Apr 30 2014, 7:12pm CDT | by Forbes
From time to time people ask me why chips cost so much, and can’t anything be done to make them cheaper. The argument goes like this:
Silicon is the second most abundant element on the earth’s crust, making up about 28% of its mass. It’s the basis for most of our rocks, clay, and sand. This implies that it should be extremely cheap to make a chip. After all a 300mm silicon wafer that is used to make dozens to tens of thousands of chips weighs only about 100 grams and a 50-pound bag of sand at the Home Depot weighs more than 200 of these wafers and sells for under $5. Why, then, can Intel charge $100 for a processor chip that uses only a fraction of one of these wafers?
This sounds like a big rip-off.
This same question surrounded photovoltaics and the polysilicon shortage that industry experienced in 2007-8. Yes, silicon is one of the earth’s most abundant elements, but that doesn’t automatically make chips cheap. That silicon must be refined to a purity that was not thought possible a few decades ago.
These refining steps are costly. The bulk of the cost of a raw wafer (~$500) is in this process, and the capital equipment and energy used to produce it.
This is a pretty elaborate process. You can read about it in detail in a 46-page slideshow pdf from Simon Fraser University, but if you’re less patient you might want to look at the lighthearted ’80s-era rendition from Humphrey Products on the Objective Analysis website. It might be old, but it’s still accurate and is actually fun to read. That’s where this post’s image comes from.
But $500/wafer is a far cry from the $1,600 or so that a finished memory chip wafer costs, or the $5,000-odd cost of a finished high-end processor wafer. Of that cost half or more is capital depreciation on the equipment that converts the raw wafer to finished chips.
The lion’s share of the capital depreciation is lithography equipment, basically cameras on steroids. These cameras have lenses that are about 5-10 feet long and around a foot in diameter, and project features onto the wafer with a resolution as fine as 38 nanometers.
How big is 38 nanometers? I’m glad you asked. It’s 38 billionths of a meter, or the length of about 190 silicon atoms.
Another big chunk goes to the materials processing equipment, which consists of exotic tools like plasma etchers that very precisely burn things off the wafer, deposition equipment that puts atoms-thick layers of new materials onto the wafer, furnaces that are used to coax foreign atoms into the crystal structure of the silicon, and ion implanters which are like baby atom smashers.
All this equipment is extraordinarily pricey, costing millions or even tens of millions of dollars per tool. That’s why a modern wafer fabrication plant can cost as much as $10 billion to build and equip.
The odd thing is that if you were to reduce that capital depreciation by using lower-tolerance (cheaper) equipment then you would get fewer transistors on a wafer, and that would push the chip’s cost per function higher.
A blog post called: Why Facebook Can’t Get its Dream Memory Chip details this for flash memory. Facebook and Google have asked the semiconductor industry to design a flash memory chip that has most of the performance of flash at a per-byte price that approaches that of a hard disk drive. There’s no harm in asking!
In brief, though, that’s not going to happen. The processes in use today are the most cost-effective available, and even if specifications were compromised chip prices would barely budge.
Making cheaper transistors is not at all simple, and (believe it or not) today’s methods are the cheapest anyone can do.
Forbes is among the most trusted resources for the world's business and investment leaders, providing them the uncompromising commentary, concise analysis, relevant tools and real-time reporting they need to succeed at work, profit from investing and have fun with the rewards of winning.
blog comments powered by Disqus