Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Semiconductor Engineers in a Global Economy Clair Brown and Greg Linden University of California, Berkeley THE CHANGING NATURE OF dustry involves three distinct stages of productionâdesign, SEMICONDUCTOR ENGINEERING WORK fabrication, and assembly and packaging. Each stage has been affected differently by globalization and offshoring: The main forces affecting the nature of engineering work in the semiconductor industry are the evolution and global- â¢ Design: The design of integrated circuits is carried ization of technology. U.S. semiconductor firms are in many out primarily by engineers. The offshoring of design cases leading these changes both at home and abroad. But activities to low-cost locations has been accelerating with increased global competition, U.S. chip engineers must since the mid-1990s. continually upgrade their skills, deal with mobility among â¢ Fabrication: Wafer fabrication involves a large number employers, and rely upon their own resources, rather than of process and equipment engineers, who account for their employers, to manage their careers. approximately 25 percent of total direct workers at a At present, global competition does not seem strong manufacturing or fabrication facility (called a âfabâ). enough to undermine the positive employment and wage Offshoring and onshoring of IC factories appears to effects of the industryâs continued growth for most workers, have reached a relatively mature and stable stage. although job opportunities for older workers and those at â¢ Assembly and packaging: The final stage of IC manu- the bottom of the job distribution have deteriorated. Many facturing is the most labor intensive, but engineers overseas companies, such as Taiwanâs foundries and Indiaâs make up only 6 percent of the typical assembly plant design-services providers, complement U.S. companies and workforce. Assembly offshoring began in the 1960s, have lowered barriers to entry at a time when the costs of and assembly and packaging are now performed design and manufacturing are skyrocketing. This situation almost entirely abroad. Assembly and packaging are plays to the strengths of U.S. engineering by keeping viable not discussed in this paper because the employment the fabless start-up system for bringing innovation to market. implications for U.S. engineers are insignificant. The cost reductions enabled by Asian suppliers of fabrication and design services are also contributing to falling semicon- The semiconductor industry produces a wide range ductor prices, and thus supporting the continued expansion of products, from relatively simple discrete diodes and of markets, both at home and abroad. transistors all the way to complex âsystems on a chip.â The semiconductor (or integrated circuit [IC] or chip) in- Most market statistics reported here and elsewhere reflect âmerchantâ semiconductor sales, that is, sales to unrelated âThis paper was prepared for the National Academy of Engineering companies. A less visible share of the industry is devoted Workshop on the Offshoring of Engineering: Facts, Myths, Unknowns, and Implications, October 24â25, 2006, Washington, D.C. The paper is based on research conducted for a forthcoming book by Brown and Linden, Change â For an analysis of the globalization of assembly, see Brown and Linden Is the Only Constant: How the Chip Industry Deals with Crisis. (2006). 149
150 THE OFFSHORING OF ENGINEERING to âcaptiveâ chip design and manufacture internal to a com- like Intelâs Pentium 4, with 42 million transistors fabricated pany. This model is most prevalent in Japan but still exists on a 180 nm linewidth process, engaged hundreds of design in the United States, primarily at IBM, where nearly 50 engineers for the full length of a five-year project. percent of chip output in 2000 was for captive use. Other Functional integration has reached a point at which certain systems companies, such as Apple Computer or Cisco, that chips encompass most of the individual components that donât make or sell chips may nevertheless design them for populated the circuit board of earlier systems, giving rise to internal use. These chips may or may not be counted in mer- the name âsystem on a chipâ (SOC). SOC integration offers chant data depending on whether they are manufactured by the benefits of speed, power, reliability, size, and cost relative a branded ASIC company, such as LSI Logic (which would to the use of separate chips. be counted), or by a manufacturing-services âfoundry,â Although the manufacturing costs of an SOC are lower such as Taiwan Semiconductor Manufacturing Corpora- than for the separate components it replaces, the fixed costs tion (which wouldnât be included). All foundry sales are of a complex design can be significantly higher. A major excluded from this analysis to prevent double counting. reason is that system-level integration has drawn chip com- The work of engineers who design, manufacture, and panies into software development because system software market chips has been transformed by the continuous pro- should be generated in parallel with the system-level chip to gression of manufacturing technology, which has evolved ensure coherence. Chip companies also offer their customers for more than 30 years along a trajectory known as âMooreâs software-development environments, and even applications, Law,â the name given to a prediction made in a 1965 article to help differentiate their chips from those of their competi- by Gordon Moore. Moore, who co-founded Intel a few years tors. In a large chip-development project, software can now later, predicted that the cost-minimizing number of transis- account for half the engineering hours. tors that could be manufactured on a chip would double U.S. chip companies accounted for about half of the every year (later revised to every two years). The industry has industryâs revenue in 2005, with Intel alone commanding maintained this exponential pace for more than 30 years. about 15 percent of the market. The only U.S.-based firms Mooreâs prediction was based on several factors, such as in the 2005 global top 10 were Intel and Texas Instru- the ability to control manufacturing defects, but the driving ments, but the United States has a great many mid-size technological force has been a steady reduction in the size companies that account for about half of the top 50. Some of transistors. The number of transistors leading-edge pro- of these are âfablessâ companies that design and market ducers can fabricate in a given area of silicon has doubled chips but leave the manufacturing to other companies, pri- roughly every three years. From 1995 to 2003, the pace ac- marily Asian contract manufacturers known as foundries. celerated and the number doubled every two years. All new entrants to the chip industry in recent years have This relentless miniaturization is now reaching the mo- adopted the fabless model. lecular level. The smallest âlinewidthâ (feature on the chip Fabless revenue has grown much faster (compound annual surface) has shrunk from two microns in 1980 to less than growth rate of 20 percent) than the semiconductor industry as one-tenth of a micron (100 nanometers [nm]) a quarter- a whole (7 percent) over the last 10 years. In 2005, the largest century later. Viewed in cross-section, the thickness of hori- fabless companies, Qualcomm, Broadcom, and Nvidia, each zontal layers of material deposited on the silicon surface is had revenues of more than $2 billion. currently about 1.2 nm. For an idea of the scale involved, the The discussion in this paper of how the labor market for width of a human hair is about 100 microns, and the width semiconductor engineers, both domestic and worldwide, of a molecule is about 1 nm (one-thousandth of a micron). has been changing in response to changes in skill require- This progress has involved considerable expense for ments is based on our ongoing interview-based research on R&D, and the cost of each generation of factories has the globalization of the semiconductor industry. Since the steadily increased. By 2003 the price tag for a fab of mini- early 1990s, the Berkeley Sloan Semiconductor Program mum efficient scale was more than $3 billion. has collected data at semiconductor companies globally. In The Mooreâs Law trajectory has led to growing complex- the past seven years the authors have interviewed managers ity of the industryâs most important chip designs. The size of and executives at dozens of semiconductor companies (both a design team depends on the complexity of the project, the integrated and fabless) in the United States, Japan, Taiwan, speed with which it must be completed, and the resources available. Design teams can be as small as a few engineers, â Terry Costlow, âComms held Pentium 4 team together,â EE Times, and project duration can vary from months to years. A chip November 1, 2000. âLinewidthâ refers to the size of the features etched on a wafer during the fabrication process. Each semiconductor process genera- â IC Insights data reported in Russ Arensman, âBig Blue Silicon,â Elec- tion is named for the smallest feature that can be produced. â The Competitive Semiconductor Manufacturing Program is a multi- tronic Business, November 2001. â The revision occurred in 1975 (John Oates, âMooreâs Law is 40,â The disciplinary study of the semiconductor industry established in 1991 by Register, April 13, 2005). a grant from the Alfred P. Sloan Foundation with additional support from â Mark LaPedus, âITRS chip roadmap returns to three-year cycle,â Silicon the semiconductor industry. Further details are available at esrc.berkeley. Strategies, January 21, 2004. edu/csm/ and iir.berkeley.edu/worktech/.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 151 India, China, and Europe. We also use data from the Bureau s Â urveys, provides not only detailed educational characteris- of Labor Statistics, the Semiconductor Industry Association, tics of workers, but also occupational and industry charac- and the Institute of Electrical and Electronic Engineers, as teristics of their jobs. Thus ACS is much better suited to our well as other published and proprietary sources (e.g., indus- labor market analysis. However, the sample size for ACS try consultants). for 1996â2002 is too small for detailed analysis. ForÂ these We begin by looking in detail at data sets on employ- reasons, we look at both the OES and ACS data sets in our ment and earnings of U.S. semiconductor engineers, H-1B analysis. Because they yield somewhat different results, workers, and overseas engineers. We then discuss the factors however, we caution the reader against drawing strong con- affecting the U.S. labor market for semiconductor engineers, clusions based on either data set alone. The inconsistencies including technological change, immigration policy, and and gaps reflect a need for better data collection by govern- higher education practices. A discussion of globalization ment agencies. follows in terms of offshoring by U.S. companies, the We also use the very large Census Longitudinal Employer- availability and quality of low-cost engineers in Asia, and Household Dynamics (LEHD) data set that links employees the development of the semiconductor industry in Taiwan, and employers to describe semiconductor career paths and China, and India. In the final section we consider the outlook firm job ladders between 1992 and 2002. This enables us to for the U.S. chip-industry workforce. look at how workers form career paths by piecing together jobs offered by semiconductor firms. THE U.S. LABOR MARKET FOR ENGINEERS Employment and Earnings (OES Data) Factors that have affected the semiconductor industry in the past six years include a severe recession during 2001, a We begin by looking at employment levels and annual recovery that stalled in 2004, a large decline in venture fund- earnings for selected engineering jobs in 2000 and 2005, ing for start-ups that picked up again in 2006, changes in the based on OES data. For the semiconductor industry, we use number of H-1B visas, and a drop and subsequent recovery the North American Industry Classification System (NAICS) in foreign student applications to U.S. graduate engineering âSemiconductor and Other Electronic Component Manu- schools since 9/11. In light of these changes in government facturingâ (NAICS four-digit level 3344), which includes policies and swings in the business cycle, disentangling an relatively low-value components such as resistors and con- underlying, long-term trend in the offshoring of engineering nectors. The most relevant subcategory, âSemiconductor jobs is extremely difficult. Readers should keep this caveat in and Related Device Manufacturingâ (NAICS 334413), mind when reading the following analysis of the U.S. labor accounted for 39 percent of employees (and 45 percent of market for semiconductor engineers, as well as the discus- nonproduction workers) in the 3344 category in 2003, but sion of engineering jobs in selected countries. occupation-specific data are not available at this level of Because of inadequacies and gaps in the available data, industry detail. we use more than one source for our analysis. To identify In 2005, 2.4 million people were employed nationally in trends in the employment levels and earnings of semiconduc- âengineering and architectureâ occupations, with average tor engineers, we use two major national data sets that have annual earnings of $63,920 (see Table 1). Another 2.9 mil- different strengths and weaknesses. The Bureau of Labor lion people were employed in âcomputer and mathematicalâ Statisticsâ Occupational Employment Statistics (OES) (www. occupations, with average annual earnings of $67,100. Na- bls.gov/oes/home.htm) provides a large job sample collected tional employment in engineering and architecture fell 7.5 from establishments that report detailed occupational char- percent from 2000 to 2005, and average annual earnings of acteristics. However, comparisons of data from different these workers rose 18.2 percent (more than the CPI-urban, years are not exact because OES is designed for cross-section which rose 13.4 percent).10 Computer and mathematical comparisons rather than comparisons over time. Moreover, jobs increased slightly (0.7 percent) from 2000 to 2005, and OES does not provide educational characteristics. average annual earnings of these workers rose 15.6 percent, The American Community Survey (ACS) (http://www. slightly more than inflation. census.gov/acs/www/), a relatively new household survey The semiconductor industry (NAICS 3344) employed started in 1996 to update the census between decennial 450,000 workers in 2005, with 21 percent in engineering and architecture occupations (36 percent of them as technicians â or drafters) and 6.4 percent in computer and math occupa- The OES survey methodology is designed to create detailed cross- tions (40 percent of them in computer support or administra- sectional employment and wage estimates for the U.S. by industry. It is less useful for comparisons of two or more points in time because of changes tive positions). These two groups do not include managers, in the occupational, industrial, and geographical classification systems, changes in the way data are collected, changes in the survey reference pe- â U.S. Census Bureau, âStatistics for Industry Groups and Industries: riod, and changes in mean wage estimation methodology, as well as perma- 2003,â Annual Survey of Manufactures, April 2005. nent features of the methodology. More details can be found at http://www. â This is the broad occupational category used for engineers in the OES. bls.gov/oes/oes_ques.htm#Ques27. 10âhttp://data.bls.gov/cgi-bin/surveymost?cu.
