Internet of Things - Company and Research Focus

The company’s 18A node seems to be making early strides. TechPowerUp, citing Intel’s engineering manager Pankaj Marria, reports that initial 18A wafers are already rolling out from the Arizona plant. The progress, according to ijiwei and Commercial Times, implies that Intel’s 18A mass production is likely to be pulled forward to mid-year 2025. Originally planned to begin 18A tape-outs in early 2025, with HVM production following in the H2 2025.  Marria’s now-unavailable LinkedIn post, cited by TechPowerUp, revealed that customers are evaluating Team Blue’s initial 18A wafer lots. He wrote, “The Eagle has landed, and this is just the beginning,” hinting at promising early results. The TechPowerUp report explains that the current progress means Intel’s 18A node PDK (Process design kit) has reached version 1.0, with customers already testing custom ICs. 

NVIDIA and Broadcom are conducting manufacturing tests with Intel’s 18A, as well AMD is also assessing Intel’s 18A process.

In addition to securing key external clients, the success of Intel’s Panther Lake, its 1st major in-house product with 18A, also crucial, as Intel plans to use its own cutting-edge facilities, like Fab 52 and Fab 62 in Arizona, to produce the compute chiplet with 18A technology.

- TSMC would take a stake of no more than 50% in a JV operating Intel Foundry 
- Talks are at an early stage-sources
- Qualcomm has also received TMSC's pitch-sources
TSMC has pitched U.S. chip designers Nvidia, AMD and Broadcom about taking stakes in a JV (joint venture) that would operate Intel's factories, according to 4 sources familiar with the matter. Under the proposal, TSMC would run the operations of Intel's foundry division, but it would not own more than 50%. Qualcomm has also been pitched by TSMC.  Nvidia and Broadcom are running mfg tests with Intel, using the company's most advanced production techniques, known as 18A. AMD is also evaluating whether Intel's 18A manufacturing process is suitable for it. But 18A has been an area of contention in negotiations between Intel and TSMC. During talks in February, Intel executives told TSMC that its advanced 18A manufacturing technology was superior to TSMC's 2-nm process.  Intel's foundry division, was a crucial part of former CEO Pat Gelsinger's effort to save Intel. Gelsinger was forced out by the BoD in December, which named two interim co-CEOs who have mothballed its forthcoming AI chip. Any deals between historical rivals TSMC and Intel would face major challenges and be costly and laborious. The two companies currently use vastly different processes, chemicals, and chipmaking tool setups at their fabs -- The talks, which are at an early stage, come after U.S. President Donald Trump's administration requested TSMC,  assist in turning around the troubled U.S. industrial icon, the sources said on condition of anonymity because the talks are not public.  Intel board members have backed a deal and held negotiations with TSMC, while some Intel executives are firmly opposed.

Intel's foundry business never seems far from the headlines these days, not least because former Intel CEO Craig Barrett trashed ex-Intel directors for suggesting the idea of splitting the Intel Foundry from the rest of Intel entirely in the wake of well-publicised financial woes. Speaking at a technology conference yesterday, John Pitzer, Intel's VP of Corporate Planning and Investor Relations, revealed that, despite previous plans to the contrary, 30% of its wafers are still outsourced to TSMC. "I think a year ago we were talking about trying to get that to zero as quickly as possible, but that is no longer the strategy,". "We think it is always good to have at least some of our wafers with TSMC. They are a great supplier. This creates healthy competition between them and Intel Foundry."  "[We are] not quite sure what the right sort of level is," said Pitzer, in reference to the 30% figure. "Is it 20%? Is it 15%? We are working through that. "But we will use, I think, external foundry suppliers longer under this new strategy."

