Optical frequency combs have transformed the way scientists measure light and time, earning the Nobel Prize in Physics in 2005 and underpinning technologies ranging from atomic clocks to exoplanet detection. Yet the miniature on-chip versions of these combs, known as soliton microcombs, have long suffered from a frustrating paradox: the most stable and useful operating regimes have also been the most fragile and difficult to sustain. A newly published study in Light: Science & Applications reports a significant advance toward resolving this tension. Researchers led by J. Tang, J. Yang and E. Yan describe a multimodal locking-enabled soliton microcomb that combines robust, self-starting operation with an exceptionally low repetition rate on the scale of a day of stable measurement time, opening a path toward practical high-precision metrology systems that can run unattended for extended periods.

At the heart of the advance is a technique the authors call multimodal locking, a scheme that stabilizes not just a single resonance of the optical microresonator but multiple modes simultaneously, locking the comb dynamics to a regime in which dissipative Kerr solitons can form and persist without delicate manual tuning. In conventional microcomb systems, soliton formation typically requires a precise and rapid traversal of a “chaotic” modulation instability regime, a step that is notoriously sensitive to thermal transients in the resonator chip. Laser frequency must be swept with exquisite control while the soliton’s own heat dissipation shifts the resonator’s resonances, creating a feedback loop that can destroy the soliton state within microseconds. The multimodal locking approach circumvents this fragility by engaging several mode families of the resonator at once, so that the interplay among them pins the comb to a stable soliton state even as environmental and thermal conditions fluctuate.

The significance of achieving a low repetition rate in the same device cannot be overstated. A frequency comb’s repetition rate is the spacing between its individual comb teeth, and for many applications, coarser spacing is a decisive advantage. In laser ranging and distance metrology, for example, the ambiguity range of an interferometric measurement is set by the inverse of the comb’s free spectral range. Microcombs, generated in whispering-gallery-mode or ring resonators only a few millimeters across, usually exhibit repetition rates of tens to hundreds of gigahertz, which severely limits the unambiguous measurement range available to a comb-tooth-resolved distance measurement. By engineering a low-repetition-rate soliton state, the research team extends this ambiguity range dramatically, making the microcomb directly useful for absolute distance measurement, multi-wavelength interferometry and coherent communications, where each comb line must be spectrally resolvable and addressable.

Achieving both low repetition rate and robust soliton generation in a single microresonator is a formidable engineering challenge. Low repetition rates demand large resonator diameters, which in turn produce dense spectra of resonances and increased susceptibility to mode crossings with higher-order transverse and polarization mode families. Such crossings historically destabilize soliton formation, causing comb states to collapse or preventing single-soliton operation altogether. The multimodal locking strategy turns this liability into an asset: rather than engineering mode crossings away, the researchers harness the coupling among mode families to lock the comb dynamics into a favorable basin of attraction. The result is a soliton microcomb whose low repetition rate is preserved over day-scale timeframes, a duration that places the device among the most operationally stable microcomb demonstrations reported to date.

Day-scale stability matters because it transforms the microcomb from a laboratory demonstration into an instrument. Many high-precision experiments, from astronomical spectrograph calibration to the accumulation of statistical confidence in atomic spectroscopy, require comb sources that remain locked and stable across many hours or even days of continuous operation. Conventional fiber-laser-based frequency combs achieve this, but at the cost of table-top size, high power consumption and substantial maintenance overhead. Microcombs promised to shrink the technology onto a photonic chip, yet until now their operational fragility has confined most demonstrations to short observation windows. The robust low-repetition-rate soliton microcomb described in the new work demonstrates that chip-scale frequency combs can, with appropriate locking architecture, sustain the kind of long-duration, hands-off operation that real metrology campaigns demand.

The physics underlying soliton microcombs rests on the Kerr nonlinearity of the resonator material, typically silicon nitride, magnesium fluoride or fused silica. When continuous-wave pump light resonantly builds up inside the ring, the Kerr effect produces a four-wave-mixing process that seeds sidebands spaced by the resonator’s free spectral range. Under the right conditions of pump detuning and intracavity power, these sidebands evolve into a coherent train of ultrashort optical pulses, dissipative Kerr solitons, that circulate around the resonator and emit a broadband comb spectrum with tooth spacing equal to the repetition rate. The comb’s output frequencies are determined by two radio-frequency quantities, the pump laser frequency and the repetition rate, which is why the combs can serve as rulers for measuring optical frequencies against a reference. Any drift in either quantity propagates directly into measurement error, and the multimodal locking mechanism suppresses precisely this drift by anchoring the comb’s underlying mode structure to the resonator’s intrinsic stability.

Applications for such a device span an unusually broad swath of science and engineering. In high-precision laser ranging, low-repetition-rate combs with tooth spacings in the low-gigahertz or even megahertz regime allow interferometric distance measurements over ranges of meters to kilometers without the need for auxiliary synthetic-wavelength techniques. In astronomical spectroscopy, comb calibration sources with well-controlled line spacing enable the detection of Earth-like exoplanets through radial velocity shifts measured in centimeters per second. In telecommunications, low-repetition-rate combs align naturally with standard channel grids used in wavelength-division multiplexed fiber networks, making them attractive as multi-carrier sources. In spectroscopy, dual-comb techniques benefit from repetition rates that are low enough to permit resolution of closely spaced molecular absorption features. A microcomb that is simultaneously robust, low-repetition-rate and day-scale stable therefore functions as a versatile enabling component across these domains rather than a single-purpose device.

The demonstration also speaks to a broader trend in photonics: the migration from delicate, expert-operated laboratory setups to self-contained, self-stabilizing systems. The self-starting character implied by multimodal locking is particularly important for this transition. A comb that requires a skilled operator to initiate soliton formation, sweeping the laser with practiced timing past the chaotic regime, cannot be deployed in an observatory dome, on a satellite platform or inside an industrial instrument. A comb that engages its soliton state reliably on demand and then holds it through thermal cycling, laser drift and mechanical perturbation removes the last practical barrier to integration. The authors’ demonstration of day-scale continuous operation is, in effect, a stress test passed by a technology that researchers have been attempting to harden for more than a decade.

The implications extend to metrology standards and timing as well. Optical atomic clocks, which now achieve fractional frequency uncertainties below one part in 10^18, rely on frequency combs to divide down their optical oscillations to countable microwave rates. Miniaturizing the comb stage of such clocks is a prerequisite for portable optical timekeeping, which would benefit navigation in GPS-denied environments, geodesy that monitors sea level and crustal motion with millimeter precision, and fundamental physics tests that track the constancy of natural constants over months and years. A robust chip-scale soliton microcomb with day-scale stability contributes a key subsystem to this vision, and the multimodal locking principle may generalize to other resonator platforms, materials and spectral regions, including the mid-infrared where molecular fingerprinting awaits compact comb sources.

