"The global semiconductor foundry market
grew 4.4 percent in 2015 to reach $48.8 billion in value, according to market
research company Gartner" (more below).
Semiconductor foundries were able
to still grow their business in 2015 even though the overall demand for semiconductors shrunk.
Of the foundries based in China, the largest foundry- SMIC had the largest
increase in growth of 13.1%.
At the same time, the overall semiconductor market
shrunk more than 2% in 2015 - See tables below.
The global semiconductor foundry market grew 4.4 percent in 2015 to reach $48.8 billion in value, according to market research company Gartner Inc.
The percentage increase was low, coming after three years of double-digit percentage growth, but still was greater than the overall chip market which contracted by 2.3 percent in 2015, according to Gartner (see Chip market fell 2.3% in 2015, says Gartner ).
TSMC was market leader and is it has been for many years, and grew by 5.5 percent in 2015 driven by the success of its 20nm planar CMOS and 16nm FinFET manufacturing processes serving the needs of application processors and baseband modem chips.
Globalfoundries was able to outgrow UMC and swapped places with the Taiwanese foundry in second position, mainly due to the contribution of its IBM acqusitions. Samsung and SMIC grew faster than both of these companies but were unable to improve their rankings.
Top 10 semiconductor foundries ranked by 2015 revenue in $millions. Source: Gartner.
Tower Semiconductor Ltd., which trades as TowerJazz, grew strongly in 2015 but remained in 7th spot. Fujitsu which was 10th in 2014 climbed to position 8 in 2015 while Vanguard and Hua Semiconductor both dropped a single ranking position.
Price competition in advanced process technologies in 2015 was exceptionally strong, not only on the 28 nm node, as more foundry suppliers have started the production volume of 28 nm polySiON technology, but also on 65 nm and 40 nm. In contrast to the highly utilized 200 mm fabs from fingerprint ID chips and power management ICs, the low 300 mm fab utilization rates at some large foundries have triggered their willingness to run more 0.18-micron wafers in the 300 mm fabs.
The implication of a ruling by "U.S. Court of Appeals for the Federal Circuit—the federal appellate court responsible for all patent law appeals—made the entire industry gasp with itsCarnegie Mellon University v. Marvelldecision in August. In its ruling, the court suggested that a chip merelydesignedin the United States may infringe a U.S. patent, even if the contract is inked overseas and the chip is made, delivered and used in another country, never once touching American soil. If this suggestion becomes law.." is discussed below. Time will tell how this decision will impact future semiconductor patent litigation cases. On the surface, based on this case any product that was designed in the US is infringed even if all sales are outside of USA. The determining factor will be the meaning of "design win" (see below). Ron Insightful, timely, and accurate semiconductor consulting. Semiconductor information and news at - http://www.maltiel-consulting.com/
Benjamin T. Horton and Cameron B. Pick, Corporate Counsel
The United States is the undisputed leader of the semiconductor world. It’s a $340 billion industry that spent $35 billion on research and development in 2014, and has spent at least 15 percent of the industry’s revenue on R&D in each of the last 15 years. Seventeen of the top 25 global semiconductor design companies—and nine of the top 10—are based in the United States, according to the Committee on Comparative National Innovation Policies. There are roughly 250,000 domestic semiconductor jobs, with another 1 million supporting jobs. As the Semiconductor Industry Association says, “It All Starts Here.”
It is understandable, then, that the U.S. Court of Appeals for the Federal Circuit—the federal appellate court responsible for all patent law appeals—made the entire industry gasp with its Carnegie Mellon University v. Marvell decision in August. In its ruling, the court suggested that a chip merely designed in the United States may infringe a U.S. patent, even if the contract is inked overseas and the chip is made, delivered and used in another country, never once touching American soil. If this suggestion becomes law, the impact would be seismic.
