新闻

A Manhattan Supreme Court judge will admit the 3D-printed gun and the notebook prosecutors dubbed a manifesto that were found in Luigi Mangione's backpack when he was arrested. The mixed ruling concludes the evidentiary hearing ahead of Mangione's state trial, in which he is charged with the shooting death of UnitedHealthcare CEO Brian Thompson outside the New York Midtown Hilton in Manhattan. The outcome dealt a blow to Mangione's defense, which had sought to exclude all the contents of the backpack that police recovered during his arrest. Manhattan Supreme Court Justice Gregory Carro did find that Pennsylvania officers violated Mangione's rights in searching him at the Altoona McDonald's where they arrested him, but that infraction will not affect two of the most critical pieces of evidence that police uncovered when they searched him in the station. "Items found in the backpack at the McDonald's should be suppressed," Carro said. "Items recovered at the station should not be suppressed." That means the items that Altoona police found when they searched Mangione's backpack at the McDonald's must not be considered at trial -- including the magazine, cellphone, passport, wallet and computer chip. However, the gun, red notebook and silencer, which were recovered after officers took Mangione and his belongings back to the police station, are fair game. Mangione was arrested on Dec. 9, 2024, in Altoona in the midst of a massive manhunt for the person who assassinated Thompson outside of a Midtown corporate convention five days prior. In December, prosecutors and defense attorneys scrutinized key evidence against Mangione during a multiday hearing. Carro validated Mangione's contention that the search of his belongings in the McDonald's was a violation of search and seizure law. Carro raised a legal precedent finding that officers are only allowed to search bags within the defendant's "immediate control" in order to prevent the destruction of evidence or access to a weapon. During the hearings, members of the Altoona Police Department said they were compelled to search the backpack to make sure it didn't contain any explosives, but they also said it was routine to search all bags in a defendant's possession when they are arrested. Carro did not find the justification persuasive. "The officers' actions were inconsistent with merely performing a safety search. The area where the police searched the backpack was open to the public and to employees, both of whom passed by the area on their way to the bathrooms, and in the case of employees, to gain access to storage closets," he wrote in his decision. But once the officers took Mangione back to the station, Carro concluded that the inventory search that uncovered a loaded gun and a notebook was valid. Prosecutors in Mangione's federal trial, which is scheduled to come after the state trial, have said Mangione wrote about his intent to "wack" the CEO of the insurance company at its investor conference. The state trial is set to begin starting in September.

Customer Impact: Those targets would give AI-chip buyers more wafer and advanced-packaging capacity as accelerator designs require larger packages and more HBM. TSMC's Chip-on-Wafer-on-Substrate (CoWoS) advanced-packaging yield has topped 98 percent, and the company expects to add five fabs in 2026, extending claims it laid out during its May 14 Hsinchu symposium session. Management is pairing that yield update with a broader argument that AI accelerators will need more wafer output and bigger package formats at the same time. CoWoS is TSMC's advanced-packaging technology for combining processor dies with high-bandwidth memory inside a single AI accelerator package. The approach lets chips move data faster and support larger memory loads, which is increasingly important for training and running AI models. But because these packages are larger, more expensive, and more complex than conventional chips, yield rates and capacity expansion are closely watched by customers trying to secure reliable supply. In remarks from the May 14 symposium session last week, Yuan Li-pen, TSMC vice president for business development, summed up the yield point plainly: "The 5.5-reticle-size CoWoS technology has achieved yields exceeding 98 percent." Because CoWoS links compute dies and stacked HBM memory inside one AI accelerator, the yield claim reaches beyond a single packaging line. TSMC expects to launch five new fabs in 2026 to support AI and 2-nanometer demand. It also projects a 70 percent annual 2nm growth rate through 2028. Combined with that node ramp, TSMC is asking customers to judge packaging yield and wafer output together instead of as separate milestones. Advanced-packaging growth is moving on a parallel track. TSMC projects a more than 80 percent annual rate for CoWoS and SoIC capacity through 2027. At the same symposium session, TSMC used those manufacturing targets to frame a wider AI roadmap. Executives also unveiled A13, A12, and N2U, linking the packaging push to denser, more memory-hungry systems. Larger accelerator packages only help when the associated node roadmap can move in step. Recent manufacturing milestones give the roadmap a near-term base. Third-generation CoWoS volume production started in 2026, bringing a larger interposer that can support 12 HBM chips. At that interposer size, yield matters to buyers. Bigger package sizes are only commercially useful when they can move through volume manufacturing without turning memory-heavy designs into low-yield custom builds. Wafer supply is rising with that packaging lane. The 2-nanometer ramp at five factories in Hsinchu and Kaohsiung is giving TSMC matching front-end support this year. Roadmap detail stretches further into 2028 and 2029. TSMC mapped a 14-reticle CoWoS platform for 20 HBM chips in 2028. Support for 24 HBM chips in 2029 points to accelerator layouts that keep adding memory rather than shrinking package ambition. Process timing still shapes how quickly those packages can ship at scale. Symposium material showed A16 moving to 2027, while A13 and A12 target 2029. On that timeline, node progress and package growth stay aligned instead of splitting into unrelated upgrade cycles. Current execution is already underway on the node side. N2 is in production now with more than 20 customer tape-outs received so far and more than 70 additional projects in the pipeline. Comparator figures from the symposium package show why TSMC is stressing manufacturing execution, not only a broad roadmap. Intel's EMIB-T packaging yield of about 90 percent gives customers one alternative advanced-packaging benchmark. Samsung provides the node-side contrast. Its reported mid-50-percent range in 2nm yields leaves a visible gap with TSMC's packaging update. Against that backdrop, accelerator buyers still care about package yield and node yield because both shape how quickly large AI chips can move from roadmap slides into sustained supply. Market size is part of the same pitch. TSMC projects semiconductor revenue of US$1.5 trillion by 2030. It also sees more than US$1 trillion in 2026. Expansion since 2022 helps explain why the company is making that case now. CoWoS capacity has been growing at a 90 percent CoWoS CAGR from 2022 to 2027, and TSMC had already been extending the CoWoS and SoW roadmap toward larger formats through 2029. TSMC deputy cochief operating officer Kevin Zhang used the market language more directly: "We are still at the beginning of the AI revolution, but we already see a significant impact across the whole semiconductor ecosystem." The practical checkpoint is whether TSMC can turn those yield, capacity, and node-ramp targets into shipment volume as AI accelerators grow larger and more memory-intensive.