At the SPIE Advanced Lithography conference here, Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) outlined more details about its 450-mm fab plans. SAN JOSE, Calif. – At the SPIE Advanced Lithography conference here, Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) outlined more details about its 450-mm fab plans.
The silicon foundry giant hopes to process 14-nm FinFET devices in full production on 450-mm wafers by 2015 or 2016, said Shang-Yi Chiang, senior vice president of R&D at TSMC.
As reported, Intel, Samsung and TSMC are pushing hard for 450-mm fabs. Intel has already announced two ”450-mm ready’’ fabs. The fab tool vendors are warming up to 450-mm development, but most are still behind schedule with the technology. Some believe that 450-mm will cause confusion in the supply chain.
Recently, TSMC said it plans to install its first 450-mm line in Taiwan by 2013 to 2014. It will process wafers at the 20-nm node on 450-mm substrates. Many of the details were not disclosed when TSMC made that initial announcement.
In an interview at SPIE after his keynote, Chiang elaborated on those plans. Initially, TSMC hopes to install a 450-mm pilot line in Fab 12 in Hsinchu, Taiwan. The line will process wafers at the 20-nm node. It hopes to get the pilot line up and running by 2013 to 2014.
Then, TSMC plans to bring up its first 450-mm production fab in Taichung, Taiwan, which will process devices at the 14-nm node. The Taichung plant is called Fab 15.
At 14-nm, TSMC plans to make a switch in transistor structures. At the 20-nm node and above, TSMC will continue to use traditional planar transistors based on bulk CMOS. At 14-nm, the company plans to make the switch from bulk CMOS to FinFET structures, he said.
So, the company will produce 14-nm FinFETs in production in Fab 15. Production is slated for 2015 to 2016.
The TSMC technologist said 450-mm wafers enable a 2.25- to 2.40-fold productivity gain over 300-mm wafers. But he acknowledged there are several challenges with 450-mm, namely to get the equipment vendors on board.
At one time, most fab tool vendors were reluctant to invest in 450-mm. Many believe it is too expensive and there is little or no return-on-investment.
Now, fab tool vendors are warming up to the idea for several reasons. First, Sematech, which is leading the charge in 450-mm, is providing some funding for fab tool vendors in 450-mm. Second, the world’s largest chip makers are pushing hard for 450-mm and fab tool vendors don’t want to lose out on some business.
Chiang in a question and answer session said that ”the government would pay for half of the cost’’ of 450-mm tool R&D, but he did not elaborate.
”We see a bit more willingness on the part of equipment makers’’ to embrace 450-mm, said C.J. Muse, an analyst with Barclays Capital, in a recent report. ”We then think by 2016-2018, we will see adoption of 450-mm.’’
Lam Research Corp. is reportedly beginning to invest in 450-mm. Other fab tool vendors are also quietly developing tools, but 450-mm won’t be cheap.
Drop in the (450-mm) bucket
It could cost about $12 billion in R&D investments for 450-mm, Muse said. ”The move to 300-mm was very much more expensive than the prior wafer transitions. While estimates from VLSI/Sematech suggest the 125-mm transition cost only ~$250-300 million and the 150-mm transition ~$700 million, the 300mm cost ~$12 billion,’’ he said.
”It is assumed that the 450-mm transition will not be cheap, and clearly equipment companies are reluctant to pay the full tab. We will likely see a chicken and egg game, but we do expect chipmakers to help support the tool development efforts with equipment companies, at the same time, sharing some of the higher dollars received in the current golden era of capital intensity,’’ he said.
Apple is reportedly looking to outsource the production of its A4 processor as well as the next-generation ARM Cortex-A9-based A5 processor to Taiwan Semiconductor Manufacturing Company (TSMC), according to industry sources. The Apple A4 processor is currently exclusively produced by Samsung Electronics, and the previous S5PC100 used in the iPhone 3GS was also developed and manufactured by the Korean company.
TSMC declined to comment on the report.
With Samsung now competing directly with Apple with its own smartphones and tablet PC, Apple is reportedly concerned about leakage of its processor technology to a major rival in the end-use market.
In fact, Apple already began handling some A4 orders to TSMC in 2010 when Samsung’s capacity was unable to fulfill strong demand of Apple devices, the sources said, adding that the move at the time was perhaps to test TSMC’s capability.
According to Digitimes Research, the iPad 2 will support an enhanced version of the A4 and the A5 will power the iPhone 5. TSMC will initially produce the improved A4, and could likely become the exclusive manufacturer of the A5.
