Next week Intel will be presenting several papers at the IEDM in San Francisco. Mark Bohr, Intel Senior Fellow and some of the research team previewed three of the topics being presented. The common theme through all three papers reflects Intel’s drive to push conventional technology as far as it will possibly go. The company feels that’s easier to do rather than introduce a whole new set of materials at this time. That being said, Intel also pursues new material research. The plan is to have options and solutions available and ready to implement when traditional silicon processing runs out of steam.

Following is a quick review of three papers that Intel will be presenting.

32nm Logic Technology

In September 2007, Intel announced their fully functional 32nm SRAM with a 0.18um² cell size. Today they are touting a 291Mbit SRAM, the same density as before, but with a 0.171um² cell size. This SRAM chip features greater than 1.9 billion transistors operating at 3.8GHz. Intel is on track for production readiness in Q4 2009.

One change that Intel is implementing with their 32nm process is the use of immersion lithography, one generation later than other manufacturers. There were no specific comments about any challenges or benefits of immersion technology, only that it was at the appropriate level of maturity for Intel. Although Intel stated there is still a lot of work to get 32nm technology into production, there is nothing out of the ordinary in the scope of work to have their 32nm process ready for production in 2009.

45nm Low Power SOC Technology

Intel’s CPU usually operates at leakage values of 10+ nanoamps per micron. SOC applications demand a much lower leakage. Intel has developed a derivative version of their 45nm high-k, metal gate process that can get the leakage values down to the 10 to 0.01 nanoamp per micron range. In addition to the low leakage, Intel added new device elements to accommodate all the analog circuitry that is usually added to an SOC.

One of the changes to the technology is a thicker dielectric as part of the gate insulating stack. This allows the transistor to operate at a higher than normal voltage. It also provides a wider range of I/O transistors and I/O types. And something that was not forgotten, an RFCMOS device element, that can accommodate the high frequencies that CMOS transistors can achieve.

High Performance Indium Antimonide (InSb) Transistors

As part of Intel’s ongoing research effort, the company is working with III-V materials such as Indium Antimonide to achieve voltage scaling while achieving the same performance at 10x power reduction. As a high mobility material, Intel has used InSb to cut voltages in half. The process involves building stacks on a GaAs wafer. There is a different lattice spacing at the bottom of the stack compared to the top. The ability to even explore these new options is made available because of the deposition techniques introduced at the 45nm technology node. Moving forward, Intel recognizes that a lot of work needs to be done to achieve comparable costs with silicon. In addition, Intel is also trying to reuse as much conventional design expertise as possible. Eventually Intel wants to transfer this to a silicon process.

-Joanne Itow

On 12/04/08, NXP Semiconductor announced their new, single-chip LCD TV platform, the NXP TV550.  This is the first TV platform manufactured in 45nm CMOS low-power process technology. Engineering samples scheduled for 1Q09.

The TV550 platform incorporates the PNX85500 processor and integrates NXP’s proprietary Motion Accurate Picture Processing (MAPP2) technology. This single SoC handles all the new High Definition audio codecs like Dolby Digital TrueHD and DTS-HD Master Audio along with the more familiar codecs from Dolby and DTS. The part also handles the video codecs for the decompression necessary to process MPEG2 and MPEG4 encoded signals.  In addition, one of the target markets for the chip are those LCD TVs that are looking to increase their frame rates to 100 Hz (Europe) and 120Hz (North America) for better rendering of fast moving video action sequences. Consideration was also given to the current trend towards LCD backlighting in LCD TVs with sectional backlight dimming enabled.

Since this is a platform, other functions can be added as more features are required.

A further twist is the ability to add NXP’s latest hybrid silicon tuner, the TDA18272, which adds the ability to receive and decode standard definition NTSC and high definition ATSC signals. To top all this off, NXP has also added hooks into the platform so that their TDA9996, HDMI 4-port switch can be added if desired.

Semico Spin

All-in-all, it is an impressive accomplishment to be able to integrate all this functionality onto a single chipset and also offer the ability to add more features through additions to the base platform. This starts to look like the way other silicon manufacturers will go in the future to address the needs of TV manufacturers to reduce costs while continuing to add capability and new feature sets.

All this functionality is aimed at bringing the rich feature sets found on high-end HDTVs downstream to more moderately priced LCD HDTVs. With today’s deteriorating economic conditions, and the prospect that consumers worldwide will reign in their spending this holiday season, a product that both reduces chip count and cost while offering features found only on higher-prices HDTVs will be greatly appreciated by HDTV manufacturers. It appears NXP has succeeded in this area.

