Tyler O'Neal, Staff Editor CLOUD

University of Tokyo creates material for spintronics in magnetic memory

Computers and smartphones have different kinds of memory, which vary in speed and power efficiency depending on where they are used in the system. Typically, larger supercomputers, especially those in data centers, will use a lot of magnetic hard drives, which are less common in consumer systems now. The magnetic technology these are based on provides very high capacity but lacks the speed of solid state system memory. Devices based on upcoming spintronic technology may be able to bridge that gap and radically improve upon even the theoretical performance of classical electronic devices. Mn3Sn. (Left) A cross-sectional transmission electronic microscope image of the research material on a layer of tungsten (W) and magnesium oxide (MgO). (Right) A top down view of the material with an inset image showing manganese atoms in red and tin atoms in light blue. ©2022 Nakatsuji et al.

Professor Satoru Nakatsuji and Project Associate Professor Tomoya Higo from the Department of Physics at the University of Tokyo, together with their team, explore the world of spintronics and other related areas of solid state physics — broadly speaking, the physics of things that function without moving. Over the years, they have studied special kinds of magnetic materials, some of which have very unusual properties. You’ll be familiar with ferromagnets, as these are the kinds that exist in many everyday applications like computer hard drives and electric motors — you probably even have some stuck to your refrigerator. However, of greater interest to the team are more obscure magnetic materials called antiferromagnets.

“Like ferromagnets, antiferromagnets’ magnetic properties arise from the collective behavior of their component particles, in particular the spins of their electrons, something analogous to angular momentum,” said Nakatsuji. “Both materials can be used to encode information by changing localized groups of constituent particles. However, antiferromagnets have a distinct advantage in the high speed at which these changes to the information-storing spin states can be made, at the cost of increased complexity.”

“Some spintronic memory devices already exist. MRAM (magnetoresistive random access memory) has been commercialized and can replace electronic memory in some situations, but it is based on ferromagnetic switching,” said Higo. “After considerable trial and error, I believe we are the first to report the successful switching of spin states in antiferromagnetic material Mn3Sn by using the same method as that used for ferromagnets in the MRAM, meaning we have coaxed the antiferromagnetic substance into acting as a simple memory device.”

This method of switching is called spin-orbit torque (SOT) switching and it’s cause for excitement in the technology sector. It uses a fraction of the power to change the state of a bit (1 or 0) in memory, and although the researchers’ experiments involved switching their Mn3Sn sample in as little as a few milliseconds (thousandth of a second), they are confident that SOT switching could occur on the picosecond (trillionth of a second) scale, which would be orders of magnitude faster than the switching speed of current state-of-the-art electronic computer chips.

“We achieved this due to the unique material Mn3Sn,” said Nakatsuji. “It proved far easier to work with in this way than other antiferromagnetic materials may have been.”

“There is no rule book on how to fabricate this material. We aim to create a pure, flat crystal lattice of Mn3Sn from manganese and tin using a process called molecular beam epitaxy,” said Higo. “There are many parameters to this process that have to be fine-tuned, and we are still refining the process to see how it might be scaled up if it’s to become an industrial method one day.”

University of Barcelona uses ML to predict the consequences of genomic changes over time that make humans emerge

Tyler O'Neal, Staff Editor CLOUD

The study of the genomes of our closest relatives, the Neanderthals, and Denisovans, has opened up new research paths that can broaden our understanding of the evolutionary history of Homo sapiens. A study led by the University of Barcelona has made an estimation of the time when some of the genetic variants that characterize our species emerged. It does so by analyzing mutations that are very frequent in modern human populations, but not in these other species of archaic humans. 58fcb985 54f2 49ae a391 1c7a5dcc629c humbcronologiajpg1095776142 73902

The results show two moments in which mutations accumulated: one around 40,000 years ago, associated with the growth of the Homo sapiens population and its departure from Africa, and an older one, more than 100,000 years ago, related to the time of the greatest diversity of types of Homo sapiens in Africa.

"The understanding of the deep history of our species is expanding rapidly. However, it is difficult to determine when the genetic variants that distinguish us from other human species emerged. In this study, we have placed species-specific variants on a timeline. We have discovered how these variants accumulate over time, reflecting events such as the point of divergence between Homo sapiens and other human species around 100,000 years ago", says Alejandro Andirkó, first author of this article, which was part of his doctoral thesis at the UB.

The study, led by Cedric Boeckx, ICREA research professor in the section of General Linguistics and member of the Institute of Complex Systems of the UB (UBICS), included the participation of Juan Moriano, UB researcher, Alessandro Vitriolo, and Giuseppe Testa, experts from the University of Milan and the European Institute of Oncology.

The predominance of behavioral and facial-related variations

The results of the research study also show differences between evolutionary periods. Specifically, they highlight the predominance of genetic variants related to behavior and facial structure —key characteristics in the differentiation of our species from other human species— more than 300,000 years ago, a date that coincides with the available fossil and archaeological evidence. "We have discovered sets of genetic variants which affect the evolution of the face and which we have dated between 300,000 and 500,000 years ago, the period just prior to the dating of the earliest fossils of our species, such as the ones discovered at the Jebel Irhoud archaeological site in Morocco", notes Andirkó.