152 THE OFFSHORING OF ENGINEERING TABLE 1â Employment Levels and Earnings for Engineers in All Industries and in the Semiconductor Industry, 2000 and 2005 2000 2005 Average Average Percentage Percentage Annual Annual Change in Change in Employment Earnings Employment Earnings Employment Earnings Architecture and Engineering Occupations (total) 2,575,620 $54,060 2,382,480 $63,920 â7.50% 18.24% âin Semiconductors 132,150 $52,100 95,520 $68,720 â27.72% 31.90% Electrical Engineers (total) 162,400 $66,320 144,920 $76,060 â10.76% 14.69% âin Semiconductors 10,050 $69,560 10,620 $82,400 5.67% 18.46% Electronic Engineers (total) 123,690 $66,490 130,050 $79,990 5.14% 20.30% âin Semiconductors 14,170 $65,400 15,700 $82,430 10.80% 26.04% Aerospace Engineers (total) 71,550 $69,040 81,100 $85,450 13.35% 23.77% Chemical Engineers (total) 31,530 $67,160 27,550 $79,230 â2.62% 17.97% Civil Engineers (total) 207,080 $58,380 229,700 $69,480 10.92% 19.01% Computer Hardware Engineers (total) 63,680 $70,100 78,580 $87,170 23.40% 24.35% âin Semiconductors 5,990 $70,780 14,440 $89,870 141.07% 26.97% Industrial Engineers (total) 171,810 $59,900 191,640 $68,500 11.54% 14.36% âin Semiconductors 12,580 $64,420 11,030 $74,250 â2.32% 15.26% Mechanical Engineers (total) 207,300 $60,860 220,750 $70,000 6.49% 15.02% Computer and Mathematical Occupations (total) 2,932,810 $58,050 2,952,740 $67,100 0.68% 15.59% âin Semiconductors 27,080 $66,660 28,770 $77,800 6.24% 16.71% Computer Programmers (total) 530,730 $60,970 389,090 $67,400 â6.69% 10.55% Software Engineers, Applications (total) 374,640 $70,300 455,980 $79,540 21.71% 13.14% âin Semiconductors 5,890 $72,680 8,250 $86,860 40.07% 19.51% Computer Software Engineers, Systems (total) 264,610 $70,890 320,720 $84,310 21.20% 18.93% âin Semiconductors 8,280 $76,660 7,090 $90,820 â14.37% 18.47% who represent 8.2 percent of semiconductor employees. by 40 percent, while jobs for software-systems engineers fell Nationally, some 12 percent of electronics engineers, 7.3 by 14 percent. percent of electrical engineers, 18 percent of computer- On average, engineers in the semiconductor industry hardware engineers, 5.8 percent of industrial engineers, and command higher salaries than their counterparts in other approximately 2 percent of computer-software engineers industries. In 2005, semiconductor industry engineers earned (applications and systems) are employed in the semiconduc- 7.5 percent more than engineers nationally, and software tor industry. Together these six occupations account for 54 engineers in the semiconductor industry earned 16 percent percent of engineering jobs in the semiconductor industry more than software engineers nationally. In any given spe- (or 85 percent if techs, drafters, and computer-support jobs cialty, engineers in the semiconductor industry had average are excluded). annual earnings of 3 percent (for electronics engineers) to Engineering jobs (âArchitecture and Engineering Oc- 9 percent (for computer software engineers, applications) cupationsâ) in the semiconductor industry fell a surprising higher than engineers in other industries. Engineers in the six 28 percent between 2000 and 2005 (Table 1, line 2).11 How- main semiconductor engineering specialties all experienced ever, if we look at the major categories for semiconductor average growth in real earnings (i.e., above the inflation rate engineers, jobs increased for electrical engineers (6 percent), of 13.4 percent for the period), ranging from 1.9 percent for electronics engineers (11 percent), and computer hardware industrial engineers to 14 percent for computer-hardware engineers (141 percent). Semiconductor jobs for industrial engineers. Note that these comparisons are not adjusted for engineers fell 2 percent, the only specialty in which job education or experience, which are taken into consideration growth for semiconductor engineers was lower than for in the next section using a different data set. engineers nationally. Of course, employment levels between 2000 and 2005 Jobs for software engineers (âComputer and Mathemati- did not increase continuously. Applications software engi- cal Occupationsâ) in the semiconductor industry increased neers experienced a dip in employment in 2004 after strong by 6 percent between 2000 and 2005, while all jobs in these employment growth in 2003, and electrical and electronics occupations increased less than 1 percent nationally. The in- engineers experienced a dip in employment in 2003 followed creases were unevenly distributed, however. Semiconductor by very strong growth in 2004. This is consistent with the industry jobs for software-applications engineers increased jump in the national unemployment rate for electrical and electronics engineers to 6.2 percent in 2003, as it converged 11â Comparison of 2000 and 2005 is not exact because SIC 367 was used for the first time in 30 years with the general unemployment in 2000 for the industry code and NAICS 334400 was used in 2005.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 153 rate, before falling back in 2004 to a more typical rate of The age-earnings profiles for the B.S. (Figures 1 and 2) and 2.2 percent.12 M.S./Ph.D. groups (Figures 3 and 4) show how the annual Overall we can say that the labor market for semiconduc- earnings of semiconductor engineers increase with knowl- tor engineers appeared to be relatively strong in the five years edge and skill levels (educational level) and experience (age) after the dot-com bust in 2000, when earnings nationally for 2000 and 2004. were mostly stagnant during the economic recovery, with The results are also given in Table 2, which shows earn- income gains going mainly to the top decile (especially the ings profiles for all three educational levels for 2000, 2002, top 1 percent). Semiconductor engineers even experienced and 2004, with earnings adjusted for inflation (in 2004 dol- better job and earnings growth than engineers in the same lars using CPI-urban).15 One cautionary note: because the specialties in other industries. Although employment for sample size for 2000 is small, the results for that year are industrial engineers and software-systems engineers in less reliable than for 2002 and 2004. Also some of the age- the semiconductor industry fell, employment for the other education groups were too small to show full results.16 four specialties increased. Although earnings growth was relatively high only for computer-hardware engineers and Returns-to-Experience electronics engineers in the semiconductor industry, all six specialties had relatively high average annual earnings Median and average real earnings increased with expe- in 2005, ranging from $74,250 for industrial engineers to rience (age) for all educational groups through the prime $90,820 for software-systems engineers. ages. After that, median (but not necessarily average) earnings declined for older workers (age 51â65). However, average earnings did not decline for older workers in any Age-Earnings Profiles by Education education group in 2000 or for older M.S./Ph.D.-level and Experience (ACS Data) workers in 2002, and median earnings did not decline for To analyze the earnings structures of U.S. semiconduc- older < B.S. workers in 2004. The general increase and tor engineers by education and experience, we use another subsequent decline in median earnings implies that these data set, the ACS (http://www.census.gov/acs/www/). We engineers typically received a positive return-to-experience calculated age-earnings profiles for three educational levels, until they were in their fifties and sixties, when earnings less than a bachelorâs degree (< B.S.), a bachelorâs degree for many of them declined. The decline can be explained, (B.S.), and a graduate degree (M.S./Ph.D.),13 using ACS data at least in part, by the number of weeks worked (Table 3). for 2000, 2002, and 2004 for a sample of workers defined as Workers older than 50 were much more likely than younger follows: workers to work less than a full year (defined, conserva- tively, as less than 48 weeks of paid work). â¢ age 21 to 65 Comparing degrees, engineers with B.S. degrees typically â¢ industry code 339 (electronics components and prod- had higher returns-to-experience than engineers with ad- ucts, comparable to NAICS 3344 and 3346) vanced degrees. B.S. holders earned one-half to three-fourths â¢ occupation codes (selected electrical and electronics, more in their peak years (age 41â50) than in their entry years software, and other engineering occupations and se- (age 21â30). Engineers with graduate degrees (M.S./Ph.D.) lected managerial occupations)14 earned 10 to 20 percent more in their peak years (age 41â50) than they did a decade earlier (age 31â40), shortly after their 12â entry-level years. Data were provided by Ron Hira. BLS redefined occupations begin- ning with the 2000 survey covering 1999, but there is no evidence that the The variance in earnings increased with age for prime- redefinition has contributed to the post-bubble unemployment rise. See also aged and older engineers (see 90/10 ratio in Table 2). The in- Kumagai (2003). crease in variance is typically thought to reflect faster grow- 13â BS includes workers with a high school degree or GED but no B.S. < ing pay for higher performers, and pay for top earners would degree (the proportion of this group that did not have an associate degree be expected to increase as engineers become Â managers. was 41 percent in 2000, 27 percent in 2002, and 13 percent in 2004); BS includes college graduates who do not have a higher degree; MS/PhD in- cludes workers with a Masters or Ph.D. degree (the proportion of this group 15â Earnings for n percent represents the earnings where n percent of obser- that had only a Masters was 90 percent in 2000, 81 percent in 2002, and 82 vations are below this value and (100 â n) percent of observations are above percent in 2004). Workers without a high school degree and workers with this value. Earnings for the 50th percentile represent the median. professional degrees (e.g., MD, DDS, LLB, JD, DVM) are excluded. 16â For education-age-year cells (3 Ã 4 Ã 3 = 36) with fewer than 10 ob- 14â We used several different samples of occupation codes in order to servations, no results are shown (two cells). For cells with fewer than 20 test for sensitivity of age-earning profiles to the definition of semiconduc- observations (and at least 10 observations), only mean and median income tor engineer occupations. In the results presented here, we included SOC and full weeks worked are shown (six cells). 172070, 172061, 151021, 151030, 151081, 172131, 172110, 172041, â The sample sizes by year and education (not age) are as follows: 119041, 113021, 111021, 112020, 113051, and 113061. When we restricted 2000 2002 2004 the sample to fewer occupation codes, the age-earnings profiles remained < BS â 44 129 127 mostly stable, with the earnings of the top 10 percent increasing for older BS 151 367 363 groups with the inclusion of more managerial occupations. MS/PhD â 78 250 271
154 THE OFFSHORING OF ENGINEERING $250,000 90% 50% $200,000 10% Earnings $150,000 $100,000 $50,000 21â30 31â40 41â50 51â65 Age FIGURE 1â Age-earnings profile for B.S. holders in 2000. Brown -Linden other way, in 2002 and 2004, a typical young engineer (age However, the increase in variance between prime-age and Figure 1 older engineers reflects a sharp drop in pay at the bottom end 21â30) with a B.S. degree earned the same pay as a typical of the scale (the 10th percentile group), especially in 2004. engineer without a B.S. but with 10 years more experience These profiles indicate that many older engineers are facing (age 31â40). declining and inadequate job opportunities. The graduate-degree premiums over a B.S. (median earn- ings for M.S./Ph.D. compared to B.S.) were not stable over the short time period shown, so it is difficult to determine Returns-to-Education the trend for returns for graduate education. The graduate- As expected, median and average earnings increased degree premium for the youngest group, when many were with education. Comparing real median earnings for the still in school, was 36 percent in 2002, but fell to 8 percent in younger groups, we see that the return for a B.S. degree has 2004. The graduate-degree premium for workers in the early been fairly high, with college graduates typically earning stages of their careers (age 31â40) was 7 percent in 2000, 20 percent to 65 percent more (depending on age and year) then shot up to 25 percent in 2002 and 36 percent in 2004, than those who finished high school but not college. Put an- confirming our interview-based findings that the relative $250,000 90% $200,000 50% 10% $150,000 Earnings $100,000 $50,000 21â30 31â40 41â50 51â65 Age FIGURE 2â Age-earnings profile for B.S. holders in 2004. Brown -Linden Figure 2
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 155 $250,000 90% 50% $200,000 10% $150,000 Earnings $100,000 $50,000 21â30 31â40 41â50 51â65 Age FIGURE 3â Age-earnings profile for M.S./Ph.D. holders in 2000. Brown -Linden Figure 3 demand for younger M.S. and Ph.D. holders is increasing as pursue graduate degrees, even though our fieldwork indicates a result of increasing technical complexity in manufacturing that the industry needs them. and design. A typical engineer (age 31â40) with an M.S. or The variance in earnings was higher for engineers with Ph.D. earned slightly less than the average engineer with a graduate degrees than for engineers with B.S. degrees in B.S. but with 10 years more experience (age 41â50). 2004. In both 2002 and 2004, the variance in earnings for For workers in their peak years (age 41â50), the graduate- older engineers with B.S. and graduate degrees was very degree premium fell from 16â19 percent in 2000 and 2002 to high, with the 90/10 ratio ranging from 4.3 to 7.6. 9 percent in 2004. For the oldest workers, the graduate-degree premium fell even more dramatically, from 38â49 percent in Earnings over Time 2000 and 2002 to 13 percent in 2004. For engineers older than 40 in 2004, the graduate degree premium was only 10 The ACS earnings profiles showed slower growth of percent, indicating weak incentives for domestic workers to average earnings between 2000 and 2004 than the OES data $250,000 $200,000 90% 50% $150,000 10% Earnings $100,000 $50,000 21â30 31â40 41â50 51â65 Age FIGURE 4 Age-earnings profile for M.S./Ph.D. holders in 2004. Brown -Linden Figure 4
156 TABLE 2â Age-Earnings Profiles (adjusted for inflation) for 2000, 2002, and 2004 a 2000 2002 2004 Age 21â30 31â40 41â50 51â65 21â30 31â40 41â50 51â65 21â30 31â40 41â50 51â65 Less than a Bachelorâs Degree 10th percentile $6,051 $2,9245 $23,194 $32,270 $32,421 $35,461 $34,448 50th percentile $34,966 $60,899 $48,973 $48,405 $57,481 $57,481 $49,515 $40,526 $60,790 $68,895 $70,415 90th percentile $90,759 $80,675 $85,717 $72,607 $121,579 $193,513 $7,770 90/10 ratio 15.00 2.76 3.70 2.25 3.75 5.46 2.84 Mean $4,606 $53,693 $70,505 $46,649 $57,127 $56,069 $52,402 $41,612 $68,819 $84,736 $64,523 Bachelorâs Degree 10th percentile $20,710 $53,444 $44,536 $30,496 $37,061 $49,026 $32,825 $24,316 $36,575 $60,790 $50,658 50th percentile $52,052 $83,505 $91,299 $72,372 $58,239 $72,005 $88,946 $70,945 $58,763 $70,921 $97,263 $89,665 90th percentile $96,867 $130,270 $158,104 $95,299 $127,066 $158,832 $158,832 $81,053 $109,421 $217,829 $217,829 90/10 ratio 4.68 2.44 3.55 3.12 3.43 3.24 4.84 3.33 2.99 3.58 4.30 Mean $58,127 $89,949 $107,758 $109,566 $60,867 $79,222 $104,635 $87,555 $57,470 $76,809 $116,220 $109,410 Masterâs Degree or Ph.D. 10th percentile $61,238 $61,238 $61,945 $55,062 $63,533 $45,002 $21,276 $60,790 $60,790 $32,320 50th percentile $89,073 $106,331 $100,207 $79,417 $90,005 $105,888 $105,888 $63,322 $96,250 $106,382 $101,316 90th percentile $111,341 $155,878 $95,299 $137,654 $158,832 $339,901 $91,184 $210,737 $217,829 $217,829 90/10 ratio 1.82 2.55 1.54 2.50 2.50 7.55 4.29 3.47 3.58 6.74 Mean $89,360 $114,175 $121,988 $79,769 $95,060 $120,872 $127,819 $61,167 $112,238 $127,075 $124,065 aThe repetition of earnings in some cells, especially for the 90th percentile group, appears to be a coincidence and not a mistake. A check of the data indicates that many workers with different levels of education and in different occupations reported the same earnings, which are not top coded.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 157 TABLE 3â Engineers Working Less Than a Full Year (48 that older engineers may be finding fewer high-quality job Weeks), by Degree Level, for 2000, 2002, and 2004 opportunities. Age Ranges 21â30 31â40 41â50 51â65 Career Paths for Semiconductor 2000 Professionals (LEHD Data) Less than a Bachelorâs Degree a 10% 0 35.71% We now look briefly at how the jobs and earnings of semi- Bachelorâs Degree 25% 3.28% 2.56% 10.53% Masterâs Degree or Ph.D. a 3.23% 4.55% 12.5% conductor workers, including engineers, changed from 1992 2002 to 2001 based on a very large linked employer-employee Less than a Bachelorâs Degree 14.81% 0 14.89% 31.82% data set, the Census Bureauâs LEHD.17 The data cover all Bachelorâs Degree 13.7% 11.11% 9.24% 28.57% occupations, engineers as well as office workers, techni- Masterâs Degree or Ph.D. 13.33% 16.13% 3.7% 26.09% cians, managers, and others. We focus here on prime-age 2004 Less than a Bachelorâs Degree 35.71% 7.69% 3.70% 20% male and female workers (ages 35â54) in two educational Bachelorâs Degree 15.85% 10.62% 9.82% 10.71% groupsâmedium (some college) and high (college graduate Masterâs Degree or Ph.D. 25% 7.34% 12.35% 17.78% and above). Note: The value in each cell is the proportion of engineers in that age group The career paths are shown for modal groups, that is, the with the indicated degree who worked less than 48 weeks in the indicated largest groups of workers who had held one, two, or three year. jobs, with at least one job in a semiconductor establishment a<10 observations (not shown) during the decade. Other (smaller) groups of workers also changed jobs but had different career paths. For those who had held two jobs; the first job was outside showed between 2000 and 2005, primarily because the ACS the semiconductor industry and the second job in it. For earnings were higher than in OES data in 2000 and com- those who had held three jobs, the first two were outside parable in 2004 and 2005. However patterns varied across the semiconductor industry, and the last one was in the occupations. In the ACS data, average computer science industry. earnings grew much faster than average electrical and elec- tronics earnings, where growth did not keep up with inflation Career Paths (not shown in tables). In comparison, the OES data showed comparable positive earnings growth for these occupations Semiconductor workers followed two distinct types of between 2000 and 2005. career pathsâloyalist and job changer (see Table 4). Work- Although ACS data were developed to be compared over ers who already worked for semiconductor employers and time, while OES data were not, the small sample sizes of the had good job ladders (high initial earnings and good earn- ACS data make them less representative and less reliable ings growth) tended to become loyalists, that is, they did not than the OES data. For these reasons, we cannot say with change jobs during the period studied. The career paths of confidence how much earnings by semiconductor engineers loyalists were considerably better than the career paths of grew from 2000 to 2005. job changers. Workers on inferior job ladders outside the semiconductor industry tended to become job changers, and most of them Summary eventually ended up on a relatively good job ladder. Job Overall the earnings data indicate potential problems in changers had relatively low initial earnings in jobs outside the high-tech engineering market. Although the graduate- the semiconductor industry and experienced substantial degree premium appears to be adequate for younger workers, earnings growth (usually 20 to 30 percent for younger and the low returns-to-experience for engineers with graduate 10 to 20 percent for older workers) by taking jobs in the degrees make returns on investment in a graduate degree semiconductor industry. Among job changers, two-jobbers inadequate over an engineerâs entire career, especially the began with higher pay outside the industry and were able to returns implied by the 2004 ACS data. The returns to a BS de- enter the semiconductor industry sooner than three-jobbers. gree were adequate for engineers younger than 50. However, Although highly educated three-jobbers experienced healthy older workers at all three educational levels experienced a earnings increases when they changed jobs outside the semi- troubling drop in median real earnings. The data also indicate conductor industry, the increase was smaller than when they that the variance in earnings for high-tech engineers is grow- got jobs in the industry. Because the overall earnings growth ing, partly because earnings at the bottom of the distribution 17â This material is taken from the Sloan-Census project that produced are rising very slowly, or even falling, as engineers age. Thus, the book Economic Turbulence by Brown et al. (2006) and related papers although the high-tech engineering labor market appears to (see www.economicturbulence.com). See Chapter 5 for an overview of job be strong nationally, data by age and education indicate that ladders and Chapter 6 for an overview of career paths in the semiconductor engineering jobs at the bottom end may be deteriorating and and four other industries (software, finance, trucking, and retail food).