Looking further into the future of the HDTV market, some trends and areas for focus become apparent.

• As the US transitions from analog TV broadcasting to full digital broadcasting, people are going to want to be able to view two or more HD signals at the same time. This is possible today, but not while using picture-in-picture (PiP) features. (Only one standard definition and one HD picture can be viewed using PiP now). It would seem therefore, that HDTV  sets that would allow the viewing of more than two ‘pictures’ at the same time using PiP would be a feature people are going to be looking for and willing to pay for. This will require two or more silicon HD-tuners in each set and is an opportunity for new designs and additional integration.
• Meeting the needs of LCD HDTV manufacturers to support 100 and 120 Hz frame rates is important, but it is doubtful this will be as far as manufacturers will go. What about 200Hz and 240Hz frame rates? Some manufacturers are already experimenting with advanced frame rates. This would seem to be another area for further new designs and silicon.
• Given that the dream of many people today is to eventually set up their own home theater – with a HDTV, high definition DVD player, high definition satellite or cable STB and all connected to an audio-video receiver with at least a 5.1 surround sound system.  Such a system brings its own set of issues to the living room. One of which is the time lag incurred when switching from one TV channel to another. This is due to the need to decompress the audio and video signals to be displayed on the HDTV. It takes anywhere from 2-4 seconds for this to happen depending on how the system is connected, over component cables, over HDMI, etc. One area of improvement for HDTV aficionados would be to reduce this to a more manageable level – to maybe one second. This will not happen unless silicon manufacturers are able to look at the entire system (the home theater) and how people use it. Admittedly, this is not a show-stopper for most people, but it can be irritating if you do a lot of channel-surfing (who doesn’t).

All the above areas and more become possible as the amount of processing power continues to rise in silicon targeted at the HDTV market. The last point made above is a good example of this since the time lag mentioned is mostly due to the need to process the signal into its correct form and also to perform the handshaking between the connected devices for HDMI, HDCP, etc.

This makes a good argument for increasing use of multicore-based silicon and fully integrated platforms like the NXP TV550.

Semico believes this will be a path many silicon solution providers will likely pursue.

The foundry market players continue to provide almost as much nail-biting suspense as a  TV reality program.  Drawing most of the media attention is the battle for large volume IDM and advanced technology customers.  Companies such as TSMC and the members of the Common Platform are developing strategies which involve unique advanced technologies.  The competition involves a fight over the biggest and strongest semiconductor manufacturers .

But that is not the only show in the foundry arena.  Smaller foundry suppliers such as Tower Semiconductor are finding ways to grow faster than the overall semiconductor industry by focusing on products that don’t require the bleeding edge technologies but do require differentiated technologies specifically designed for small to medium sized manufacturers.  The so-called niche players are actually targeting markets such as MEMS, RF CMOS, telematics, and portable medical that are growing faster than the overall industry.

Tower Semiconductor and Jazz Semiconductor have taken a big step which sets them up to uniquely meet the needs of many companies looking for a foundry partner to provide a solution to their manufacturing challenges.  The merger of Tower and Jazz brings together a number of complementary ingredients.  Tower has a strong presence in embedded non-volatile memory and image sensor technologies while Jazz has always been known for its BiCMOS, SiGe, RF CMOS, and power/BCD technologies.  The customers of Tower and Jazz had very little overlap making the combined business much more broadly based.  In addition, the combined manufacturing capacity now has a worldwide presence with an opportunity to operate in a much more efficient manner.  Tower also has a proven track record of successful technology transfers for both specialty and legacy products. 

Since Russell Ellwanger took charge of Tower in May 2005, the company has consistently improved its manufacturing operation and expanded its customer base.  The merger with Jazz will positioned the company to be an even stronger player in the very fast growing specialty foundry arena.  The name of the game is definitely survival of the fittest. 

Joanne Itow

Frequent readers of the Semico Spin know that I am a fan of ComicCon and attend it in San Diego, CA during my family vacation.  This is the fifth year in a row if I remember correctly which we attended.  As I say every year, there were more geeks than you can shake a light saber at.

 For those unfamiliar with ComicCon , it is the largest convention for comic books in the world.  However, it covers a great deal more.  Science fiction and fantasy TV and movies are heavily represented.  These are usually tied in with comics and animation.  There is also a strong tie in with video games.  For the first time in its history ComicCon was sold out for all four days with attendance of 125,000 each day.  It is the largest convention through out the year for the city of San Diego.

