WordPress database error: [Got error 134 from table handler]
SELECT * FROM wp_bas_visitors, wp_bas_refer, wp_bas_ua, wp_bas_os WHERE referer = referer_id AND osystem = os_id AND useragent = ua_id AND lasthere > DATE_SUB(NOW(), INTERVAL 20 MINUTE) AND visit_ip = 65998638 AND ua_string = 'claudebot' ORDER BY lasthere DESC LIMIT 1

The Cargo Cult of Business » Won’t Get Fooled Again

Won’t Get Fooled Again

Published on 9 Jun 2006 at 1:37 am | No Comments | Trackback
Filed under The Engine Room, Technopolitical, Information Technology, Economics and the Economy.

There are some ideas, and some predictions, which sound appealing and yet, over and over again, manage to not come true. Will they, some day? Some day is a long time, and experience teaches a person never to say never, so let’s define an upper bound: current thinking says the sun will explode in about 8 billion years, so if we can safely say something hasn’t happened by then, it ain’t gonna happen.

Armed with a pint of Fuller’s finest ale, this writer predicts that among the things that will not have happened by then are:

- video conferencing as a popular, ubiquitous means of communication.
- depletion of petroleum supplies.
- the death of Moore’s Law
- multi-chip packages, where multi is greater than two.

The first two are obvious. The last two require a bit of explanation. As the supply of ale is short, I shall focus on number 4 for now, and leave Dr. Moore for another day.

Almost since the invention of ICs (and to some extent before), people have thought it would be neat if you could package up a bunch of silicon chips into a package, and then encapsulate it, rather than packing each chip individually. It would indeed be neat, and useful. The latest proponent is a nice fellow from Georgia Tech named Rao Tummula. Writing in the June 2006 issue of IEEE Spectrum, he waxes eloquently about the potential of this technology. “By 2010,” he says, “the more than Moore’s Law movement, which focuses on system integration rather than transistor density, will lead to revolutionary megafunction electronics.”

Sounds good. The article continues in a somewhat breathless tone, rather more Popular Science than peer-reviewed science. It shows a picture of a device with silicon and gallium-arsenide devices, featuring both optical and electrical interconnects. And indeed, it would be a wonderful thing to build. But….

Among the challenges are:

- thermal stress management
- heat dissipation
- component fabrication precision
- design validation and verification
- testability

Dr. Tummula goes on to say “we need not compromise speed, cost, time to market, or reliability.” Wow. The flip side of that is the old proverb “Good, Fast, Cheap – pick any two.”

So why won’t this nirvana come to pass? There are a number of challenges to be solved. Moreover, many bright people have worked hard on solving them, and founded companies based on the current state of technology, and failed. Utterly. Why?

One reason is commonly referred to as Known Good Die, or KGD. ICs are tested, via probes, and the wafer level, then diced, then packaged. Packaging affects the exact electrical characteristics of the die; parts can be faster, or slower, as a result of packaging. As a result, it’s tough to design a circuit that depends on specific electrical characteristics of the device. Designs which work when built with discretely-packaged parts may not work if built with bare-die multi-chip modules.

Another factor is thermal stress. Silicon has a coefficient of expansion. Other materials are different. Direct bonding of silicon to such materials leads to mechanical stresses and chip death. Plastic IC packages absorb the difference. You can use silicon as a substrate, but silicon, even ultra-pure, is enough of a conductor to be sub-optimum.

It’s possible to fabricate resistors and capacitors in the substrate materials. However, it’s not easy to fabricate them to the levels of precision we’ve come to expect. Once upon a time, 20% was considered the standard tolerance for most resistor and capacitors; today 5% is the benchmark for discrete parts. Thin-film techniques can’t yet match that.

Furthermore, most system designs still involve a bit of cut-and-try, in one form or another. Discrete technology allows rework, integrated technology doesn’t. Wrong capacitor value? Fab a whole new substrate.

Individually, these are all solvable problems, and indeed, there are specialty makers of hybrid modules that prosper even today. Why, then, do I predict another round of failure?

The companies that survive are in niche markets, mostly military. These markets can afford the cost. No company that has ventured into the mainstream commercial market has survived. MMC? gone. Strand Interconnect? gone.

Bet not your money. Better to invest in a pint and stick with Moore’s Law, the end of which is not in sight.

-- Oliver
Computer Recyclers
Pre-Owned Macintosh Computers, Parts, & Service
Clearance
iMac G4s & G5s
10% off on all PPC imacs in stock during February.

Comments are closed.

Blogroll

Technopolitical

Networking Technology

General Interest

Design, Interface, and Usability

Business and Corporation Related

Blogosphere

Apple Computer Related