Above: Dummy features?identified on TI OMAP 2420 Applications Processor (Source: Semiconductor Insights)    More Information Download a complimentary Insight Report about Texas Instruments OMAP2420 Applications Processor (requires registration)

Aren’t fruits and vegetables by definition organic? DFM - or design for manufacturability - is a common buzzword these days. Am I alone on this one too, or were we not always designing for manufacturability? If not, why did designers even use design rule manuals?

What really happened is that the semiconductor industry reached a peak where the planets aligned and there was so much standardization and repeatability in process technology and design tools that creating working IC?s became relatively simple. No one would say that getting a state-of-the-art IC to work today is a simple task. Why? Repeatability has been replaced by process variability across the wafer and within the IC. This is not suggesting that the fab guys have turned their attention to recreational drugs and the technology has gone to pot. Moore?s Law has simply pushed our industry so close to the physical limits that purely statistical fluctuations create many of the challenges to achieving acceptable device yields.

Enter DFM.

I don?t think DFM would be the media darling it is today if not for the massive industry shift to fabless or fab-light models. Why? Responsibilities are split across corporate boundaries. You create a design, then the corresponding layout files that can be written onto a mask. They take the mask design, print and pattern the wafer. What if the circuit operates below target specs? What if clock timings are out of sync? What if the test chips you get are complete duds?

Well, if your design data has not just been thrown over a wall, but also over the largest ocean on the planet along with some major language barriers, chances are that something will go wrong. So count on a long cycle from conception to volume production of working chips. DFM ? whether it?s a new EDA tool or simply a new teamwork process ? will doubtless be the difference between functional devices and a pile of useless junk. But the real success of the design will ultimately be measured by how well you hit the market window for your product.

Perhaps DFM needs to be design for marketability.

From our analysis, a lot of DFM is starting to ?stick.? Control and consistency of CMP process has required so-called dummy features for a long time, but these are becoming more critical as low-k dielectrics become softer. Dummy metal fill is also necessary to keep the chip’s layers from peeling apart.

Free Report Download an Insight Report about this device from our website.

It seems like a long time ago in a job far far away, but I do remember the infamous memory bus wars between Rambus DRAM (RDRAM) and the Synchronous Link DRAM (SLDRAM) interface.  At the time, both were going to save the day by alleviating the bottleneck between CPU’s and DRAM on motherboards.  As so often happens, the winner (RDRAM) was ultimately determined by allegiances, market clout, and politics more than technical merit.  Once Intel threw its weight behind Rambus, SLDRAM quietly sulked it’s way to wherever all those Beta video tapes went.  Of course we all know that RDRAM died a couple of years later but that’s another story. 

Fast forward a couple of years and another bus-war emerges, pitting RapidIO against PCI-Express.  Both replace traditional Northbridge chips with switching fabrics, sometimes integrated into the CPU.  I won’t put my neck out and pick a winner for fear of losing friends.  Let’s just say I know which side the borg, I mean, Intel is on – “resistance is futile”!  RapidIO will probably enjoy healthy “niche” status in PowerPC systems, likely in the comms space primarily.

As someone once said, “the beauty of standards is that there are so many to chose from”.  A new bus-war has emerged, this one in the mobile space rather than the desktop/server market.  The companies who market display drivers and LCD controller ICs today are enjoying all the pleasures (high volumes) and pitfalls (low margins) typical of a mass market.  Display driver IC volumes are now comparable to many memory types – just think a moment about how many DRAM chips versus display drivers you personally own.  There are probably ten to fifteen in your computer monitor alone.  Add to that your new flat panel TV, a couple of cell phones, your MP3 player, your car’s navigation system, and you own a lot of display driver ICs.

There are many features that used to differentiate these chips in the market; things like internal display RAM, picture-in-picture, fancy gamma correction, and integrated secondary display drivers, not to mention pixel density and colour depth of course.  The problem is that these design elements are largely digital and easily integrated into the next spin.

Recent standards have emerged that seek to address a significant design challenge in mobile phones: The interface between the controller IC and the display driver circuits, which is often through the hinge of clamshell phones.  All these new intra-panel interconnects are designed to send more bits (pixels and colours) over a narrower interface, reducing the amount of chip-chip interconnect between the controller and the panel, reducing the size of the bezel around LCD displays, and perhaps most importantly, reducing system costs.

In April NEC announced with the launch of their µPD106290 240RGB source driver with a whopping 68 billion colours (12 bit).  The new driver uses an interface called Point-Point mini-LVDS (PPmL), which has also been embraced by TI and Thine for their timing controller chips.  This is presumably to replace the “Current Mode Average Differential Signalling (CMADS) scheme announced by NEC in 2004.  The first CMADS drivers were launched in November 2005, and the first handsets just reached production in February of this year on the NEC FOMA N902i handset.  This tells you something about how long it takes a newly announced standard to appear in the palm of your hand.