ComicCon 2008

 No, I did NOT go in costume, though there were plenty of interesting sights.  There were the usual Star Wars storm troopers of various types, including the Elvis storm trooper and this year the Burger King storm trooper.  At ComicCon one sees the rich source material for new entertainment.  This is very important for the video game market.  The fans of the sci-fi/comic genre want cutting edge graphics and special effects.  They represent a key segment of the video game market.  ComicCon is a venue for previewing many new upcoming games for the holiday season.
I can always count on my 18-year old son for in-depth research.  He spent the entire time in the video game area just like every other year.

 In 2007 one could find gaming, both consoles and PCs, in the central portion of the exhibit area.  This year all gaming was pushed to the far left side.  There was not as strong an emphasis on the hardware.  Last year I said that content was king.  It was even more so in 2008.  Various game vendors had demos for all of the gaming consoles, but the consoles themselves were not prominently displayed.

Last year Sony and Nintendo were quite prominent with their handheld games.  This year I could not find a single demo featuring PSP or Nintendo DS.  I guess the vendors see no need to keep promoting the platforms.

In 2007 Sony had several areas for different product lines including PS2, PS3, Sony Ericsson cell phones, PSP, and Blu-Ray.  This year Sony focused on upcoming movies from Paramount.  There was a PS3 area.  However, there were no signs of PS2 anywhere.  Clearly Sony is intent on phasing out this platform since there were no games featured for it.  Little mention of Blu-Ray except for DVD releases.  Sony has won that battle, no need to promote the hardware at this show.
There was an Xbox 360 booth.  It appeared that this was mainly for demoing the Xbox Live service.

The big story in the gaming area was Capcom.  Like last year, Capcom’s booth was an arena hosting tournaments played on the PS3 and Xbox 360.  The big news is the release of Capcom’s “Streetfighter 4” title.  In fact the “Streetfighter 4” presentation was in one of the larger rooms in the convention center holding over 1,000 people and was at capacity.  It was like being at a preview for a major Hollywood movie.

As always I could count on finding my son around the Maximum PC booth.  Several PC games on Alienware platforms were featured.  There were other PC games featured around the exhibit floor, most notably World of Warcraft.  At the Maximum PC booth there were tie-ins with hardware vendors such as Sapphire and Antec.  The big raffle item people were salivating over were the Intel quad core Extreme MPUs in boxed sets.  My son missed out on an MPU by a number or two more than once.  He had to satisfy himself with other PC components he won.

What was quite notable in the PC gaming area was the lack of any gaming laptops.  These platforms are advertised on line and are being promoted by Intel, AMD and their key customers.  However, at ComicCon 2008 it appears that gaming laptops have yet to catch on.  The gamers still love their big impressive desktop cases.  The fact is that this upper end of the PC market likes to get into the inner workings of a PC and be able to “supe-up” their machines like hot rodders.  A laptop does not lend itself well to this.

The last couple of years the gaming consoles and handheld consoles were heavily promoted.  It appears the hardware vendors have established their foothold and will grow based on the demand for new game titles. Based on what I saw at ComicCon 2008 this coming holiday season should see healthy activity for the video games.  The driving force will be content – new game titles and films.

Tony Massimini, Chief of Technology
tonym@semico.com

Why is it critical to look at DRAM and NAND memories together?  As the target application for memory technologies continues to shift away from desktop personal computing, the value propositions of memory technologies have also changed.  In this report, we examine the strategic issues of the total memory market where manufacturing and technology decisions decide the profitability and health of participants and end use implementers.

What immediate business impact do DRAM and NAND have? DRAM and NAND manufacturing decisions affect all aspects of the semiconductor and electronics markets.  Cost per megabit and consumption by the PC market used to bound the memory market that was predominantly DRAM.  The growth of cell phones and the consumer market transformed NAND Flash into the trendsetter for costs and manufacturing technologies. The impact of this shift in the value proposition of memory technologies is magnified by the observation that approximately 60% of the 300mm fabs belong to companies with heavy investments in memory production and almost 60% of semiconductor Capex goes towards tools for memory production. “Yet memory revenue still continues to fluctuate on a quarterly basis, and the profit margins for the past several years have continued to be low, making it difficult to fund this Capex,” stated Bob Merritt, Semico VP of Memory Products. 

Who are the Players?  DRAM and NAND Flash vendors have changed the manufacturing landscape and subsequently the memory market.  NAND Flash is the technology driver for memory manufacturers that previously were dedicated DRAM manufacturers.  Major memory vendors have accomplished some economies by having fabs capable of converting from DRAM to NAND Flash and vice versa.  Flexible capacity works when these two markets are out of phase, but when the cycles align, supply-demand challenges are extraordinary.  Memory manufacturers are looking at foundry agreements to increase flexibility and reduce Capex investment.  However, orchestration and concept are not always soothing music.