158 THE OFFSHORING OF ENGINEERING TABLE 4â Semiconductor Career Paths, Workers Age 35â54 Males Females Loyalists Two Jobs Three Jobs Loyalists Two Jobs Three Jobs Medium Education A $32,564 $15,046 $12,458 $13,084 $8,148o $7,314 B .054 .056 .058 .039 .030 .041 C $55,780 $25,926 $21,998 $19,641 $10,999 $10,999 High Education A $36,084 $22,893 $18,197 $14,990 $10,132 $9298 B .059 .048 .047 .044 .028 .030 C $65,207 $36,925 $29,068 $23,569 $13,356 $12,570 Notes: = mean initial earnings (2005 dollars, inflated from 2001 dollars using the CPI-urban). A B = net annualized earnings growth rate (in log points) over the 10-year simulated career path. C = simulated 2001 final average earnings (2005 dollars). Source: Adapted from Economic Turbulence (Brown et al., 2006), Chapter 6, Table 6.1. Original calculations by authors from Census LEHD data. These career paths are for all workers in all occupations in the industry. They include engineers, as well as office workers, technicians, managers, and other occupations. of two-jobbers and three-jobbers was about the same over In 1983, IBM offered workers at five locations a voluntary the 10-year period, the two-jobbers usually maintained their early retirement program in which workers with 25 or more initial earnings advantage. years of experience would receive two years of pay over a Although job changers usually experienced higher earn- four-year period. IBM offered voluntary retirement programs ings growth over the decade than loyalists, the growth did not again in 1986 and 1989.18 Because these programs were vol- offset their much lower initial earnings. Thus loyalists ended untary for the general workforce, rather than for targeted job the period with substantially higher earnings. The legendary titles or divisions, the change in workforce usually did not job hoppers in Silicon Valley (engineers who left good jobs turn out as the company might have chosen: better workers for even better ones), constituted a smaller group than the often opted to leave, and weaker workers, without good job job changers shown here, who left relatively low-wage jobs opportunities elsewhere, often opted to stay. for jobs that paid slightly more. The deep recession in the early 1990s finally pushed IBM, DEC, and Motorola, once known for providing employment security, to make layoffs.19 The new approach to downsizing Job Ladders included voluntary programs for targeted workers. If these Data (not shown here) indicate that large firms provided workers did not accept the termination program, they could 85 percent of semiconductor jobs. Firm fortune matters in be subject to layoffs, making the program less than voluntary the job ladders offered by large, low-turnover firms, as we in reality. In 1991 and 1992, IBM selected workers eligible see by comparing firms with growing employment to firms for termination, which included a bonus of up to a yearâs with shrinking employment. Large growing firms with low salary. In this way, more than 40,000 workers were âtransi- turnover provided 50 percent of the jobs in the industry, and tionedâ out of the company. Downsizing continued through these firms are typically known for providing good jobs. 1993, and by 1994 IBM was actually laying off workers.20 Semiconductor jobs in these firms tended to last a relatively With the dot-com bust in the early 2000s, semiconductor long timeâ27 percent lasted for at least five years during companies undertook massive layoffs. By the end of 2001, the decade studied. Motorola had laid off more than 48,000 workers from its Large shrinking firms with low turnover provided an peak of 150,000 employees in 2000.21 As swings in demand interesting contrast. Even though these firms were reduc- became more volatile, the idea of lifetime employment in ing employment, new hires still accounted for 30 percent the semiconductor industry became a thing of the past, al- of jobs; however, less than 20 percent of jobs lasted more though selected workers still had excellent job ladders and than five years. Thus these firms appeared to be replacing long careers. experienced workers with less-expensive new hires. When The data in Table 5 show that for large firms with low we compared ongoing and completed long-term (more than turnover, growing firms offered higher initial earnings than five years) jobs, we found that shrinking large firms tended shrinking firms to both men and women (by 7 to 37 percent), to shed experienced workers with lower earnings growth, because annualized earnings growth was higher (by half a 18â http://www.allianceibm.org/news/jobactions.htm. 19â Some of the observations about specific firms here most likely reflect percentage point) in ongoing jobs than in completed jobs for all groups. divisions of these large, complex firms beyond their production of semicon- ductors. We think the patterns discussed reflect the impact of globalization These patterns marked a change in the way big companies on high-tech firms. deal with difficulties. IBM provides a good example of how 20â http://www.allianceibm.org/news/jobactions.htm. downsizing programs evolved from the 1980s to the 1990s. 21â http://www.bizjournals.com/austin/stories/2001/12/17/daily22.html.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 159 TABLE 5â Job Ladders for Semiconductor Industry Workers, Age 35â54 Growing, Large Firms Shrinking, Large Firms Growing, Large Firms Growing, Small Firms Growing, Small Firms with Low Turnover with Low Turnover with High Turnover with Low Turnover with High Turnover Males Medium Educated A $21,462 $18,012 $14,810 $15,517 $17,115 B .054 .061 .063 .068 .076 C $36,592 $33,266 $27,860 $30,771 $36,592 Highly Educated A $23,057 $21,541 $21,388 $21,070 $20,600 B .059 .061 .040 .075 .055 C $41,582 $39,503 $32,018 $44,493 $35,761 Females Medium Educated A $13,024 $9519 $10,589 $8,506 $8,879 B .039 .036 .021 .048 .085 C $19,128 $13,722 $12,890 $13,722 $20,791 Highly Educated A $14,080 $10,334 $12,424 $10,692 $9897 B .044 .036 â.002 .054 .064 C $22,038 $14,970 $12,059 $18,296 $18,712 Notes: = mean initial earnings (2005 dollars, inflated from 2001 using the CPI-urban). A B = net annualized earnings growth rate (in log points) across the simulated career path. C = simulated 2001 final average earnings (2005 dollars). Source: Economic Turbulence (Brown et al., 2006), Chapter 5, Table 5.1. Original calculations by authors from Census LEHD data. The career paths are for all workers in all occupations in the semiconductor industry, including engineers, office workers, technicians, managers, and other occupations. and the growing firms compared to shrinking firms offered educated workers. Although these firms offered relatively lower earnings growth to men and higher earnings growth low initial earnings, their earnings growth was high. After to women. Overall menâs job ladders are more similar in 10 years, earnings at these companies surpassed earnings of growing and shrinking firms than womenâs job ladders, and experienced workers in large shrinking firms and were close so men seem more protected from economic turbulence than to earnings at large growing firms with low turnover. Small, women. A comparison of âstayersâ (i.e., ongoing long jobs) growing firms may be an increasingly important source of and âmoversâ (i.e., completed 1â3 year jobs) shows that an- good job ladders. nualized earnings growth for short jobs was only two-thirds Overall, economic turbulence has had negative effects on that of long jobs in both growing and shrinking large firms. job ladders. Over the decade studied, growing large firms These results indicate that growing firms used high initial with low turnover allowed highly paid new hires to compete earnings to attract talented workers, among whom only a for access to long job ladders with career development, while select group was given access to career development with shrinking large firms with low turnover forced experienced long, steep job ladders. workers to compete to keep their jobs, which were either Compared to growing firms, large shrinking firms paid being eliminated or being filled by new hires paid at market lower initial earnings but offered higher earnings growth rates. In any case, the era of lifetime jobs with career devel- for short jobs; the job ladders for younger men were better opment appears to be over, and many workers must improve relative to those of older men. These results indicate that their job prospects through mobility. large firms, both growing and shrinking, used market-driven compensation systems based on salaries in the spot market FACTORS THAT INFLUENCE for engineers. Growing firms appeared to provide long job ENGINEERING WORK AND WAGES ladders with career development for a select group, while other workers faced either a plateau or âup or out.â Pos- The U.S. labor market for engineers is affected by a va- sibly workers not on the fast track left voluntarily for better riety of long-term forces, including technological change, jobs elsewhere. Shrinking firms appeared to keep selected immigration policy, and educational practices. In this section experienced workers and replaced the others with new hires we consider the effects of each of these. at market rates. New hires appeared not to have access to long job ladders with career development, even though the Technological Change: Wafer Size older workers still had long job ladders. These findings are consistent with changes we observed in our fieldwork at large Engineering jobs in chip fabs have evolved over the last U.S. companies in the 1990s. several technology generations, driven primarily by simulta- Small growing firms with low turnover were likely to be neous increases in wafer size and automation, both of which early-stage fabless companies that hired mainly technical have been important for raising productivity and keeping personnel and offered relatively good job ladders for college- the industry on its Mooreâs Law trajectory. We look at how
160 THE OFFSHORING OF ENGINEERING engineering work within the fab changed during the transi- TABLE 6â Workforce Composition (mean head count in tion from 150 mm to 200 mm wafers. Our analysis is based matched 150 mm and 200 mm fabs) on detailed data gathered in the mid-1990s by the Berkeley 150 mm Fabs 200 mm Fabs Competitive Semiconductor Manufacturing (CSM) Program Operators 547 (73%) 470 (62%) at a sample of fabs running 150 mm and 200 mm wafers in Technicians â 91 (12%) 107 (14%) four countries.22 Engineers 114 (15%) 181 (24%) Larger wafer size requires major reengineering of equip- Total 752 758 ment and process technology. In addition, materials handling Source: Brown and Campbell, 2001. and information systems must be highly automated to handle the increased weight and value of each wafer safely and to minimize human error. Automation changes the composition of the workforce by increasing the need for engineers and TABLE 7â Workforce Compensation (mean wage or salary decreasing the need for operators. In the CSM data, the per- in matched 150 mm and 200 mm fabs) centage of engineers increased from 15 to 24 percent of the 150 mm Fabs 200 mm Fabs total workforce between 150 mm- and 200 mm-generation plants; at the same time the percentage of operators declined Initial Maximum Initial Maximum Pay Pay Pay Pay from 73 to 62 percent (see Table 6). The overall employment level of the fab stayed approximately the same at about 750 Operators (hourly) $5.88 $15.47 $7.12 $18.44 workers. Technicians (hourly) $6.68 $11.50 $9.12 $15.83 Engineers (monthly) $1,785 $5,019 $2,381 $4,689 The shift in jobs from operators to engineers resulted in an increase in higher paying, high-skilled jobs at the expense Source: Brown and Campbell, 2001. of lower paying, low-skilled jobs. However, the earnings structure across occupations also changed (see Table 7). The initial pay of technicians and engineers was more than one- dropped by human handlersânot to mention the ergonomic third higher in the 200 mm fabs than in the 150 mm fabs, risk to humans. and their pay premium over operators increased. Because new 300 mm fabs process advanced circuits, In terms of returns-to-experience (i.e., maximum pay such as circuits with 90 nm or 65 nm processes, the amount compared to initial pay), experienced engineers fared poorly, of inspection, number of metrology steps, and number of as their ratio of maximum to initial pay fell from 2.8 (150 mm in-line engineering-related activities are significantly higher fabs) to 2.0 (200 mm fabs). The returns-to-experience for than for their older 200 mm counterparts for the same wafer technicians and operators remained stable, as the experi- throughput. As a result, most of the labor savings achieved enced techs and operators had the same pay improvement in through the automation of materials handling, which requires the 200 mm fab as the new hires. approximately 30 percent less labor input, is reapplied Over time, experienced engineers lost out as their average to new engineering tasks, which are much higher value- maximum real salary was actually lower in the 200 mm fabs added and more intellectually challenging and require more than in the 150 mm fabs. In interviews, we learned that fabs troubleshooting. were more interested in having young engineers with knowl- The overall number of workers is not reduced as a result edge of new technology than they were worried about losing of advanced factory automation. Instead, there is a shift in older engineers. Consequently, they were willing to increase task composition. The percentage of workers with higher wages of new hires without raising the wages of experienced engineering and technical problem-solving skills is greatly engineers. With rapidly changing technology, an ample sup- increased, while the percentage of workers required for wafer ply of new hires, and low turnover, companies were able to movement and equipment starting and stopping is greatly flatten engineersâ career ladders (see, for example, Figure 4, decreased. However, the proportion of engineers remains above) with no adverse consequences. the same.23 We do not have comparable data for 300 mm fabs, which have completely automated materials handling and wafer processing. Complete automation is necessary because of the H-1B Visas high value of each 300 mm wafer, which has an area 2.25 U.S. visa and educational policies directly impact the sup- times that of a 200 mm wafer. The 300 mm wafer is heavier ply of engineers, especially those with advanced degrees, in and more awkward to handle, which raises the risk of it being the domestic market. In this section, we look at the earnings of H-1B visa holders. The H-1B visa is used by foreigners 22â Twenty-three fabs in four countries were part of the CSM survey. For employed temporarily in positions that require specialized this table, the 150 mm wafer fabs were matched to the 200 mm wafer fabs by company, so that human resource policies could be compared for the two groups. This reduced our sample to 14. 23â Personal communication, April 2005.