Looking forward, the study suggests several directions that the photonics community is likely to pursue. Combining multimodal locking with on-chip pump lasers and integrated electronics would complete the transformation toward fully self-contained comb modules with no free-space optics. Extending the approach to lower repetition rates still, into the megahertz regime where fiber-laser combs operate, would bring microcombs into direct competition with their bulkier predecessors for the most demanding ranging and spectroscopy tasks. Improvements in resonator fabrication, reducing mode-volume disorder and engineering dispersion with greater precision, will further expand the parameter space in which robust soliton states can be locked. The present work demonstrates that the fragility once considered intrinsic to soliton microcombs is not a fundamental limit but an engineering problem, and one that can be solved by embracing, rather than avoiding, the multimodal structure of real optical resonators.

The research, published in Light: Science & Applications, marks a milestone in the maturation of microcomb technology. By uniting three properties that had previously been achieved only separately, self-starting robustness, low repetition rate and day-scale operational stability, the work delivers a soliton microcomb genuinely suited to high-precision metrology. As the demand for compact frequency references accelerates across navigation, astronomy, communications and fundamental science, the multimodal locking strategy reported by Tang, Yang, Yan and colleagues offers a clear and technically grounded route from chip-scale optical chaos to dependable, precision measurement light.

Energy storage is becoming critical to grid resilience and electricity affordability because battery systems can help balance supply

💾

Researchers at the U.S. Department of Energy’s Oak Ridge National Laboratory’s National (ORNL) Transportation Research Center have developed

Startup Bluecore Energy has raised $50 million in seed funding to develop floating nuclear power plants intended to supply ports and other locations with rapidly growing energy demands. The California-based company said the Silverton Partners-led round, which includes $10 million in previously announced pre-seed funding, brought in several new investors. It plans to use the money to engineer and test its compact nuclear power system and pursue regulatory approval and maritime classification. Its initial design is intended to generate about 10 MWe – megawatts of electrical output – continuously using light-water reactor technology. Bluecore says the maritime system would be able to operate for years at a time and require refueling only once every few years. The company has secured its first barge and is developing a non-fueled, electrically heated reactor module prototype at its Port of Long Beach headquarters to test monitoring, sensor, and control systems. Bluecore is targeting ports, AI datacenters, and coastal infrastructure, with the reactors potentially deployed offshore and connected to customers by subsea cables. Remote islands and coastal settlements in places such as Alaska may also lack practical connections to a wider electricity grid. Bluecore claims its approach could provide reliable, zero-emission power to energy-intensive locations and open additional markets for nuclear generation. "Six months ago, we were building the foundation. Today, we have the capital, the team, the hardware, and some of the most important institutions in nuclear and maritime working alongside us," said Bluecore founder and CEO Kofi Asante. "Our focus now is simple: create and deliver zero-emission energy as safely and quickly as possible." The move follows the launch of the International Atomic Energy Agency's (IAEA) Atomic Technologies Licensed for Applications at Sea (ATLAS) initiative in the US capital last month. Secretary of Energy Chris Wright linked that event to the Department of Energy's own efforts to rapidly expand the provisioning of atomic energy generating capacity in order to meet the requirements of AI datacenters and other industries. Bluecore has begun discussions with the US Nuclear Regulatory Commission (NRC) and Coast Guard. The two agencies recently signed a memorandum of understanding to coordinate oversight of the design, construction, and operation of civilian floating nuclear power plants. The NRC says it is ready to license maritime reactors under existing federal regulations, while it develops an additional framework for microreactors and similarly low-risk designs. Bluecore remains years from commercial deployment. It has yet to build a fueled reactor or apply for the approvals required to operate one, and its Long Beach proposal must also navigate California's restrictions on new nuclear power plants. While some may have concerns about waterborne atomic power plants, America has a long history of maritime nuclear generation. All of the US Navy's active aircraft carriers are nuclear-powered, but the technology first went to sea aboard the submarine USS Nautilus in 1955. USS Enterprise became the first nuclear-powered aircraft carrier in 1961, while the cruiser USS Long Beach, commissioned the same year, was the first nuclear-powered surface combatant. Bluecore says its team includes former US Navy nuclear submarine officers and leaders with experience across organizations including SpaceX, Northrop Grumman, Toyota, Rivian, and Uber. Russia currently operates the world's only floating nuclear power plant. The barge-like Akademik Lomonosov is considerably larger than Bluecore's proposed design and can generate up to 70 MWe. ®

ASML and chipmaking giant TSMC have launched an industry-wide push toward 12-inch photomasks, saying the larger format could lower costs and overcome some limitations of high-NA EUV lithography. Dutch firm ASML, the only commercial supplier of EUV lithography systems, says it and TSMC have established a collaboration intended to lead the transition to larger-format photomasks for extreme ultraviolet (EUV) lithography. Intel Foundry and Samsung Electronics have also voiced their support. The initiative aims to establish a 12-inch mask pilot line by 2031 and have the supporting lithography systems ready for advanced node production by 2033. According to ASML, high-NA EUV will initially enter production using current 6-inch masks, but moving to 12-inch masks could increase fab productivity, lower costs, and remove stitching constraints. The push for larger masks stems from the anamorphic optical design ASML adopted for its first generation of high-NA EUV machines so chipmakers could continue using existing 6-inch masks. The "NA" in high-NA EUV stands for numerical aperture – a measure of an optical system's ability to collect and focus light. Achieving a higher NA required larger optics, which increased the angle at which light strikes the reticle, or mask, creating shadowing and contrast problems. ASML says it could have used optics that reduced the mask pattern by 8x in both directions, rather than the 4x used in existing systems, but that would have required larger masks. Instead, it adopted anamorphic optics that reduce the image by 4x in one axis and 8x in the other, allowing chipmakers to continue using 6-inch masks. The trade-off is an exposure field half the size of that in previous machines. Large dies that exceed this field may therefore require two separately exposed patterns to be stitched together, adding complexity and reducing productivity. ASML and TSMC now see larger masks as a way to restore the full exposure field as the industry pushes toward denser chips and smaller process nodes. "We expect the adoption of High NA EUV to increase progressively along the device scaling roadmap, first using current 6-inch masks and then further supported by 12-inch masks, which enable greater scanner productivity and allow the industry to meet the demand for smaller, faster and more energy-efficient chips," said ASML president and CEO Christophe Fouquet. TSMC has so far not used high-NA EUV, including for its 2 nm process, but says it intends to deploy the technology in high-volume manufacturing for advanced nodes beginning in 2030. It expects the number of layers requiring high-NA EUV will rise as new process nodes are introduced, driven by increasingly complex transistor architectures required for AI applications. Intel Foundry signaled its support for the initiative, with EVP Naga Chandrasekaran saying: "Within our lithography capabilities, Intel Foundry is focused on near term enablement of high-NA on 6-inch masks with or without stitching, and making the transition to 6x12-inch masks. We will continue to work closely with ASML and the entire industry to enable this transition." Samsung is also backing the initiative and says it plans to become the first chipmaker to introduce ASML's high-NA EUV technology into high-volume DRAM manufacturing by 2028. "The AI era is transforming the semiconductor industry and increasing the importance of technological innovation across the entire value chain," said Samsung Electronics vice chairman and CEO Young Hyun Jun. "By further strengthening our collaboration with ASML, we are helping lay the foundation for the next generation of AI and semiconductor innovation." IDC senior research director Andrew Buss told The Register that high-NA EUV creates design and manufacturing challenges for future products. Although reticle stitching and chiplet designs can mitigate these, he said, the move to 6 x 12-inch photomasks is a necessary step comparable to the industry's transition from 200 mm to 300 mm wafers. Buss said the transition would require toolmakers, chip manufacturers, mask suppliers, and designers to align their technologies and processes, making broad industry participation essential. ®