The end markets previously defined the differences between DRAM and NAND functionality, with NAND Flash implemented in consumer products and DRAM in computing.  With the introduction of SSD (solid-state drives), combined with USB, computing again becomes the major consumer of both DRAM and NAND. 

Why be interested in Memory?  DRAM and NAND are the major memory market today.  The development of new memories (and the adoption in applications of those new memories) will yet again extend the importance of the memory market to equipment vendors, electronics OEMs, semiconductor designers and manufacturers, system and software designers at a minimum. 

This report provides a tutorial on the characteristics of the types of memory, reviews and anticipates the applications for DRAM and NAND, breaks out the investment required to participate in this market, discusses the key participants, and highlights the disruptive and evolutionary memories that are the future. 

When are DRAM and NAND One Driver?  Viewing this from the perspective of the total semiconductor market, the DRAM and NAND markets are just one market to the OEMs.  They are dependant upon manufacturing decisions that determine supply that equates to price.  Semico presents an analysis of DRAM and NAND and how, both independently and cooperatively, these two memory types affect the market.  The memory market is one of perception regarding the exact status of supply and demand.  This could be a wonderfully graceful tango if it didn’t keep devolving into a particularly vicious game of musical chairs.

To purchase this study, or to receive a copy of the Table of Contents, including a list of exhibits, please contact Susan Cadel at 607-368-7600 or susanc@semico.com and reference CC315-08, DRAM and NAND Flash:  Doing the Memory Tango.

Several Semico analysts attended Semicon West this year.  Although the forecast for 2008 capital expenditures is for less spending, there was still a lot to get excited about at the show this year. 

We talked to a few companies with innovative improvements to image sensor technology.  Sarnoff has developed a back illumination technology that utilizes SOI wafers.  Tessera continues to find ways to improve optics with their wafer level camera technology that allows for high performance imaging combined with innovative packaging to produce an ultra thin camera module.  Digital camera technology will continue to grow in cell phone and standalone camera applications but it will also increase in numbers for industrial, automotive and medical applications as well.  These improvements from Sarnoff and Tessera will enhance and broaden image sensor applications.    

Novellus is expecting to win big as the memory manufacturers go into high production with their mega-fabs which can process 140,000 to 160,000 wafers per month compared to the 40,000 or 50,000 wafers per month that an average logic fab is capable of processing.  Even at 1 or 2 copper layers, that’s a lot of material and processing steps which translates into more equipment.  I have to keep reminding myself that these are also 300mm wafers!  If unit demand continues to grow at 10% CAGR, wafer demand could double!  Who says we don’t need 450mm wafers?

In addition, this year we seemed to have talked to more companies who are finding ways to improve defect detection and cleaning steps.  Qcept is a privately held company that was founded in 2002.  They have a product that can scan wafers and utilize software enhanced imaging to detect non-visual defects.  They already have seven tools installed.  Both Novellus and ATMI focused their presentations on their cleaning solutions. 

Of course, the solar displays were probably the most popular but one of the most unique and ground-breaking presentations we had was from a company call Quantum Sphere.   The company manufactures materials for electrode systems and can lower the cost and raise the performance of products such as batteries, fuel cells, and hydrogen generation.  A few of their biggest application breakthroughs are still proprietary but expect to hear some new announcements from this company in 2009.

All in all a very good SemicoWest show.¼br> ¼/p>

During the week of June 16^th, 2008 at the VLSI Technology Symposium, there were a lot of compelling papers presented.  One in particular caught my attention because the findings had a very subtle underlying message.  I’m referring to the Intel paper on floating body cell (FBC).  The floating body cell is not new.  Intel talked about a non-planar floating body cell in 2006.  This year Intel touted a much smaller device using a planar design on SOI.   The device could have 3-4x more bits per area.  The benefit is faster computational rates.   

It is well known that Intel does not currently use SOI for volume production. When considering the use of SOI for microprocessors, Intel has long claimed that its benefits diminished with each technology node shrink.  The floating body cell announcement endorses the use of SOI but of course leaves a few openings for Intel to continue to walk a fine line around SOI.  The buried oxide (BOX) thickness in Intel’s device is only 10nm compared to >100nm for many SOI uses.  Intel feels that this technology is suitable for the 15nm node and beyond.  IBM and ISS both have an SOI memory technology which they’ve shown as beneficial at 45nm and 32nm process technology. 

