recognition is by no means finished. It’s not clear which technology has the most promise, and novel ideas are frequently gaining commercial attention. It also seems likely that this field, like so many others, will be altered in light of the coronavirus pandemic, with wearable sensors that can detect speech through vibrations of the throat potentially providing important diagnostic tools for diseases, such as COVID-19, that can that affect the vocal cords.
From speech to text
(Courtesy: Shutterstock/SergeyBitos)
To generate text from live speech, two things need to happen: an acoustic sensor has to convert the incoming sound waves into an electrical signal, and then software must be used to figure out what words have been said.
For the second stage, the electrical signal is traditionally first converted from analogue to digital, before being analysed using a fast Fourier transform technique to find the variation in amplitude of different frequencies over time. Little time sections of the graphs, representing small sounds known as “phones”, are matched to the ideal short sound, or “phoneme”, that they probably represent. Then algorithms build the phonemes into full speech.
Because voices vary so much, these programmes cannot just rely on a phonetic dictionary to piece together the phonemes, which is why machine learning is so useful to improve accuracy over time. When using Alexa, for example, it will “remember” when we correct it on what we said, and so become more accurate at interpreting our individual voices. Most algorithms also use a probabilistic approach to work out which phonemes are most likely to follow each other – this is called a hidden Markov model.
A single-step approach, called “end-to-end deep learning”, is now gaining popularity and is used in current voice-recognition technology. This uses a single learning algorithm to go from the electrical signal representing audio to a text transcript without extracting the phones.
Audrey and the capacitors
The story of how acoustic sensors reached such extraordinary sensitivities begins in the late 19th century when the first acoustic sensor, the carbon or “contact” microphone, was developed independently by three inventors – Emile Berliner and Thomas Edison in the US, and David Hughes in the UK. These devices consist of carbon granules pressed between two metal contact plates with a voltage applied across the plates. Incoming sound waves cause one of the plates, the diaphragm, to vibrate. During compression, the graphite granules deform, increasing the contact area between the plates so that the resistance of the set-up drops, and current increases. These changes as the diaphragm moves mean that the sound is encoded in an electrical current.
However, it wasn’t until 1952 that voice-recognition technology was first developed. A team at Bell Telephone Laboratories (now Nokia Bell Labs) in the US created a program called “Audrey” – the Automatic Digit Recognition machine – that could understand the digits 0–9 spoken into a standard telephone (which most likely featured a carbon microphone). Audrey could be used for hands-free dialling, but had to be trained to the user’s voice and required a room full of electronics to run.
While the computing side of voice recognition has obviously come a long way since Audrey, acoustic sensors have also gone through rigorous development. Various designs, such as ribbon, dynamic and carbon microphones, have come in and out of fashion, but the one that has prevailed for voice recognition is the capacitor, or “condenser”, sensor. Originally developed in 1916 by E C Wente of Western Electric Engineering Department in the US (later becoming Bell Telephone Laboratories), the design hinges on the fact that the voltage across a capacitor depends on the distance between the plates. To this end, the sensor features a stationary backplate and a moving diaphragm, charged up by an external voltage. As the diaphragm vibrates from the incoming sound waves, the capacitance – and hence the voltage across the capacitor – varies with diaphragm displacement, from which the amplitude variation of different frequencies in time can be calculated.
The inventors Gerhard Sessler and Jim West at Bell Telephone Laboratories further developed the capacitor sensor in 1962 by using an electret diaphragm, creating the electret condenser microphone (ECM). Electret materials, such as Teflon, have a pre-existing surface charge, which means they keep a permanent voltage across the capacitor, thereby reducing the power input needed. Roughly 3–10 mm in diameter, ECMs dominated the general microphone market for nearly 50 years, but the move towards compact devices led to drops in signal-to-noise ratio and decreased stability, particularly in variable temperature environments.
