Silicon Valley, What Next?"—In the Proper 120 That Market Lead Steadily Eroded, with the Context
Total Page:16
File Type:pdf, Size:1020Kb
MATERIAL MATTERS of $150 billion a year. The automotive in- dustry—or any industry—would love to Silicon Valley, experience growth like this. With this kind of curve, if you just hold your market share, you have wonderful growth opportunities in front of you. Un- What Next? fortunately, the semiconductor industry has been prone to price wars, something akin to what is going on in the airline industry in Craig R. Barrett the United States today. In periods of excess capacity it has been a profitless industry for many of the participants. But it is, on aver- The following text is based on the plenary Worldwide Semiconductor Revenues age, growing rapidly. The United States address given at the 1993 MRS Spring Meeting 180 was a world leader and 20 years ago held in San Francisco, April 12. 160 some 65 percent or so of the worldwide Let me begin by putting tonight's topic— 140 semiconductor market share (see Figure 2). "Silicon Valley, What Next?"—in the proper 120 That market lead steadily eroded, with the context. We can talk about Silicon Valley, 100 Japanese market share steadily increasing Silicon Gulch, Silicon Desert, Silicon Glen, 80 during the '80s. Western European market or Silicon Island: the topic is all the same. 60 share went down slightly. What I really want to talk about is the future 40 Over the last three or four years, the U.S. of the semiconductor industry, an industry 20 market share has rebounded. As I men- which only a few years ago was declared 1966 1976 1986 1996 tioned before, around 1988 many economic almost legally dead in the United States, analysts and academic theorists projected with no future. Today, that industry is alive that the U.S. semiconductor industry would and well and I hope to demonstrate that it Figure 1. Worldwide semiconductor revalue die a natural death. There was no way it serves as the foundation for the most vi- growth. Source: WSTS, Dataquest. could compete in the world marketplace; brant manufacturing industry in the world the only hope for the U.S. electronics indus- today—namely, the electronics industry. try was in design alone, not in manufactur- To accomplish this, I will draw upon a ing. Fortunately, that scenario proved to be few of the materials principles I learned Worldwide Semiconductor Market Share false. We began to regain market share in some thirty years ago. I'll talk about some of 70 the late 1980s, and the United States and the major industry trends, some of the tech- 60 Japan are now locked in a tight struggle to nology challenges, and my projections for •U.S. 41.1% determine who will be number one in the what will happen over the next half decade Japan 42.8% marketplace in 1993. I think both countries Europe 10.2% have viable competitors, and competition is or so. APAC 5.9% In the worldwide market place, the elec- going to be strong as we move forward. tronics industry is the largest manufactur- The major growth curve in Figure 2 is ing industry in the world, and by far the really from the Asia-Pacific (APAC) coun- fastest growing. It is currently estimated to tries. In countries such as Korea or Taiwan, be in the range of a $900 billion industry, 70 72 74 76 78 80 82 84 86 88 90 92 it is a priority for them to get into the semi- much larger than automotive or steel. The conductor business to fuel their electronic semiconductor portion of the electronics Figure 2. Worldwide semiconductor market businesses. APAC has gone from virtually industry is in the $60-$70 billion range, but share for Japan, the United States, Europe, no market share to a 6 or 7 percent market it really forms the basis of electronics. The and APAC. Source: Dataquest. share. They are doing to the Japanese semi- argument these days is that you can't have conductor companies precisely what these a viable electronics industry—the computer manufacturers did to the U.S. semiconduc- industry, home electronics, or any sort of Moore's Law tor manufacturers a decade ago. That is, electronics industry—without owning the they are coming into the low end of the base semiconductor industry that feeds it. business, the dynamic random access So, most governments subscribe to the phi- memory (DRAM) area, buying market losophy that you need to have a vibrant share, and making it uncomfortable for their semiconductor industry to succeed, espe- competitors. Fortunately, for companies cially if you are interested in export market- like Intel, we left the DRAM marketplace places. Electronics is the biggest export mar- many years