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 161 knowledge and at least a bachelorâs degree. Hâ1B visas are tions, of which 14,035 were granted to U.S. firms. Overall, granted to companies (rather than workers). A company 49 percent stated a specific salary rate, and 51 percent stated must submit an application that includes a job title and the a minimum-maximum salary range (reported separately intended wage or earnings, which must reflect the prevailing in Table 8). We analyzed four occupational groups, which wage rate. With various application fees and legal expenses, represent most of the semiconductor applications: electrical the initial cost to an employer is in the range of $2,500 to engineering, computer-related jobs, manufacturing-related $8,000 per application.24 H-1B employees can work only jobs, and business and administrative jobs. Since most H-1B for the sponsoring U.S. employer25 and only do the activi- applications were made by U.S. firms, we focus on these. ties described in the application. A foreigner can work for a More of the applications by non-U.S firms were for business maximum of six continuous years (including one extension) and support jobs (15 percent) or for other kinds of engineer- on an H-1B visa. ing jobs (18 percent); 80 percent of these stated earnings The current law limits the number of H-1B visas that rates. Compared to the earnings stated by U.S. companies, may be certified to 65,000 per fiscal year. Many companies the earnings stated by non-U.S. companies for EE and CS think this number is too low, and businesses have lobbied applications tended to be slightly higher on average with a for higher limits. The numerical limitation was temporar- larger 90/10 ratio, but lower on average for non-EECS jobs ily raised to 195,000 in FY2001, FY2002, and FY2003.26 with a larger 90/10 ratio. Note that only initial applications are included in the annual U.S. chip companies were most likely to apply for H-1B limit; requests for extensions beyond the initial three-years visas for EE jobs (37 percent with average rate $77,560 or are not included. Applications by universities and nonprofit average minimum of $66,944) or CS jobs (52 percent with research institutions are also not counted against the cap. In average rate $78,537 or average minimum of $75,685). The addition, there are 20,000 special exemptions for foreigners other applications were primarily for other engineering jobs with masterâs and Ph.D. degrees from U.S. universities. Even (8 percent with an average rate $79,806, or average minimum in 2003, before U.S. graduates with advanced degrees were of $65,425). exempted, many H-1B visa holders had advanced degrees EE applications primarily stated a specific rate; the dis- (M.S. 29 percent; Ph.D. 14 percent; professional degree tribution tended to be approximately 15 percent above the 6 percent).27 Hâ1B visas are granted for a wide range of distribution for the minimum when a range was given. In occupations, including engineering, medicine, law, social contrast, CS applications primarily stated a range, whose sciences, education, business specialties, and the arts. minimum had a distribution close to the distribution of the We collected data from the H-1B applications certified28 specific earnings rates, where they were used. A possible to the top 10 U.S. chip vendors and the top 10 non-U.S. chip interpretation, consistent with the OES data in Table 1, is that companies (referred to here, for convenience, as the top 20 the high computer science minimum indicates that software companies) from 2001 through 2005 (U.S. government fis- programmers in the chip industry are receiving a premium. cal years). Companies can provide either a specific proposed We checked the applications in 2005 by all other compa- pay rate or a minimum and maximum of the proposed pay nies and industries (called âother firmsâ here) for EE and CS range, and pay can be annual, monthly, weekly, or hourly.29 jobs to see if they used comparable rates and ranges, since The reasons for choosing either a specific rate or a range are H-1B visas might be functioning differently in different in- worth exploring in future research. One possibility is that a dustries. The top chip companies accounted for 56 percent of specific rate may be stated when a company has a specific all EE applications but only 5 percent of CS applications. individual in mind for the visa; a range may be used when We can compare H-1B application rates to actual earn- an individual has not yet been identified. ings for EE-CS engineers. In the ACS data, EE-CS engineers During the five-year period, the 20 companies in our earned, on average, $69,000 to $96,000 (overall average sample were granted approval of 15,784 H-1B visa applica- $86,000) from 2000 to 2004, and in the OES data they earned $66,000 to $84,000 (overall average $74,000) from 2000 to 24â GAO (2003). 2005. The average rates on H-1B visa applications granted 25â The U.S. employer may place the H-1B visa worker with another to the top 20 semiconductor companies fell between these employer if certain rules are followed. two averages. However, it is difficult to make comparisons 26â http://www.uscis.gov/graphics/howdoi/h1b.htm. 27â of these earnings independent of worker experience and edu- DHS (2004). 28â During this five year period, 1.6 percent of the applications were denied cation, because many semiconductor companies hired H-1B (including a small number that were put on hold). These applications are not visa workers as new EE-CS graduates, often with graduate included in our analysis. We also dropped one outlier, which was probably degrees, from U.S. universities. an input error, an application naming $10.6M as the pay for a senior test Interestingly, the âother firmsâ mostly specified earnings engineer. The prevailing wage was given as $93,330. rates in their H-1B applications for both EE and CS jobs. The 29â The two methods of applying (rate and range) are reported separately here. Most applications (95 percent) use annual earnings. Monthly, weekly, rates used on EE applications by âother firmsâ have a lower and hourly rates were converted to annual rates (12 months, 52 weeks, or mean and 10th percentile compared to the top chip firms; the 2,000 hours). rates used on CS applications by âother firmsâ have a con-
162 THE OFFSHORING OF ENGINEERING TABLE 8â H-1B Visa Applications Approved, 2001â2005 Observations (%) Mean Standard Deviation 10% 90% Top Ten U.S. Chip Firms Electrical and Electronics Engineering Job Codes Rate given 3436 (24%) 77,560 16255 62,400 96,160 Range given min 66,944 13991 52,800 85,225 max 1792 (13%) 102,992 23410 73,375 130,000 Computer Science Job Codes Rate given 2106 (15%) 78,537 18275 61,302 98,239 Range given min 75,685 18318 56,277 100,000 max 5234 (37%) 96,118 19662 75,000 125,000 Manufacturing Engineering Job Codes Rate given 649 â (5%) 79,806 16801 58,200 96,000 Range given min 65,425 14609 48,788 85,000 max 403 â (3%) 104,798 25202 73,006 130,000 Business, Marketing, Administrative Support Job Codes Rate given 163 â (1%) 87,533 40824 50,400 130,000 Range given min 73,549 24725 44,200 106,000 max 252 â (2%) 101,535 34193 64,900 140,000 Top Ten Non-U.S. Chip Firms Electrical and Electronics Engineering Job Codes Rate given 430 (25%) 80,161 18941 59,527 105,694 Range given min 77,580 18627 55,104 99,808 max 188 (11%) 106,911 31388 70,900 154,300 Computer Science Job Codes Rate given 432 (25%) 79,525 18476 57,500 101,100 Range given min 68,712 13843 52,361 86,606 max 124 â (7%) 91,773 22201 64,676 120,000 Manufacturing Engineering Job Codes Rate given 292 (17%) 73,458 16419 53,600 95,000 Range given min 69,070 16997 53,100 102,168 max 19 â (1%) 86,217 25232 60,270 132,000 Business, Marketing, Administrative Support Job Codes Rate given 230 (13%) 81,882 39447 42,150 134,838 Range given min 60,406 24271 39,145 88,486 max 34 â (2%) 82,882 36511 50,000 140,000 Other Chip and Non-Chip Firms Electrical and Electronics Engineering Job Codes Rate given 7701 â (6%) 69,302 24175 45,000 100,000 Range given min 67,737 20807 45,000 95,256 max 2098 â (2%) 84,710 28592 50,000 124,000 Computer Science Job Codes Rate given 96720 (71%) 60,698 20371 42,000 87,250 Range given min 58,523 16860 42,000 81,600 max 29964 (22%) 77,277 25747 50,000 120,000 Note: companies can submit applications with a specific proposed rate to be paid or can provide a range (min, max). No duplicates were submitted. Source: U.S. Department of Labor. Available online at http://www.flcdatacenter.com/CaseH1B.aspx. siderably lower distribution compared to the top chip firms. electronics engineers (EEs), to a sample of U.S. workers Once again, consistent with the wage data in Table 8, H-1B using the Census Departmentâs Current Population Survey applications for EE-CS jobs in the chip industry appear to of 2002. As GAO notes, for a variety of reasons the annual carry a premium compared to other industries. salary comparisons are not exact. For one thing, we do not A GAO study in 2003 of H-1B visa holders compared know if the visas were actually used. the annual pay for selected occupations, including electrical/ The GAO comparison of EEs with H-1B visas and U.S.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 163 citizenship in 2002 showed that the H-1Bs were younger IBM applications stated a range of earnings, with an average (32 years vs. 41 years; 62 percent under 35 years old vs. minimum ($82,072) that was considerably higher than the 28 percent) and much more likely to have graduate degrees average minimum stated by the other four companies. Since (50 percent vs. 20 percent) (GAO, 2003, pp. 14, 15). When IBM is now more of a services company then a hardware median annual salary of EEs aged 31 to 50 years old were company, we assume that many of these applications were compared, H-1Bs earned less than citizens (H-1Bs with for jobs that were not chip-related. graduate degrees earned $77,000, while citizens earned Like IBM, Motorola most often stated a range of earn- $88,000; H-1Bs with less than a graduate degree earned ings. Motorolaâs average minimum ($62,866) was 25 percent $65,000, while citizens earned $70,000) (GAO, 2003, p. 42). lower than IBMâs. Even so, Motorolaâs rates were slightly For younger EEs (age 18 to 30) without graduate degrees, higher than the national EE-CS salaries in the OES. however, H-1Bs earned more than citizens ($60,000 vs. The three companies that focus on semiconductors, Intel, $52,000) (GAO, 2003, p. 42). These data indicate that H-1B Qualcomm, and Texas Instruments, tended to state actual visa holders may be having a downward impact on labor- earnings, and their average earnings were within 2 percent market opportunities for mature engineers, but probably not of each other. Stated earnings for Intel showed less variance for young engineering college graduates. than for QualComm or Texas Instruments. The three compa- nies applied for Hâ1B visas to fill jobs that required a variety of skills and experience. Overall, their rates seemed to reflect Applications for Selected Companies the national EE-CS salaries in the ACS. H-1B visa applications for five large U.S. companies, By comparing Freescaleâs applications to Motorolaâs IBM, Intel, Motorola/Freescale, Qualcomm, and Texas for 2004 and 2005, we can estimate the extent to which Instruments, together accounted for 76 percent of the H-1B Motorolaâs applications were for engineers in their semicon- granted applications in our sample (Table 9). Since Motorola ductor business. Freescale was granted 11 percent as many spun off its chip operations as an independent company, H-1B visas as Motorola in 2004 and 18 percent as many in Freescale, in 2004, we combined applications granted to 2005. Freescaleâs pay rate had a much narrower range than Motorola and Freescale. Motorolaâs, with the ratio of Freescaleâs sample maximum to Of these five companies, IBM used the most H-1B visas; minimum rates between 2.5 and 2.7 (compared to Motorolaâs almost 4,000 were granted during the five-year period. Most ratio of 5.0 to 4.6). However Freescaleâs averages for the minimum and maximum rates were very close to Motorolaâs averages in 2004 and 6 percent higher than Motorolaâs in TABLE 9â H-1B Visas Granted to the Top Five U.S. 2005. This indicates that semiconductor engineers had aver- Companies, 2001â2005 age earnings compared to the broad range of other workers at Motorola. Variable Observations Mean Standard Deviation The proposed wages for the top-20 companies, as well as IBM for these specific companies, indicate that H-1B visas were Rate given 395 $88,353.9 33462.38 Range given min 3599 $82,071.5 18307.34 issued for a wide range of jobs, some of them high-level max $96,150.2 19493.39 jobs that paid well over $100,000, and some low-level jobs that paid less than $50,000. To what extent the lower paying Motorola/Freescale Rate given 264 $66,472.4 28978.98 jobs are being used to keep semiconductor earnings low for Range given min 2256 $62,910.4 12993.04 domestic new hires, and to what extent the higher paying max $92,573.9 18760.1 jobs are going to foreigners at the expense of qualified expe- Intel rienced U.S. engineers cannot be determined These remain Rate given 1574 $78,065.1 11673.03 important policy questions. Range given min 1122 $65,921.4 10107.71 max $121,519.6 19650.69 Inter-year Comparisons.â If we compare H-1B visas grant- Qualcomm ed by year, we note that the number granted to each of these Rate given 1632 $76,775.5 14152.01 five companies jumped, either in 2003 or 2004, and remained Range given min 0 0 0 high, even as the national limit and fee dropped dramatically. max 0 0 The semiconductor companies seemed to take advantage Texas Instruments of the additional 20,000 H-1Bs available for workers with Rate given 1076 $76,754.3 15717.13 Range given min 61 $73,352.4 20891.36 graduate degrees from U.S. universities that went into effect max $91,727.7 22138.06 in 2004. Sixty-one percent of the H-1B visas awarded to the top-20 companies were awarded during the last two years Note: Companies can submit applications with a specific proposed rate to be paid or can provide a range (min, max). No duplicates were submitted. (2004 and 2005) of the five-year period. Source: U.S. Department of Labor, H-1B Program Data. Available online Intelâs applications for H-1B visas increased dramatically at http://www.flcdatacenter.com/CaseH1B.aspx. during the five-year period. One-quarter of its H-1Bs were
