Behind the Boom:
It all comes down to Moore's Law
By Dean Takahashi
06/15/98
The Wall Street Journal
Page R4
(Copyright (c) 1998, Dow Jones & Company, Inc.)
The nascent boom in digital consumer electronics is being driven by the compounding of Moore's Law.
For three decades, Moore's Law has held that the number of transistors on a chip of a given size, roughly a gauge of chip performance, doubles every 18 months or so. Named after Intel Corp. Chairman Gordon Moore, the law is the cornerstone of the process by which electronic goods are getting cheaper, faster and better -- to the point that some gadgets, such as calculators and cellular phones, cost a fraction of what they did when they first hit the market.
Now, that process is pushing the electronics industry to new planes. The number of transistors that can dance on the head of a chip has become so vast that each doubling allows for not merely incremental improvement, but fundamental change.
Designers are getting better at placing many formerly separate functions on a single chip. Such feats of integration already allow advanced video-game players and digital cameras to have all of their capabilities on one silicon chip. And manufacturers promise more sophisticated "system-on-a-chip" devices in a year or two, including personal computers that have all their functions except memory on a single chip.
Here's how it happens. Chip makers use finer and finer lenses and etching materials to miniaturize their circuits -- collections of transistors that perform specific functions of a device -- thereby shrinking the distance between circuits. Leading-edge chips now have circuits that are 0.25 micron wide, or roughly 1/400th the width of a human hair.
With such narrow dimensions, electrical signals have shorter distances to travel. Because the signals can reach their destinations much faster, the chip's speed multiplies. Designers can reduce costs by making the chip smaller and expand capability by cramming more circuitry with new features on each chip.
It's no surprise that, despite weakness now in the memory-chip sector, the world-wide chip industry is forecast to grow to $232 billion in annual revenue by 2000 from $162.6 billion in 1998, according to the Semiconductor Industry Association, a trade group in San Jose, Calif.
Nowhere is the impact on consumer electronics more visible than in the progress of chips that encode video in "real time," or instantaneously. Just as cheap chips helped turn office copiers into consumer goods, the phenomenon of chip integration is transforming expensive encoders for the television industry into personal broadcasting devices for consumers, an entirely new product.
Massive computing power is necessary for video encoding, which is crucial to shaping video data so that it can be transmitted efficiently over the airwaves or cable into homes. An encoder scans blocks of video for redundancies that can be compressed into a smaller amount of information that can be reconstructed by the TV set without any loss of clarity.
The pioneering video-encoding concern C - Cube Microsystems Inc. in Milpitas, Calif., created its first encoder-chip set in 1993 under the MPEG-2 standard (for Motion Picture Experts Group, the industry committee that sets standards for video transmission). The eight-chip set -- which was used by Hughes Electronics Corp.'s DirecTV unit to broadcast the first digital satellite-TV images -- cost $23,000 and was manufactured with a process that created circuits 1.0 micron wide.
In 1994, C - Cube created a 14-chip MPEG-2 encoder for an improved format, at a $10,000 price. In 1996, C - Cube integrated the chip set into seven chips costing $6,000.
By 1997, C - Cube used a 0.35-micron manufacturing process to put both the encoder and decoder in a single $1,800 chip. This year, the company expects an 0.25-micron version of the chip to be cheap enough to go into a $300 add-on board for personal computers.
At that price, the video-encoding board could function as a digital videocassette recorder, taking a live video feed or a movie playing off a digital video disk and storing it as data on the computer's hard-disk drive. As such, the technology promises to turn the average PC user into a video editor or Internet broadcaster.
"Whenever the price of a technology falls below $300, it finds mass-market applications," says Alex Balkanski, chairman and chief executive of C - Cube .
In a demonstration of C - Cube 's newest chip earlier this year at a hardware-engineering conference in Orlando, Microsoft Corp. Chairman Bill Gates said: "This technology has moved much faster than I ever thought it would."
Indeed, Moore's Law is turning into a vortex. By 2005 or so, companies should be able to put a billion transistors on a microprocessor chip -- depending on the mix of memory transistors for storing data and logic transistors for processing that data -- compared with about 7.5 million on Intel Corp.'s fastest chip today.
Single-chip systems that pack all the sophistication of a supercomputer from a decade ago onto a thumbnail slice of silicon are set to transform consumer gadgets in every market. Brian Halla, the chief executive officer of National Semiconductor Corp., a Santa Clara, Calif., chip maker, often talks of PCs so cheap that they can be buried inside the dashboards of cars or the innards of refrigerators, sensing performance and allowing machines to converse with their owners.
Credit cards -- already getting smart -- will get even smarter, perhaps with features such as touch-sensitive fingerprint identification. And, says C-Corp.'s Mr. Balkanski, "we think cheap encoding will lead to a more intelligent television, something that subsumes a lot of functions of other devices."
Even when a technology moves to a single chip, there is still room for improvement. TeraLogic Corp. in Mountain View, Calif., is developing a next-generation video-encoding method -- compression algorithms that take a snapshot of an entire image, rather than just a block within it, and then look to eliminate redundancies across a much larger amount of data. These so-called wavelet-based compression chips could work five or 10 times as efficiently as MPEG-2 compression, allowing for faster processing.
"If you design the chip to be more efficient at processing," says Peng Ang, chief executive officer of TeraLogic, "then you make better use of the additional transistors you get from Moore's Law."
---
Mr. Takahashi is a staff reporter in The Wall Street Journal's San Francisco bureau.
---
Digital Driver
Moore's Law at work, as seen in the the number of transistors on
Intel Corp.'s newest microprocessor:
YEAR Transistors
1972 3,500
1974 6,000
1978 29,000
1982 134,000
1985 275,000
1989 1,200,000
1993 3,100,000
1995 5,500,000
1997 7,500,000
Source: Consumer Electronics Manufacturers Association
|