Though it offered this suggestion, the court did not actually decide whether Marvell’s international sales infringed Carnegie Mellon’s U.S. patents, so all is not yet lost. Rather, all is open for debate. Title 35 of the United States Code says, “whoever without authority makes, uses … or sells … within the United States … infringes the patent.” 35 U.S.C. § 271(a). The court seemed content that chips cannot infringe U.S. patents if they were internationally manufactured, shipped and used. Whether those same chips are considered “sold” in the United States, however, was not quite so simple a question for the court. The court’s uncertainty comes as a shock. Until now, most of the patent world had been quite certain that chips made, shipped and used outside the United States are very much extraterritorial.
The Court’s Earlier Views
The Federal Circuit was not always so unsure of the boundaries of extraterritoriality when it came to semiconductors. In fact, it used to be relatively straightforward. In Power Integrations v. Fairchild Semiconductor (2013), for example, the Federal Circuit decided that the patent owner could not recover lost profits based on the defendant’s foreign sales of chips made and shipped abroad, even assuming that the foreign sales were “the direct, foreseeable result of Fairchild’s domestic infringement.” There, the court stated, “foreign exploitation of a patented invention ... is not infringement at all.” Similarly, in Halo Electronics v. Pulse Electronics (2014), the Federal Circuit found that the defendant met regularly with Cisco design engineers in the U.S., sent product samples to Cisco for pre-approval in the U.S., attended sales meetings with customers in the U.S. and provided post-sale support for products in the U.S. The court decided that this did not constitute a sale within the U.S., because the products were “manufactured, ordered, invoiced, shipped, and delivered abroad.” In the most recent decision, however, the court seemed to be going in a different direction.
In Carnegie Mellon, the Federal Circuit reiterated some well-known concepts on the location of a sale, e.g. the place of the legal commitment to buy and sell, the place of delivery and the place of ordering. But the court further stated that “[t]he standards for determining where a sale may be said to occur do not pinpoint a single, universally applicable fact that determines the answer, and it is not even settled whether a sale can have more than one location.” The court went on to broadly suggest that a sale may occur at the place where substantial activities of the sales transactions occurred. This “substantial activities” language seems to embody the court’s recent uncertainty on extraterritoriality. As part of that uncertainty, the court suggested that the design of a chip could be part of the chip’s sales cycle and, therefore, may be considered “substantial activity,” especially if the activity is custom design, or a “design win,” something the court considered more closely integrated with the sales cycle. According toQuality-Adjusted Price Measurement: A New Approach With Evidence from Semiconductors, custom design wins may account for as much as 25 percent of the semiconductor industry, making the court’s suggestion an expensive one.
The Meaning of a ‘Design Win’
So what, exactly, does the Federal Circuit consider a design win? According to the court, a design win occurs when a designer’s custom chip is purchased and enters mass production. Generally, a design win results in exclusive use of the customized chip for a certain period. To secure a design win, the design is tailored to the customer’s product. This typically encompasses a lengthy sales cycle involving extensive joint work over several years, which may include designing, simulating, testing, evaluating and qualifying the chips, and likely providing samples. With all these considerations, a design win may involve a bevy of new steps, factors and locations.
So the Federal Circuit is now fascinated with design wins. What does that mean? It means that products and transactions previously thought to be extraterritorial, and therefore beyond the reach of U.S. patents, are vulnerable. The court’s discussion of what activity constitutes a design win, however, may provide clues as to how a semiconductor company might shift (from a legal standpoint) design wins outside the United States, even if some portion of the design is done in the United States. For each design win factor located outside the United States, the sale of the chip may be more likely to be considered extraterritorial, and therefore outside the scope of United States patents.
Possible Solutions
So how can a company with U.S. design activity, particularly custom design activity, move “design wins” overseas? Here are a few suggestions based on the court’s discussion:
• Simulations Perform system simulations and post-layout simulations for custom designs on overseas machines, and provide waveforms from the system simulations to overseas customers, even if unit and subsystem simulations occur in the United States and/or the test software for the simulations is developed in the United States.
• Manufacture/Marking Relocate manufacturing of sample custom products to foreign foundries, and move or outsource the sample marking process overseas.
• Storage Store samples (even temporarily) overseas before they ultimately reach the foreign customer. This way, no product ships directly from the United States.
