SemiSerious

It is a perplexing thing that semiconductor technology requires so mucn investment and innovation, but it is a simple mechanical design that makes one product more useful or attractive to the consumer than another. Think about the swivel disk USB drives, or stiletto type where the connector is pushed out of a sleeve. Convenience is something that is not about the technology node or wafer size but whether your flash drive has a cover you can’t lose.

This confuses the semiconductor analyst and many of the engineers who cram features into smart phones but find little interest outside of the tech community. It’s a problem with a long history amongst techies and geeks. The products we often drool over seem to find little broad-based market appeal. I think it’s high time to tip our hats to the mechanical engineers and designers out there. These are the people that design fantastic plastic cases and switches that not only house the insane feature sets going into consumer products but sometimes even stay working well into the devices’ obsolescence.

Getting back to the USB drive for a second, I shudder to think what ancient technology my 256MB drive (received free at a conference) is built with. But it stores much more than I need to  quickly transfer files between work and home. (I hope my manager is reading this!) But what I might appreciate are the really tangible things. How it feels in my hand or how attractive it is could influence future purchases.

That look and feel is the result of the industrial designers and mechanical engineers. Consider the Sandisk Cruzer Contour drive. It even comes with an atractive faux leather carrying case. But it’s the hideaway USB connector that is the killer technology.

But innovation of design doesn’t require quite so much flash (excuse that pun). Even the Memorex Traveldrive has a cap that stays with the unit so you can’t lose it. I wonder now if there are any retractable pen patents still alive.

Steve Jobs may have a golden touch, but who would have ever connected Apple’s iPhone to the sixties Bond classic, Goldfinger? Well, some folks at Cambridge Silicon Radio (CSR) did. Some CSR chip designers were either great James Bond fans or not quite fans of one particular project. (I trust they weren’t talking about Steve.) The design is the Bluecore 4 chip found in the iPhone that provides the iPhone’s Bluetooth capability. One of the die markings is “Oddjob” and there is also die art of that character from Goldfinger. Being an admirer of the Bond series more than a fanatic, I have to thank CSR for adding the written reference to Oddjob.

But there are even some technically interesting features of one of the iPhone’s radio modules. The Wireless LAN (WLAN) IC is particularly interesting. The Marvell 88W8686 transceiver is the first device on the market that integrates RF functionality on a 90nm logic IC.  (The 90nm technology node is confirmed by a 6T SRAM cell size of 1.14 square microns.) The Marvell 88W8686 chip is advertised as the world’s first ultra-low-power, 90nm (WLAN) single-chip solution. The device integrates an ARM compliant CPU, and high-speed serial host interfaces. The advanced RF transceiver supports the IEEE 802.11 a/b/g standards. Marvel uses a thick Media Access Control (MAC) architecture in their design to offload the host CPU resulting in lower power consumption and increased system performance. I suppose it is no surprise that we gave it an Insight Award for innovation this year.

It is interesting to note that while both the Marvell 88W8686 and the CSR Bluecore 4 die are found separately on the iPhone, they are again bundled as a WLAN and Bluetooth system-in-package (SiP) solution in the Wi2Wi W2CBW003. The Marvell 88W8686 device provides for both traditional bond wires as well as flip-chip packaging options to allow the form factor of the SoC to be as small as possible depending on the users specific bonding requirements.

A third option allows for wafer scale packaging or mounting another packaged device directly onto the large metal areas before going ahead with wire bonding to the traditional IO ring. These bump pads are 200 microns in diameter and are placed throughout the chip. I guess it’s that package-on-package flexibility that makes it right at home in the iPhone.
 

Confusion abounds about the manufacturer and type of microprocessor found in Apple’s iPhone. With over a million units shipped to date and Apple on the verge of launching their revolutionary computing platform into Europe, the timing is right to reveal the “secret.” Actually, I am just revisiting Semiconductor Insights’ revelation from July as reported in EETimes. Our own Greg Quirk was quoted identifying an Apple branded Samsung stacked die package containing the S5L8900 processor.