One of Intel’s key statements is that FBC could “potentially be less costly” than embedded DRAM.  Very few companies can reach the economies of scale in manufacturing that Intel and commodity memory vendors reach.   Intel will continue to study the cost impact.   The real message here is that all leading edge logic companies endorse SOI’s technology benefits. 

-Joanne Itow

.  

So far in 2008, the NAND market has not experienced the “Apple effect” seen in previous years, despite the upcoming 3G iPhone (with up to 16GB of storage) and the SSD option for the MacBook Air.  Consumer confidence is low due to the repercussions of the US sub-prime mortgage crisis and consumer disposable income is hampered by rising prices at the gas pumps.  These factors are having a dampening effect on consumer electronics sales.  Despite slower growth in NAND sales this year, the market can still be characterized as a growth market.  NAND unit shipments are expected to reach 3,528.5 million units in 2008 compared to 2,508.6 in 2007.  NAND revenues will grow 13% in 2008 compared to 25% in 2007. 

Removable flash applications—cards for cell phones, digital cameras and other devices, USB drives—drive the bulk of today’s NAND market.  NAND growth through 2012 will be driven by the computing segment — USB drives, hybrid drives, and solid-state drives going into notebooks, UMPCs, servers, mobile and enterprise storage and data centers.  These markets will drive high gigabyte growth, which will need to be fed by NAND manufacturers.  SanDisk/Toshiba, Samsung, Hynix and IMFlash are all racing to conquer 40nm production for the cost advantages it will bring. 

The report presents a detailed NAND forecast, highlighting important market events and trends, as well as upcoming changes in capacity and an applications analysis.  NAND and NOR are compared in terms of revenues, units, gigabytes, and ASP.  Finally, quarterly price information from our NAND price survey is followed by the appendix, which contains all the data represented in charts throughout the report. An Excel spreadsheet also contains all the data in the report.

To purchase this study, or to receive a copy of the Table of Contents, including a list of tables and figures, please contact Susan Cadel at 607-368-7600 or susanc@semico.com and reference the NV101-08 or NAND Market – Where’s Apple?  To buy online, visit our website at:  http://semico.com/studies/category.asp?id=8#1086

The legal issues between Rambus and the DRAM manufacturers continues to move through the US court systems. Rambus won a patent infringement lawsuit three months ago against Hynix for the sale of Jedec-standard DRAM.  While Hynix is the first DRAM manufacturer to be addressed, Rambus is also pursuing patent infringement claims against most of the remaining DRAM manufacturers. Yesterday Rambus requested a court-mandated license from Hynix for past shipments of Jedec-standard DRAM sold from 2001 to 2007 and permanent injunction to stop Hynix from shipping DRAM into the US during the appeal process of the patent-infringement verdict.

Even if the courts decide to deny or delay this injunction requested by Rambus, the possibility of other disruptions to the US supply of Jedec-standard DRAM still looms on the horizon as other DRAM manufacturers take their turn in the box. 

And the alternative?  One investment company estimates that the cost to Hynix will be as much as $550 million.  Presumably most of the remaining DRAM manufacturers would also owe a similar ratio of penalties relative to their volume shipments. 

Semico’s recent report “The Introduction of DDR3 and the Rambus Tipping Point” anticipated that the issues between Rambus and the DRAM manufacturers would ultimately lead to this kind of show-down.  Semico believes that OEMs will inevitably see higher DRAM costs in the future as a result of the current litigation path that is being followed.  Other short-term impacts are also identified in the report. 

There is no such thing as too thin, too rich, too much horsepower or too much data transmission speed.  That is why semiconductor companies spend millions of dollars developing faster ICs, network processors and countless other ways of increasing transmission speed.  But, there might be a less expensive alternative. 

 On June 23, 2008 NetExÒ, announced its HyperPipe™ bandwidth optimization technology enabling ISPs, phone and cable providers to deliver next-generation wideband services over existing broadband IP networks.  HyperPipe™ is a software solution that accelerates traffic on an IP network.  The software intercepts TCP packets, which are then aggregated and sent over the network optimized to achieve greater throughput.  HyperPipe protects data packets from variations in line quality conditions that cause latency, jitter, and bit errors that constrain the speed of typical IP networks by requiring packets to be re-sent.   Semico Spin 

Although the just announced HyperPipe™ technology applies to wired or fiber optic connections, there is no reason that the underlying technology could not be applied to wireless networks.  The need for speed on WiMAX or other wireless networks could be satisfied with much less expense than by using faster, more expensive semiconductors manufactured on advanced technology nodes.