Arm's latest smartphone platform is optimized for running AI agents on mobile devices and introduces neural graphics that it claims can deliver desktop-class gaming within a smartphone's power constraints. Announced at the Arm Everywhere China event in Shanghai today, Compute Subsystem (CSS) for Mobile 2 succeeds last year's Lumex CSS and reflects Arm's embrace of AI as a strategy for future growth. Arm sees itself as central to this shift, arguing that AI workloads currently run largely in the cloud but will increasingly move to devices at the network edge. "When every interaction has to travel to a datacenter and back, there's implications for latency, connectivity, the cost of delivering those services at enormous scale, so this is where we're seeing it becoming more distributed, and I think few doubt now that basically everything you can push to the edge you will," says Arm's EVP of Edge AI, Chris Bergey. AI has evolved from background capabilities such as speech recognition and image processing into tools people actively engage with, including generative assistants, Bergey claims. Agentic AI goes further by interpreting objectives, coordinating applications and models, and executing sequences of actions. "So you have repeated inference, application logic, system orchestration, multiple workloads operating concurrently, and so that means you don't just need more CPU performance. You need multiple high-performance CPUs working in parallel to keep all those different workloads running," Bergey explains. With this in mind, the C2 CPU cluster features a new top-end C2-Ultra core that Arm claims delivers up to 15 percent higher single-thread performance than last year's C1-Ultra. The C2-Ultra cores have a larger execution engine to keep more instructions in flight, and Arm claims more accurate branch prediction and better fetch and target prediction to keep the instruction pipelines fed. Arm's chipmaker licensees can mix and match components to suit their requirements, as usual, but the firm touts a flagship cluster with two C2-Ultra cores, six efficiency-focused C2-Pro cores, and two SME2 units (Scalable Matrix Extensions), compared with one SME2 unit in last year's platform. Doubling the SME2 capability enables up to a 70 percent speedup on the latest small language models, according to Arm, while the cluster topology supports private caches on the individual CPUs plus a large shared L3 cache. All of this is tied together by the DSU, or the DynamIQ Shared Unit. Arm says the cluster has been optimized to reduce latency and keep concurrent agentic workloads responsive. CSS for Mobile 2 also introduces dedicated neural acceleration for graphics in the Mali G2-Ultra NX, which Arm calls its first AI-native Mali GPU. Arm previewed the neural technology last year, giving developers time to prepare for the hardware. The GPU introduces an all-new execution engine, neural accelerator hardware, and a third-generation ray tracing unit, built for full-scene lighting and shadows. The hardware supports Neural Super Sampling (NSS), which upscales game graphics from 540p to 1080p, and Neural Frame Rate Upscaling (NFRU), which generates intermediate frames to increase the frame rate. Using both techniques means the GPU renders one eighth of the pixels contained in the final sequence of displayed frames, with the neural hardware reconstructing the rest. Arm claims this makes desktop-class graphics possible within a 1 W power budget. "The G2-Ultra NX GPU is able to sustain a 30 frame per second gaming session on content with very rich and intricate ray traced scenery. And this is really enabled thanks to a 70 percent reduction in the overall ray tracing workload, which then translates to a 30 percent boost in performance," says GPU Product Manager Deyan Lazarov. Arm also demonstrated other games running at 60 frames per second. Like the CPU cores, Arm is offering Premium and Pro variants of the Mali G2 GPU, allowing its chipmaker customers to trade off performance against cost. To encourage uptake, Arm has published its neural graphics SDK and sample code on GitHub, with support for the cross-platform Vulkan graphics API. As with previous Arm smartphone platforms, chipmakers must now turn the designs into working silicon before smartphone vendors can bring devices to market, possibly as early as next year. ®

Huawei has launched a new processor it claims is free of US technology and is using it to drive a new three-screen folding phone. The Chinese giant dubbed the new processor the Kirin 9050 Pro and says its LinxiCore processor cores, GPU, and neural processing unit all feature its own designs. The SoC into which Huawei added them uses the "Tau Scaling” architecture that Huawei says arranges components closer to each other to reduce die size and improve performance without needing the kind of advanced chipmaking processes practiced by TSMC and Intel. As The Register wrote when Huawei announced Tau Scaling last May, analysts think Huawei has done interesting things with its silicon but don’t think the company has made any breakthroughs. Huawei begs to differ and says its new silicon offers 42 percent higher performance than the chip it used in the three-screen phone it delivered last year. The Chinese company uses the Kirin 9050 Pro in its new three-screen phone, the $2,979 Mate XT2. Analyst firm Counterpoint thinks Huawei will exit 2026 with 24 percent of the total folding phone market, which it believes accounted for 1.6 percent of smartphone sales last year. Huawei therefore has itself a chip that targets the high end of a market niche – and announced it a couple of days before pundits believe Apple will debut a folding iPhone that Counterpoint predicts will win 25 percent of the folding phone market by the end of the year. Huawei’s Kirin 9050 Pro will therefore hardly make a dent in the smartphone market. At its launch event, Huawei execs pointed to its totally-not-Android versions of HarmonyOS now running on 85 million devices. That’s strong uptake since Huawei delivered the OS in October 2024. Again, however, it is a fraction of the overall smartphone and smart device market. If Huawei can use the tech it created for the Kirin 9050 Pro in other roles that would allow local orgs to avoid comparable products from AMD, Intel, or Nvidia, its new chip will show that Chinese firms can create tech that reduces the nation’s reliance on imports. But amid all the hoopla of a phone launch, Huawei hasn’t said anything about its intentions to adapt its new chip for different roles. The Tau Scaling architecture is therefore not going to upset Huawei’s rivals for some time, although western chipmakers do know that Beijing is steering local buyers away from their products and encouraging adoption of local alternatives. Whether any of those Chinese chipmakers succeed beyond the Middle Kingdom is the wider question. China emphasizes export of consumer products as a tactic to fuel domestic growth, and its smartphone makers Xiaomi, Oppo, and Honor have delivered on that mission by winning solid sales across Europe, Africa, and Latin America – with phones running Android and tied to Google services. Huawei’s hardware and HarmonyOS ecosystem, by contrast, have not racked up notable success outside China. ®

It is surprising to me that the authors of these studies did not consider it more of a priority to understand the mechanics of robustness.