EW YORK – In response to various network operators’ diverging demands for small to large cells, Freescale Semiconductor and Texas Instruments are unveiling this week at the Mobile World Congress their respective visions for a “base station on a chip.” Freescale is rolling out a scalable, multimode wireless base station processor family, dubbed QorIQ Qonverge. The new family of products, designed to scale from small cells (Femto and Pico) to large cells (Metro and Macro), share a common architecture consisting of Freescale’s proven multi-core communication processor, multi-core DSPs and baseband accelerators. Freescale’s new baseband SoC is also playing a critical role in lightRadio technology, recently announced by Alcatel-Lucent. LightRadio technology, which Alcatel-Lucent is working on with Hewlett Packard and Freescale, is designed to help create mobile phone wireless base stations for carriers that are said to be “barely bigger than a golf ball.” Lisa Su, senior vice president and general manager of at Freescale’s networking and multimedia group, said, “Our new baseband SoC is in it.” Texas Instruments, on the other hand, has developed a new multimode wireless base station chip, called TMS320TCI668, delivering “double the LTE performance of any existing 40nm SoC,” according to the company. TI has added hardware accelerators to the company’s recently announced base station SoC, called TCI6616. Both TCI6618 and TCI6616 use TSM320C66x – TI’s new DSP featuring floating point and fixed point math in every core. Facing exponentially increasing data traffic, network operators have been scrambling to find new solutions to their networks. Freescale’s Su bluntly put: “Most operators can’t keep up with data traffic today.” Operators want network solutions that are “multi-mode” and “future proof,” she explained. While the transition to LTE could help, LTEs are still in early stage, said Su, despite a number of trials. If operators are still building out a 3G network, they want that equipment “to be 4G capable,” she said. In explaining the wireless network architecture’s current state of flux, she added: “Femto cells, deemed an ‘interesting solution’ six months ago, are now a part of the solution many operators are looking at.” Network operators want network architecture “optimized for cost, performance and capacity,” she added. Many in the industry agree that there is no one-size-fits-all answer to the wireless network architecture of tomorrow. “Everyone is designing their own vision of network architecture right now,” observed Brian Glinsman, general manager of TI’s communications infrastructure business. “Solutions proposed by equipment vendors are colored by their top five customers,” he added. This trend, in turn, influences semiconductor suppliers’ base station SoCs. “Any operator who says they know what client devices will demand in flavors of 802.11, WiMax, LTE, various flavors of 4G…is lying, overly optimistic, or both!” noted Rick Doherty, co-founder and director, at The Envisioneering Group. “So the only sane survival method is build cell systems with agile software radio support until 4G ‘stratifies’ into clear winners… again, driven by the consumer, business and institutional device mix and demand.” TI’s strategy is squarely focused on “spectral efficiency.” The new hardware acceleration integrated in the TCI6618 is responsible for handling the high numbers of bits flowing through base stations, while freeing the programmable DSP cores’ processing power to execute customer differentiation chores like scheduling and multiple-input and multiple-output (MIMO) antenna processing. TI claims the new TCI6618 enables gains “up to 40 percent spectral efficiency.” By making TCI6618 pin and software compatible with TCI6616, TI offers customers flexibility in designing multimode base stations supporting all 2G, 3G and 4G standards, according to the company. TI’s TCI6618 base-station SoC does not come with a RISC processor — necessary for network processing. The company won’t be detailing such a base station SoC complete with a cluster of ARM cores until mid-2011. As an interim step, in collaboration with Axcom Technology, TI is offering a new 3G/4G small cell base station platform in the second quarter of 2011. The platform consists of TCI6616 SoC for PHY and Layer 2 processing; C6A8167 Integra DSP+ARM processor for Layer 3 processing; GC5330 transmit/receive processor for digital radio front-end processing; and NaviLink 6.0 solution GPS for clock synchronization. “We are offering such a platform now so that developers can start writing code,” explained Glinsman. In contrast, Freescale’s plan is to start offering a family of base station processors integrated with their proven network processor. Well-established CPU and DSP technology Freescale’s QorIQ Qonverge processors combine on a single chip: multiple Power Architecture cores; StarCore DSPs with MAPLE packet processing acceleration engines; and interconnect fabric. Noting that there will always be waste in a system using discrete components, Su pointed out the efficiency of the QorIQ Qonverge processor, in particular, comes from its multi-core fabric. “We spent a lot of time developing it.” “The key strength of Freescale is that it has both well-established CPU and DSP technology,” noted Joseph Byrne is a senior analyst at The Linley Group. “Nobody else is in the same position.” According to Byrne, “Freescale’s embedded-processor business has been stronger than its DSP business, which creates a particularly good opportunity for the company.” He explained, “Freescale is well-placed to lure OEMs that had been using TI DSPs with Freescale embedded processors, eliminating TI from these designs.” TI, of course, will try to do the reverse but [the company] is not a well-established supplier of embedded processors, he added. Freescale’s Picocells/Enterprise-Femtocells base station SoC In all fairness, the timing for the availability of complete base station SoCs – both from Freescale and TI — may not differ much in the end. Both are aiming at the second half of 2011. But analysts believe Freescale may have an edge. “We think Freescale’s exisitng and new customers will get to the market faster because Freescale offers more tools and endorsed, trusted third party solutions (like performance monitoring) than TI,” said Doherty. “Time to market, flexibility to change designs as market demands (more so on enterprise cell than femto cell) is criucial.” Freescale is seeing fundamental changes in base station design and deployment. Freescale’s Su described the expected proliferation of tiny base stations enabled by Alcatel-Lucent’s lightRadio technology as akin to cloud-computing. “Instead of racks of servers, we now see a network of desktop connected to cloud,” she said. Similarly, by combining Alcatel-Lucent’s antenna and RF communications with Freescale’s digital baseband unit, “you will soon see a network of small base stations that are the size of a Rubik’s cube,” enabling networks. The Linley Group’s Byrne agreed. “The big-picture is that mobile broadband requires a dense network of base stations, but carrier’s capital expenditure is limited. Thus, some kind of solution that provides density economically is required.” He said that lightRadio looks like the kind of architecture that can do the trick.
