Signal Frontier @SignalFrontierX

Faster AI chips can still spend a surprising amount of time waiting. An accelerator only runs at full speed if weights and activations can reach the compute units fast enough. As arithmetic throughput rises, the pressure shifts to HBM, advanced packaging and chip-to-chip interconnects. That is also why simply adding more compute can give disappointing gains. Once data movement becomes the limit, extra arithmetic units are capacity the system cannot fully feed. The bottleneck moves as soon as the fastest component gets faster.

@Christinany6666 is that real account ?

🔴 A 1.27-billion-year-old Martian meteorite may be preserving chemistry from a much older Mars. NWA 13441 is a shergottite, but its age falls into a strange gap. Most known shergottites are either younger than ~600 million years or around 2.4 billion years old. The isotope chemistry is the better clue. Its neodymium signature is close to chondritic and unlike other known shergottites, suggesting the magma tapped a previously unsampled Martian mantle reservoir. With no Earth-like plate tectonics constantly remixing the interior, Mars may preserve chemically distinct reservoirs for much longer. So this meteorite does more than fill an age gap. It may be sampling a part of the Martian mantle that our meteorite collection has simply never seen before. Source: “Entirely Surprising” – Scientists Have Found a Martian Meteorite Unlike Any Known Before. SciTechDaily

@elonmusk another reset !

@thefrogmaxi Yes sir

I read X’s newly public ranking code so you don’t have to. The biggest surprise: "LIKES ARE CHEAP". Sharing, replies, follows, and reaching the right audience matter much more. Here’s what that means for reach and verified impressions. x.com/i/article/2088…

@thefrogmaxi Of course, bro !

⚛️ Slowing electrons down can make their interactions stronger. In the van der Waals magnet Fe₅GeTe₂, UChicago researchers found an unusually flat electronic band near the Fermi level. The electrons barely disperse, yet remain coherent enough to form a √3×√3 charge-ordered state. A flatter band means less kinetic energy competing with electron–electron interactions. And unlike many flat-band systems built with twisted moiré structures, this state appears to emerge from interactions inside the material itself. The coherent phase survives to about 100 K. That is still far from room-temperature electronics, and a useful memory device would need reliable switching between states. Here, faster is not better. Slowing the electrons down is what gives their interactions enough weight to reorganize the material. Source: Electrons Mysteriously Slow to a Crawl Inside This Magnetic Material. SciTechDaily

♻️ Illinois researchers built a molecule that turns voltage into a control knob for metal separation. The extractant does three jobs at once: it selectively binds metal ions, carries charge through the organic phase, and changes its chemistry when a voltage is applied. That means capture and release can be controlled electrically rather than by repeatedly adding additional chemical reagents. In e-waste tests, the system recovered 89% of the gold while cutting chemical use by roughly 10–100×. It is still a lab-scale platform, and scaling liquid-liquid extraction without losing selectivity or efficiency will be the harder test. But the design points to a useful shift: not just electrifying the machinery around a chemical process, but making the chemistry itself electrically switchable. Source: Electricity-driven molecule cuts chemicals 100x, recovers 89% gold from e-waste. Interesting Engineering

🔦 A waveguide usually exists because we fabricate one. In MoOCl₂, the crystal itself can become the routing rule. Researchers at Stuttgart and IIT launched mid-infrared plasmon polaritons from a nanoscale gold antenna. Around 4.2 μm, the waves collapsed into a narrow, nearly diffractionless channel without etching a waveguide into the material. At shorter wavelengths, the propagation is hyperbolic; at longer wavelengths it becomes elliptical. The canalized beam appears right at that transition. There is another useful knob: changing the flake thickness shifted the canalization wavelength by more than 1 μm. The limitation is equally clear. This is not arbitrary routing—the preferred direction comes from the crystal’s anisotropy, and the experiment still relies on nanoscale optical launchers. Not every optical road may need to be etched. Some can be built into the material’s dispersion. Source: Scientists Steer Infrared Light Through an “Invisible” Waveguide. SciTechDaily