Source

German chemical conglomerate BASF has filed a patent infringement lawsuit against Apple in a U.S. federal court, accusing the iPhone maker of using its technology in Face ID and related facial authentication systems without a license.

The complaint, lodged in the United States, claims Apple’s TrueDepth camera system, Secure Enclave processing, and liveness-detection features in iPhones, iPads, and other devices infringe multiple BASF-owned patents covering methods for secure biometric matching, three-dimensional facial mapping, and anti-spoofing analysis. BASF alleges these technologies have been incorporated into Apple products since the introduction of Face ID in 2017 and continue to be used for device unlocking, Apple Pay, and other authentication functions.

BASF, best known for chemicals, materials, and agricultural products, maintains a sizable intellectual-property portfolio that extends beyond its core industries. The company asserts the patents in question describe techniques for capturing and verifying facial geometry in a way that resists photographs, masks, or other presentation attacks — capabilities central to modern smartphone security.

Apple has not publicly commented on the specific filing. The company has faced a series of Face ID and Touch ID-related patent cases over the years from various rights holders and has generally defended its biometric systems as independently developed and implemented within its own secure hardware architecture.

The suit seeks an injunction against further sales of accused devices, monetary damages for past infringement, and other relief. Patent cases of this type often take years to resolve and frequently involve challenges to patent validity at the U.S. Patent and Trademark Office in parallel with district-court litigation.

The dispute highlights how companies outside the traditional consumer-electronics sector sometimes assert older or adjacent patents against high-volume products such as smartphones. Whether BASF’s claims will survive early motions or claim-construction proceedings remains to be seen. Apple routinely contests such assertions and has prevailed in several prior biometric-patent matters.

MacDailyNews Take: The case is yet another in the crowded field of Face ID-related litigation that has dogged Apple since the feature launched nearly a decade ago. Courts and the patent office will now determine whether BASF’s patents actually cover the specific combination of sensors, algorithms, and secure processing Apple deploys.

And, yes, we haven’t thought of BASF since:

Also, yes, we’re old.

---

‎
Please help support MacDailyNews — and enjoy subscriber-only articles, comments, chat, and more — by subscribing to our Substack: macdailynews.substack.com. Thank you!

Support MacDailyNews at no extra cost to you by using this link to shop at Amazon.

The post BASF sues Apple over alleged face authentication patent violations appeared first on MacDailyNews.

Since the formation of the Earth 4.2 billion years ago, hot rock from the interior of the planet has cooled off and developed crystals that accumulated over millions of years, forming surface rocks with a layered composition, referred to as continental crust. Geologic evidence of how continental crust formed comes from large, ancient crustal blocks, called cratons. Cratons have existed for billions of years and form the cores of continents. 

Scientists have shown that the upper 10 to 15 kilometers (6 to 9 miles) of continental crust contains high concentrations of elements like uranium and thorium that decay and release heat over time, referred to as radiogenic elements. However, the lower 20 kilometers (12 miles) of crust has very few radiogenic elements. The cause of this planetary-scale differentiation is highly questioned, as geoscientists don’t know what formation processes could produce this unique characteristic. 

Earlier researchers suggested that changes in the concentration of radiogenic elements could occur when tectonic plates collide. Plate collision induces temperatures in the lower crust high enough to melt minerals that contain uranium and thorium, like zircon and monazite. The newly melted minerals form magma, which then rises and crystallizes in the upper crust. To test whether this process formed the layered continental crust, researchers Andrew J. Smye and Peter B. Keleman incorporated geochemical data from studies worldwide into a mathematical model comparing how concentrations of radiogenic elements vary under different temperatures. 

The team separated their global analyses into 2 categories of rocks that were buried deep in the crust and subjected to intense pressures and temperatures, via a process known as metamorphism. The first category included rocks formed from the accumulation of eroded grains buried in the deep crust that are subjected to high pressures and temperatures, called metasedimentary rocks. The second category included rocks formed when magma rises from the deep Earth and solidifies before experiencing high pressures and temperatures, called metaigneous rocks.

The researchers analyzed 3 subcategories from each of the 2 types of deep-crustal rocks. The first was rocks that experienced no intense pressure or temperature, the second was rocks subjected to 650 to 900°C (about 1,200 to 1,650°F), and the third was rocks subjected to 900 to 1,100°C (about 1,650 to 2,000°F). The researchers modeled the geochemistry of the 3 rock types at different temperatures to determine whether ultra-high-temperature metamorphism could cause crustal differentiation of radiogenic elements.

Smye and Keleman found that monazite doesn’t melt in either metasedimentary or metaigneous rocks at temperatures below 900°C, as indicated by the relatively high thorium content in these deep-crustal rocks. Alternatively, they found that above 900°C, uranium and thorium concentrations decreased in metasedimentary rocks, suggesting that minerals like zircon and monazite were melting. The melted minerals enter magma and rise into the upper crust, leaving behind metasedimentary rocks depleted in radiogenic elements by around 0.10 ppm.  

The team concluded that the enrichment of uranium and thorium in the upper continental crust requires temperatures greater than 900°C to initiate the melting of zircon and monazite in the lower crust. Temperatures above 900°C are linked to the movement and collision of continental plates. This result suggests that evidence of Earth’s differentiation has been preserved for more than 3 billion years, spanning most of the planet’s geological history. Since stable continental crust formed, uranium and thorium enrichment in the upper crust has consistently recorded layering produced by ultra-high temperature metamorphism.

The post Why is Earth’s crust layered? appeared first on Sciworthy.

Since the 1850s, plowing and clear-cutting in modern agriculture have increased how fast soil erodes in the Upper Mississippi River Valley. Across the globe, farming strips soil up to 100 times faster than natural processes. It also removes farmable land and clogs waterways, threatening long-term environmental sustainability. However, scientists in the past have struggled to separate the effects of land-use changes from the transition out of the last ice age over the past 20,000 years. 

To determine whether current farming practices are sustainable, researchers from Minnesota recently measured the speed at which soil eroded during the ice age transition and compared it with modern agricultural erosion. They focused on the Trout Creek tributary of the Mississippi River. This tributary has 2 step-like formations, or terraces, on its bank that represent past floodplains before the river eroded to its current level. 

The oldest, upper terrace is about 14 meters (50 feet) above the lower terrace, which is about 8 meters (26 feet) above the modern floodplain. Scientists studying nearby areas have suggested that the terraces contain sediment from the last ice age, 20,000 years ago, referred to as the Last Glacial Maximum, to today. 

In the upper terrace, the researchers pushed an empty cylinder into the ground, like a straw into a layer cake, to collect a cylindrical sample of sediments about 4 centimeters (1.5 inches) wide and 6 meters (20 feet) deep, called a sediment core. They collected slices of this core from 2 meters (6 feet) and 6 meters (20 feet) to find the terrace’s age. 

Some grains in the sediment trap electrons while buried and release them when exposed to light. So the team shone light on the sediment slices to stimulate these electrons and measure the intensity of light emitted, known as their optically stimulated luminescence. Brighter light emissions indicate the sediment is older and was buried for a longer period of time. Using this method, they confirmed that the deep sediment slice was 22,000 years old, and the shallow sediment slice was 20,000 years old, dating the terrace back to the Last Glacial Maximum.

