Latest news with #3nm
Android Authority
24-06-2025
- Android Authority
Next year's Tensor G6 could be an even larger leap forward than the Pixel 10's Tensor G5
Luka Mlinar / Android Authority TL;DR Google's expected to shift to TSMC for production of the Tensor G5 chip in the Pixel 10. While the Tensor G5 will be fabricated on a 3nm process, industry rumors claim the G6 will already move down to 2nm. With the Tensor G6 in next year's Pixel 11, Google could potentially offer one of the first 2nm flagships around. Google's next smartphones will be here before you know it, and the latest rumors have pointed to a late-August debut for the Pixel 10 series. If you've been paying half as much attention to this upcoming hardware as we have, you're already excited about the new generation of Tensor chips set to power the phones. While Google's been doing its own chips for years now, the Tensor G5 that Pixel 10 phones will run represents a significant step forward for the series, moving to both a new manufacturing partner and a new 3nm fabrication process. But today we're setting our sights even further out, as we get a new report on what to expect from the Tensor G6. With the Tensor G5, Google is moving from Samsung to industry-leader TSMC to actually make its chips, and jump from the 4nm fabrication level down to 3nm. Right now, though, we're already solidly in the era of the 3nm chip, with examples like Qualcomm's Snapdragon 8 Elite already powering this year's most popular handsets. Even Samsung's in on the 3nm action, just announcing its own Exynos 2500 earlier this week. Next year, though, Google might not be playing catch-up. Taiwan's Commercial Times reports that with the Tensor G6, TSMC will already be moving forward to 2nm fabrication (via 9to5Google). And while we haven't yet seen how the rest of the industry will be evolving, it's fairly likely that most other players will still be on 3nm next year. Of course, 'first' doesn't always mean 'best,' and a race to adopt new technologies can frequently involve some early bumps along the road. That said, TSMC has a solid reputation, and hearing this report has us cautiously optimistic that we could be getting to a point where we find ourselves talking about Google and its Tensor chips leading SoC advancement, rather than chasing Snapdragon's tail. Got a tip? Talk to us! Email our staff at Email our staff at news@ . You can stay anonymous or get credit for the info, it's your choice.
GSM Arena
23-06-2025
- GSM Arena
Exynos 2500 unveiled: 3nm GAA, Cortex-X925, bigger GPU and satellite messaging support
The Exynos 2500 was not ready in time for the Galaxy S25 series, but Samsung has put the finishing touches on it and made it official – just in time for the Galaxy Z Flip7, which will debut on July 9. The Exynos 2500 is fabbed on Samsung's latest 3nm Gate All Around (GAA) node for improved power efficiency and uses new packaging – Fan-Out Wafer-Level Packaging or FOWLP. This allows for a reduced chip thickness and improved heat dissipation. The CPU is a 10-core (deca-core) design with what Samsung calls a 1+7+2 arrangement. In reality, it is 1+2+5+2 as the mid-cores are split into two to reduce power consumption. Samsung is claiming a 15% higher 'big-core' performance. The CPU features one Cortex-X925 core running at 3.3GHz, two fast Cortex-A725 at 2.74GHz and five more A725 at 2.36GHz, plus a pair of Cortex-A520 cores at 1.8GHz. You can see a detailed breakdown in the table below, plus a comparison with the old Exynos 2400 and the Dimensity 9400 (non-plus). Exynos 2400 Exynos 2500 Dimensity 9400 Node Samsung 4nm 4LPP+ Samsung 3nm GAA TSMC 3nm N3E CPU Prime 1x Cortex-X4 @ 3.2GHz 1x Cortex-X925 @ 3.3GHz 1x Cortex-X925 @ 3.62GHz CPU Big 1 2x Cortex-A720 @ 2.9GHz 2x Cortex-A720 @ 2.74GHz 3x Cortex-X4 @ 3.3GHz CPU Big 2 5x Cortex-A720 @ 2.6GHz 5x Cortex-A720 @ 2.36GHz 4x Cortex-A620 @ 2.4GHz CPU Small 2x Cortex-A520 @ 1.95GHz 2x Cortex-A520 @ 1.8GHz - GPU Xclipse 940 (RDNA 3) Xclipse 950 (RDNA 3) Immortalis-G925 GPU spec 6 WGP, 4 RB 8 WGP, 8 RB MC12 NPU 17K MAC 24K MAC, 59 TOPS MT NPU 890 Storage UFS 4.0 UFS 4.0 UFS 4.0 5G Rel. 179.6Gbps FR1 12.1Gbps FR2 Rel. 179.6Gbps FR112.1Gbps FR2 NTN Rel. 17 Connectivity Wi-Fi 6E, Bluetooth 5.3 Wi-Fi 7, Bluetooth 5.4 Wi-Fi 7, Bluetooth 6 Camera 320MP, 8K @ 30fps 320MP, 8K @ 30fps 320MP, 8K @ 30fps The GPU is an Xclipse 950, the fourth generation