• Program/Configure Relocate the programming or configuration of sample devices overseas.
• Test Move the testing process, or at least a component of the testing process (e.g. burn-in or ESD), overseas. Even if the protocol or software is developed in the United States, an oven or lab in another country can shift a valuable component of the sales cycle.
Currently, there is no timetable as to when these questions about U.S. design activity and U.S. patents may be conclusively resolved. In the meantime, your authors, and an entire industry, will be watching and waiting.
This article is for informational purposes only and is not legal advice.
Samsung benefits a lot in its relationships
with Apple from the consolidation of DRAM memory fabs. The consolidation keeps
DRAM prices high, while limiting the choice of Apple of vendors for its
products. While Samsung and Apple compete on cell phones, Apple still needs
Samsung for its processors and memory chips.
The article below discusses Samsung's
semiconductor production investments and it shows the preference of investing
more in DRAM than NAND -
"we invested around ..$9.36 billion in
memory...We have set DRAM and NAND production ratio to 7 to 3"
The article also shows the path Samsung is
taking this year in its transition from 2D to 3D on SSD, while mobile will use
2D NAND this year.
Summary: Samsung plans to
pump up to $10 billion into its semiconductor business, which drives the South
Korean tech giant's best-performing business division.
By Cho
Mu-hyun | April 30, 2015 -- 00:08 GMT (17:08 PDT)
Samsung plans to inject as much as $10
billion during 2015 into its semiconductor business, as unprecedented demand
continues to fuel growth in the company's best-performing division.
"Last year, we invested around 10
trillion won [$9.36 billion] in memory semiconductors," said Jeeho Baek,
senior vice president of Samsung's semiconductor division, during a conference
call with analysts. "We plan to maintain similar volume this year. We have
set DRAM and NAND production ratio to 7 to 3, but will manage this flexibly
depending on demand, going forward."
Samsung is increasing its production
capacity for memory chips. An additional line that will produce DRAMs is being
built at Hwaseong, its main hub for memory chip production in South Korea,
which will be completed by the end of the year.
The South Korean tech giant is planning to
start construction of a new plant in Pyeongtaek, South Korea, next month. It
plans to invest 15.6 trillion won in the new factory by 2017, the same year
that it is tentatively set to start production.
The world's second-largest semiconductor
vendor also plans to increase offerings that use 3D V-NAND. Only Samsung and
Japan's Toshiba have commercialised 3D V-NAND, which stacks cells vertically
within a chip to increase storage capacity in the same space.
Baek said the company will offer new 3D
V-NAND solid state drives (SSDs) aimed at consumers by the second half of the
year, as well as next-generation offers for enterprise-aimed goods.
Samsung cemented its leadership in SSDs
last year. According to market research firm IHS, Samsung Electronics held 34
percent market share for 2014, with revenues of $3.996 billion -- double that
of runner-up Intel, which posted $1.99 billion in the segment.
Last year, Samsung launched a variety of 3D
V-NAND SSDs for the enterprise. The senior vice president said talks with
clients were going smoothly, and he expected a surge in orders in the second
half of the year.
Samsung insiders said the firm has clinched
supply deals with Google and Amazon to supply SSDs for the latter's
datacentres. Samsung SDS, the IT service arm of Samsung Group, will also use
Samsung-made NANDs when building servers for clients.
Rival makers Toshiba, SK Hynix, and Micron
have just begun ramping up efforts to commercialise 3D V-NANDs, but are yet to
join the fray, which will likely contribute to prices being stable, allowing
Samsung to continue to reap high margins in the area.
For mobile clients, however, supply will
still be 2D NANDs. Samsung believes 2D NANDs are still more appropriate for mobile
set products, said Lee Myung-jin, head of IR, during the conference call.
"For high-integration,
high-reliability NAND flashes, we will use V-NANDs, and for mobile and consumer
products, we will use 2D NANDs and secure competence in both product lines,”
said Lee.