keyboard_arrow_left Previous keyboard_arrow_right Next

🧠 We may have learned one rule of brain aging too well from mice. Microglia are usually described as a self-renewing immune population established early in life. But Stanford researchers traced naturally accumulated DNA mutations in blood and brain cells from 20 older people. In every one, bone-marrow-derived cells had entered the brain. Some became microglia-like and could make up a large share of the microglial population. The surprise is that this influx appears common in aging humans but does not seem to occur in the same way in mice or nonhuman primates. It can begin as early as middle age. That does not make animal models useless. It means they may miss a human-specific route by which the brain’s immune system changes with age. Human brain aging may involve not just changing old microglia, but replacing part of the population from outside the brain. Source: Immune cells flood into the aging brain, Stanford scientists discover. Science Daily

☢️ A nuclear reactor can shut down without going completely silent. Double Chooz analyzed 17.2 days with both Chooz B reactors off. A detector 400 m away recorded 106 ± 18 antineutrino candidates in the key energy range, a 5.9σ signal consistent with simulations of long-lived fission products in the cores and spent-fuel pools. The signal also carries history. Its spectrum changes as different isotopes decay, so it contains information about the fuel itself. In this dataset, spent-fuel pools supplied 44% of the predicted residual signal. There is a catch: this residual flux is less than 1% of the normal reactor signal, which is why detecting it requires very low backgrounds and substantial detector infrastructure. So this is not a plug-and-play reactor monitor yet. A reactor can stop producing power long before its fuel stops producing information. Source: Scientists detect a nuclear reactor’s ghostly afterglow for the first time, Science Daily

⚛️ China just entangled two quantum memories across 420 km of optical fiber. The distance is easy to headline. The harder result is what happened to fiber loss. The USTC team stored quantum states in laser-cooled rubidium memories, shifted the photons into the telecom S-band, then used single-photon interference to entangle the remote memories. At long distance, the protocol exceeded the repeaterless capacity of the same lossy channel. One caveat matters: 420 km was the fiber path length, not two cities sitting 420 km apart. Much of it came from coiled fiber, alongside a 10.1 km deployed field link. So this is not a 260-mile quantum internet. It is evidence that quantum memory can start changing the scaling problem that makes direct fiber links collapse with distance. The next milestone is not a longer spool of fiber. It is a repeater that can work across real city-scale networks. Source: China pushes quantum memory entanglement 4x farther, reaching 260 miles over fiber. Interesting Engineering

@memphis_ea 😄

📡 WiFi identified people with 99.5% accuracy in a new study. Then researchers changed the radio perspective. In several cases, accuracy fell below 5%. KIT researchers used beamforming feedback already exchanged by ordinary WiFi devices to recognize people from how their bodies disturb radio propagation. That huge accuracy drop reveals something more interesting than the headline: the system is not learning you alone. It is learning you + the multipath environment around you. Change the geometry, reflections or observation angle, and the “biometric” signature can change too. ⚠️ But that does not remove the privacy problem. You do not necessarily need to carry the sensing device yourself. The wireless infrastructure around you can provide the signal. So the real test now is not whether WiFi can recognize a person in one controlled environment. It is whether that identity survives new rooms, moving furniture and crowded spaces. If it does, WiFi stops being just communication infrastructure. It becomes ambient biometric infrastructure. Source: Ordinary WiFi can now identify you with near-perfect accuracy, Science Daily

⚡ For charge-based memory, one electron per bit is about as far as miniaturization can go. The hard part is making that single electron easy to read at room temperature. Researchers at Fudan University built a 2D non-volatile memory device where adding or removing one electron shifts the electrical signal by about 0.5 V. A similar experiment in 1997 produced only ~55 mV and held the state for around five seconds. Here, the states remained stable through a 5,000-second test. That signal margin is what makes the result stand out. If a one-electron state cannot be separated cleanly from noise and leakage, the tiny storage element is not very useful. There is still a long way to go. Programming requires roughly 29–30 V, and making one clean device is very different from fabricating billions of uniform cells with acceptable yield. 🧠 The physics is starting to look convincing. The manufacturing problem is now the harder question. One electron per bit may be the physical limit. The real test is whether an industry-scale memory array can ever get close to it. Source: This 2D Memory Device Stores Data With Just One Electron, ZME Science