of AMD-based graphics. These are based on RDNA 3 (like on the previous chipset) and support hardware-accelerated ray tracing. While it's based on the same architecture, the Xclipse 950 is a bigger GPU with 8 work group processors (WGP) and 8 render backends (RB, these contain the ROPs). For comparison, the Exynos 2400 GPU had 6 work group processors and 4 render backends. This should deliver 28% higher FPS in ray tracing-enabled games. The latest chipset also brings a significant speedup to the NPU – AI tasks can leverage up to 59 TOPS, 39% more than the Exynos 2400 had to offer. This should enable more features to work on device. In terms of hardware, Samsung built a design with two 12K MAC clusters (for a total of 24K MAC), up from the 17K MAC of the old chip. Samsung beefed up the camera hardware as well – the Exynos 2500 supports AI-based image processing for sensors up to 320MP. Photos can be shot with multi-layer noise reduction and Dynamic Range Compression (DRC). The video block can record at 8K resolution at up to 30 fps. Video playback goes up to 8K 60fps and features AV1 decoding. A key upgrade to the built-in Exynos modem is support for non-terrestrial network (NTN) capabilities – this opens the door for satellite-based messaging. The 5G modem supports the latest 3GPP Rel. 17 and offers up to 9.6Gbps speeds for FR1 and up to 12.1Gbps for FR2. Local connectivity includes Wi-Fi 7 and Bluetooth 5.4. Samsung also says that it has improved the analog GNSS interface (for global positioning). The official Exynos 2500 page reads 'Product status: mass production'. Samsung should have enough units for the new Galaxy Z Flip7. The cheaper Z Flip7 FE might use the older Exynos 2400. Source
CNBC
06-06-2025
- Business
- CNBC
Here's what Wall Street has to say about Broadcom's second-quarter results
Analysts from several major Wall Street banks walked away from Broadcom 's latest quarterly report with more conviction on the stock. The chipmaker reported fiscal second-quarter revenue and profits on Thursday that exceeded analysts' estimates, and also gave strong guidance for the current quarter. Broadcom's adjusted earnings per share of $1.58 on revenue of $15 billion beat the expected $1.56 per share and $14.99 billion from analysts polled by LSEG. Broadcom saw $4.4 billion in AI revenue during the quarter and said it expects $5.1 billion in AI chip sales in its fiscal third quarter. Shares dipped 2% in premarket trading as the company gave a weaker forecast for its non-AI semiconductor revenue, where analysts were hoping to see signs of a cyclical recovery. The stock has soared 45% in the past three months. Analysts held to a long-term bullish outlook on Broadcom and praised what they called strong growth visibility heading into its next fiscal year, confident in management reaffirming its hyperscale customers and highlighting a well-positioned networking business amid continued AI demand. Bank of America, for example, said to "ignore quarterly noise" and focus on Broadcom's strong AI growth targets. Take a look at what some had to say: JPMorgan: overweight, price target $325 from $250 Analyst Harlan Sur said Broadcom remains his top pick in semiconductors. His bullish price target indicates roughly 25% upside ahead for the stock over the next year. "We are encouraged by team's strong line of sight to FY26 AI revenue profile driven by strong cloud/hyperscaler capex spending trends with continued focus on AI training combined with accelerating AI inference workloads, continued ramp of Google's next-gen TPU v6/v7 3nm AI accelerator ASICs, ramp up of Meta 3nm, and continued strong adoption of ethernet networking," he wrote in a note to clients. Overall, the team continues to drive a solid revenue growth profile even in a period of macro volatility given its portfolio breadth/diversification/product cycles." UBS: buy, lifts price target to $290 Analyst Timothy Arcuri bumped up his price target on Broadcom and called it "a clear AI winner." But the stock could consolidate some of its recent gains in the near term given already high investor expectations, he said. "Beyond the near-term, AVGO is likely to win on multiple fronts as hyperscaler customers look to create large heterogeneous compute clusters using either custom ASIC (the vast majority of which we think will be made by AVGO) or AMD alongside NVDA GPUs," Arcuri said