According to Gartner, the worldwide
semiconductor market grew 7.9 percent last year to $339.811 billion. Memory
segment grew 16.9 percent and led the growth, it said.
Samsung's contract chip business, called
System LSI, will likely enjoy a surge in profit as well. Its migration to
14-nanometer FinFet process ahead of rivals has helped it clinch new clients.
Samsung will produce chips for Apple,
Qualcomm, and Nvidia, both in 14-nanometer and 20-nanometer processes, for the
latter's next-generation chips, insiders have said.
It is interesting that while Samsung
is continuing to develop 3D devices and processing on a single chip -
"Samsung described a second generation of
its 128 Gbit 3-D NAND flash, a 3-bit/cell version with 32 layers now in
production on a 68.9mm2 die. The Korean giant is ahead of the
pack in dense flash and also is expected to beat TSMC to market with 14/16nm
FinFET logic later this year, a fact that help it retain archrival Apple as a
foundry customer."
ITRS and Intel focus on developing chip
stacks -
"Paolo Gargini who leads the ITRS
semiconductor road map effort said chip stacks are today what high-K metal gate
design was in 2007, the next big requirement for progress.
Intel fellow Mark Bohr disagreed. “We need
vertical interconnects one or two orders of magnitude more dense than today’s
through-silicon vias,” he said"
Ron
Insightful, timely, and accurate semiconductor consulting.
Semiconductor information and news at - http://www.maltiel-consulting.com/
SAN FRANCISCO — This year’s International Solid-State Circuits Conference once again served up a smorgasbord of silicon innovations. Despite the rising costs and complexity of chasing Moore’s Law, engineers cooked up smaller, faster, more media-rich devices and the strange new world of ultra-low power design delivered a few exotic dishes at the annual gathering of chip designers.
My time at the event started with a reception where for the first time I met Lynn Conway, co-author of one of the landmark texts in chip design. In 1979, Conway’s MPC79 event paved the way for quickly turning software files into running chip prototypes, establishing a model of how to spread enabling abstractions quickly to enable innovation.
Samsung described a second generation of its 128 Gbit 3-D NAND flash, a 3-bit/cell version with 32 layers now in production on a 68.9mm2 die. The Korean giant is ahead of the pack in dense flash and also is expected to beat TSMC to market with 14/16nm FinFET logic later this year, a fact that help it retain archrival Apple as a foundry customer.
Only a handful of papers at ISSCC described designs using TSMC’s 16nm process. Expect many more next year as well as perhaps the first papers of foundry customers using Intel’s 14nm FinFET process.
In the hallways at ISSCC, I sought out topic experts to get their opinions on paper sessions they attended. Behzad Razavi of UCLA said he saw a trend to soft radio with transmitters reduced to DSPs linked to small ADCs and antennas, although receivers still require more complex analog circuits.
You don’t have to go to the cloud for analytics, said Chris van Hook, a medical electronics specialist at the Imec institute outside Brussels. More self-learning algorithms are getting embedded at node-level chips in papers he saw.
Jim Warnock, a designer of IBM mainframe processors, aid he was most fascinated by papers in the low-power digital track. “Some of them are running at a few hertz off picowatts with way different kinds of circuit designs…it’s a different world,” he said.
Indeed, I ran into one post-doc from Berkeley who told me aboutstartup Psikick that is designing integrated IoT chips that aim to run off energy harvesters. He said he hopes to launch his own IoT chip startup eventually.
Back at the high end, analyst David Kanter of the Linley Group said IBM’s mainframe chip designers will be challenged to continue differentiate their processors as they increasingly move to off-the-shelf technologies. If a landmark fab sale goes through as expected, the group will design its next zSeries chips in a 14nm process run by Globalfoundries.
IBM’s Warnock expressed confidence he has a laundry list of accelerator blocks for analytics and mainframe functions he can imagine integrating into future mainframe chips. And he said he hopes to design chips made in Globalfoundries’ giant Malta, New York fab.
The event echoed with groans of designers trying to squeeze more out of increasingly complex and costly process nodes. “We should keep on scaling to 7 and 5nm but we need to harvest the sweet nodes,” said Geoffrey Yeap, vice president of technology at Qualcomm, speaking on an evening panel on Moore’s Law.