Next, the team drilled into the upper terrace with an auger to collect 25-centimeter (10-inch) wide plastic tubes of sediment. They pulled 4 samples from 0 to 2 meters (0 to 6 feet), and 1 sample from 6 meters (20 feet) deep. They explained that minerals in the sediment called quartz produce the radioactive form of the element beryllium, known as ¹⁰Be, when exposed to the Sun, so low ¹⁰Be means the sediment was buried quickly and had little sun exposure. The team measured the concentration of ¹⁰Be in each sediment sample and used an online calculator to estimate how quickly it eroded. The researchers calculated that during the Last Glacial Maximum, 0.073 millimeters (0.003 inches) of surface sediment were removed per year.

The researchers collected similar samples from the lower terrace and used optically stimulated luminescence to determine that it was between 22,000 and 14,000 years old. This terrace was the floodplain during the transition out of the Last Glacial Maximum, or deglacial period, where temperatures warmed and permafrost retreated. They measured the concentration of ¹⁰Be in samples from this terrace and determined that 0.049 millimeters (0.002 inches) of surface sediment were removed per year. 

Finally, the researchers measured ¹⁰Be in a sediment sample from the modern floodplain, deposited between 14,00 years ago and today. They found that the sediment had eroded 0.053 millimeters (0.002 inches) per year during that interval, at a rate slower than it had during the glacial period, but similar to the deglacial period. 

The researchers compared their findings to the average soil loss caused by modern agriculture, as reported by 4 previous scientists studying the area. They saw that modern farming removes an average of 0.60 millimeters (0.02 inches) of soil per year, which is 8 to 12 times greater than all the natural erosion rates they calculated. 

They concluded that the recent spike in soil loss from agriculture greatly exceeds that caused by the largest natural climate transition in recent Earth history. Human-driven erosion shows an upward “hockey stick” trend like the one seen in global temperatures, suggesting that similar trends could exist throughout Earth systems. The team concluded that modern agricultural practices aren’t currently sustainable. While measures like native plant reintroduction can help curb this trend, more research is required to determine how best to limit long-term erosion.

The post Is modern farming sustainable? appeared first on Sciworthy.

Temperatures in the Arctic are increasing at a rate 4 times faster than the average global temperature. Scientists have documented the global effects of climate change, but there are still regions where these effects are less well known. This is the case for freshwater ecosystems in the Arctic.  

When the ground has been permanently frozen for 2 years, scientists call it permafrost. Permafrost contains soil, sand, rocks, and organic matter held together by ice. As temperatures rise, permafrost thaws, which can change groundwater flow paths. It can also alter a stream’s temperature, flow patterns, and nutrient loads by introducing new material from the permafrost.  

To determine how thawing permafrost affects Arctic ecosystems, scientists from the U.S. Geological Survey recently studied 10 headwater streams within the Brooks Range in the Noatak National Preserve of Alaska. The scientists studied streams with different amounts of permafrost and ice to test whether connections exist between permafrost, food webs, and populations of fish called the Dolly Varden and Arctic Graylings. They focused on these 2 species of fish because their migrations often indicate ecological shifts in the area.    

The scientists collected fish from the streams by temporarily stunning them, using a method called electrofishing, or by catching them in minnow traps. They manually identified each species they collected, then froze the fish and shipped them to a lab. Once in the lab, the scientists measured the length of each fish and used documented growth patterns to estimate its biomass. To calculate fish abundances, they divided their total biomass by the time spent electrofishing or trapping. Finally, they measured the energy density of each fish by burning it in a device called a bomb calorimeter. 

At each stream, the scientists took manual temperature measurements and deployed loggers that recorded the temperature every 15 minutes. They also deployed equipment at the bottom of each stream to record water pressure in 15-minute intervals and converted the data to depth, since pressure increases with depth. Then, they used wading rods to measure how much water was flowing through each stream at a given spot. 

They combined the depth and wading rod data to determine the flow rate of water in the streams over time. They used these data to identify unstable streams where water rose and fell rapidly. They also collected water samples from each stream and measured their nutrient concentrations, including dissolved organic carbon, nitrogen, and phosphorus, to determine water quality.   

Finally, they measured 2 types of smaller organisms that allow energy to flow through the food web to the fish, called microbial biofilms and macroinvertebrates. They used a hand-held device to estimate the types and amounts of biofilms, and nets to collect macroinvertebrates.

Once they had compiled these data, the team conducted a statistical analysis to compare the studied parameters and fish populations to the extent of permafrost in each area. They found that streams with more permafrost in the area had warmer temperatures, because the permafrost prevented water from infiltrating the soil to cool off. The streams with warmer temperatures, in turn, had lower abundances and biomass of Dolly Varden fish and Arctic Grayling with lower energy densities. They suggested that these results meant that the high temperatures had pushed the fish above their optimum temperature range. 

They also learned that the limited infiltration in permafrost areas caused rainfall to flow straight into the stream. This process created instability as the stream levels rose and fell rapidly. Fish in these unstable streams had lower energy densities, likely due to the harsher environment. 

Areas with greater permafrost also had higher concentrations of dissolved organic carbon and phosphorus, which stimulated the formation of biofilms at the bottom of the food chain. They noted that areas with more biofilms also had more macroinvertebrates and more Dolly Varden biomass. Since the Dolly Varden eats macroinvertebrates, this finding suggested that the lowest levels of the food web shape the health of species higher up. This is called the bottom-up effect, and means that areas with more biofilms and macroinvertebrates will likely have healthier and larger fish populations.  

The scientists explained that knowing how these factors influence Arctic fish populations can help researchers understand how thawing permafrost might affect Arctic ecosystems in the future. They anticipated that their data could help future researchers predict how these and other ecosystems will be impacted by changing landscape dynamics caused by global warming. 

The post Thawing permafrost is affecting fish health appeared first on Sciworthy.

Have you ever experienced a severe weather event, like a hurricane or a tornado? Researchers have shown that climate change is causing such weather events to become more extreme and more frequent, destroying communities and infrastructure. To determine which areas are most at risk, researchers from Canada studied the environmental effects of a powerful flooding event that occurred in southwestern British Columbia (BC) in November 2021. 

They focused on a type of severe flooding event that is characteristic of the North American Pacific coastal region, called an atmospheric river. An atmospheric river occurs when a long, narrow vapor cloud flows toward land from the Pacific Ocean, causing heavy rainstorms. In November 2021, an atmospheric river in southwestern BC destroyed numerous regional highways, pipelines, power lines, and railways within its path. Over 18,000 people were evacuated due to flooding, and 6 people died. 