164 THE OFFSHORING OF ENGINEERING granted in the first three years and three-quarters in the last U.S. Education of Foreign Students two years. The company also shifted from stating minimum- Higher education has played an important role in the de- maximum ranges to actual wage rates, although the earnings velopment of the U.S. semiconductor industry, and M.S. and rates remained comparable. Ph.D. engineering graduates provide the essential workforce for semiconductor companies. Engineers who graduate at the The H-1B Share of the Workforce.â We now look at how highest level, the Ph.D., have attained not only state-of-the- H-1B visa applications compare to employment levels at art knowledge, but also the ability to conduct research and to Intel, Motorola, and IBM. In 2005, Intel employed approxi- keep abreast of the latest technology during their careers. mately 99,900 people worldwide, with more than 50 percent Many U.S. graduate engineering students are foreign located in the United States. Motorola employed 69,000 nationals. Figure 5 shows the number of engineering Ph.D.s employees, with more than half employed outside the United (not including computer science) awarded at U.S. universi- States, and with 24,000 eligible for stock options. IBM ties to students from five key Asian countries over a 12-year employed 329,000 worldwide, approximately 40 percent of period. As the figure clearly shows, China has sent a large whom were eligible for the U.S. retirement plan (at the end and growing number of doctoral engineers to the United of 2004, when the plan was discontinued).30 States. At the other extreme, Japan sent very few students If we assume that Intel had 50,000 domestic employees during the same period. and used its 1,280 H-1B visas to hire new workers in 2005, The number of students from Taiwan, which relied on then approximately 2.6 percent of Intelâs domestic employ- U.S-educated Ph.D.s to develop its semiconductor industry, ees were newly hired H-1B visa holders. If most H-1B visa has declined since 1994. In our fieldwork in Taiwan in Feb- holders work for Intel for five years, then approximately 5.4 ruary 2005, many semiconductor experts raised concerns percent of Intelâs 2005 domestic employees (and an even about decreasing interest in U.S. graduate study because larger percentage of engineers) were H-1B visa holders. they still considered Taiwanese doctoral training inferior to If we assume that Motorola used its 728 H-1B visas to hire U.S. training. The number of advanced engineering graduate new workers in 2005 and that these professional specialists students from India and Korea also declined in the late 1990s, held similar jobs to those of Motorola employees eligible for although both have been increasing again since 2002. stock options, then almost 3 percent of Motorolaâs domestic We also looked at the granting of Ph.D.s in electrical professionals were newly hired H-1B visa holders. If most engineering and computer science to U.S.-born and foreign- H-1B visa holders work for Motorola for five years, then 8 born students. Figure 6 shows electrical engineering Ph.D.s percent of Motorolaâs domestic professionals were H-1B by citizenship and gender for 1995 to 2004. Noncitizen visa holders (or 6 percent of Motorolaâs domestic workforce, male students earned significantly more diplomas than their if one-half of the workforce was domestic) in 2005.31 Com- U.S. counterparts throughout the period. Noncitizen female parable calculations for IBM indicate that IBM hired 1,150 students earned more degrees than U.S. women beginning H-1B visa holders in 2005, or 0.8 percent of the domestic in 1998. workforce, and 2.8 percent of its domestic workforce (and The same data for computer science students (Figure 7) a larger percentage of its professional domestic workforce) show that the numbers of degrees awarded to citizens and were H-1B visa holders. noncitizens are much closer, although once again noncitizen These data indicate that semiconductor companies use male students were awarded more Ph.D.s than their U.S. H-1B visas strategically in hiring and managing their engi- counterparts nearly every year. neering employees. In large U.S. companies, H-1B visa hold- These figures clearly show that the United States is train- ers comprise an important part of the domestic professional ing hundreds of foreign advanced engineers every year, thus workforce. One reason for the importance of H-1B visas is increasing the ability of foreign chip firms to compete with that major U.S. universities provide engineering graduate U.S. companies and making it easier for U.S. firms to find education to many foreign students, and upon graduation, qualified personnel, either for their U.S. operations (when these students are in great demand by U.S. companies. noncitizens can remain in the country) or their offshore subsidiaries. In our earlier discussion of returns-to-education, we noted that the earnings premium for a domestic BSEE who pursues 30â These employment figures are from the companyâs 10-K reports to a graduate degree was relatively low. For foreign BSEEs, the SEC: Intel at http://finance.yahoo.com/q/sec?s=INTC, Motorola at however, the financial incentive to pursue a U.S. graduate http://finance.yahoo.com/q/sec?s=MOT, and IBM at http://finance.yahoo. degree is much greater. A U.S. graduate degree opens the com/q/sec?s=IBM. door for these students to high-paid jobs both in the United 31â This percentage was adjusted downward for the Freescale spinoff. We assumed that 15 percent of the H-1B visa holders hired in 2001, 2002, and States and at home. In our fieldwork, we found that advanced 2003 (the proportion of H-1B applications by Freescale compared to Mo- degree holders in semiconductor centers like Shanghai and torola in 2004 and 2005) worked for Freescale (not Motorola) in 2005. Bangalore, especially if they have some U.S. work experi-
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 165 1,400 China India 1,200 Taiwan Korea Japan 1,000 Number of Ph.D.s 800 600 400 200 0 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Year FIGURE 5 Engineering Ph.D.s in the United States by country of origin, 1993â2004. Source: National Science Foundation, Division of Science Resources Statistics, Science and Engineering Doctorate Awards: 2002 (App. Table 5), 2003 (App. Table 11), and 2004 (App. Table 11). 1000 900 Brown -Linden Figure 5 Number of Ph.Ds in electrical engineering 800 700 600 500 400 Non-citizen, male U.S. citizen, male Non-citizen, female 300 US citizen, female 200 100 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Year FIGURE 6â Electrical engineering Ph.D.s by gender and citizenship status, 1995â2004. Source: National Science Foundation, Division of Science Resources Statistics, Science and Engineering Doctorate Awards: 2004 (App. Table 3). ence, earn comparable salaries to their U.S. counterparts. engineers, and then profile the state of the chip industry in Locally educated EEs earn much less. Taiwan, China, and India. Brown -Linden Figure 6 GLOBALIZATION Offshoring by U.S. Semiconductor Firms32 Globalization is one of the primary forces affecting the The three primary reasons for locating value-chain activi- work and rewards of U.S. semiconductor engineers. In this section we briefly describe offshore investments by 32â See Brown and Linden (2006) for a more detailed discussion of offshor- U.S. semiconductor companies, provide some data on chip ing by U.S. semiconductor firms.
166 THE OFFSHORING OF ENGINEERING 500 450 Number of Ph.Ds in computer science 400 350 300 250 Non-citizen, male U.S. citizen, male 200 Non-citizen, female U.S. citizen, female 150 100 50 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Year FIGURE 7â Computer science Ph.D.s by gender and citizenship status, 1995â2004. Source: National Science Foundation, Division of Sci- ence Resources Statistics, Science and Engineering Doctorate Awards: 2004 (App. Table 3). ties are (1) access to location-specific resources, especially TABLE 10â Distribution of U.S-Owned Fab engineering talent; (2) cost reduction; and (3) access and Capacity, 2001 Brown- Linden Figure 7 development of a local market. Often, all three reasons influ- North America 65.4% ence the decision of a company to move an activity to a new Europe/Middle East 18.6% location via internal investment or outsourcing. For example, Japan 13.0% a company may move chip design to China to take advantage Asia (except Japan) 3.0% of low-cost engineering talent with knowledge of customized Source: Calculations courtesy of Rob Leachman. solutions for regional Chinese telecommunication systems, as well as to gain government approval for market access. Because chip manufacturing is so capital intensive, offshore investments in chip fabrication have been driven historically by concerns about market access, particularly Corporation (TSMC), is still the largest. If TSMC sold chips tariffs, more than by cost reduction. Thus most U.S.-owned under its own name, it would have been on the chip industry offshore fabs are located in developed countries, such as top 10 list in 2005, with $8.2 billion in revenue. Foundries are Japan. In 2001, approximately one-third of U.S.-owned excluded from our calculations, however, to avoid double- capacity was located offshore (Table 10). Conversely, about counting of their and their customersâ chips. Because the 22 percent of the fab capacity located in North America was foundry price accounts for about one-third of the final chip owned by companies based in other regions (not shown). value, TSMC actually manufactured nearly $25 billion worth Foreign companies still find the United States an attractive of chips, which would make it number two (after Intel) in the place to invest, as evidenced by Samsungâs recent commit- overall chip industry. ment to a new, multibillion-dollar fab in Austin, Texas. 33 On the one hand, the emergence of the foundry model in One factor that limits fab investment abroad by U.S. com- Asia has meant that less production capacity has been built in panies is the availability of high-quality fabrication services the United States. On the other hand, the foundry model has for hire. In 2005, the outsourced fabrication market was greatly facilitated the growth of the fabless design sector, one worth $18 billion, 34 most of it accounted for by dedicated of the industryâs growth engines, as discussed in ÂSectionÂ 1.35 contract manufacturers, known as âfoundries,â in Taiwan. As a back-of-the-envelope calculation, we estimate that if all The first foundry, Taiwan Semiconductor Manufacturing foundry production were based in the United States instead of Asia, it might add 11,000 industry jobs, of which about 33â David Lammers, âAnalysis: Samsung fab deal ends drought for Austin,â EE Times, April 14, 2006. 35â See Macher et al. (1998) for a discussion of the factors leading to 34â Gartner Dataquest estimate reported in âFoundry Revenue Drops in the U.S. industryâs resurgence after its loss of global market share in the 2005, Gartner Reports,â Electronic News, March 27, 2006. mid-1980s.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 167 2,600 would be highly paid engineering jobs.36 Keep in Silicon Valley area. But even these salaries are much higher mind, however, that not all foundry sales are to U.S. custom- than salaries in India and elsewhere, as discussed below. ers. In 2003, for example, half of TSMCâs gross revenues Cost-driven, in-house offshoring incurs non-wage costs came from non-U.S. sources.37 that partially offset the difference in salaries, especially As a point of comparison, the Fabless Semiconductor during the early stages of establishing an offshore design Association reported that publicly traded fabless companies center. One non-wage cost that is often mentioned is the in North America employed approximately 45,000 workers lower quality and productivity of inexperienced engineers, as of December 2004.38 A review of company information which also adds monitoring costs. Another is the time and suggests that more than half of these employees were soft- inconvenience of communicating across time zones, which ware or hardware engineers. The proportion that was located can be considerable. Finally, additional control mechanisms offshore is not known. may be necessary to protect key intellectual property. Ac- On the design side, U.S. chip firms have opened increas- cording to a venture capitalist, the actual savings from going ing numbers of offshore design subsidiaries in Asia over the offshore is more likely to be 25 to 50 percent than the 80 to last decade. Specialized skills are an important reason U.S. 90 percent suggested by a simple comparison of salaries.42 semiconductor companies are investing overseas, particu- Nevertheless, U.S. firms are investing regularly in low- larly in Europe. Britain, for example, has developed expertise cost design centers in Asia, especially in India. The first U.S. in consumer multimedia, and Scandinavian countries are company to establish design operations there was Texas noted for skills in wireless network technology. U.S. firms Instruments in 1985. Among the top 20 U.S. semiconductor regularly acquire small European companies to obtain both companies, only two (Micron and Atmel) have not estab- application know-how and teams of pre-trained engineers. lished design centers in India. Nine of these top companies As is true of fabrication, design offshoring works both have opened Indian operations since 2004. The size of these ways, and many foreign companies maintain design cen- operations varies widely, with Intel employing about 3,000 ters in Silicon Valley or elsewhere in the United States to engineers and smaller companies, like Marvell, employing take advantage of the high skills and productivity available fewer than 100 engineers. there as well as to be closer to U.S. customers. Philips of The net impact on U.S. jobs from the offshoring and the Netherlands, for example, bought VLSI Technology, outsourcing of fabrication is hard to assess. As we said a major ASIC company with more than 2,000 employees above, offshore fabs have been at least partially balanced by (about one-third of whom were fab workers), in 1999 for foreign-owned fabs in the United States. And although Asian nearly $1 billion.39 Hitachi Semiconductor has a U.S. design foundries have probably contributed to a long-term reduction group of several hundred strong.40 Toshiba has a network of in U.S. chip manufacturing, the net loss of engineering jobs seven ASIC design centers around the United States.41 Even has probably been offset, at least partly, by the increase in foreign start-up companies may need a U.S. design team to design jobs at fabless companies. The government data in work with U.S. customers or to access leading-edge analog Table 1 suggest that despite the increase in offshoring in re- design skills. cent years, the number of engineers employed in the industry However, a reason for design offshoring that has gener- has increased. ated a great deal of attention in the industry is cost reduc- Data from the Semiconductor Industry Association (SIA) tion. For Silicon Valley firms, some cost reduction can be provide further support for this argument (see Table 11). The achieved by opening satellite design centers elsewhere in SIA data are based on an annual survey of large and medium- the United States, because some locations have average sized U.S. semiconductor companies, which together ac- engineering salaries as much as 20 percent lower than in the count for approximately 80 percent of the U.S. industryâs sales. The results are then extrapolated to represent all U.S. semiconductor firms. Although the data may not be strictly comparable from year to year, they can be used to discuss 36â TSMC, which accounts for about half the foundry industry, has one general trends and confirm other data. The total engineer- 150mm, one 300mm, and five-and-a-half 200mm fabs outside the United States. These fabs probably have different rated capacities, but we can ap- ing employment at the top 20 companies has increased proximate employment by calculating 750 workers per plant, which works significantly over the period in question, with the offshore out to 5,625. Doubling that to approximate the entire foundry sector brings engineering staff growing slightly faster in most years. The us to 11,250. number of engineers located in the United States increased 37â Note 27c of Form 20-F filed by TSMC with the Securities and Exchange sharply at the end of the 1990s, before the recession caused Commission for fiscal year ended December 31, 2003. 38â FSA (2005). an employment slump in the early 2000s. Another sharp in- 39â âPhilips to acquire VLSI Technology for $953 million,â Semiconductor crease in U.S. employment is shown between 2004 and 2005, Business News, May 3, 1999. 40â âHitachi Forms North America Semiconductor Systems Solutions Unit,â Hitachi Press Release, September 2, 1998. 41â âToshiba Expands Soc Design Support Network with Opening of San Diego Design Center,â Toshiba Press Release, November 26, 2002. 42â Interview, May 2004.