Given Intel's current limbo-like state, and with interim CEOs Dave Zinsner and Michelle Johnston Holthaus at the temporary reins, it's perhaps no surprise that this dependency has yet to meaningfully change.  Upcoming Panther Lake mobile CPUs are planned to be built on Intel Foundry's 18A process node, the current Arrow Lake and Lunar Lake CPU series primarily use TSMC-produced silicon, with final assembly and packaging handled at Intel's US-based facilities. While Intel is still reliant on TSMC's advanced silicon to manufacture its products, making use of TSMCs silicon will likely affect the margins more than if it were made in-house. Near enough a year ago, ex-CEO Pat Gelsinger said that he hoped Intel would reduce its dependency on TSMC wafers from 30% to 20%, although much has occurred since then—not least Gelsinger's departure. As the world awaits the eventual outcome of Intel's current troubles, it remains to be seen if the company will meaningfully decrease its use of TSMC wafers in the near future, or whether the two will perhaps become tied ever closer. Rumours have suggested 

TSMC Chairman C.C. Wei recently held a press conference with US President Donald Trump at the White House, announcing an additional investment of US$100 billion in the US. Shortly after, he convened another press conference with Taiwan's President Ching-te Lai to clarify TSMC's intentions regarding its investments in the US.

More Qulaity hits at NVIDIA, they're having their own Strategic Inflection Point and they don't even know it.

Backed by state subsidies and a self-sufficient supply chain, China is ramping up SiC substrate and mature process chip production at an unprecedented pace. The expansion is fueling fears of a global oversupply and intensifying price competition for industry leaders like Rohm Semiconductor and Wolfspeed.

Infineon Technologies said it will be launching a new family of automotive microcontrollers (MCUs) based on RISC-V, extending the company’s existing AURIX portfolio. While the date of availability is yet to be announced, Infineon said it will make a virtual prototype starter kit at embedded world 2025 in Nuremberg, Germany, next week, for Infineon’s partners to begin pre-silicon software development.  In its announcement, Infineon said it is committed to “making RISC-V the open standard for automotive,” and in this respect highlighted it is aiming to develop an ecosystem to enable this to happen. To facilitate the adoption of its future product family, the company said it is working closely with its software and tool partners. To start with, the virtual prototype starter kit, to be presented in Germany at embedded world 2025, is based on the tool suite of its strategic partner Synopsys. This will help Infineon’s partners to start developing their software and tool products for Infineon’s RISC-V architecture, before the MCU hardware is available. In addition to Synopsys, several partners like IAR, Elektrobit, Green Hills Software, HighTec, Lauterbach, PLS and Tasking have already started to use the software development kit and will also showcase their first solutions next week, with more partners to follow during the course of 2025. Based on their solutions, the virtual prototype will evolve into a full-fledged digital twin of Infineon’s future MCU family. The new products will become part of Infineon’s AURIX brand and extend its existing automotive MCU portfolio based on TriCore (AURIX TC family) and Arm (TRAVEO family, PSOC family). The RISC-V MCUs will cover entry-level MCUs up to high-performance MCUs.

Investors are starting to examine what they've long ignored: Tesla's bedrock value minus the "Musk Premium".In the weeks bracketing Donald Trump’s victory on Nov. 5, Tesla’s stock enjoyed one of the most explosive rides in the annals of publicly traded equities, gaining over 50% and almost half a trillion dollars in valuation in the span of just over a month. That surge reversed four years of poor performance for Tesla’s shares as investors soured on the EV maker’s weakening fundamentals and CEO Elon Musk’s serial promises of fully self-driving cars and an inexpensive mass-market model that proved an ever-receding horizon.