ago. ket in the world. Many countries see the semiconductor industry as the lifeblood of the electronics Semiconductor Industry 1970 1975 1980 1985 1990 1995 2000 industry, as represented by any one of their If you look at the industry in terms of its respective critical technology lists. Key tech- revenue growth over the years (Figure 1)— Time • nologies in each major industrial block are keeping in mind that exponential growth virutally the same and they all include elec- Figure 3. The "integration trend" given by hardly continues forever—you will see that Moore's law, which shows doubling of the tronics, photoelectronics, or electronic ma- this industry has an enviable growth rate. number of transistors in an integrated terials. The fact that these technologies are Currently in the $60 to $70 billion range, it is circuit every 18 months for the past 30 virtually the same has led me to an amusing projected by the year 2000 to be in the range years. Source: Intel Corp. conclusion about industrial policy and pick- MRS BULLETIN/JULY 1993 MATERIAL MATTERS ing winners and losers. Regardless of whose Semiconductor Industry Productivity list of winners you pick, you are essentially a) b) picking the same 20 or so industries or tech- 1017 nologies, so it doesn't make any difference Quantity of Bits Produced (DRAMS) which list you pick. Just pick a list, and if you back those technologies, you probably will be backing the correct technologies for the next decade. Regardless of the source, the lists all cite microelectronics and elec- tronic materials as key technologies for the next decade, so we can take some comfort in that. Industry Trends Let's look at some of the trends that drive the industry. They are the integration trend, 0.001 the anti-inflationary trend, and the capital- 76 78 80 82 84 86 90 92 94 96 intensive trend. Figure 3 shows the integra- tion trend. Moore's Law was first formu- Figure 4. Anti-inflationary trend showing semiconductor industry productivity relative to (a) price lated by Gordon Moore, founder of both per DRAM bit and (b) quantity of DRAM bits produced. Source: Dataquest. Fairchild Semiconductor and then Intel. Moore, in the late 1960s and early 1970s, projected that every 18 months or so the number of transistors in an integrated cir- cuit would double. He kept plotting this Semiconductors: A Capital-Intensive Industry trend on his semi-logarithmic plot for years Capital Cost per 1,000 Wafer Starts per Week and he kept expecting the curve to bend /uu over, but it hasn't bent over for the past 30 2" Wafer 3" Wafer 4" Wafer 6" Wafer 8" Wafer 12" Wafer years. It still continues on its trend. If you 0.25MJ carry Moore's projection out to the year 600 2000, it predicts a gigabit RAM—a micro- processor with about 100 million transis- tors, and a computational speed giving 500 - about two billion instructions per second. i/i That will be your standard desktop com- CZ „ / puter. | 400 Figure 3 shows how Intel microproces- sors follow Moore's law, but the same curve applies to static RAMs, DRAMs, Motorola 300 - 0.4M/ microprocessors, etc. They are all roughly parallel. There are no physical limitations that would prevent Moore's law from con- 200 - // tinuing. As I will point out later, to maintain Moore's law, what is needed is a few billion V dollars of research and development, and 100 - \ much of it is materials oriented. 10M 8M 5M 2M Figure 4 shows the anti-inflationary n •I i trend. From the production of the first com- 1970 1975 1980 1985 1990 1995 2000 mercially available DRAM (the 1103) in 1971 or 1972, to today's four-megabit or 16-mega- •Year bit DRAMs, the cost per bit has gone down by over a factor of 1,000 (Figure 4a). I con- Figure 5. Growth of capital cost per 1,000 wafer starts per week in the semiconductor industry. sider that to be relatively anti-inflationary Source: Intel Corp. compared to most things today, including my tax rates. Figure 4b shows the number of DRAM bits that have been produced. This year industry is shown in Figure 5. We judge lion. Recently, we announced the building there will be something like 1,016 memory capital intensity by the cost of building a of such a plant in Albuquerque. A 5,000- bits produced worldwide. If you divide that manufacturing plant. Typically, we build wafer-a-week plant, building chips on by the number of people—every man, manufacturing plants that produce about eight-inch wafers with a 0.5- or 0.4-micron woman, and child on the face of the earth— 5,000 wafers a week. Today, it is 5,000 eight- line-width dimension, has a capital expense you'll come up with a startling statistic: This inch silicon wafers a week.