168 THE OFFSHORING OF ENGINEERING TABLE 11â U.S. Semiconductor Engineers by Location, 1997â2005 1997 1998 1999 2000 2001 2002 2003 2004 2005 U.S.-based Engineers 49,702 46,704 61,856 76,129 72,564 72,860 71,991 66,581 83,167 Offshore Engineers 7,253 19,692 17,446 19,964 27,226 29,813 30,876 34,632 42,193 Total 58,952 68,394 81,301 98,093 101,791 104,675 104,870 103,217 127,365 % in U.S. 87.3% 70.3% 77.9% 79.2% 72.7% 70.9% 69.9% 65.8% 66.3% Source: David R Ferrell, âSIA Workforce Strategy Overview,â ECEDHA Presentation March 2005; 2004 and 2005 data: unpublished SIA survey results provided by Ferrell. although the OES data for those two years do not confirm gineering will develop in India and China as the semiconduc- this trend.43 tor industry in those countries matures, with the important The number of offshore engineers increased sharply in difference that Taiwan is a much smaller country. 1998, and again in 2001, and again in 2005. Even with the The semiconductor industry in India and China is still ups and downs, the percentage of the workforce in the United quite young in terms of design, although both countries are States tended to hover between 70 and 80 percent from 1998 active in this area. In China, domestic companies, often with to 2003; it then fell to 66 percent in 2004â2005. These data personnel and funds from Taiwan, are major players in the indicate a mild shift in employment of engineers offshore development of semiconductor design. In Chinaâs fabrica- relative to the United States. If it continues, this shift could tion sector, both multinational companies (MNCs) and do- have a depressive effect on U.S. engineering employment mestic companies (again with input from Taiwan) are very and earnings. important players. In India, where subsidiaries of MNCs are the major players in the development of the semiconductor industry, fabrication has not yet begun. The Semiconductor Industry in Japan, Taiwan, China, and India Semiconductor Engineering in Asia Engineers in the U.S. semiconductor industry have long been accustomed to competition from abroad. However, With the caveat that comparisons of semiconductor engi- the competition may now be within a single company, for neers in the United States, Japan, Taiwan, China, and India example, between two design groups in different countries. involve comparing engineers with different education and In this section, we look at the availability, quality, and cost experiences, Table 12 provides rough estimates (based on a of chip engineers outside the United States. combination of published sources and interviews) of salaries, A major problem with comparing semiconductor engi- worldwide fab investment by local companies, and the num- neering talent in different countries is that the engineers in ber of active chip designers (excluding embedded software). China and India, and to a lesser extent in Taiwan, are younger We also provide an index of protection of intellectual prop- and have less education than engineers in the United States erty (IP), which is an important consideration in deciding and Japan. In India and China, technicians with two-year which engineering activities might be moved outside the degrees are often classified as engineers (this happens much United States. However, the intellectual property protection less often in the United States and Japan). Relatively little rating covers all industries; thus low scores in the table may graduate training is available in semiconductor engineering reflect lapses in specific sectors, such as pharmaceuticals, in India and China, and what is available is not comparable trademark goods, or recorded media, which are not relevant to graduate programs in the United States and Japan. Taiwan to the semiconductor industry. is an intermediate case; undergraduate and masterâs level The salary figures suggest that engineers in the United engineering programs are comparable to those in the United States and Japan earn much more than most Asian engineers. States and Japan, although Ph.D. programs are still catch- These data, however, are imprecise and have high variance; ing up. thus they provide only a general guide. The salaries are for Taiwanâs semiconductor industry was built largely by engineers with at least five years of experience in the United Ph.D. engineers who returned to Taiwan after receiving States and for engineers aged 40 in Japan, the approximate degrees and valuable work experience in the United States. age they leave the union and begin to receive higher salaries. A similar process is occurring in China and India. Thus we Note that 40 is the age at which the salary trajectory for U.S. think Taiwan may provide a model of how semiconductor en- engineers begins to level out. Semiconductor engineers in the other countries tend to be younger and less experienced; thus 43â The OES total for all software and other engineer categories was 73,650 the salaries for engineers in China and India are for individu- in the May 2004 data and 76,300 in the May 2005 data. als with one to three years of experience.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 169 TABLE 12â Estimates for Selected Countries In the United States, benefits, including health insurance, Value of Fabs Intellectual Social Security, and stock options, also make comparisons Annual Constructed, Property difficult. Salaries by Country of Number Protection, The value of fab construction over the past decade pro- for EE/CS Ownership, of Chip 2004 vides a general idea of the presence of this part of the value Engineers 1995â2006 Designers (10 = high) chain in each country. China, at $26 billion, has made signifi- United States $82,000 $74 billion 45,000 9.0 cant inroads since its early public-private joint ventures with Japan $60,000 $66 billion âa 7.2 Japanâs NEC in the mid-1990s. In India, in sharp contrast, Taiwan $30,000 $72 billion 14,000 6.5 India $15,000 $0 7,000 5.0 not a single commercial-scale fab has been constructed, China $12,000 $26 billion 5,000 3.7 although several have been proposed. aWe have been unable to obtain an estimate for the number of chip We also estimate the number of chip designers, a group designers in Japan. that is critical to the development of the semiconductor in- Sources: U.S. salary from 2004 BLS Occupational Employment Statistics dustry. According to some sources, about 400 chip designers web site (average for electronics and software engineers in NAICS 3344); are being added each year in India and China.47 However, Japan salary (average for circuit designer and embedded software engineers that number can be misleading, because there is some con- aged 40 years old) from Intelligence Corporationâs data on job offers in fusion about the definition of âchip designer.â One industry 2003; Taiwan salary information from March 2005 interview with U.S. executive in Taiwan; China and India salaries are estimated based on a com- executive claimed that there were only 500 âqualified IC bination of interviews, business literature and online job offerings; value designersâ in China in 2004.48 A Taiwanese consultant didnât of fabs (when fully equipped) from Strategic Marketing Associates (www. even consider the later (and lower skilled) stage of physical scfab.com), reported in âChipmaking in the United States,â Semiconduc- design, called âplace and route,â to be part of chip design.49 tor International, August 1, 2006; number of chip designers in U.S. from By those criteria, about 30 percent of the Taiwanese design- iSuppli as reported in âAnother Lure Of Outsourcing: Job Expertise,â WSJ. com, April 12, 2004; number of chip designers in Taiwan from interview ers shown in the table would be eliminated. with Taiwan government consultant to industry, March 2005; number of chip designers in India and China are author estimates based on conflicting published sources and discussions with industry analysts in 2005; intel- Estimates of Higher Education lectual property protection data from Gwartney et al., 2006, Chapter 3. All As we discussed above, engineering programs in U.S. numbers rounded to reflect lack of precision. universities have attracted large numbers of foreign students. The United States leads the world in higher education, es- pecially in graduate training, as the Academic Ranking of World Universities (http://ed.sjtu.edu.cn/ranking.htm) by As the semiconductor industry quickly expands in China Shanghai Jiao Tong University shows (see Table 13). Fifty- and India, wages are reportedly rising rapidly. For example, three of the top 100 universities are located in the United the salary range offered by SanDisk in Bangalore (JobStreet. States; five are located in Japan. Of the top 500 universities, com, June 2005) for a design engineer with one to three years 168 are in the United States, 34 are in Japan, and only 21 are of experience was $9,200 to $18,400.44 in China, Taiwan, and India combined. The salary gap is narrower for comparable key employ- The numbers for bachelor of science engineering degrees ees. One report claimed in 1999 that the salary ratio between in Table 13 must be treated with caution, because the qual- the United States and India for experienced design engineers ity of education varies widely from country to country. The or managers was only 3-to-1.45 Senior managers with foreign numbers may indicate political and social commitment to experience are paid a large premium that eliminates any advancing technical education rather than actual capability. cost advantage; this reflects the critical importance of these Also, these numbers are changing as India, and especially managers in implementing new technology and projects.46 China, expand their engineering degree programs. According The overall differential between Indian and U.S. salaries to a widely cited Duke University study, the annual number has been declining as Indian salaries rise, and the earnings of new EE-CS-IT bachelorâs degrees in China in 2004 had of domestically trained Indian engineers has been doubling reached 350,000 (Gereffi and Wadhwa, 2005). But it is an in their first five years on the job. open question how long it will take these new programs to Salaries are also difficult to compare because of dif- develop quality teaching programs. ferent compensation packages. In the United States and Although China and India have large numbers of engi- Taiwan, profit-sharing bonuses that vary with the business neering graduates, according to our interviews graduates cycle can be an important part of a compensation package. from U.S. universities are better trained, especially in 44â Convertedat 43.52 Indian rupees to the dollar. 45â âSpecial report: India awakens as potential chip-design giant,â EE 47â For India: âDesigns on the future,â Express Computer (India), February Times, January 22, 1999. 10, 2003; for China: PriceWaterhouseCoopers (2004), p. 7. 46â Interviews at 15 semiconductor design centers in Bangalore in No- 48âPriceWaterhouseCoopers (2004), p. 7. vember 2005. 49âE-mail exchange, March 2005.
170 THE OFFSHORING OF ENGINEERING TABLE 13â Estimates of Higher Education for Taiwan Selected Countries Taiwan has the best-established semiconductor industry Academic Ranking of of the three Asian countries. According to Taiwanâs Ministry World Universities, 2005 of Economic Affairs, the country ranked third (behind the Universities Universities Engineering U.S. and Japan) in semiconductor-related U.S. patents.51 in Top 100 in Top 500 B.S. Degrees, 2001 The foundry model originated in Taiwan in 1987, and three U.S. 53 168 110,000 of the top five foundries are located there. Taiwan also has Japan 5 34 110,000 rapidly growing production of memory chips and numerous Taiwan 0 5 35,000 successful fabless chip companies, four of which reported China 0 13 220,000 India 0 3 110,000 revenues of more than $500 million in 2005.52 Table 14 shows the value of Taiwanâs semiconductor Source: Academic Ranking of World Universities values tabulated by authors from ARWU 2005 Edition, accessible at http://ed.sjtu.edu.cn/ industry output by stage of production for 2005. Fabrication, ranking2005.htm; engineer B.S. degrees tabulated by authors for âEngineer- at $18.9 billion, accounts for the largest share of the $34.8 ingâ and âMath/Computer Scienceâ from Appendix Table 2-33, âScience billion total, followed by chip design at $8.6 billion. Simi- and Engineering Indicators 2004,â National Science Foundation except for lar analyses are not possible in most major chip-producing India, which is an estimate for 2003â2004 from Appendix âUSA-China- countries where all stages of production are performed by Indiaâ in Gereffi and Wadhwa, 2005. large integrated producers. Taiwanese companies, however, have embraced the disaggregated business model, and only a handful of companies are involved in multiple steps in the value chain. teamwork on projects and in using tools and equipment. For Since the late 1970s, Taiwan has benefited from focused example, undergraduate students in India and China usually government programs and the return of U.S.-educated and have no chance to work with automated chip design (EDA) trained engineers.53 In 1980, the government created the tools, while EE students in the United States do. According Hsinchu Science-Based Industrial Park, which is still the to McKinsey, only 10 percent of Chinese and 25 percent of islandâs largest concentration of semiconductor firms. Hsin- Indian engineering graduates are likely to be suitable for chu is also home to two of Taiwanâs leading engineering employment by U.S. MNCs (McKinsey Global Institute, universities, and the governmentâs microelectronics lab, 2005).50 ERSO, which played a pioneering role in the development However, as we have already pointed out, the competi- of the industry, including the creation of chip companies tion is not only between U.S. students trained in the United such as TSMC and UMC. ERSO conducts some of the most States and foreign students trained abroad. A large number advanced research in the country, and its thousands of alumni of foreign students receive training in the United States. are encouraged to commercialize technology via local start- up companies. Country Profiles The Taiwanese chip-design sector is mostly locally owned, although a few MNCs also operate design subsidiar- Next we look at the evolution of the semiconductor ies there. Taiwanese companies have embraced the fabless industries in Taiwan, India, and China and compare the model, and some 60 fabless companies were listed on the technology capabilities of these countries with those of the Taiwan Stock Exchange in December 2004.54 By compari- United States. On the design side, the quality of engineers son, about 70 fabless companies were listed on NASDAQ in in Asian countries, both in universities and in companies, 2004. In 2001, the Taiwanese government renewed its efforts has been improving, as is clear from papers submitted to (Si-Soft) to improve local chip-design capabilities. As part the International Solid-State Circuits Conference (ISSCC), of this initiative, the faculty teaching chip design more than which is IEEEâs global forum for presenting advances in doubled, from 200 in 2001 to more than 400 by 2005.55 chip design (see Figure 8). From 2001 to 2006, submissions One advantage for Taiwanâs fabless firms is the availabil- from China, India, and especially Taiwan increased notice- ity of an important local market. Many Taiwanese systems ably. The number of acceptances for Taiwan also increased companies design, assemble, and procure components for dramatically, even as the overall acceptance rate fell from 53 computers, communications equipment, and consumer elec- percent to 38 percent, and we expect that acceptances from India and China will increase in the near future as the quality of their university engineering programs improves. 51â Cited in âTaiwan ranks 4th in the world in US patents received,â Taipei Times, Oct. 17, 2006. 52â âData Snapshot,â Semiconductor Insights: Asia (FSA), Issue 1, 2006. 50â These figures were arrived at by McKinsey based on a survey of HR 53â Saxenian (2002). managers at multinational subsidiaries in these and other countries that 54â FSA (2005). asked the question: âOf 100 graduates with the correct degree, how many 55â Chikashi Horikiri, âTaiwan Transforms into IC Development Center,â could you employ if you had demand for all?â Nikkei Electronics Asia, February 2006.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 171 Average annual rejections Average annual acceptances U.S. Japan Korea Taiwan China India FIGURE 8â ISSCC acceptances and rejections by country, 2001â2006. Source: Tabulated from unpublished ISSCC data. TABLE 14â Value of Taiwanâs Semiconductor In the early stage of development of its semiconductor Industry, 2005 industry, Taiwan depended upon graduate training in the Output Value United States. Since the mid-1990s, the number of Taiwanese (US$ billions) Growth Since 2004 receiving Ph.D.s in engineering has declined steadily, and Brown -Linden today only a few are pursuing graduate training in the United Figure 8 IC design $8.63 5.8% Foundry services $18.90 â3.0% States. Although graduate education has improved in Taiwan, IC packaging $5.21 6.4% "fixed image" heard some concerns in our interviews about declining we IC testing $2.04 13.0% numbers of returnees from the United States. Past returnees Source: IEK-IT IS data, reported in âTaiwan IC production value reached brought with them both graduate training and work experi- US$34.8 billion in 2005, says government agency,â DigiTimes.com, ence that included management skills as well as practical J Â anuaryÂ 19, 2006. knowledge. The Taiwanese government has instituted several pro- grams to improve the local design sector, including a plan tronics for world-famous brands, including Hewlett-Packard, to train several thousand new design engineers in Taiwanâs Nokia, and Sony. In 1999, 62 percent of Taiwanâs chip-design universities, the creation of an exchange where local chip- revenue came from local sales.56 Taiwan is second only to design houses can license reusable functional blocks, and an the United States in fabless firms by revenue, with firms incubator where early-stage start-ups can share infrastructure specializing in cost-down, fast-follower capabilities. From and services.58 Another initiative is intended to attract chip- a U.S. perspective, Taiwanese competition has shortened design subsidiaries of major semiconductor companies; early the market window during which U.S. chip companies can takers include Sony and Broadcom (a major U.S. fabless recoup their investments in chips before similar products are company). In 2000, a government research institute created produced at a lower price. the SoC Technology Center (STC) to design functional Taiwanâs design teams were praised in our interviews for blocks that can be licensed to local companies, a model their execution, a vital trait in an industry where time-to- Taiwan has used successfully in other segments of the elec- market often means the difference between profit and loss. A tronics industry. STC has more than 200 engineers, most of frequent criticism, however, was that they were not yet truly whom have masterâs degrees or better.59 innovative. Ironically, Taiwanese companies are locked in as For the Taiwanese semiconductor industry, China pres- technology followers by their reliance on business from local ents both a major challenge and a major opportunity. The systems firms, which are as much as a generation behind the leading-edge technology.57 5 8 ââ Tr e n d s i n S O C d e s i g n u n t h aw a t S O C 2 0 0 4 ,â E D N , December 9, 2004. 56â Data from Taiwanâs Industrial Technology Research Institute cited in 59âSoC Technology Center interview, March 2005. âSoCâ is a common Table 5, Chang and Tsai (2002). industry acronym for âsystem-on-a-chipâ meaning a complex semiconduc- 57â Breznitz (2005). tor. integrating multiple functions.