The sad story of the Exynos 2500 chip is well known. Samsung designed it to power the Galaxy S25 series, but low yield rates at its 3nm factories will prevent its implementation. Instead, Samsung had to resort to the expensive Snapdragon 8 Elite for all models globally, losing significant profit margins. Samsung wants to prevent a similar situation with the 2nm Exynos 2600 SoC and is setting up a dedicated team to get it ready for the Galaxy S26 lineup.As reported by The Financial News, there’s a new “task force” (TF) at Samsung Foundry that will look to ensure the Exynos 2600 chip is ready for the Galaxy S26 series. If you’re not aware, Samsung’s Semiconductor division has two main branches: Samsung System LSI and Samsung Foundry. System LSI is responsible for designing the chips, while Samsung Foundry produces them. The failure of the Exynos 2500 chip was mainly due to low yield issues in Samsung Foundry’s 3nm wafers. So, the South Korean giant is directly attacking the situation with a new dedicated team in the Foundry’s division. The task force group has two main goals: optimizing the yield rate of the 2nm wafers and optimizing the performance of the Exynos 2600 chip itself. It’s not that Samsung is having problems with its 2nm process. In fact, initial tests in February showed positive results with a promising 30% yield rate. While the figure may seem low, these are initial tests. From there, the figure can only improve. Semiconductor factories typically aim for yield rates of 60–70% to make chip production economically viable. In comparison, Samsung’s 3nm wafers had a yield rate as low as 10% at its worst. According to reports, the firm already managed to optimize it, reaching at least 50%.  The Financial News report also mentions that Samsung is pleased with the progress on its 2nm wafers. “The yield rate of the Exynos 2600, which uses the SF2 process, which is a 2nm process, is significantly higher than that of the previous generation, the Exynos 2500,” an official told.  If all goes well, Samsung will begin mass production of the Exynos 2600 in May. Much of the financial future of Samsung’s Semiconductor division could depend on this chip. It could help the company save millions in costs by being able to implement its own chips again in its devices. It could also attract the attention of third parties to contract Samsung Foundry’s services again. In recent years, the division has lost major clients such as NVIDIA and Qualcomm, among others.

TSMC's chief executive and chairman, C.C. Wei, said at a press conference in Taipei on Thursday that the company is expanding its production capacity in the U.S. due to customer demand and extraordinary booking through late 2027. According to Reuters, Wei stressed that the expansion in the U.S. would not affect the foundry's expansion plans in Taiwan. TSMC's American customers — such as Apple, AMD, Broadcom, Nvidia, and Qualcomm — asked TSMC to boost U.S.-based capacity despite 25% - 30% higher quotes on chips produced in America, according to C.C. Wei. He did not reveal whether demand for U.S. capacity from some of America's largest fabless semiconductor companies picked up before or after President Donald Trump threatened to impose 25% – 100% tariffs on chips made in Taiwan. While it is generally unlikely that a U.S. administration would make chips 2X more expensive for customers in the US overnight, political uncertainty certainly affected TSMC's decision making.  -- The company announced this week that it intends to build five additional Fab 21 modules, two advanced packaging facilities, and an R&D center in the U.S. for around $165 billion over an unspecified period of time. In addition, the company will continue to build its second JASM fab module in Japan in partnership with Sony and Toyota and its first ESMC fab aimed at customers in the automotive industry near Dresden, Germany.

CEO C.C. Wei emphasized that TSMC's $165 billion U.S. investment plan will not reduce its focus on Taiwan's semiconductor sector. The company remains committed to increasing domestic production, acknowledging that current capacity is still insufficient.

Indeed, Taiwan will remain TSMC's home turf, though, as the company preps to launch 11 new production lines on the island this year alone. This probably includes the company's Fab 20 near Baoshan in Hsinchu County, which will be the company's primary fab for production chips on N2 and A16 fabrication processes (2nm-class, 1.6nm-class) before its second N2-capable fab near Kaohsiung in Southern Taiwan Science Park kicks off in 2026. In addition, the company is expanding production on its advanced (7nm and below), mature, and specialty nodes in Taiwan as demand for all of its process technologies remains at record levels.

