Worldwide Production of Semiconductors

Excepting the annus horribilis of 2001, the worldwide market for semiconductors recently passed $200 billion per year. That's about the same as the national output of Saudi Arabia or Switzerland. Figure 5.1 shows month-by-month sales of semiconductors for 312 consecutive months (26 years), from the beginning of 1976 through the end of 2001. Note that these are monthly, not yearly, figures, so sales have been in the range of $10 billion to $15 billion per month—or $400 million per day—for several years.

Figure 5.2 shows the same data graphed yearly, so the vertical scale is 10 times higher. Although some years have been volatile, the overall growth in semiconductors is evident, and has been averaging about 15 percent compounded growth per year.

Sales by Revenue

The year 2001 saw the worst downturn in the comparatively brief history of the industry, with total revenue falling by an astounding 32 percent from the previous year's level of over $204 billion. Memory chips led the decline, with sales falling by half, but all sectors suffered.

In good years and in bad, the proportion of revenue contributed by each type of component remains about the same. Microprocessors, microcontrollers, digital signal processors (DSPs), and peripheral chips account for about one-third of the total revenue (but not of the total unit volume). The next largest contributor to vendors' coffers is memory. DRAMs, SRAMs, and ROMs of all types account for 20 percent to 25 percent of the revenue share. Analog components and miscellaneous digital-logic devices each contribute 15 percent to 17 percent to the total. Discrete components, such as resistors and capacitors, and optoelectronic components, such as LEDs and optical sensors, both make single-digit contributions. These segments are summarized in Figure 5.3. Bear in mind that, barring the first few years of the 21st century, every percentage point represents about $2 billion in revenue. That's not small potatoes.

Sales by Unit

Using data from 2001 as our yardstick, the huge majority of unit sales are in discrete components. Like the insects of the semiconductor realm, these tiny devices are almost invisible, even to industry insiders. At an astounding 200 billion units per year, it's clear that diodes, transistors, rectifiers, and the like are ubiquitous. (See Figure 5.4.)

Next up, but almost an order of magnitude lower in unit sales, are the analog components. Analog amplifiers, D/A and A/D converters, voltage regulators, and numerous other nondigital components make up this category.

Optoelectronics are not far behind their analog cousins, including LEDs, laser diodes, CCD image sensors, and other things that light up. The majority of optoelectronic components are used in consumer electronics, making up the LED displays of alarm clocks, the laser diodes of CD players, and the “film” of digital cameras.

Miscellaneous digital logic components take fourth place. This category includes low-end logic chips such as AND gates and OR gates, but also much higher value devices such as programmable logic (CPLDs and FPGAs) and custom ASIC chips. These latter chips tend to be expensive so they give the logic segment a bigger slice of the revenue pie than they do of the unit-volume pie.

It's interesting that the number of microprocessors and microcontrollers sold is not much lower than the number of memory chips. Because every microprocessor typically requires several memory chips as “support” components, you'd expect this ratio to be more lopsided. Yet the number of chips in the CPU and DSP category is only about 20 percent smaller than the number of memory chips. In part, this is because the microprocessor category also includes some peripheral controllers, which aren't really microprocessors and don't require external memory chips. Many low-end microprocessors and microcontrollers have some memory of their own built in, so they often don't need extra memory chips, either. The high volume of these one-chip microcontrollers balances out the relatively few high-end microprocessors that need lots of extra memory.

Taking another big jump down the volume chart, we see the sensors. This category includes silicon acceleration sensors for automobile airbags, temperature sensors used in cars and industrial applications, magnetic sensors, and other specialized chips.

Bringing up the rear are the so-called bipolar components. Bipolar components aren't really a separate category from sensors, microprocessors, and so forth, but they are manufactured using a different process than all the other devices we've covered. Market research usually classifies them separately for that reason. As you can see, there are very few bipolar components made or sold, and their numbers are dwindling slowly.

Average Selling Prices

The average selling price (ASP) of a class of components can be both useful and misleading. It is useful because it gives us an idea of what the high-value components are and where the most value is being added. It is misleading because the statistics group many different types of components together, and not all chips within each category have similar selling prices. The mix of components can significantly skew the perception of the value of an entire segment.

Having said all that, the discrete components clearly have the lowest ASP, at about $0.06 per device. Some components, such as microwave or radio frequency (RF) transistors, raise this average while low-value components, such as diodes and rectifiers, lower it. Still, when discrete components account for nearly half of the unit volume but only about 8 percent of the revenue, you know you're dealing with inexpensive parts.

Optoelectronics are the only other major category of components with ASPs under half a dollar, at $0.37. Bipolar components and sensors both hover at about $0.50 each. Then there's a modest $0.75 for analog components, as Figure 5.5 shows.

Large-scale digital chips clearly command a price premium. Part of this is perception and part is actual component cost. Some large-scale digital chips, such as FPGAs and ASICs, have huge development costs that must be amortized across comparatively few components. Other logic devices in this same category, such as individual logic gates, are commodities, sold almost by the pound. Overall, this category commands an ASP of $1.50.

