Drug-resistent yeast, synthetic synapses on the nano scale, and memristor research.
For the last decade or so, bacteria that are immune to the effects of antibiotics have been a persistent and growing threat in medicine. Ultimately, the problem goes back to the antibiotic not being administered long enough to kill off the entire colony. The few survivors that managed to make it through the increasing toxicity of their environment because they either had a gene which rendered them immune (and the toxins released when the other bacteria died weren't enough to poison them) or assembled one and survived long enough to breed and pass the gene along to other bacteria. This means that the pharmaceutical industry has been scrambling to find new antibiotics that won't harm the patient any more than they absolutely have to... except that now we're seeing antibiotic resistant yeasts in the wild, also. A strain of the yeast candida auris was discovered in 2009.ev in Japan that is resistent to every commonly used drug used to treat fungal infections, including caspofungin, amphotericin B, and fluconazole. Since that time, the dangerous strain of c.auris has spread to the United States, India, South Africa, Pakistan, Kuwait, South Korea, Colombia, the UK, and Venezuela. The fungus is known to invade the body through open wounds in an opportunistic fashion and take up residence in the bloodstream, where it subsequently causes organ failure. It is also known to infect the lungs to some degree, as evidenced by having been extracted and cultured from same. The US Center for Disease Control published a bulletin on 24 June 2016 describes the outbreak in more detail, including the risk factors for contracting the infection (diabetes, recent surgery and antibiotic use (both of which impact the integrity of the body overall), and the presence of large venous catheters). Unfortunately, c.auris is difficult to differentiate from several other less-critical fungal species without extensive testing so it can be misdiagnosed until it is too late; the CDC advises the use of MALDI-TOF mass spectrometry or DNA sequencing (analyzing the D1-D2 region of the 28s rDNA) to confirm infection.
And now I'm going to blither on about the newest and tastiest obsession in the fields of artificial intelligence and machine learning, neural networks. More stuff has transpired that's very interesting and appears to represent significant advances in the fields from a hardware perspective (whereas the previous stuff was largely in the software realm). Multiple research teams around the world are developing hardware implementations of neural networks, because when you get right down to it implementing something in hardware that's optimized for the task is always going to be faster than implementing it in software and running it on general-purpose iron. First, a research team at the Pohang University of Science and Technology in South Korea announced on 18 June 2016.ev that they had fabricated nanoscale synaptic transistors that are partially organic in construction. They placed 144 of them on a four inch square wafer and connected them in a 2D mesh with very fine filaments (200-300 nanometers each) that are composed of two different organic materials, which simulates the structure of organic axons and dendrites. Their experiments showed some fascinating results: For starters, their semi-organic neural network used only a bit more power to operate than a comparable neural network in an organic brain; synthetic neural nets tend to use as much power as silicon, which is orders of magnatude more than "real" neural networks. Second, they were able to demonstrate that their implementation could evidence all of the characteristic electrical activity of organic brains, which is to say that it showed signs of multiple classes of both plasticity and pulse facilitation, which one would expect to see when examining a relatively complex organic brain.
In a related project in Russia, a separate research team implemented a prototype neural network hardware using nanoscale memristors, electronic components which combine the electrical capabilities of capacitors, resistors, and inductors, and are capable of retaining state after the power is disconnected. The team at the Laboratory of Functional Materials and Devices for Nanoelectronics (English page is kind of bare at the moment) at the Moscow Institute of Physics and Technology is working to replicate the synaptic structure of organic brains using these weird components. Their prototype device consists of matrices of halfnium oxide memristors 40 nanometers on a side (to give you a sense of scale, each memristor was about the size of your average virus) and linked together. Like the other research team in South Korea, their proof-of-concept implementation also evidenced electrical properties commonly encountered in organic neural networks, including plasticity, potentiation, and depression. In short, the team's memristor-based neural networks are capable of a form of associative learning, just like humans. The research team's open access peer-reviewed paper, Crossbar Nanoscale HfO2-Based Electronic Synapses, is available at PubMed.
Finally, two separate laboratories associated with the US Department of Energy are working to figure out just how memristors function at the atomic scale. Teams at SLAC and LBNL have been using synchrotron X-ray imaging to watch the movements of atoms within memristors while they were put through their paces in the hopes of identifying and mitigating potential failure modes to make them more reliable. Right now a big problem involves relatively high switching voltages resulting in areas of individual memristors permanently deficient or overloaded with oxygen atoms, which make them much less likely to store charges properly and much more likely to refuse to switch states; either way, the memristor is functionally dead in such a state. Since beginning this research ten years ago with different memristor implementations (defined as being fabricated out of different semiconducting materials), they've made some interesting discoveries that will hopefully be useful soon. Of all of the different materials used, they noted that tantalum oxide seems to result in the most tractable memristor designs. They also noted that reversing the direction electrical currents pass through memristors seems to result in less permanent damage at the atomic level by causing oxygen atoms to scatter more slowly over time; the first thing that came to mind when I read that was that they were working with what amounts to purely electrical reversible computing rather than electromechanical but I might be reading a bit too much into it. The conclusions drawn from their research thus ar are that tantalum oxide-based memristors with carefully controlled low-power switching voltages can function for around a billion operational cycles, ideal for commercial applications and a couple of zeroes greater than the results they'd been getting with other fabrication materials. These experimental results strongly imply that memristors could be a viable component in both practical neural computing as well as general-purpose computing architectures a few generations of hardware from now.