[Note: More capering struck out.]
Wednesday, April 22, 1998
by Rob Landley
Austin, TX (Apr. 22, 1998) --
Yesterday, in my exploration into how
a microprocessor gets made, I started out by confessing the following, "Intel
may well be the most attractive Cash-King business to buy right now (I think
And today, the market responded. As of this writing, the stock is up over
$5, just topping $84 per share. Of course, volume is ringing in at 30 million
shares, or over $2.4 billion in dollar volume on the day, so I can't really
take credit for this one. Or can I?
In yesterday's report, I gave a fairly technical explanation of microprocessor manufacturing. Now I'd like to talk about a key element of Intel's business model: Churning out faster and faster chips on a frequent basis. Intel's core products are dedicated to speeding up computer tasking; likewise, Intel's mission is to speedily deliver one great chip after another. (Note: We in the computer industry have found that very large amounts of caffeine really help with understanding all this stuff. Your mileage may vary.)
Let's begin at the beginning. If Intel couldn't make faster chips, it would be in trouble. Serious trouble. Why? Well, microchips aren't like light bulbs. A microchip can easily last for decades. So if Intel wants to sell millions of them every year, the ones it makes this year have to be so much better than the ones it made last year that consumers are willing to literally throw away what they've already been sold. To date, the most effective way to do this is to dramatically increase the speed of each new chip.
The speed at which a microchip runs is controlled by a clocking crystal, which is basically a little fragment of quartz. Quartz has this nifty property called "piezoelectricity" -- a word that you might want to toss out at the next cocktail party.
Piezoelectricity? Think of the quartz crystal as an electric sponge. If you squeeze the crystal like a sponge, it gives off a tiny jolt of electricity. Conversely, if you shoot electricity into the crystal, it swells up. So the quartz acts as a vehicle for electricity. And, it turns out that if you cut and polish one of these crystals just right, you can get it to vibrate when you run an electric current across it. The result is the creation of an electrical pulse emanating at regular intervals, as the sponge-like crystal swells and shrinks with electricity.
Now, have I gone off the deep end? Am I re-living old television episodes
of Land of the Lost? Will I next suggest that you wear quartz around your
ankles to ward off evil spirits? No. Or at least, not yet! The reason I'm
introducing all this is that it is the electrical pulse in the quartz crystal
that controls your computer's speed. This "clock pulse" tells your
central-processing unit (CPU) when to execute the next instruction. And the
speed of the pulse is measured in "megahertz," with 1 megahertz equaling
1 million pulses per second. Close your eyes and think about this for a second
-- Dell just introduced a 400 megahertz computer. That's 400 million electrical
pulses per second. Zing!
Since I can't imagine that any of you have fallen behind here, I'll press
The clock tied to this pulsing electricity is there because the speed of electrical flow has to be measured and managed. Quartz crystals are cheap, and it's easy to make a really, really fast clocking crystal. But it takes a certain amount of time for an electrical signal to go down a given length of wire. And if you feed in a clock pulse that's too fast for a given circuit, that circuit will start missing signals, skipping operations, producing incorrect results, and generally going nuts. The center cannot hold; things fall apart.
Now, Fools, let's bring it all home to roost and explain why the business
that Intel has chosen to be in is a great one. Our microchipper has got the
laws of physics on its side. A critical aspect to speeding up computers is
the shortening of wires to enable electrical pulses to whir from point to
point rapidly. If you make a wire shorter, current gets from one end of it
to the other faster. And, if you make a microchip smaller, all the wires
in it are shorter. You see, the smaller the microchip, the faster it can
And for Intel, investing in smaller and smaller products yields significant
cost savings over time. It's why Microsoft doesn't make personal computers.
It's why Coca-Cola doesn't own supermarkets. It's why Pfizer doesn't buy
hospitals. These companies concentrate on creating lighter and lighter products,
driving costs down methodically... and that's good business.
But how do you make a microchip smaller?
Here we jump back quickly to yesterday's report on Intel and the making of microprocessor chips (cf. Inside Intel). Remember that chips are made by focusing light onto a surface. Naturally, then, to shrink the chip, you just re-focus everything to make a smaller picture. That sounds easy. Piece a cake. Well, almost... but not quite.
For starters, you need a bigger lens to focus a smaller picture, otherwise you get fuzzy lines that don't etch a good groove (and thus produce bad wires that don't conduct electricity very well). Imagine being asked to put a road map of the United States, showing every sidewalk and pothole, onto the back of a dime. You'd need a REALLY BIG lens. Humongous! Think about the Hubble Space Telescope. Now think about the contact lens they put on the Hubble. Now think bigger than that, and just keep going. And going.
Building that huge lens isn't Intel's only challenge, though. After getting it all in place, it may turn out that the surface it is etching the wires onto isn't smooth enough to handle significantly smaller wires. So, Intel may have to pay chemists to come up with better chemicals to coat it with. Then weird interference patterns can crop up in the signals -- due to quantum physics, which we won't get into -- so they have hire physicists to figure out where the wires should be placed to avoid the interference. The chemists work hard to better insulate the wires and, when they can't, the physicists are brought in to rearrange the design.
After solving all those problems (and more), Intel has a smaller chip that executes instructions faster than ever before. The signals travel down shorter wires between each transistor and get to all of their destinations faster. Consequently, technologists can begin feeding in a faster clock pulse (with a higher megahertz) without losing any signals.
There are a few other ways to generate more speed out of a chip: 1) By turning up the juice and cramming more electricity down the wires (which can cause the heating problems in personal computers), and 2) through more elegant designs of the circuitry. But generally, microchips get faster as they get smaller. In the process, though, they often get much more complex. Chemistry and physics are brought in along the way to solve the manufacturing problems, and a lot of money is spent on upgrading manufacturing facilities to create even more delicate circuitry.
All of this comes together to give us Moore's Law, which states that the power and speed of computers should double about every 18 months. The beautiful thing about Intel's business is that its chief concerns are 1) attracting very bright people, and 2) investing in the plants that can make finer and finer microchips.
What are Intel's rewards? The market just demands that they make smaller and smaller product, the result of which is the methodical reduction of material costs. Yes, Intel's upfront investments are onerous, but the operational costs of designing and delivering smaller and smaller chips are light.
Ok, tomorrow I'm going to talk more specifically about the costs associated with manufacturing central-processing units (CPUs), and how this affects the pricing of Intel's finished chips. It'll be considerably less technical than my last two reports. Of course, I hope you enjoy all of them.
Finally, to see how far these CPUs have come, visit: Intel's processor hall of fame.
- Rob Landley