The Department of Veterans Affairs (VA) is planning a reorganization that includes cutting over 80,000 jobs from the agency that provides health care for retired military members, according to an internal memo obtained by The Associated Press. The VA’s chief of staff, Christopher Syrek, told top-level officials at the agency that it had an objective to cut enough employees to return to 2019 staffing levels of just under 400,000. That would require terminating 10s of 1000s of employees after the VA expanded during the Biden administration, as well as to cover veterans impacted by burn pits under the 2022 PACT Act. The memo instructs top-level staff to prepare for an agency-wide reorganization in August to “resize and tailor the workforce to the mission and revised structure.”  Veterans have already been speaking out against the cuts at the VA that so far had included a few 1000 employees and 100s of contracts. More than 25% of the VA’s workforce is comprised of veterans.The plans underway at the VA showed how the Trump administration’s DOGE initiative, led by billionaire Elon Musk, is not holding back on an all-out effort to slash federal agencies, even for those that have traditionally enjoyed bipartisan support.  Blumenthal, top Democrat on the Senate committee that oversees veteran’s affairs, said the Trump. “has launched an all-out assault” against progress the VA has made in expanding its services as the number of covered veterans grows and includes those impacted by toxic burn pits. “Their plan prioritizes private sector profits over veterans’ care, balancing the budget on the backs of those who served. It’s a shameful betrayal, and veterans will pay the price for their unforgivable corruption, incompetence, and immorality,” Blumenthal said. “Democrats are here to say in unison we will not allow our veterans to be defined as government waste,” said Rep. Katherine Clark, the No. 2 ranking House Democrat.

The Snapdragon X85 is Qualcomm’s newest entrant of the company’s top-end 5G modems, and it leverages AI to not just boost signal strength but also to reach unparalleled downlink speeds of 12.5Gbps. The latest release certainly delivers a large performance gap when compared with Apple’s custom C1 5G modem, which is why Qualcomm’s CEO, Christiano Amon, does not feel threatened by its rival’s efforts. New Snapdragon X85 also claimed to be the 1st 5G modem integrated with heaps of AI, with Qualcomm’s CEO believing that even premium Android devices do not stand a chance. 

The firm had estimated to stop supplying Apple with 5G modems in 2027, but Qualcomm likely has a roadmap planned for how it aims to offset those losses. Bear in mind that there is still a massive client list on the Android side that continues to rely on these modems. Even if Christiano Amon appears confident in the Snapdragon X85’s capabilities, Apple is already said to be testing the C2, which may bridge that performance gap with Qualcomm’s future high-end offerings.

The Army and Navy may soon use alternative methods for building barracks, including 3D printing, in an effort to reduce the financial burden.  Top engineers for the respective services met with lawmakers on Capitol Hill to discuss the potential use of 3D printing and Mass Timber Construction as a means of Maximizing cost Efficiency. The group also discussed the use of high-performance cement, geosynthetics, composite materials, tension fabric structures, and carbon fiber-reinforced polymers. The Navy had already employed the use of mass timber, Hamilton said, to build a child development center in Hampton Roads, Virginia. The practice of mass timber construction involves sourcing structural beams that come from smaller lumber. The 1st Army barracks made using mass timber structural elements was recently completed, with another potential mass timber project on the horizon at Mountain Home Air Force Base. Tyndall Air Force Base, Florida, and Fort Bliss, Texas, have already both used 3D printing, he pointed out, the latter specifically for projects involving barracks. 

Nvidia CEO Jensen Huang recently announced the company's fourth-quarter fiscal 2025 (4QFY25) financial results, reinforcing investor confidence in AI's long-term growth, even amid industry shifts sparked by DeepSeek's emergence.

Taiwan has historically attained significant self-sufficiency in panel production equipment and materials; however, its self-sufficiency in semiconductor equipment and materials is somewhat limited. As a result, when US President Donald Trump claimed Taiwan had stolen the US chip manufacturing industry and threatened to put a 100% tariff on Taiwanese chips, Taiwanese companies had no bargaining power with the US government because critical equipment remains under US control.