Liam Madden rallied for the 2.5-D chip stacks he is helping create at Xilinx, the latest of which will use a 20nm process pack 19 billion transistors including 10 ARM A9 cores. With two new metal layers added in the latest nodes, “the RC delays are killing you routing across chip,” he said.
Paolo Gargini who leads the ITRS semiconductor road map effort said chip stacks are today what high-K metal gate design was in 2007, the next big requirement for progress.
Intel fellow Mark Bohr disagreed. “We need vertical interconnects one or two orders of magnitude more dense than today’s through-silicon vias,” he said
.
Madden agreed but expressed confidence TSV density “will go up at least an order of magnitude in the next two or three years, but getting beyond that will be tricky.”
Despite the pains of progress, engineers showed plenty of silicon innovations at the event. The follow pages cull out a few of them.
Apple chose to manufacture most of the A8 processor by TSMC. Out of the total volume of A8 chips, Samsung is producing around 30 percent, while TSMC is making 70 percent,. It is preferable in the electronic industry to have more than one vendor producing the semiconductor chips. If you have multiple sources for your chips you are not too dependent on only one supplier for your products. You also have more flexibility in price negotiations. It is interesting that earlier iPhone were dependent on only Samsung producing the microprocessor running the cell phone.
However, since fab lines are very expensive (see below and at Samsung To Invest Nearly $15B In New Chip Plant ) often vendor are stuck using only one foundry source. Different foundries processes are not identical and it will take long time to copy chip design from one vendor to another.
Construction of the New Chip Plant Will Begin Next Year
By
MIN-JEONG LEE
Updated Oct. 6, 2014 1:54 a.m. ET
Samsung said it will invest in a new chip plant in South Korea, as the company seeks growth beyond smartphones. Bloomberg News
SEOUL— Samsung Electronics Co. plans to invest 15.6 trillion won ($14.7 billion) to build a new cutting-edge chip-manufacturing plant in South Korea as it seeks to maintain its lead in memory chips and grow beyond smartphones.
Samsung’s chip unit has become increasingly important to its profit growth as its smartphone business—which accounts for about 60% of its total operating profit—continues to suffer from intensifying competition from low-cost Chinese handset makers. Analysts expect demand for Samsung’s memory chips to stay strong for the rest of this year amid tight global supply.
Samsung said Monday that construction of the new chip plant will begin next year, with operations to start in 2017. The new plant will be located in Pyeongtaek, a city south of Seoul.
Samsung hasn’t yet decided what types of chips it will produce at the plant. Samsung’s competition with Taiwan Semiconductor Manufacturing Co.s likely to intensify in coming years as they compete to make logic chips for Apple Inc. iPhones. Logic chips are semiconductors that process decisions, as opposed to storing memory, to operate gadgets.
Microprocessors, considered to be the brain of a smartphone, are a type of logic chip.
The investment “will significantly influence the shaping of Samsung’s future semiconductor business,” Kwon Oh-hyun, co-chief and a vice chairman of Samsung Electronics, said.
Samsung typically outspends rivals when investing in high-tech chip-manufacturing facilities.
The ability to spend heavily, thanks to its hefty cash pile of $60 billion, has allowed the company to maintain its lead in memory-chip production.By moving quickly to finer chip-process technology, Samsung is able to cut chip-manufacturing costs.
Samsung is the world’s biggest producerof dynamic random access memory chips, used in personal computers and smartphones.
It also makes NAND flash-memory chips, which are used in mobile devices including digital cameras.
The investment comes as Samsung, which is to report earnings estimates for the third-quarter Tuesday, is expected to say that its operating profit likely halved from a year earlier due to the poor performance of its mobile business.
Some analysts expect the company’s chip-division profit to exceed mobile phones for the first time in three years.
Samsung posted an operating profit of 10.2 trillion won in the third quarter of 2013. About a third of its profit come from chips, televisions and home appliances.