When it comes to voice recognition, ECMs have therefore been mostly replaced by micro-electro-mechanical system (MEMS) capacitive microphones. At only 20–1000 μm in diameter, these devices are typically what you find in smart speakers. MEMS sensors differ from ECMs in the internal circuitry, notably in that they convert the signal from analogue to digital while still inside the microphone. Apart from being less susceptible to electrical noise than ECMs, the design is smaller and easier to make, since the silicon wafers required can be made on semiconductor-manufacturing lines. The drawback with MEMS microphones is that they are not very durable, so do not cope well with harsh environments. This is because dust particles tend to gather under the diaphragm, limiting its vibration, and rain can also damage it. Even in ambient settings, the trapped layer of air between the diaphragm and backplate makes it harder for the diaphragm to vibrate at maximum amplitude, limiting the device’s sensitivity.
A new way?
Although capacitor sensors have dominated the industry for decades, the technology might not hold all the answers for the future. US firm Vesper Technologies is paving the way in the commercial development of piezoelectric acoustic sensors, with the endorsement of Amazon Alexa. Founded in 2014, the company’s initial designs were based on the PhD research of Bobby Littrell, the company’s chief technology officer, and the work has since won many awards.
These devices work using a diaphragm made of a piezoelectric material, such as lead zirconate titanate, that directly converts the mechanical energy from sound waves into an electrical response. When the piezoelectric diaphragm stretches as a sound wave hits, the distances between ions are increased, creating small electric dipoles in the new most energetically stable ionic arrangement. The lack of a centre of symmetry in the crystal’s unit cell means no equivalent dipoles are formed on the other side of a centre of symmetry (which would cancel out, to leave no net dipole). The cumulative effect of all of these tiny dipoles across the crystal is the generation of a voltage, which varies in time as the strain in the crystal varies.
Changing times There have been multiple acoustic-sensor designs since the carbon microphone was first developed by Emile Berliner (left, courtesy Granger Historical Picture Archive/Alamy Stock Photo), Thomas Edison and David Hughes in the late 19th century, including electret condenser microphones (middle, courtesy CUI Devices) and micro-electromechanical system capacitive microphones (right, courtesy CUI Devices).
Compared with capacitor acoustic sensors, piezoelectric devices have the distinct advantage of containing just a single layer, meaning they do not trap dirt, air or rain and so are much more durable. The devices are also self-powered, meaning the range of applications, particularly when there is limited room for a battery, is much wider.
However, thin-film devices like these – and the capacitive designs – tend to be quite difficult to make. “You need a high or ultrahigh vacuum,” explains Judy Wu, a physicist at the University of Kansas in the US, “and you need to select a good substrate because [otherwise] you will not be able to get epitaxial growth.” Epitaxial growth is what happens when a thin film grows as a single crystal with one orientation of the unit cells. This is needed so that the dipoles formed under mechanical strain all point in the same direction. “You have to really raise the temperature,” Wu continues, “to give the thermal energy and the mobility when you put the atoms on the substrate [for them] to find the minimum energy position to form a perfect lattice.”
As Vesper has shown, these conditions can be produced, but they do limit the device applications. For instance, growing a thin film of a single crystal on a flexible substrate is difficult, since single crystals have to grow on an ordered structure to be ordered themselves and most flexible materials are not crystalline. “You cannot provide a perfect lattice there – it’s just amorphous material,” explains Wu.
However, Wu and her team are working on a potential solution. They have used graphene in solution to grow piezoelectric zinc oxide nanowire arrays for a strain sensor (ACS Applied Nano Materials3 6711). As the graphene is flexible but crystalline, the nanowires grown are still piezoelectric crystals. The difficulty is that it is very delicate. The researchers overcame this by using a solution to grow the nanowires, so as not to destroy the graphene’s perfect structure with sputtering techniques, which would reduce its conductivity. Not only this, but the ambient pressure and relatively low temperature used (90 °C) mean the process is very cheap.
Their strain sensor works by detecting changes to graphene’s conductivity, which occur as a result of the extra surface charge that develops on the zinc oxide nanowire array when it is mechanically strained. Wu says they are also working on a flexible version of the sensor, encased in PET plastic. The research is still in fairly early stages, and the sensor does not have a definite application just yet, but in the hope of filling some niche function needing a sensitive, flexible sensor, they have patented the graphene process. “We want to do something the ceramic sensor cannot do,” Wu explains, “[Our sensor] might recognize your skin, or recognize your voice, because it’s very sensitive to acoustics.”