Texas Instruments’ believes its OMAP 5 platform is expected to change the concept of ‘mobile’ by driving disruptive mobile computing experiences providing stereoscopic 3D, gesture recognition and computational photography based on multi-core processing, including ARM Cortex-A15 MPCore processors
The 28 nanometer OMAP 5 applications processors carry on the OMAP family tradition of delivering increases in performance and functionality, while lowering power consumption compared to their predecessors. They offer up to 3x processing performance and five-fold 3D graphics improvement, provide a nearly 60 percent average power reduction compared to a sample user experience on the OMAP 4 platform.
The OMAP 5 platform’s software is designed for maximum reuse to ease migration from the OMAP 4 platform.
The OMAP 5 processor uses two ARM Cortex-A15 MPCores capable of speeds of up to 2 GHz per core in the OMAP 5 implementation. The 50 percent boost in performance over the Cortex-A9 core (at the same clock frequency), is combined with up to 8GB of dynamic memory access and hardware virtualization support.
In addition to the two Cortex-A15 cores, the OMAP 5 processor includes individual, dedicated engines for: video, imaging and vision, DSP, 3D graphics, 2D graphics, display and security.
The processor also includes two ARM Cortex-M4 processors for offloading real-time processing from the Cortex-A15 cores to improve low-level control and responsiveness of mobile devices.
The OMAP 5 processor can support up to four cameras in parallel, as well as record and play back S3D video in 1080p quality, and perform real-time conversion of 2D content to S3D at 1080p resolution. The processor can also deliver advanced short- and long-range gesturing applications, as well as full-body and multi-body interactive gestures, utilizing either 2D or S3D cameras.
The OMAP 5 processor, coupled with a TI DLP Pico projector and a camera, can also enable interactive projection where the user can actually “touch and drag” projected images on both a table top or wall.
Additionally, the OMAP 5 processor can interface with and leverage a variety of sensor technologies to enable touchless sensing, such as proximity sensing, capacitive sensing and ultrasonic sensing.
The OMAP 5 processor includes hardware and software resources that enable the development and deployment of computational algorithms to improve picture and video quality from built-in cameras. These provide camera stabilization, motion blur reduction, noise reduction, high dynamic range and face-based processing.
The latest processor goes a step further by using the same OMAP 5 hardware resources with vision algorithms to extract features and data from the picture, in order to implement applications such as face recognition, object recognition and text recognition. These vision capabilities can also be used as the foundation for augmented reality applications.
TI’s OMAP 5 platform is expected to sample in the second half of 2011, with devices on the market in the second half of 2012. The OMAP5430 processor is offered in a 14x14mm package-on-package with LPDDR2 memory support. The OMAP5432 processor is offered in a 17x17mm BGA package with DDR3/DDR3L memory support. These products are intended for high-volume mobile OEMs and ODMs and are not available through distributors.
TI also plans to develop compatible ARM Cortex-A15 processor-based solutions for broader market applications across TI’s product portfolio.
Semiconductor capital spending is expected to hit $59.070 billion in 2011, up 15 percent over 2010, according to IC Insights Inc. SUNNYVALE, Calif. – Semiconductor capital spending is expected to hit $59.070 billion in 2011, up 15 percent over 2010, according to IC Insights Inc.
The forecasted 2011 top 25 semiconductor capital spenders are shown below. ”There are five companies that are expected to spend at least $3.0 billion in 2011, the same number as in 2010 and three more than in 2009,” according to IC Insights.
The members of the $3 billion ”capex club” are Samsung, Intel, TSMC, Globalfoundries and Hynix,” according to the firm.
“Five of the top 10 increases are expected to come from major DRAM and flash memory suppliers. However, the top increase in 2011 is forecast to come from Intel, a massive $3.8 billion surge,” according to the firm.