The team explained that before the storm, BC experienced a severe drought and an unprecedented heatwave, followed by widespread thunderstorm activity. In 2021, lightning and human activity caused nearly 1,000 wildfires, which led to the third-worst wildfire season ever recorded in BC, partially because trees grow back slowly in forests that have been cut down or harvested. Poor vegetation recovery makes harvested forests more vulnerable to wildfires than old-growth forests. The November atmospheric river occurred shortly after this wildfire season. 

To determine how the atmospheric river impacted southwestern BC, the researchers analyzed approximately 70,000 square kilometers (27,027 square miles) of affected area. They compared satellite images of southwestern BC from Google Earth, Sentinel, and Landsat taken before and after the event to detect changes in river channels and identify geohazards, like landslides and debris flows. They also used satellite images and Canadian government data to identify which forests had been harvested within the study area. The researchers then traversed the area to confirm their remote analyses. 

The researchers used a type of 3D remote sensing, called light detection and ranging, or lidar, to measure how the 2021 atmospheric river redistributed sediment in the Coquihalla and Coldwater River valleys. With the help of lidar, the researchers discovered that both rivers had severely unstable banks due to erosion from the flooding event. Along the Coquihalla River, approximately 53,000 cubic meters (1,871,677 cubic feet) of sediment were eroded, while nearly 50,000 cubic meters (1,765,733 cubic feet) of sediment were eroded along the Coldwater River. 

The researchers also found that hillslope geohazards increased the volume of sediment in both rivers. And in some cases, upstream areas with a history of forest harvesting and wildfires also affected stable banks downstream. For example, they identified debris flows on an unstable hillslope that triggered a landslide on a riverbank downstream. They observed that the river and stream channels shifted due to a sudden addition of sediment from hillslope geohazards and floodwater during the event.

In total, the researchers documented over 1,300 geohazards along 47 kilometers (about 30 miles) of the Coquihalla River and 19 kilometers (about 12 miles) of the Coldwater River and nearby creeks. Nearly half of the geohazards originated in an area with a history of wildfires or forest harvesting. Of the 155 geohazards at the Coquilhalla River, 43% occurred below forest harvesting resource roads, and of the 117 geohazards observed at the Coldwater River, 29% originated in areas with past wildfires.

The researchers suggested that their findings could help scientists predict where severe geohazards will occur during extreme weather events, since both watersheds had geohazards originating in harvested forests. Before 2000, clear-cutting was a widespread practice in BC, and the team’s results demonstrate that unsustainable harvesting practices like this can have lasting environmental impacts.

The researchers concluded that both forest harvesting and wildfires contributed to the severe impacts of the November 2021 atmospheric river, causing catastrophic landslides and debris flows to occur in traditionally dry regions. They suggested that more accurate risk assessments can better equip communities for severe weather events like this. While the next atmospheric river that lands in the Pacific coastal region may be worse than the last, this study can inform scientists and communities about which areas it might affect most.

The post Wildfires and forest harvesting worsen severe weather impacts appeared first on Sciworthy.

The IEEE is warning that China’s export controls on rare earth minerals could start to impact datacenters, as shipments of some key materials are down 75 percent and Beijing is due to review its policies this November. The techie org's official publication, IEEE Spectrum, focuses on two rare earths in particular, erbium and yttrium, which are used in components for IT equipment. Exports of yttrium are down by about 75 percent this year in comparison with 2025. Yttrium was one of seven medium and heavy rare earth elements that China placed under mandatory licensing in its first wave of restrictions in 2025. Yttrium is used in white LED lights and displays, microwave filters, and solid-state lasers, and also in ceramic capacitors found in pretty much all electronic kit. Erbium is used in the amplifier circuits that boost optical signals in fiber-optic links. The Chinese government enacted restrictions on the overseas sale of rare earth minerals last April, as reported by The Register. This was part of retaliatory measures against the Trump administration’s slapdash trade tariffs. It upped the ante later in the year with tougher measures, including laws that require companies to apply for a license before they ship any goods containing rare earths sourced in China to anywhere in the outside world. This latter rule is a mirror of US laws like those used to block the export of ASML chipmaking equipment to China. That second wave of measures – which was to have included increased export controls of five new elements including erbium – was suspended in November 2025 for 12 months, following trade talks between the US and China. When the year is up this November, Beijing will decide whether to drop the tougher measures or re-apply them, perhaps extending them to all or most of the rare earth minerals it controls. According to the International Energy Agency (IEA), the supply of rare earths remains among the least geographically diversified among all critical minerals. For those minerals used in magnets, China accounted for about 60 percent of global mining output in 2024, and its dominance is even greater in the separation and refining stages, making up about 91 percent of global production. But for yttrium and erbium used in IT kit, China is effectively the sole source, IEEE Spectrum claims. The country mines 90 percent of the world’s yttrium, but it processes all of the world’s mined ore into an industrially usable form such as yttrium oxide, and likewise supplies all erbium oxide. Washington has been taking some steps to try and offset China’s virtual monopoly of key rare earth materials. Recently, the US Department of Defense announced a $750 million investment to secure supplies of rare earth elements for magnets from a production site in central Brazil, while the Department of Energy announced $500 million for seven selected projects to expand critical mineral and material processing, battery manufacturing, and recycling. The US signed agreements with representatives of 54 countries at the 2026 Critical Minerals Ministerial meeting in February this year, as well as talking up commitments to build secure and resilient supply chains for such materials. Also in February, President Trump launched Project Vault, a supply chain security initiative establishing an American Strategic Critical Minerals Reserve stockpile, with funding of up to $10 billion from the federal Export-Import Bank of the United States. But many of these initiatives will take time to have any impact, and China’s impending decision will put these to the test, should it decide to take a tougher stance. Relations between Washingon and Beijing remain strained because of trade and other factors, and a lot could be riding on a summit between President Trump and President Xi Jinping next month. ®