Following nature
Adding to the whirlpool of ideas in the voice-recognition field, Keon Jae Lee and his team at the Korea Advanced Institute of Science and Technology (KAIST) have been developing a new piezoelectric sensor design (Advanced Materials 32 1904020) that mimics human hearing. Speaking to Physics World, Lee light-heartedly explained his faith in the concept, “If nature is doing it, it’s probably the most efficient way.”
Their piezoelectric sensors have a similar shape to the basilar membrane in our ear (see box “How ears have inspired a new kind of voice-recognition device“), thereby allowing about twice as much information to be harvested than conventional capacitor sensors. This advantage comes from the fact that, rather than collecting a single signal containing all the frequencies and analysing that to figure out frequency amplitudes, many signals (in Lee’s case, seven) are analysed from various positions along the membrane. This wealth of information makes voice predictions more accurate. Lee and colleagues found their 2018 design exhibited 97.5% accuracy in speech recognition; a 75% improvement on a reference MEMS condenser microphone. “I think the two advantages are accuracy and sensitivity,” Lee concludes. “We can pick up sound from a long way away and recognize individual voices.”
The tricky part of their research is analysing the data from the channels to give the relative amplitudes of different frequencies, since amplitudes are modulated by the resonance behaviour of the channel. It’s why Lee thinks so few research groups have taken up the idea. But his group seems to have cracked it, and even founded a spin-off company manufacturing its unique sensors, Fronics, in 2016. Lee is optimistic for the future of the design commercially and believes the team has found the right number of channels for the sensor. It’s a fine balance, he explains, between improving your accuracy by collecting more data, and needing a bulky machine to process it all.
How ears have inspired a new kind of voice-recognition device
(Courtesy: Claus Lunau/Science Photo Library)
Keon Jae Lee and his team at Korea Advanced Institute of Science and Technology are developing a new voice-recognition sensor that draws inspiration from hearing in nature. It has a piezoelectric membrane that mimics the basilar membrane found in ears, which in humans is curled into a spiral inside our cochlea. Lee’s design, however, is more similar to the straight membrane found in the ears of many birds and reptiles.
Shaped like a trapezium when viewed from above, a basilar membrane is tapered: it is narrow and thick at the base (located in the centre of the spiral in our ears) and becomes wide and thin. The membrane can be thought of as a series of oscillating strings lying perpendicular to the axis of symmetry of the trapezium, except that the strings really form a continuous spectrum that waves can pass between.
An incoming sound creates a travelling wave that passes down the membrane, beginning at the wide, thin end. Essentially a superposition of individual waves of different frequencies, as each wave travels down the membrane, it will reach one of the oscillating strings that has a resonant frequency equal to the wave’s own frequency and will vibrate most strongly here. The membrane is covered by more than 10,000 hair cells, which transmit information to the brain about the amplitude of the frequency at their position along the membrane.
Ears are more complex than this membrane alone: for example the hair cells also amplify the “correct” frequency for that position on the basilar membrane. But the concept of a collection of resonating sections of a membrane is the key to Lee’s design.
Neck sensors and COVID-19
Voice-recognition technology is not limited to a device sitting in the corner of your room or in your pocket. Sensors that work using neck vibrations, instead of sound waves travelling though air, would be very useful where sound propagation is almost entirely prohibited, such as noisy industrial environments or when people have to wear bulky equipment like gas masks. A breakthrough occurred late last year when Yoonyoung Chung and his team from Pohang University of Science and Technology, South Korea, reported creating the first flexible and skin-attachable capacitor sensor, which can solve this problem by perceiving human voices through neck-skin vibrations on the cricoid cartilage (part of the larynx). Demonstrating that skin acceleration on the neck is linearly correlated with voice pressure, they realized that they could measure the variation of voice pressure in time by designing a device that detects skin acceleration through changes in capacitance (Nature Communications10 2468).