The researchers also analyzed variants related to the brain, the organ that can best help explain key features of the rich repertoire of behaviors associated with Homo sapiens. Specifically, they dated variants that medical studies conducted in present-day humans have linked to the volume of the cerebellum, corpus callosum, and other structures. "We found that brain tissues have a particular genomic expression profile at different times in our history; that is, certain genes related to neural development were more highly expressed at certain times," says the researcher.

Supporting the mosaic nature of the evolution of Homo sapiens

These results complement an idea that is dominant in evolutionary anthropology: that there is no linear history of the human species, but that different branches of our evolutionary tree coexisted and often intersected. "The breadth of the range of human diversity in the past has surprised anthropologists. Even within Homo sapiens, there are fossils, such as the ones I mentioned earlier from Jebel Irhoud, which, because of their features, were thought to belong to another species. That's why we say that human beings have lived a mosaic evolution," he notes.

“Our results,” the researcher continues, “offer a picture of how our genetics changed, which fits this idea, as we found no evidence of evolutionary changes that depended on one or several key mutations," he says.

Application of machine learning techniques

The methodology used in the study was based on a Genealogical Estimation of Variant Age method, developed by researchers at the University of Oxford. Once they had this estimation, they applied a machine learning tool to predict which genes have changed the most in certain time windows and which tissues these genes may have impacted. Specifically, they used ExPecto, a deep learning tool that uses a convolutional network — a type of computational model — to predict gene expression levels and function from a DNA sequence.

"Since there are no data on the genomic expression of variants in the past, this tool is an approach to a problem that has not been addressed until now. Although the use of machine learning prediction is increasingly common in the clinical world, as far as we know, nobody has tried to predict the consequences of genomic changes over time,” notes Andirkó.

The importance of the perinatal phase in the brain development of our species

In a previous study, the same UB team, together with the researcher Raül Gómez Buisán, used genomic information from archaic humans. In that study, they analyzed genomic deserts, regions of the genome of our species where there are no genetic fragments of Neanderthals or Denisovans, and which, moreover, have been subjected to positive pressure in our species: that is, they have accumulated more mutations than would have been expected by neutral evolution. The researchers studied the expression of genes — i.e., which proteins code for different functions — found in desert regions throughout brain development, from prenatal to adult stages, covering sixteen brain structures. The results showed differences in gene expression in the cerebellum, striatum, and thalamus. "These results bring into focus the relevance of brain structures beyond the neocortex, which has traditionally dominated research on the evolution of the human brain," says Juan Moriano.

Moreover, the most striking differences between brain structures were found at prenatal stages. "These findings add new evidence to the hypothesis of a species-specific trajectory of brain development taking place at perinatal stages — the period from 22 weeks to the end of the first four weeks of neonatal life — that would result in a more globular head shape in modern humans, in contrast to the more elongated shape seen in Neanderthals," concludes Moriano.

Chinese researchers propose an approach for detecting LDoS attack based on cloud model

Tyler O'Neal, Staff Editor CLOUD

Cybersecurity has always been the focus of Internet research. An LDoS attack is an intelligent type of DoS attack, which reduces the quality of network service by periodically sending high-speed but short-pulse attack traffic. The existing LDoS attack detection methods generally have the problems of high FPR and FNR. The processing flow of LDoS detection

To solve the problems, a research team led by Wei SHI published their new research on 02 April 2022 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.

The team proposed a cloud model-based LDoS attack detection method using a classifier based on SVM to train and classify the feature parameters. The detection method is verified and tested in the NS2 simulation platform and Test-bed network environment. Compared with the existing research results, the proposed method requires fewer samples, and it has lower FPR and FNR.

In the research, they analyze the abnormal changes in network traffic caused by the LDoS attack and use the cloud model to compare the difference between the normal state of the network and the state of the LDoS attack. To more accurately judge whether the network is under LDoS attack, they use the cloud model to obtain the feature parameters in two states and then use the Support Vector Machine (SVM)-based LDoS attack detection classifier to train and classify the obtained feature parameters, detect whether there is an LDoS attack on the network.

Firstly, the cloud model is used to analyze network traffic. The reverse cloud generation algorithm analyzes the network traffic in the bottleneck link to obtain feature values of the cloud model, and analyzes the changes of the feature values under the LDoS attack, then uses the SVM with a “small sample” learning ability to establish LDoS attack detection classifier to judge whether the LDoS attack occurs. The experiment is performed in the NS2 and the Test-bed. The experimental data shows that compared with the existing research methods, the proposed method requires fewer sample data and has the characteristics of a high Accuracy, low FNR, and low FPR value.

Future work can focus on finding more suitable public datasets containing the LDoS attack, expanding the experimental platform, and designing a more effective method for accurately detecting the LDoS attack.