Shots were fired at the Hot Chips conference in California this week as Nvidia announced its new Groq-3-based LPX racks had entered production, with early tests showing the systems churning an eye-watering 3,400 tokens a second in Gemma 4 31B. That's four times faster than rival Cerebras. A day later, Cerebras fired back, touting nearly equivalent performance from its next-gen CS-4 accelerators revealed last week. These top-line performance figures make inference feel instantaneous relative to the chatbots we've grown accustomed to over the past four years. But the two companies are essentially arguing over numbers their customers will probably never see in production. To be clear, neither is lying. If you wanted to recreate these results, you certainly could — Artificial Analysis, the team responsible for both sets of benchmarks, knows what they're doing — but beyond a marketing gimmick, no inference-as-a-service model operator in their right mind would run either system this way. Not unless you've somehow figured out how to make a profit serving one request at a time. In reality, these numbers are more like the top speed on a race car. It makes for great marketing but you probably aren't driving that fast on a regular basis, and if you did, you wouldn't get very far before your tank runs dry. These aren't the numbers that matter Unsurprisingly, the economics of "premium" or "ultra-low latency" inference are a bit more nuanced, but ultimately come down to how efficiently you can scale that performance at the rightmost end of the Pareto frontier. We've shown the graphic below before. It charts the performance characteristics of how different Nvidia B300 configurations perform across a Pareto front. As you can see, GPUs are great for high-throughput, low-interactivity applications, but run out of steam quite quickly as per-user generation rates climb. Chips like Nvidia's Groq-3 LPUs or Cerebras' dinner plate-sized accelerators offer massive bandwidth thanks to their SRAM-heavy architectures, so they can extend and push the Pareto curve out to the right. Unfortunately, the same thing that makes them so fast also limits their ability to scale that performance to large numbers of users. To understand why we need to walk through a little math. We're going to pick on Gemma 4 31B, in part because we have rough numbers for both Nvidia and Cerebras, but also because it's a small model and therefore presents a best-case scenario. Each Groq-3 accelerator is equipped with 500 MB of on-chip memory capacity. Nvidia achieved its 3,400 tok/s per user run at 8-bit precision. So to run a 31 billion-parameter model, we need about 31 GB of memory or about 64 LPUs. But that's just to hold the model's weights. Since it's running as a standalone system, we also have to account for the KV caches generated during the prefill stage in which the prompt is processed. According to LMcache's KV calculator, at 8-bit precision, a 100,000-token input sequence is going to chew up another 8 GB of memory, and that's per sequence. So for every additional 100,000 token input, that's another 16 LPUs worth of memory. By our estimate, that means the most a single LPX rack with its 256 LPUs can manage on its own before running out of memory is a batch size of 12 at a 100,000 token input length each. The numbers look a little different for Cerebras' CS4, but the conclusion is largely unchanged. Each CS4 rack contains three massive WSE-3T accelerators with 44 GB of SRAM apiece, totaling 132 GB of capacity. That leaves the system a bit more headroom, but we're still looking at a maximum batch size of 12. To be clear, that's not a compute limitation. The systems simply don't have enough memory to support larger batches without adding additional racks or cutting the prompt size. Sure, we're oversimplifying a bit. We're assuming batch inference with fixed input sequences, not concurrent users hitting the systems asynchronously, and with different input and output sequences each time. For a chatbot, especially one with a limited context window — think the model's short term memory — a single CS4 or LPX rack could conceivably serve far more users than the 12 noted here. Lifting the curve Even if both systems can hit 3,400 tok/s, the better platform is going to be the one that delivers better scale at a lower cost and power consumption. The benchmark figures touted by Nvidia and Cerebras this week are really measuring the top speed the systems could run on, not the speed customers will actually run them. In fact, customers probably won’t run these systems on their own. These SRAM-heavy architectures work a lot better as decode accelerators, which is exactly why Nvidia spent $20B to license Groq's tech and poach its engineering team. It's also why Cerebras recently teamed up with AWS and AMD. We've discussed these kinds of heterogeneous compute architectures in the past, but in a nutshell they involve GPUs or other compute heavy XPUs, like Amazon's Trainium chips, with something like Groq's LPUs, Cerebras' waferscale accelerators, or SambaNova's SN-series parts. This combination largely eliminates, or at least minimizes, the KV cache and compute overheads that made scaling Cerebras' and Groq's chips so difficult. As decode accelerators, they only really need to hold the model weights, as almost everything else can get on just fine on a GPU. Again, top line AI performance makes for great headlines, but the far right of the Pareto is rarely where you'll want to be. As is usually the case with Pareto curves, the sweet spot usually presents itself as a notable hump or knee pushing toward the upper right corner of the chart. Here you'll be trading some per-user interactivity for better concurrency, throughput, and economics. Combining GPUs with Cerebras or Groq accelerators is what Nvidia, AMD, and AWS are all counting on to make "premium inference" cost-effective. There's a better benchmark for this We've yet to see benchmarks combining Cerebras' or Groq's chips with Nvidia, AMD, or AWS' respective GPUs and XPU platforms, but these are the metrics that are actually going to matter. SemiAnalysis' InferenceX offers a better representation of whether or not adding a CS4 or LPX rack would even help, as we can see exactly how each system performs across the full Pareto, ranging from high-volume low-interactivity "bulk" tokens on the left, to ultra-low-latency but also lower-volume tokens on the right. In fact, while Nvidia and Cerebras opted for Artificial Analysis' simpler test methodology, OpenAI actually opted for InferenceX to show how its spicy new Jalapeño accelerators perform across bulk and premium inference regimes. So why didn't Nvidia and Cerebras do the same? If we had to guess, it's because their LPX and CS4 racks are still best suited toward low-latency inference and wouldn't look as impressive on the left and middle sections of the Pareto curve. We fully expect that both companies will share InferenceX results for their combined GPU and LPX/CS4 systems when they're ready, but no marketer ever got a bonus for making their product look worse than it really is, so we'll just have to wait. In any case, the next time you see someone boasting of some impossibly fast inference accelerator, it's worth asking where on the Pareto that number came from. ®