Although memory chips are sometimes characterized as the semiconductor commodity, their prices suggest otherwise. The ASP for all memory chips in 2001 was $3.25, a fair number, but one that is very volatile. Because so many memory chips are interchangeable and demand is so cyclical, price competition is ferocious. Memory production might be unconscionably profitable one year and diabolically costly the next. Each downturn tends to shake out one or two memory vendors, slowly reducing their ranks.

At the top of the pricing chain are, of course, the microprocessors. Intel didn't get to its exalted position by making commodity components. Intel's well-known PC processors account for an increasingly small proportion of total microprocessor unit sales, but its success nonetheless pumps up the average for the whole category. The only thing preventing the $6.12 ASP from being contaminated by this anomaly is the enormous unit volume of the other components.

Average Microprocessor Prices

The ASP for microprocessors and related components is so lopsided that it warrants further examination. The data in Table 5.1 are once again taken from the World Semiconductor Trade Statistics (WSTS) organization for the calendar year 2001. Although overall semiconductor sales were badly depressed in that year, the ratio of ASPs and the relationship among components is the same in any year.

In Table 5.1, the top line (32-bit microprocessors) represents only a few vendors' components. Specifically, it is overwhelmingly Intel's latest generation Pentium processors used in PCs. This segment also includes AMD's PC-compatible processors and the PowerPC processors made by Motorola and IBM, primarily for Apple's Macintosh computers. Macintosh commands about a 4 percent market share of desktop computers compared to 95 percent for the “Wintel” PC, so the unit volume (and hence, the ASP) is comparatively unaffected by Macs.

Table 5.1. The average selling price for microprocessors, microcontrollers, and DSPs varies considerably. The top-end PC processors skew the ASP for the category with their abnormally high prices. Courtesy of World Semiconductor Trade Statistics. Used with permission.

Device Type	Average Selling Price
Microprocessor, 32-bit	$92.89
Microprocessor, 16-bit	$6.87
Microprocessor, 8-bit	$3.72
Microcontroller, 32-bit	$7.57
Microcontroller, 16-bit	$4.23
Microcontroller, 8-bit	$1.44
Microcontroller, 4-bit	$0.78
Peripherals	$6.08
Digital signal processor (DSP)	$6.44
Average price for category	$6.12

Other desktop computers, such as engineering workstations from Sun Microsystems, Silicon Graphics (SGI), Hewlett-Packard, and others merely represent rounding errors in this segment. Their unit volumes are “in the noise margin,” as electrical engineers might say. As proud as these computer companies are of their high-end microprocessor technology, those processors have next to no effect on the economy of semiconductors.

With an ASP fully 13 times higher than the average for the rest of the category—which is itself the highest in the semiconductor industry—it's no wonder that Intel attracts so many imitators. Intel's processors are difficult to clone because of their aging, baroque design. Indeed, it's very likely Intel itself would have abandoned the product line years ago had not IBM serendipitously chosen it for its first IBM PC back in 1981. Intel's Pentium family has succeeded completely in spite of its technology, not because of it.

Geographic Breakdown of Production

For geographic and political purposes, semiconductor manufacturing is generally divided into four major regions: North America, Europe, Japan, and Asia and the Pacific region, as shown in Figure 5.6. Roughly speaking, production is evenly divided among these four areas, with North America (mostly the United States) maintaining a slight lead. Throughout the late 1980s and 1990s and into the following decade, the United States produced a little more than 30 percent of the world's semiconductors (measured by revenue), with the other three regions each accounting for about 20 percent to 24 percent of the total. During those two decades, Japan's share of production eroded gradually, losing ground mostly to vendors in Taiwan, Korea, and Singapore.

Because these statistics measure revenue and not units, they're skewed toward higher priced parts. Thus, the United States is somewhat overrepresented because of its preponderance of microprocessor companies. Texas Instruments, Motorola, Intel, AMD, and the dozens of other microprocessor, DSP, and microcontroller companies based in the United States contribute revenue out of proportion to their impact in simple unit volume.

The United States consumes most of its own chips, swallowing three-quarters of its domestic production. Table 5.2 shows that Europe and the Far East show somewhat less provincialism, but Japan keeps a generous 82 percent of its chips within its borders.

The European and Far Eastern regions both ship about half of their semiconductors to the United States. The exception again is Japan, which gets only about 20 percent of its revenue from North American customers; the rest stays in the country.

Table 5.2. Average consum†ption of domestic semiconductor production, 1992–2000. Courtesy of World Semiconductor Trade Statistics. Used with permission.

Region	Percent
Asia	28%
Europe	30%
Japan	82%
North America	74%

If we examine the data from the demand side, shown in Figure 5.7, North America buys about half of the world's production of semiconductors. Japanese consumption dropped in the late 1990s because of the country's economic recession, whereas other Far Eastern and European countries increased consumption slightly. The rise of the Asia and Pacific region vendors' percentage of consumption was not merely a statistical side effect of the Japanese reduction. These countries really did purchase more chips on an absolute revenue basis, not simply in relation to Japan's shrinking share.