Bolt Graphics, a GPU startup from Sunnyvale, California, this week introduced its Zeus GPU platform that is designed for gaming, rendering, and supercomputer simulations. The company says that its Zeus GPU not only supports features like upgradeable memory and built-in Ethernet interfaces, but it can also beat Nvidia's GeForce RTX 5090 by around 10 times in path tracing workloads, according to slide published by ServeTheHome. There is one major catch: Zeus can only beat the RTX 5090 GPU in path tracing and FP64 compute workloads because it does not support traditional rendering techniques. This means it has little of no chance to become one of the best graphics cards. Like many processors these days, Zeus relies on a multi-chiplet design. The entry-level Zeus 1c26-032 features a single processing unit that is equipped with 32 GB of LPDDR5X memory at 273 GB/s and can be mated with up to 128 GB of DDR5 memory using two SO-DIMMs (at 80 GB/s). The Zeus GPU is also paired with an I/O chiplet that packs a QSFP-DD port for 400GbE/800GbE, two PCIe Gen5 x16 slots with CXL 3.0 on top (enabling efficient memory sharing across multiple cards), and a GbE port for BMC. The GPU chiplet connects to its I/O chiplet at 256 GB/s.

The more advanced Zeus 2c26-064/128 uses two Zeus processing units, an I/O chiplet, and supports either 64 GB or 128 GB of LPDDR5X memory. The most powerful — Zeus 4c26-256 — implementation integrates four processing units, four I/O chiplets, 256 GB LPDDR5X and up to 2 TB of DDR5 memory. The quad-chiplet Zeus implementation is not a card, but rather is a server.

Unlike high-end GPUs that prioritize bandwidth, Bolt is evidently focusing on greater memory size to handle larger datasets for rendering or simulations. Also, built-in 400GbE and 800GbE ports to enabling faster data transfer across networked GPUs indicates datacenter focus of Zeus.

RISC-V multi-chiplet GPU  -  Unlike GPUs from AMD, Intel, and Nvidia that rely on proprietary instruction set architectures, Bolt's Zeus relies on the open-source RISC-V ISA, according to slides published by ServeTheHome. The Zeus core relies on an open-source out-of-order general-purpose RVA23 scalar core mated with FP64 ALUs and the RVV 1.0 (RISC-V Vector Extension Version 1.0) that can handle 8-bit, 16-bit, 32-bit, and 64-bit data types as well as Bolt's additional proprietary extensions designed for acceleration of scientific workloads.

Bolt Graphics - As Zeus is aimed at path tracing rendering technique as well as compute workloads, it does not seem to have traditional fixed-function GPU hardware like texture units (TMUs) and raster operation units (ROPs), so it has to rely on compute shaders (or similar methods) for texture sampling and graphics outputs. This saves precious silicon real estate for compute elements. Nonetheless, each Zeus GPU has one DisplayPort 2.1a and one HDMI 2.1b output.

The big picture: Western Digital, a titan in the storage industry long renowned for its hard disk drives and solid-state drives, has officially separated its NAND flash memory business, effectively ending the company's direct involvement in SSD production and sales. Western Digital's exit from the market leaves behind a legacy of innovation and quality that has significantly impacted the PC gaming community.

This move, which had been in development for some time, was finalized last week. The SSD division has been fully spun off into SanDisk, leaving Western Digital to focus exclusively on hard disk drive technology. This separation marks a pivotal moment for the company, which has been a household name in both HDD and SSD markets, competing alongside industry giants like Samsung and Crucial. Meanwhile, Western Digital is focusing on the evolving opportunities in the HDD market. "As AI accelerates and impacts industries around the world, and as companies generate and store more data, HDD exabyte shipments are expected to increase," CEO Irving Tan said. He also points out that much of the data stored by cloud service providers, such as native cloud application data, AI data lakes, media, and machine learning data, runs on HDDs. Tan said the near-term plan is to deliver HAMR when it reaches economic crossover. "Beyond HDD, our teams are exploring new growth opportunities that leverage our core capabilities in magnetics and materials science, and with their dedication, the possibilities for future applications of our technologies are truly endless."