172 THE OFFSHORING OF ENGINEERING challenge is competition in the foundry and fabless sectors, TABLE 15â Major Fabs in China, 2006 especially for low-cost designs using older technology, as Capacity well as competition for talented engineers to work in China First (wafers per and bring with them their knowledge of advanced technology Year of month, 8-inch in design and manufacturing. The opportunity is the chance Company Fab Location Production equivalent) to partner with Chinese companies elsewhere in the value Advanced Shanghai 1995 25,000 chain, enabling Taiwanese companies to provide high-end Semiconductor design services. In addition, Taiwanese companies would Manufacturing Corp (ASMC) have access to Chinaâs rapidly growing markets. Shanghai Hua Hong Shanghai 1999 50,000 So far, political issues have made it difficult for Taiwan- NEC Electronics ese chip companies to develop partnerships and markets in (HHNEC) China, even as they lose experienced engineers to Chinese Semiconductor Shanghai, 2001 150,000 competitors. Taiwan-born engineers are an important force Manufacturing Tianjin, and International Corp Beijing in technology development in China, in much the same way (SMIC) that the United States was an important force in technol- Grace Semiconductor Shanghai 2003 27,000 ogy development in Taiwan. Although China seems to be Manufacturing Corp benefiting more than Taiwan from the flow of engineers, (GSMC) capital, and business activities between the two countries, He Jian Technology Suzhou 2003 42,000 Taiwan Semiconductor Shanghai 2004 15,000; this may change over time if the Taiwanese government Manufacturing Co (40,000 planned) changes its policy. (TSMC) Source: iSuppli data, reported in Cage Chao and Esther Lam, âDespite China China-based foundries reporting full utilization rates in 1Q, Taiwan players not overly impressed,â Digitimes.com, March 22, 2006. China appears to be following a similar patternâ government sponsorship, local access to system firms (such as Haier, Huawei, and TCL) that are increasingly engaged in world markets, and active involvement of expatriates fabrication is now firmly established in China and will gradu- returning from the United States or experienced engineers ally expand. Although Chinaâs fabs pose a growing chal- relocating from Taiwan.60 In little more than a decade, with lenge to Taiwanese foundries, from the perspective of U.S. the help of foreign companies (as investors or as technology chip firms they add welcome competition to the market for licensors) and the Chinese government, Chinese firms have wafer processing. developed impressive fabrication capability. A potentially more worrisome development for U.S. firms Table 15 shows the main chip fabs in China, based pri- is the emergence of a fabless design sector in China. Since marily in Shanghai. The most striking feature is that they are 2003, China has claimed to have more than 400 chip-design all foundries working under contract rather than companies firms. Many are small, poorly managed companies that that design and manufacture their own products. U.S.-based deplete their seed money before they can bring a product to chip companies have few high-profile deals with Chinese market. Others offer design services rather than their own foundriesâthe major exception being Texas Instruments, products.64 One interviewee, echoed by others, claimed that which began working with Semiconductor Manufacturing many, if not most, firms outside the top 10 are engaged in International Corp (SMIC) in 2002 and added a deal to various types of reverse engineering, which is often illegal.65 co-develop SMICâs 90 nm process in 2004.61 Executives Foreign firms are generally reluctant to bring lawsuits, how- with U.S. experience have also played key roles. For ex- ever, for fear of displeasing the authorities and the likelihood ample, the CEOs of ASMC and HHNEC previously worked of losing in Chinese courts. But at least two U.S. companies at AMD.62 are suing Chinese rivals in export markets for intellectual Apart from SMIC, Chinaâs foundries have adopted property violations.66 modest growth plans, especially compared to the headline- Chinaâs top 10 chip-design firms in 2005 had total rev- grabbing predictions of two or three years ago.63 But chip enues of more than $1 billion, $400 million of which was from Hong Kong-based Solomon Systec, a designer of LCD 60â Saxenian (2002). 61â Mark LaPedus, âTI, SMIC sign deal to develop 90-nm technology by 64â Assessment of Byron Wu, iSuppli analyst, reported in âAnalyst: Chinaâs Q1 â05,â Silicon Strategies, Oct.28, 2004. IC design houses struggling for survival,â EE Times, May 20, 2004. 62â Chintay Shih, âExperience on developing Taiwan high-tech clus- 65â Interview with a European chip executive, conducted by Elena ter,â presentation at 4th ITEC International Forum, Doshisha University, Obukhova in Shanghai, December 2003. June 17, 2006. 66â See âAn offshore test of IP rights,â Electronic Business, May 2004; 63â Mike Clendenin, âDeflated expectations in Chinaâs IC biz,â EE Times, and âSigmaTel Sues Chinese Chipmaker over IP,â Electronic News, August 28, 2006. January 6, 2005.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 173 drivers that was spun off from Motorola in 1999.67 The next offshoring by U.S. firms. Of the top 20 U.S. semiconductor largest firms (Actions [media player chips], $150 million, companies, only a handful had opened design centers in and Vimicro [PC camera image processors], $95 million) China (compared to 18 in India) as of June 2006. Most of had IPOs on NASDAQ in 2005. these design centers are targeting the local market for the Chinaâs large, growing domestic market provides op- time being, and, according to press reports, some are engaged portunities for Chinaâs chip design companies to grow and in software or system design rather than chip design per se. become profitable, and in the future Chinese companies may Concerns about intellectual property protection appear to be able to design products for the global marketplace. The pose a greater barrier to foreign design activity in China local systems firms provide a sizable market for local fabless than in India.75 start-ups. The best chip design work is being done by local Chip design in China is at an early stage, but the relatively systems firms and a few world-class start-ups headed by U.S. young Chinese chip-design engineers will steadily build their returnees. experience. One factor that favors the development of local The Chinese government has taken many steps to sup- design companies is that Chinese engineers prefer to work port chip-design firms, some of the largest of which are for domestic start-ups and domestic companies rather than state owned. These measures include tax reductions, ven- MNCs. Many young Chinese engineers, especially returnees, ture investing, incubators in seven major cities, and special are willing to risk working for emerging companies that may government projects.68 A value-added tax preference for earn them great wealth. Some companies, particularly those domestically designed chips was phased out under U.S. whose founders include expatriates with foreign education pressure and will reportedly be replaced by a WTO-friendly and experience, are likely to begin to impact global markets R&D fund, although this had not been announced as of this by the end of the decade. It is still too early to predict the fu- writing (September 2006).69 ture relative importance of domestically owned and foreign- The return of Chinese nationals with education and work owned chip-design activities, or to predict whether domestic experience has been an important part of Chinaâs recent firms will be involved mostly with contract services or with technology development.70 Returnees provide valuable man- creating and selling chips. agement experience and connectivity to global networks that The education of semiconductor engineers in China is tend to accelerate the development of Chinaâs chip sector.71 also at an early stage. As discussed above, the quality of According to government statistics on student returnees, in Chinese engineering graduates varies widely, and few have 2003, of the 580,000 students reported to have gone abroad the knowledge and skills necessary to work on advanced since 1978, 150,000 had returned.72 The returnees had started technology or for MNCs. However, MNCs, including chip 5,000 businesses, including more than 2,000 IT companies and EDA firms, have been involved in improving engineer- in Beijingâs Zhongguancun Science Park (one-sixth the ing education in China, and the government has been actively park total).73 China is working to attract more high-tech recruiting world-class engineering professors to Chinese returnees with a range of specially targeted incentives and universities. Over time we expect semiconductor engineer- infrastructure.74 ing education, especially at the graduate level, to continue China is not yet an important destination for design improving. For now, returnees from the United States and experienced engineers from Taiwan will continue to play an important role in transferring technology to China. 67â Chinese government data cited in Mcallight Liu, âChinaâs Semicon- ductor Market: IC Design and Applications,â Semiconductor Insights: Asia (FSA), Issue 1, 2006 and iSuppli data in Mark LaPedus, âiSuppli lists India Chinaâs top fabless IC rankings,â EE Times, April 21, 2006. 68â âSynopsys Teams with Chinaâs Ministry of Science and Technology, The semiconductor industry in India presents a very dif- SMIC,â Nikkei Electronics Asia, March 21, 2003; âAn Uneven Playing ferent picture. India faces benign neglect by the government, Field,â Electronic News, July 3, 2003; âChina nurtures home-grown semi- a lack of manufacturing for chips and systems, and fewer conductor industry,â EBN, December 8, 2003; âChina government to sup- returnees from the United States.76 Unlike Taiwan and China, port Solomon Systech, Actions and Silan,â DigiTimes, April 14, 2005. 69â âChina to form R&D fund to replace VAT rebate, says report,â EE India has no high-volume chip manufacturing, although as Times, April 15, 2005. many as five proposals to build foundries are in various 70â Saxenian (2002). stages of negotiation.77 71â âStory behind the Story: Design in China is growing, but not ex- India is estimated to have 120 chip-design firms, and ploding,â audiocast by Bill Roberts, Electronic Business, September 1, revenues from chip design in 2005 were estimated to be 2006, http://www.edn.com/article/CA6368425.html?text=%22design+in+ china%22#. 72â âMore overseas Chinese students returning home to find opportunities,â November 16, 2003, http://www.china-embassy.org/eng/gyzg/t42338.htm. 75â âSIA Pushes Steps to Better IP Protection in China,â Electronic News, 73â âMore overseas Chinese students return home,â January 1, 2004, http:// November 17, 2004. www.china-embassy.org/eng/gyzg/t57364.htm. 76â Saxenian (2002). 74â Mike Clendenin, âChina starting to lure back its best brains,â EE Times, 77âRuss Arensman, âMove over, China,â Electronic Business, January 3, 2002. March 2006.
174 THE OFFSHORING OF ENGINEERING $583 million.78 Most chip design is taking place in MNC inadequate infrastructure. As in China, the quality of Indian subsidiaries, including most of the top 20 U.S. companies engineering graduates varies greatly. This problem is exac- and many European companies. The flow of semiconduc- erbated in India because most engineers there want to study tor engineering talent to MNCs has slowed the diffusion of computer science rather than electronics, and many are not technology to local firms, and India has no major fabless aware of the job opportunities in semiconductors. Graduate companies designing chips for sale under their own brand. education in EE is in its infancy, and doctoral education Domestic chip-design companies with varied capabilities in the seven major technical universities is not up to U.S. mainly provide design services. According to a study by the standards. The very low wages paid to professors, the lack India Semiconductor Association, local design companies of expensive and constantly changing EDA tools, and the use a time- and material-based pricing method by which difficulty and expense of having sample chips fabricated, all specific tasks are allocated to be carried out within set time contribute to problems in the development of world-class lines.79 These companies tend to develop simple subsystems graduate education. based on customer specifications. In addition, India has not attracted nearly as many return- Larger independent design-services firms are much more ees as China. The low flow of new domestic graduates and sophisticated. They use a fixed-price method, are able to returnees into the EE labor supply, coupled with the need for provide end-to-end solutions that incorporate in-house pro- at least three to five years of experience for fully productive prietary intellectual property, and offer design services across chip designers, has meant that the supply of design engi- the VLSI design flow. The government is developing policies neers has not kept pace with increasing demand. As a result, to support domestic chip-design firms. wages for chip designers have been rising rapidly, both at In our fieldwork we found that Indian engineers prefer the entry level and during the first five years. As mentioned MNCs to local start-ups, which are perceived as risky by above, salaries for engineers with five years of experience engineers and their family members. This is a contrast with are double entry-level salaries. China, where engineers are relatively eager to join start-ups, Inadequate infrastructure, especially in Bangalore, also which often receive some government support. poses serious problems for chip-design centers. Because of Foreign chip companies have been attracted by Indian the lack of a stable energy supply and lack of office space, engineersâ knowledge of English and the successful Indian foreign subsidiaries must make substantial investments to software sector. Many early investments by chip companies provide both offices and electricity. Bangalore, the countryâs were focused on software, the writing of microcode that primary city for high-tech, is plagued by narrow, pothole- becomes part of a chip. Over time, Indian affiliates have filled roads that are often gridlocked, forcing employees to taken on a bigger role, eventually extending to complete chip spend long hours commuting. In addition, high-tech com- designs from specification to physical layout. This transition panies are spread throughout the city, making commuting sometimes happens quickly. Intel, for example, opened a between companies, or even between company locations, software center in Bangalore in 1999 and began building a very time consuming. design team for 32-bit microprocessors in 2002.80 In addition, the housing stock in Bangalore has not kept up Since most domestically trained engineers lack knowl- with growth, and housing prices and rents have been rising edge of the technology being transferred, the necessary rapidly. Many employees are faced with a choice of living management skills, and knowledge of the entire product in inadequate housing or living far from work. The housing cycle, American MNCs are highly dependent on returnees and schooling problems are especially severe for returnees with advanced degrees from the United States to develop from the United States, who want to replicate the quality new projects in India. So far there have been few instances of U.S. housing and schools their families know. Several of design engineers in India leaving MNCs to start their own executives told us that their cost of living in Bangalore was companies, as often happens in the United States. However, almost as high as in the United States because of the high we heard of at least two cases in the past two years at one cost of housing and international schools.82 U.S. subsidiary. We also heard that leaving an MNC to start The shortage of engineering talent and weak infrastruc- a company is becoming more acceptable among Indian en- ture have constrained the rate of growth in the semiconductor gineers, many of whom are motivated to help India develop design industry, both for foreign subsidiaries and for local rather than to accumulate great wealth.81 companies, in India generally, and in Bangalore particularly. Foreign subsidiaries face formidable problems in their Some companies have been moving operations to areas that Indian operations, including a very tight labor market and have better infrastructure and are less expensive than Ban- galore. However, the talent shortage remains, especially for 78â Data from Frost & Sullivan, in Chitra Giridhar, âIndia design firms as experienced engineers with advanced degrees. product innovators,â Electronic Business, July 18, 2006. 79â âStudy: Indian design firms prefer time and material model,â EE Times, Sept 22, 2006. 80â âIntel, TSMC Set Up Camps In Developing Asian Markets,â WSJ.com, August 30, 2002. 81â Personal communications in Bangalore, November 2005. 82â Personal communications in Bangalore, November 2005.