🦴 A microscopic “birthmark” inside a 236-million-year-old bone may push mammalian-style live birth back by nearly 100 million years. Researchers found a neonatal line in Chiniquodon theotonicus, marking the shift from embryonic to postnatal bone growth. Its estimated newborn mass was about 1.68 kg, roughly 14% of adult body mass, much closer to placental mammals than to egg-laying monotremes or reptiles. The fossil leaves two very different evolutionary histories on the table. Live birth may have evolved independently in this cynodont lineage and later in therian mammals. Or it may have appeared much earlier in our ancestry, with monotremes later returning to egg-laying. 🧬 That second possibility is the one I find harder to ignore. We often treat traits that look “primitive” as ancestral by default. Evolution does not have to work that way. If monotremes really reversed back to egg-laying, how many other supposedly ancient mammalian traits might actually be evolutionary reversals? Source: Fossil Evidence Pushes Back Origin of Live Birth by 90 Million Years, Sci. News

Just testing Grok Imagine 2.0 in creating a video from my own photo 😃

keyboard_arrow_left Previous keyboard_arrow_right Next

🌌 JWST may have found something more important than the “black hole star” nickname suggests: a reason to be careful about how we weigh black holes in the early Universe. MoM-BH*-1 existed only about 660 million years after the Big Bang. Its strange spectrum can be explained by a growing black hole buried inside a dense, turbulent, nearly dust-free gas envelope roughly 10–100 AU across. That gas may be doing two jobs at once. It can help trap and redistribute radiation, giving the black hole a route to very rapid growth. It can also scatter photons enough to broaden the hydrogen lines we observe. ⚫ That second part changes the measurement problem. Broad-line widths are often treated as tracers of gas velocity and then used to estimate black-hole mass. If scattering is doing much of the broadening, that estimate can be badly biased. But the story is not as simple as “JWST has been overestimating black holes.” Another little red dot, QSO1, recently had its mass measured dynamically. There, the conventional virial estimate worked surprisingly well, while an electron-scattering model undershot the mass by nearly two orders of magnitude. So similar-looking little red dots may not be governed by the same line-broadening physics. That is the part I would watch. Before turning a spectrum into a black-hole mass, we may first need to know what kind of object produced that spectrum in the first place. Source: Astronomers Find Black Hole Star in Early Universe, Sci News

🧬 “Life began twice” is a catchy headline. The real story is stranger, and more interesting. Researchers reconstructed 420 core metabolic reactions and found that LUCA, the common ancestor of modern cells, may have had enzymes for only about half of them. The rest may have depended on metals and chemistry already present around hydrothermal vents. That suggests early life was not fully self-contained. The environment was still doing part of the chemistry. A useful way to picture this is geochemical scaffolding: the vent environment helped support primitive metabolism until cells could take over those reactions themselves. This is where Bacteria and Archaea become especially interesting. After the two lineages split, they appear to have evolved different enzymes for some of the same essential reactions. Same metabolic job, different molecular solution. 🌋 In other words, both lineages may have inherited the same early biological foundation, yet each found its own way to become less dependent on the vent. The researchers also showed that phosphite and naturally occurring metals can drive phosphorylation reactions in water, hinting that some energy chemistry may predate modern ATP-based metabolism. So I would not say life itself originated twice. A better interpretation is that cellular independence may have evolved twice. And that makes the origin of life look less like a single moment, and more like chemistry gradually learning how to survive without the environment doing the work for it. Source: Scientists find evidence for two origins of life on Earth, Science Daily