in a Friday note. "AVGO's opportunity to link this all together using Ethernet should also scale alongside these solutions and AVGO also benefits from scaling out from cluster to cluster given its existing dominance in this market." Wells Fargo: equal weight, price target to $255 Analyst Aaron Rakers' price target implies Broadcom shares are expensive, and could fall about 2% from Thursday's close. Rakers highlighted Broadcom's "sustained AI momentum" into fiscal year 2026. "We continue to see shares representing a balanced risk/reward at current levels with significant leverage and an expectation that future acquisitions will remain a use of capital keeping us on the sidelines," Rakers said in a Thursday note. Deutsche Bank: buy, price target to $270 from $205 Analyst Ross Seymore expects the lack of recovery in Broadcom's non-AI business to remain a headwind heading into the fiscal fourth quarter, but expects greater focus on the growth trajectory of the company's AI business in the future. "Overall, the combination of secular growth (AI), continued software integration and execution (Vsphere contract conversions), and a potential cyclical rebound in the non-AI semis components should set the company up well for both the short- and longterm," Seymore said in a note. "Consequently, with our increased confidence in AVGO's ability maintain its AI growth in FY26 leading to a nearly +10% rise in our CY26 EPS ests, we maintain our buy rating." Bank of America: buy, price target to $300 from $240 Analyst Vivek Arya's new price target suggests the stock could gain another 15% in the coming year. A risk is that Broadcom is trading at the upper end of its historical range and at a premium to market leader Nvidia, he said. "While some might be disappointed AVGO did not update its FY27 $60-$90bn AI TAM, we believe it's only a matter of time especially as FY27 sell-side AI revenue cons. ests. is still well below $45bn, so plenty of headroom for upgrades," Arya wrote in a note. "Reit. Buy, a top computing and top-5 sector pick on unique mix of capital appreciation and consistent dividend growth with a blend of recurring (50% software), secular (AI) and cyclical (non-AI) profiles."
GSM Arena
30-05-2025
- Business
- GSM Arena
Huawei tipped to make 3nm chips in 2026
Huawei is developing a true 5 nm production line without the need for EUV lithography machines. Furthermore, 3 nm chips are already in the R&D stage, with plans to start production by 2026, according to a report by UDN. Huawei cannot use standard EUV technology, as it is patented by the Dutch company ASML, which is banned from working with the Chinese manufacturer. Instead, the company is using SSA800 lithography machines with multi-patterning, made by Shanghai Micro Electronics (SMEE). Research and development for 3nm chips are following two approaches: one is GAA architecture (the standard one, used by TSMC and Samsung), and the other is a carbon nanotube-based chip that has already completed lab validation and is now being adapted for SMIC's production lines. Earlier this month, the Huawei Matebook Fold arrived, powered by Kirin X90. While the company calls it a "5nm chip," it is actually a 7nm design with advanced packaging technology. It brings performance on par with other 5 nm chips, but the yield is only 50%, which is extremely low, making the chip more expensive to manufacture. Source
South China Morning Post
20-05-2025
- Business
- South China Morning Post
Tech war: Xiaomi's new 3-nanometre chip takes on rival products from Apple, Qualcomm
In a Tuesday post on Chinese microblogging site Weibo , Lei confirmed local media reports that the XRing O1 was based on the 3-nanometre lithography process in semiconductor manufacturing. He also said the new integrated circuit (IC) would be installed in the company's new 15S Pro smartphone and the Pad 7 Ultra tablet. Xiaomi did not provide more information about the XRing O1, especially which contract semiconductor manufacturer was producing the locally designed mobile SoC. Semiconductor foundries in mainland China are not able to mass produce 3-nm chips owing to US tech restrictions Still, Xiaomi has become the fourth tech company in the world to design a 3-nm mobile SoC for mass production after Apple Qualcomm and MediaTek The Beijing-based smartphone giant did not immediately respond to a request for comment on Tuesday.