TSMC's recent bold move, with a $100B investment in U.S. IC manufacturing, seemingly a win for American tech, don't expect TSMC to send its best chip tech to U.S. shores. Instead, the Taiwanese giant will focus on building new plants and an R&D center, leaving its cutting-edge production processes in Taiwan. This could be a game-changer for U.S. chip ambitions but presents an opportunity for Intel that continues to struggle with delays in its own chip plants, including its much-hyped Ohio facilities, now expected to be operational in 2030 at the earliest. Intel, meanwhile, has its eyes focused on its 18A manufacturing process. Intel is in talks with Nvidia and Broadcom, both testing Intel's latest ICs, crucial for powering AI tech. But it's a waiting game. Intel has yet to seal the deal with these major players, and after some disappointing results in 2024, the stakes are high. Even if tests go well, Intel still has a long road ahead to secure its place as the go-to IC maker for top-tier customers, especially as its AI chip timelines slip further into the future. It's clear: the semiconductor battle between Intel and TSMC is heating up, with TSMC doubling down on U.S. production, it's taking its competition with Intel to another level. Intel is not out of the game yet if it can overcome its delays and secure the high-profile contracts, it could still turn things around. Investors should keep a close eye on how these chips fall regarding Intel's ability to adapt will determine whether it can hold its own or get left behind.

Intel has new network adapters, at least in some respects, with the Intel E830 200GbE and E610 10GbE NICs. These two new series are expected to roll-out over 2025 and be Intel’s new PCIe Gen4 NICs moving forward. This is a launch where it really paid to read into the details versus the headline features. Intel says that its new Intel Ethernet E830 controllers have precision timing and security features while offering up to 200GbE of throughput, up from 100GbE currently. The Intel Ethernet E610 series is a 10Gbase-T series that supports multi-gigabit speeds and up to 50% lower power. That 50% lower is the TDP difference between the Intel E610-XAT2 and the Intel X550-AT2 dual port 10Gbase-T adapters. There are also single and quad-port options like the E610-XT4 which might actually be a dual E610-XAT2 under the hood since the spec table says it supports bifurcation and a PCIe Gen4 x8 link split into x4 and x4. Launching a 200GbE adapter in 2025  is quite interesting. In many servers, instead of onboard networking, there are OCP NIC 3.0 slots. So instead of a server having dual 1GbE, dual 10Gbase-T or other options onboard, these lower-performance NICs are moving to an OCP NIC 3.0 slot. That seems to be the main push of the Xeon E610 and E830 right now. And not going after the high-end of the market.  It is not an understatement to say that after Intel lost its bid for Mellanox (Jensen and NVIDIA beat Bob Swan and Intel for the company) Intel’s networking has never really recovered. NVIDIA has been shipping multiple 400GbE NICs while Intel is launching a 200GbE NIC (we covered Mellanox ConnectX-6 Dx at 200Gbps in 2019.)

Wearables have been trying to thread a particularly tricky needle as of late. Though funky chunky hardware will always have its appeal, you ideally want kit that sits on your body to be both reasonably powerful and lightweight. Some smart glasses bulk up their frames in order to stash impressive lenses alongside computational power, while others delegate the technical heavy-lifting to a separate box hooked up via a tangle of wires. Neither are especially elegant solutions. Then there's Orion, Meta's in-development AR glasses. In a recent blog post, Meta claimed, "Orion combines the benefits of a large holographic display and personalized AI assistance in a comfortable, all-day wearable form factor." How will the frames achieve this? With a wireless 'Compute Puck' containing "Meta-designed custom silicon for AI and machine perception."  The elongated, plastic patty "offloads Orion’s processing power to run application logic"—so, it's like a mouse and a CPU in one. This keeps the glasses lightweight, while also achieving a pocketable form factor for the puck itself.  Personally, I'm not sure I buy it. I already have a very fraught relationship with my phone, in no small part due to the screaming hellmouth that the social media platforms I willingly choose to hang out on have become. Strapping a similar device more directly to my face does not fill me with excitement. Meta's splitting of that device into three parts, with little clear idea of what exactly will be the ecosystem's 'killer use case' does not fill me with confidence either.