Around the chokepoint, back from the brink

Modern accelerators pair a logic chip with stacks of high-bandwidth memory, or HBM, sitting millimeters away. Bridging them takes a dense mat of wiring, and the standard HBM4 interface runs 2,048 data signals per stack, a count only a silicon interposer can carry.[2] The dominant way to fabricate those interposers is TSMC's CoWoS packaging, and it is effectively sold out. Reported capacity ran near 35,000 wafers a month in 2024 and is being pushed toward 115,000 to 140,000 by the end of 2026, yet the largest chip buyers are described as fully booked with severe shortages of both variants.[4]

The new document, JESD330-4, defines what JEDEC calls Standard-Package HBM4, or SPHBM4.[1] It uses the same memory dies but rewires the connection. Instead of 2,048 parallel signals it carries 512, folding four internal channels onto each external one and running them roughly four times faster, at 22.4 to 46 gigatransfers per second.[2] Fewer, faster wires reach the same aggregate speed, about 2.94 terabytes per second per stack at the top rate, slightly above regular HBM4.[3]

The point of trading width for speed is what it permits. A 512-signal link tolerates bump pitches wider than 90 micrometers and channel lengths up to 20 millimeters, coarse enough to run on an ordinary organic package substrate rather than an etched silicon interposer.[3] Organic substrates are the resin-and-copper laminates that a broad base of packaging suppliers already produces. Building HBM bandwidth on them would loosen the link between memory and the single scarce CoWoS line.[3]

The maneuver does not come free, and the tradeoffs are engineering claims rather than settled results. Serialization moves complexity into the memory's base die, which must now handle lane training, equalization and error correction, making it likely costlier than a conventional HBM4 base die.[3] Those extra steps can add latency and power draw.[3] Analysts note that dropping the interposer does not by itself make a stack cheap, since it still needs dense memory dies and advanced assembly.[2]

A published standard is a coordination point, not a shipping product, and adoption rests on memory makers and chip designers choosing to build to it. Its significance is directional. For years the response to the packaging shortage has been to build more interposer capacity. SPHBM4 is the clearest sign yet that parts of the industry would rather engineer the interposer out of the critical path than keep waiting in line for it.[1]

For years, teams walking the dark sand of Boa Vista, in the Cabo Verde archipelago, would meet only five to ten nesting females on a good night.[2] Now they count dozens. A study published this year in the journal Biological Conservation, built on nightly counts kept without interruption since 1998, reports that nesting on three of the island's beaches has climbed about 80-fold since 1998.[1] Because Cabo Verde holds one of the largest loggerhead nesting populations on Earth, the trend carries weight well beyond its own shoreline.[1]

The recovery has no single cause. Monitoring by the group Cabo Verde Natura 2000, whose field crews are almost entirely local residents, paired systematic nest counts with anti-poaching patrols, hatcheries, and a growing turtle-watching economy that gave the animals more value alive than dead.[3] Persistent pressure from conservation groups helped push the government to pass a law in 2018 that criminalised the killing and capture of turtles, with prison sentences attached.[4] On some beaches the people now guarding nests are former hunters.[5]

The global loggerhead population is estimated to have fallen about 47 percent over the past three generations, so a genuine, sustained increase is the exception rather than the rule.[3] The researchers are candid about the limits of the good news. Single seasons swing hard, and on the beaches one group monitors, nests fell from 19,188 in 2024 to roughly 9,500 in 2025, a change it calls normal for the region.[6] Crowding brings its own risks, and warming sand is a longer shadow, because a loggerhead's sex is set by the temperature of its nest and hotter beaches produce ever more females.[2]

The value of the result is in what it demonstrates. Where monitoring is constant, enforcement is real, and the people living beside a rookery have a stake in its survival, a depleted turtle population can climb back over a single human working lifetime.[1] Whether Boa Vista's turtles hold that ground will depend on the same patrols, and on a climate those patrols cannot control.[2]

The Bank of Central African States, known by its French initials BEAC, joined the Pan-African Payment and Settlement System, or PAPSS, a clearing network built by Afreximbank and the African Union so that two African currencies can settle directly against each other.[1] BEAC issues the currency used across Cameroon, the Central African Republic, Chad, the Republic of Congo, Equatorial Guinea and Gabon, a bloc of more than 72 million people.[5] Its entry folds a large piece of Francophone Central Africa into a system that recent reporting places at 28 countries and more than 190 banks and fintechs.[5]

Rather than send each payment abroad, PAPSS tallies all of a day's transactions between African currencies and nets them out before midnight, and central banks then settle only the small remaining balance among themselves.[2] That removes the third currency, and with it the layer of foreign intermediaries that most African cross-border payments still pass through. By one industry estimate, more than 80% of those payments route through banks in the United States or Europe, adding 2% to 5% to each transfer.[4]

The costs land on ordinary users. Average remittance fees for sub-Saharan Africa reached 8.78% in the first quarter of 2025, well above the global average of about 6.5%, and the same survey notes several corridors within the region where sending money still costs more than 20%.[8] When PAPSS launched in 2022, Afreximbank projected the system could eventually save African businesses about 5 billion dollars a year in such costs.[3] That figure is an aspiration, not a measured result, and take-up has been gradual since launch.[2]

The structural stake is trade itself. Commerce among African countries still sits at just under 15% of the continent's total trade, well below the levels seen within Europe or Asia.[7] Part of that gap is roads and tariffs, but part of it is plumbing, because it has often been easier to price and pay in dollars than to move naira, cedis or CFA francs directly.[4] A single central bank joining a clearing system does not change that overnight. But each member that signs on widens the set of trades two Africans can settle in their own currencies, and trims the toll a merchant in Douala or Nairobi pays to a bank on another continent for the ordinary business of trading with a neighbour.[6]