If you’re going to build a custom AI accelerator, inference is a good place to start. The clusters are smaller, the chips can be simpler, and at the end of the day they only need to do one thing: serve tokens. OpenAI’s spicy new Jalapeño inference chips are just the latest example. Social media magnate Meta is bucking this trend. Its first proper generative AI accelerator, the MTIA 400 — short for Meta Training and Inference Accelerator — is aimed squarely at LLM training. The Facebook parent is no stranger to custom silicon. But, much like Amazon and Google, its first AI accelerators weren’t built to run AI chatbots or train generative AI models. They were built to serve ads. The MTIA 400, teased earlier this year and detailed at the annual Hot Chips semiconductor development conference this week, will perform some inference duties. Just not of the GenAI variety. Instead, it’ll be saddled with running the ad recommender systems that actually pay Meta’s bills. The combination of LLM training and the deep learning recommender model (DLRM) inference used for serving ads is unusual, as the two have wildly different performance demands. LLM training is enormously compute-intensive, often requiring tens or even hundreds of thousands of accelerators to train models in a reasonable amount of time. DLRM inference, on the other hand, is a predominantly memory-bound job, which means most of the FLOPS that make the chip good at training are going to be left sitting idle. But given how much cash Meta is burning on compute these days, it’s probably not a bad thing for its chips to pull double duty – even better if one of the two use cases is highly profitable. While MTIA may play a bigger role in how Meta serves its generative AI workloads, its in-house silicon probably won’t replace AMD or Nvidia’s GPUs any time soon. The two chip designers' GPUs are currently much better suited to LLM inference and are almost certainly what Meta Superintelligence Labs is using to train frontier models like Muse Spark. Application-specific hardware has always been better suited to well-understood workloads, and Meta’s accelerators are no exception. But enough about market positioning and workloads. Let’s dig into the chip itself. Dissecting the MTIA 400 At first blush, the MTIA looks a lot like either Nvidia’s Rubin GPUs or AMD’s MI355X accelerators. The MTIA 400 is based on a heterogeneous multi-die architecture, which means it is built from a collection of different chips each with their own job. It features two compute dies, two I/O dies, and an SoC die that handles host connectivity and workload orchestration — all pretty standard stuff. Broadcom’s influence here is hard to miss. The chip almost certainly was built using the IP house’s XPU tech, which we explored in detail last year. Building a competent AI accelerator isn’t a trivial endeavor. So, if you’re trying to bring one to market quickly, it makes sense to focus on what makes your chip special and let someone like Broadcom handle the rest. Digging deeper into the MTIA 400’s heart, the compute chiplets are built on a 3nm process tech, presumably from TSMC, and feature a 6x8 grid of processing elements responsible for the bulk of the accelerator’s AI performance. Combined, the two chiplets are capable of outputting 12 petaFLOPS of MXFP4 compute at 1.7 GHz. To put that in perspective, the MTIA 400 is about 20 percent faster than Nvidia’s top-specced Blackwell accelerators at the higher precisions more commonly used to train models, while sucking back roughly the same power. But the parts don’t hold up nearly as well when compared to Nvidia and AMD’s latest chips – the MTIA 400 is between 3x and 3.3x slower than Rubin and the Instinct MI455X, respectively. Meta’s latest accelerator is fed by eight 36 GB HBM3e stacks that deliver 288 GB of memory and about 9.2 TB/s of bandwidth. Again, that’s about 15 percent faster than Nvidia and AMD’s last-gen parts, but with less than half the bandwidth of their new GPUs. This is probably why Meta is positioning the part as a training chip. It has much better and faster options for LLM inference. The MTIA 400 features a pair of I/O chiplets that provide 1.2 TB/s of chip-to-chip bandwidth over RDMA. We don’t know what transport tech Meta is using at this point, but given Broadcom’s involvement, Ethernet would be the obvious choice. Scaling up MTIA 400’s physical resemblance to Nvidia Rubin and AMD MI350-series parts also extends to the system design. Each compute blade features four MTIA 400 accelerators connected via a PCIe switch to an x86 CPU and a scale-out NIC. A single rack is equipped with 18 compute blades and eight switch blades for a total of 72 accelerators in a single unified domain. Meta shared a pic of the rack earlier this year, so while the compute blade is highly reminiscent of AMD's Helios blades, it's not using the same double wide OCP racks. Meta hasn't shown how large a cluster its chips can support just yet — the slides simply state "multi-thousand accelerator scaling" — but even at rack scale the systems should be quite competent relative to Nvidia’s GB200 and GB300 rack systems for training. More on the way While the MTIA 400 is aimed primarily at LLM training, Meta is already working on an inference-optimized version of the chip. Disclosed back in March, the MTIA 450 will double the chip’s memory bandwidth, presumably by swapping HBM3e for much faster HBM4. That part is expected to enter production next year, and should be extremely well suited to running those LLM-based recommender models Zuck and crew have been talking about the past few quarters. Meanwhile, the MTIA 500, which is also slated for 2027 release (probably in H2 if we had to guess), will increase memory bandwidth by another 50 percent — likely using 4 additional HBM4 stacks — and double the number of compute chiplets. ®

The economist John Maynard Keynes once remarked, “In the long run we are all dead.” From an astronomical standpoint, this is true for the Earth itself. In roughly 5 billion years, the Sun will exhaust the hydrogen at its core and enter the next phase of its life as a Red Giant, where it will swell up large enough to engulf and destroy the Earth.

Currently, the Sun is still in its main-sequence phase, where it fuses hydrogen into helium in its core. Although it remains relatively stable during this phase, it will gradually grow brighter with time. As the Sun shines brighter, the increased heat and radiation will raise Earth’s surface temperature, potentially pushing it past the threshold for life to survive. 

One of Earth’s geological processes that helps regulate its surface temperature is the carbon cycle. Over millions of years, carbon moves from the Earth’s interior into the atmosphere as carbon dioxide or CO₂, where it functions as a greenhouse gas that traps heat. Plants take in this CO₂ through photosynthesis and move carbon through the biosphere. 

Meanwhile, rain and rock weathering draw carbon into the oceans and rocks. By pulling CO₂ out of the atmosphere, the carbon cycle could mitigate Earth’s warming as the Sun brightens over time. Today, human activities are increasing atmospheric CO₂ and warming the planet. However, over billions of years, the carbon cycle could gradually remove enough CO₂ from the atmosphere that eventually too little would remain to support photosynthesis. 

With these challenges to life on Earth in mind, the researchers Jacob Haqq‐Misra and Eric Wolf investigated how long Earth’s biosphere could last in the distant future. The team used computer models to calculate the balance of energy that Earth would receive and emit over time, accounting for the effects of increasing sunlight and changing atmospheric CO₂ concentrations. They followed this up with calculations using the ExoCAM model to determine the temperature and atmospheric conditions of future Earth across different latitudes. They related these conditions to a habitability metric to determine their impact on the biosphere.

To estimate the biosphere’s maximum possible lifespan, the team conducted 29 simulations representing the planet’s conditions over the next 2 billion years. In one set of simulations, the Earth’s carbon cycle contributed little to regulating surface temperature, with the atmosphere’s CO₂ concentration fixed at approximately modern levels of 400 parts per million (ppm) and surface temperatures continuing to increase. In another, the carbon cycle worked to keep Earth’s average surface temperature fixed at the modern average of approximately 59℉ (15℃), no matter how much it had to reduce CO₂ in the atmosphere. 

In the scenario with a fixed CO₂ concentration, the team found that in 1.7 billion years, Earth’s surface reached an average of 122°F (50°C), which is too hot for most land plants. After 1.9 billion years, it reached 149°F (65°), becoming too hot for all land plants to survive as temperatures continued to rise. 

They also found that in the fixed average surface temperature scenario, atmospheric CO₂ dropped below the 150 ppm level that most plants require for photosynthesis within 500 million years. After 1.4 billion years, CO₂ levels fell below 10 ppm, leaving only a few land photosynthesizers and aquatic plants alive. After about 1.8 billion years, atmospheric CO₂ dropped to 1 ppm, leaving only microbes capable of photosynthesis.

The researchers’ findings generally extend the estimated lifetime of Earth’s future biosphere compared with earlier, simpler models that suggested maximum lifespans of 100 million to 1.5 billion years. However, they clarified that other scientists should verify their results by running tests with other models, especially under extreme conditions of high incoming solar radiation and low atmospheric CO₂. 

Lastly, Haqq‐Misra and Wolf noted that their model did not account for major evolutionary changes that could help life adapt to future conditions. Over time, plants could evolve to tolerate higher temperatures or lower atmospheric CO₂ levels. Future humans or another intelligent species might also engage in large-scale geoengineering projects to prevent the end of life on Earth. If life on Earth proves adaptable enough, the biosphere may endure until a Red Giant fills the daytime sky.

The post The long-term fate of Earth’s biosphere appeared first on Sciworthy.

Scientists have discovered China’s first known freshwater cave snail, an eyeless species so rare that only three living adults have been found. Deep inside a mountain cave in southwestern China, scientists found an animal that appears built for a world where sunlight never reaches. The tiny freshwater snail has no eyes, almost no color, and [...]