The sensor Chung’s team has created is less than 5 μm thick and the diaphragm is made from epoxy resin (figure 1). “We wanted to develop a flexible microphone sensor, so it was natural for us to use a polymeric material, which is intrinsically flexible,” explains Chung. The individual sensor panels are small enough that any neck curvature is negligible in them and will not affect vibrations – a bit like how the Earth’s curvature seems small enough to ignore from our perspective, since we occupy such a small proportion of its surface.
1 Electronic skin The throat sensor developed by Yoonyoung Chung and colleagues at Pohang University of Science and Technology in South Korea, illustrated above, consists of a crosslinked ultrathin polymer with a hole-patterned diaphragm structure. (CC BY 4.0 Nature Commun.10 2468/CC BY Christian Rambow)
What’s also vital for capacitive sensors is that they have a “flat frequency” response. This means that no frequencies appear falsely high in amplitude because of a resonance in the device, and it makes electrical signals much easier to analyse. In order to achieve this, the diaphragm must have a narrow resonance peak that lies well above the frequency range of the human voice. Chung achieved the narrow width by making the diaphragm from a fully cross-linked epoxy resin with a low damping ratio. Cross-linking prevents the movement of molecules past each other and the oscillation and conformational flip of the phenyl rings, which would all lead to friction between molecules. The second condition, a high resonant frequency, could be achieved by either a high stiffness or a low mass. The snag is that a device that’s less stiff is actually more sensitive because the diaphragm’s vibrations are bigger. Chung has found a solution to this stiffness dilemma: instead of reducing the stiffness of the bulk material, he added holes to the diaphragm. The holes reduce the diaphragm’s mass to satisfy the high resonant frequency, while keeping stiffness low to increase sensitivity. “This solution has also been commonly used in the silicon MEMS area in order to reduce the air resistance,” adds Chung.
The group has submitted patent applications for the design and is collaborating with several industry partners, though Chung is careful to note that commercialization is still in the early stages, since there is more research to be done in the lab. “We are now trying to make a sensor interface circuit on a flexible substrate so the entire sensory system can be attached on the neck skin without any difficulties,” he explains. The current design has a rigid circuit board, which can cause occasional problems by suppressing neck vibrations.
Throat sensors like these could also be used to diagnose illnesses like COVID-19, which have symptoms that manifest themselves in the vocal cords. Researchers from the Massachusetts Institute of Technology in the US have found that COVID-19 changes our voice signals by affecting the complexity of movement of the muscles in the throat (IEEE Open Journal of Engineering in Medicine and Biology1 203). They hypothesized this to be due to an increased coupling between muscles, preventing them from moving independently. Sensor devices paired with an app could provide early-stage screening to alert people to seek further testing for COVID-19. “It can also be used as a cough detector to diagnose people,” speculates Chung, while considering clinical uses for sensors like his. Even Vesper, though its current designs are not flexible, has shown interest in this potential outlet for innovation. “Our team is already brainstorming new ideas such as acoustic respiratory health monitors,” wrote its chief executive Matt Crowley in a blog post in March about the pandemic.
An advantage of sensors being repurposed from a voice-recognition context is that there is money available for developing the technology. Well-resourced tech companies are always on the lookout for technology that has exciting prospects for smart devices – such as sensors like Lee’s and Chung’s, which are sensitive enough to recognize individual voices in lieu of a password or fingerprint. And given the human need, as a result of the pandemic, to quickly detect respiratory illnesses on a mass scale, throat-monitoring devices will likely be on a fast-track to becoming a viable diagnostic technology. It will be interesting to see how both of these areas of innovation evolve over the next few years following the pandemic. Likewise, the debate between capacitive and piezoelectric sensors will continue to change, possibly as a result of manufacturing improvements, or the changing demand for certain features, like flexibility and resistance to harsh environments.Perhaps, one crisp December morning as you wait for the Brussels sprouts to boil, you might ask your smart speaker for an update.