The NEC announcement follows February’s announcement by Qualcomm that many of their CDMA chipsets will incorporate the Mobile Display Digital Interface (MDDI).  Seiko-Epson, Toshiba, and Renesas have hitched themselves to that wagon with display drivers that communicate across this new, narrow link.

Both NEC and Qualcomm are of course, trying to outdo National Semiconductor, who appears to have good traction with their “Point-to-Point Differential Signalling” (PPDS) scheme launched last October.  It has since been licensed by ST Microelectronics, Magnachip, Sharp and HiMax.

And of course, don’t forget last month’s announcement by Solomon Systech, launching their first products with the “mini-RGB” interface.  At this point mini-RGB only supports QCIF formats (176RGB x 144 to 220), so I think it falls a little bit short.

PPDS, mini-RGB, MDDI, CMADS, PPmL.  M-O-U-S-E.  I think the winning display driver company will be the one who comes up with a multi-lingual interface, since they all appear to be based on some variant of differential LVDS signaling.  Time will tell.  It would all be so easy to predict if Intel made cell phones.

I bet if Sony had used RDRAM and RapidIO in their video cassette recorder architecture, Beta would’ve won.  Perhaps not.

Die Photos showing Advanced Process Lithographies (click to enlarge) 

Analyzing devices has become increasingly difficult over the past years with the advancement of technology.  Previously, anyone with some tools and know-how could take a semiconductor device, rip it apart, and find out some technical details about it.  Two decades ago, feature sizes were measured in microns and an advanced process was 5um.  Today, advanced devices are pushing 50nm, a 100x improvement.  Optical analysis is no longer a feasible solution, as the tools, such as a microscope, would need to be conisderable better than one would typically have in a garage or personal lab.
 
SI has taken steps to be able to keep ahead of the technology curve in the semiconductor industry, and have successfully analyzed advanced integrated circuits in 70, 65, and 50nm lithographies from Toshiba, Samsung, and IM Flash Technologies, respectively.  This has required, and continues to require, a sizable investment of time and resources into pioneering new techniques for sample preparation, imaging, and circuit extraction to be able to handle the most advanced devices.  For our effort, SI has been awarded 6 US patents to date, with several more pending.

While companies understand the technology they utilize, SI has analyzed their technology, and that of all of their competitors.  With thousands of completed analyses, SI is in a unique position to identify, understand, and comment on innovations in the industry.

For years now, we have been subjected to prediction and speculation about the emergence of mainland China as an economic superpower. Several semi foundries were built there with leading edge 300mm wafer capabilities, and it seemed very much like the Chinese would pull much of the world’s IC production out of other sites in Asia.

China has been ramping up procedures on protecting their intellectual property rights.  However, there is still work to be done.  For example, TSMC successfully litigated against China’s pre-eminent foundry SMIC in January 2005.  But on August 27, 2006 TSMC filed another patent suit against SMIC claiming infringement regarding the settlement.

But we should not overlook several key points that have kept other countries on the competitive edge until now. Many companies realize that there are other excellent options in Asia.

Recently, I traveled to Singapore and had the chance to see first-hand the traits that might slow the rise of the Chinese Dragon. Here are some key points:

• Busiest port in the world (CIA world Facts)
• Close proximity to major producers and markets of India, Malaysia, Indonesia, Taiwan and China
• Cosmopolitan feel with 20% expatriate population
• World #3 (forecast) - pure-play foundry

Chartered once seemed weakened by competition that included China, but now seems to be on the march. Singapore’s foundry recently displaced SMIC from the number three spot in the IC Insights world foundry rankings (forecast for 2006)

Chartered worked very hard over the last two years to align themselves with important business and technology allies (think IBM and AMD). MPU shipments for AMD are ramping, and they are increasing output of video game processors for Microsoft’s X-box machines. Moreover, Chartered is expected to begin producing high volumes 65nm chips for TI beginning next year.

Soitec recently announced plans to build a 300mm fab in Singapore Soitec’s CEO cited increasing demand for advanced gaming chips like the X-Box IC manufactured at Chartered as a driver for advanced SOI technology.

Singapore is blessed not only by its prime trading location, but also its complete sheltering from tsunamis and other disasters that are fairly common in South East Asia. I think they will find a way to stay sheltered from the Chinese storm as well.

Singapore may not be recognized as a new Asian tiger, but the history and mythology of the Merlion  and the its naming as the “Lion City” suggest that it will not go quietly without putting up an epic fight.