Although nearly three months have passed since Quirk and his team first laid bare the guts of the iPhone, a search for “iPhone processor” information still yields primarily speculation and conclusions based on software tests. I’m hopeful this post will get Google taking Semi Serious a lot more seriously.

But the source of the iPhone’s brainpower is not the end of the story (fortunately for this scribe). The chip Samsung supplied to Apple contains a significant first. The iPhone processor is the first package-on-package or PoP device with significant market presence. Unlike earlier devices that contained multiple die that were stacked, bonded, and then encapsulated with plastic molds, the Samsung processor chip takes two independently packaged devices and stacks them. This decreases the board footprint of an apps processor in the same way a single package with stacked die would. However, the package gets taller. The Samsung PoP configuration is 1.3mm high (not including solder ball height at the system PCB interface). Most of the difference comes from the inter-package solder balls that are about 0.35mm in diameter. That’s still a fairly low profile package and a very small price to pay since 1.3mm presents little if any challenge for Apple’s contract manufacturers. After all, the iPhone includes a camera and the image sensor module is typically the limiting factor in keeping the final product thin.

There is a big upside to using PoP. It allows package and test of the cheap commodity DRAM independent of the more precious high performance logic IC. This creates flexibility in the test programs and allows a kind of “final” test on each packaged device before marrying them into the system-in-package (SiP). But there is a big manufacturing cost advantage because of de-risking potential loss of the logic IC. An expensive advanced process logic IC could easily end up in the junk pile because of the relatively complex assembly process required to stack memory and logic in the same package.

But wait. There’s even more to the Samsung apps processor at the die level. The device is manufactured in 90nm technology. That’s not cutting edge, but there is DRAM embedded on the die. Samsung integrates approximately 1.2M of e-DRAM along with about 375K of the more traditional SRAM SoC memory.

Has Samsung concluded the SiP versus SoC debate by bundling them all together in a single device?

Recently, copper appeared in the news as a replacement for traditional gold bonding wires due to the rising price of gold. War and upheaval has a way of attracting investors to gold. Looking at the 30-year history of gold prices, there are spikes in 1979 when the Soviet Red Army invaded Afghanistan as well as in 1982 during the Falkland Islands War. More recently, gold was the only winner following the US invasion of Iraq in 2003.

For obvious reasons, gold has always been viewed as an expensive commodity. Therefore, there were several groups working on more cost-effective alternatives such as copper dating back 30 or more years. Several issues have held copper back. Foremost among these is copper’s tendency to continuously form an oxide. Gold, after all, is the most inert metal there is. To achieve acceptable yields in the bonding, assembly would likely need to move to much cleaner environments than those used for gold today.

No doubt, there are many good engineers out there who could overcome the shortcomings of copper as a wirebond material. Ironically, it is probably cost-effectiveness that will stop it. Gold is more expensive, but it benefits from a 40+ year history on the job. It would take an unprecedented level of extremely costly retooling to get copper into the mainstream. It is hard to imagine even a copper wirebond tool maker getting a machine into production, let alone widespread adoption by assembly houses. Gold will not be displaced by copper, at least not for wires.

What will change in the industry is the decision point for moving a design from wirebond to flipchip. Typically today, flip-chip is not considered to be a viable option below about 200 pins. If gold continues the trend to be a more significant portion of the manufacturing cost, we will see lower I/O count devices avoiding the wirebond route. In this case, copper pillar technology might be used to lower the overall cost of the chip. Solder bumping devices that use copper back-end metallization on the IC require some special additional metals - under-bump metals or UBM - to ensure low resistance die-to-package connection. Therefore, copper pillar technology such as used in the Intel 65nm D 920 Dual Core MPU, might squeeze out the solder bumps usually used today. So copper may still have a future as a cost-saving measure in the packaging industry. 

The crisis in Lebanon seems to have subsided for now, but either way, there is an impact on the markets. The news from Wall Street this time seemed to be that investors have become more hardened to the idea of major crises. This left the Dow Jones relatively stable, and gold prices actually declined slightly.

It’s doubtful that the price of gold represents the end of the world as we know it, and it’s up to you whether you buy or “share” the REM song of that name. But that brings up the whole issue of digital rights management which will have to wait for another day.