Last week Samsung, moving to capitalize on cheap labor costs, said it would invest $560 million for a new TV plant in Vietnam, saying that the complex will be used for addressing mid- to long-term demand for consumer electronics globally, including emerging markets.
In South Korea, Samsung has memory-chip production lines in Hwaseong and facilities in Giheung that focus on logic chips. The company also has chip plants in China and the U.S.
The article below discuss IBM announcement that it is
selling its manufacturing operations. It will impact SOI (Silicon on Insulator), a process
that IBM specialized in.
It is not likely that Intel will buy an existing fab. Intel
always duplicates an existing process in a current fab when they build a new
fab
Following the semiconductor shock of IBM pulling out of chip manufacturing, David Manners considers who could buy and run the IBM fabs, whose processes include SOI and SiGe, and which supply manufacturing process technology to half the world’s major companies: Samsung, GloFo, UMC, ST, Renesas, NEC, Freescale, Toshiba and Infineon.
The biggest shockwave the semiconductor industry has had for years is the IBM announcement that it is pulling out of chip manufacturing.
IBM CEO Ginni Rometty has decided that IBM’s future is in software and services and chip manufacturing clearly doesn’t fall in either of these categories .
Moreover IBM’s chip manufacturing business lost $130 million last year and is slated to lose another $130 million this year while the capital costs to stay in the game are $5 billion a factory and rising.
That’s one way of looking at it.
The other way of looking at it is that IBM has been a stalwart of the world semiconductor industry, inventing the DRAM, mastering SiGe, SOI and much more besides and supplies manufacturing process technology to half the world’s major companies: Samsung, GloFo, UMC, ST, Renesas, NEC, Freescale, Toshiba and Infineon.
It is shocking to think IBM’s legendary R&D operation may be curtailed but, without the justification of a business outlet for its innovations, it may be.
One rather assumes that the obvious candidates for buying IBM’s fabs have already been tapped and that the announcement that Goldman Sachs has been appointed to look for a buyer for the fabs is to find an unobvious candidate.
The obvious candidates are Intel, TSMC, GloFo and Samsung.
However, Intel has net cash of only $7 billion and makes only CMOS chips, while IBM’s capacity includes processes like SOI and SiGe which Intel won’t need and the IBM fabs are not as advanced as Intel’s.
TSMC says it doesn’t want to operate foreign fabs; GloFo’s owners are said to be getting fed up with their capital-greedy chip investment and may not want to expose themselves further to the chip industry; Samsung already has as much operational and planned capacity in the USA as it needs.
So how about the non-obvious candidates?
Suggestions have been that Apple could enter into a jv with IBM to run the fabs. This would involve Apple putting a lot of money in to upgrade the fabs to the latest processes, but Apple has lots of money and IBM has the latest processes.
TowerJazz is another suggestion. The company has just formed a jv with Panasonic in which TowerJazz has a 51% share and which will own three Panasonic fabs in Japan.
The Panasonic deal gives TowerJazz’s already impressive array of specialty processes the addition of Panasonic’s specialty flows such as High Definition FSI for sensors and high voltage SOI power management technologies.
If TowerJazz now added IBM’s specialty processes like SiGe and SOI to its stable, it would become the world’s No.1 specialty fab.
Another suggestion is China. China has been trying, unsuccessfully, to build an indigenous chip industry for 30 years. Following the successful integration of IBM’s PC business into Lenovo, followed by the recent sale of IBM’s x86 server business to Lenovo, it would seem quite logical that a sale of IBM’s chip manufacturing would be to China.
Of course the European authorities, wedded to the idea of gaining 20% world market share in IC manufacturing, may see the opportunity to acquire IBM’s world-class process development capabilities as a sufficient reason to buy the fabs. IBM’s proficiency in SOI would give a boost to the European-developed FD-SOI process.
Then, of course, there are other emerging national tech players like Brazil and India who could see this as a once in a lifetime opportunity to acquire the basis for a chip industry.
Whatever the outcome of this, a chip industry without IBM will be a diminished industry.