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 175 OUTLOOK AND CONCLUSION in head count. Qualitative opinions were also divided, with optimists noting that reduced costs have strengthened com- The United States remains the world leader in the semi- panies and increased job security, and pessimists bemoaning conductor industry in terms of market share, development of downward pressure on wages and employment as well as the successful new companies, supply of experienced engineers, possible loss of intellectual property and, in the long run, and graduate engineering education. Moreover, the United industry leadership.86 States is the leading location for system design, the stage at We have observed that some movement of design jobs which most semiconductor purchase decisions are made.83 is related to the business cycle. There was a wave of design Our competitors, especially Japan, Korea, Taiwan, and the offshoring at the height of the dot-com bubble. Then, when European Union, look to the United States for lessons on the cascading effect of the subsequent downturn reached the encouraging innovation and start-ups in the semiconductor semiconductor industry, chip companies began cutting staff industry. Nevertheless, competition from low-cost countries, at home. Now that the recovery requires the expansion of especially China and India, which have rapidly growing and design operations, chip companies appear to be expanding potentially large markets, may pose competitive threats to design operations abroad faster than at home.87 It is too early U.S. companies and engineers in the future. to predict where this relative shift in the geographic distribu- tion of employment will find its new equilibrium. Outlook for U.S. Engineers Even experts disagree about whether or not the United States is educating too few engineers and scientists and is The job market for U.S. semiconductor engineers shows facing a shortage.88 This is partly because economists find it there is some strength in employment and earnings growth, hard to believe there can be a shortage in a labor market when but also shows evidence of labor market problems, espe- real earnings across the board are stagnant. This is partly a cially for older engineers and for the bottom 10 percent at reflection of government policies that affect the immigration all educational levels. We also observed signs of a decline and education of high-tech engineers. in the earnings premium for graduate degrees (M.S./Ph.D. compared to a B.S.), and low returns-to-experience for en- gineers with graduate degrees. The situation is especially Policy Issues difficult for older engineers whose skills can rapidly become The industryâs offshoring has gone well beyond the point obsolete. Experienced design engineers are often forced to at which blunt instruments such as trade policy can help work on mature technologies, which pay less and may pres- engineers without harming companies. Taxes or quotas on ent fewer interesting problems. For example, according to a traded activities or goods would raise costs for the many salary survey in 2004 by EE Times, the average annual salary companies that have already invested offshore in a wide for U.S. and European engineers skilled at designing for the array of design and manufacturing activities for both the latest chip-process technology was $107,000, whereas engi- foreign and domestic chip markets. Policy changes are thus neers designing for more mature analog technology averaged unlikely to improve the demand side of the labor market. $87,000.84 Industry has, however, been actively lobbying for changes Results of a regional survey of Silicon Valley, considered on the supply side in the form of changes to educational and the cradle and creative font of the semiconductor industry, immigration policies that increase the supply of high-tech reveal that the recent job climate there is difficult. Over- workers. The winter 2005 newsletter of the Semiconductor all the number of jobs in Silicon Valley has continually Industry Association includes two articles on the subject, decreased since 2001, and jobs in the semiconductor and âMaintaining Leadership as Global Competition Intensifiesâ semiconductor-equipment industries declined 23 percent by the organizationâs president and âAmerica Must Choose between 2002 and 2005, although the average wage rose to Competeâ by the outgoing CEO of Intel. 12 percent during the same period. Thus the survey paints a One of the main targets of industry analyses is education. mixed picture of the health of the industry.85 Higher education policies, which reflect both university de- Not surprisingly, industry participants disagree about the cisions and government funding, determine the number and significance of offshoring for the U.S. job market. A 2004 country of origin of students at all levels, but especially at survey by EE Times of more than 1,453 chip- and board- the graduate level. Foreign nationals in our M.S. and Ph.D. design engineers and managers showed that about half programs in science and engineering have a direct impact on believed that foreign outsourcing would lead to a reduction the supply of knowledge workers, both in the United States 83â and in China and India. Foreign graduates of U.S. universi- iSuppli data reported in Dylan McGrath, âU.S. still top design influ- encer; China, India rising fast,â EE Times, September 28, 2006. 84â âAfter 10-year surge, salaries level off at $89k,â EE Times, August 86â âItâs an outsourced world, EEs acknowledge,â EE Times, 28, 2003. August 27, 2004. 85â Joint Venture: Silicon Valley Network, â2006 Index of Silicon Valley,â 87âSee, for example, âThe perfect storm brews offshore,â Electronic Busi- available online at http://www.jointventure.org/PDF/Index%202006.pdf. ness, March 2004. The data are from state unemployment insurance data, which is the basis 88âSee, for example, Freeman (2003, 2005); Task Force on the Future of for the Census data. American Innovation (2005); NRC (2000, 2001); Butz et al. (2004).
176 THE OFFSHORING OF ENGINEERING ties must obtain temporary visas, usually H1-B visas, before their careers, both in terms of improving pay and learning they can work in the United States after graduation. Legisla- new technologies and skills. Networking with colleagues tion is under consideration to provide permanent residency from oneâs alma mater and former companies as well as status to foreigners educated in the United States. We are through professional associations is an excellent way of hopeful that this policy will be implemented soon. keeping up with job opportunities as well as learning about Government policies regulating immigration, especially new technologies. the issuance of H-1B (Non-Immigrant Professional) and Our advice to semiconductor engineers is to embrace the L-1 (Intra-Company Transfer) visas, also have a significant mobile labor market and look to job changes as a way of impact on the number of foreign engineers engaged in advancing. Each job should be chosen carefully to improve semiconductor and software work. In a delayed response to skills and take advantage of previous job experience. En- the recession, changes in policy that took effect in 2004 set gineers must continually stay in touch with their networks severe limits on the number of visas for foreign workers. and share knowledge with their colleagues about what is When the number of H-1B visas was thus reduced, many happening in the field and about job opportunities. In short, U.S. companies used the opportunity to send foreign nation- engineers today must be in charge of their careers; they can als with U.S. education and experience back to India and no longer depend on employers to provide them with the China to help build operations there. training they need to keep up their skills. An area of policy that has received less attention is com- Foreign nationals working for U.S. companies can use pensation for engineers who are harmed by offshoring. As their networks to develop careers both in the United States a result of the offshoring of chip design, consumers have and in their home countries. Returnees who are willing to benefited from lower prices and new products (although return home for short- or long-term stints can bargain for much of that benefit is received outside the United States). good salary packages from U.S. employers. U.S. nationals Some of the short-term cost of offshoring, however, is be- should also go abroad to develop contacts and expertise in ing borne by engineers in particular companies or industry specific cultures and regional markets. sectors in which companies are restructuring globally. Semiconductor engineers are known for their flexibil- Currently, white-collar workers like chip designers do not ity and ability to solve challenging problems and to learn qualify for trade-adjustment assistance from the government new technologies. The semiconductor industry is likely to when their jobs are sent abroad. It would make sense to continue to undergo constant crisis and change, and chip help these highly-skilled workers with retraining and other engineers should use these industry characteristics to their forms of assistance to enable them to remain productive. As advantage in planning their careers by seeking jobs where Federal Reserve Chair Bernanke remarked, âThe challenge they can learn about new technologies and new markets. To for policy makers is to ensure that the benefits of global be successful in the industry, an engineer must see change economic integration are sufficiently wide-sharedâfor as an opportunity rather than a problem. example, by helping displaced workers get the necessary training to take advantage of new opportunitiesâthat a con- Lessons Learned sensus for welfare-enhancing change can be obtained.â89 Finally, we need more and better data. As researchers in In its short history, the semiconductor industry has faced other industries have noted, more labor market data, both for continual challenges and has done an extraordinary job of the United States and for our trading partners, are necessary overcoming them, often in innovative ways that were not for proper assessments of the effects of offshoring.90 In the anticipated. The industry has also continually experienced meantime, national policies affecting education, labor mar- large swings in demand and prices, and we expect the cycli- kets, and innovation will continue to be based upon informed cal nature of the industry to continue, even as the long-term speculation. trend moves upward. Our predictions for the future of the industry and recommendations for setting policy must not extrapolate from conditions in the short run, especially dur- How Should U.S. Engineers Respond? ing a downturn. We must look to the long-term history of American engineers are naturally responding to the the industry to ensure that policy decisions, either by gov- impact of the changing labor market on their careers. The ernments or by companies, are made on a solid foundation. highly rewarded career path of working for one company Macro-policies that ensure a strong economy with steady for an entire career is no longer an option. Most engineers growth are critical to the development of the semiconductor today must expect to work for several firms. In fact, chang- industry, which is negatively affected by national recessions ing jobs is now the most effective way for them to advance and high interest rates. Government support for higher education, especially 89â Edmund L. Andrews, âFed Chief Sees Faster Pace for Globalization,â graduate education, should be the cornerstone of public New York Times, August 25, 2006. policy to support innovation. A strong university system 90â See the excellent study by Tim Sturgeon et al. (2006). with state-of-the-art graduate training and strong links to
SEMICONDUCTOR ENGINEERS IN A GLOBAL ECONOMY 177 companies is critical for innovation in the semiconductor Industry Association, Chintay Shih, Gary Smith, Strategic industry. U.S. universities are essential to educating Ph.D.- Marketing Associates, Yea-Huey Su, Tim Tredwell, and C-K level engineers, who are as likely to be from Asia as from the Wang for their valuable contributions. Melissa Appleyard, United States. Social networks, such as workersâ contacts at Hank Chesbrough, Jason Dedrick, Rafiq Dossani, Richard their former universities and former employers, are important Freeman, Deepak Gupta, Bradford Jensen, Ken Kraemer, adjuncts to a companyâs formal knowledge base. Company Frank Levy, B. Lindsay Lowell, Jeff Macher, Dave Mowery, awareness of this is critical to ensuring that employeesâ Tom Murtha, Tim Sturgeon, Michael Teitelbaum, and Eiichi knowledge is recognized and used rather than flowing out- Yamaguchi, as well as participants at the NAE Workshop on ward into these networks. the Offshoring of Engineering, the 2005 Brookings Trade Forum on Offshoring of White-Collar Work, the Berkeley Innovation Seminar, and the Doshisha ITEC seminar series Conclusion provided thoughtful discussions that improved the paper. We The semiconductor industry is in the intermediate stages are especially grateful to Gail Pesyna at the Sloan Foundation of the complex, dynamic process of globalization. At this for her long-running support of, and input into, our research. point it is hard to predict the impact of offshoring on the The authors are responsible for any errors. competitive position of the U.S. semiconductor industry and on the earnings and employment of domestic engineers, and REFERENCES whether the new equilibrium will be acceptable. Thus policy interventions must be flexible. Breznitz, D. 2005. Development, flexibility, and R&D performance in the Taiwanese IT industryâcapability creation and the effects of Offshoring is an important step in the integration of India state-industry co-evolution. Industrial and Corporate Change 14(1): and China into the global economy. These countries appear 153â187. to be pursuing different roles vis-Ã -vis the United States, Brown, C., and B. Campbell. 2001. Technical change, wages, and employ- with Chinaâs chip industry acting more as a competitor (e.g., ment in semiconductor manufacturing. Industrial and Labor Relations fabless start-ups) and Indiaâs playing a more complementary Review 54(2A): 450â465. Brown, C., J. Haltiwanger, and J. Lane. 2006. Economic Turbulence: Is role (e.g., design services). Both countries will certainly a Volatile Economy Good for America? Chicago, Ill.: University of become more important in high-tech industries, both as Chicago Press. markets and suppliers. However, their ability to move up the Brown, C., and G. Linden. 2006. Offshoring in the Semiconductor Indus- semiconductor technology curve is constrained at the mo- try: A Historical Perspective. Pp. 279â333 in Brookings Trade Forum ment by a lack of graduate education, undeveloped financial 2005: Offshoring White-Collar WorkâThe Issues and the Implications, L. Brainard and S.M. Collins, editors. Washington D.C.: Brookings systems, and inadequate intellectual property protection, as Institution. well as severe problems facing their political systems. Butz, W., T. Kelly, D. Adamson, G. Bloom, D. Fossum, and M. Gross. 2004. We expect that the United States will maintain its leader- Will the Scientific and Technical Workforce Meet the Requirements of ship position in the semiconductor industry, and we expect the Federal Government? Santa Monica, Calif.: RAND Corporation. the industry and its resourceful engineers to continue to find Chang, P.-L., and C.-T. Tsai. 2002. Finding the niche positionâcompetition strategy of Taiwanâs IC design industry. Technovation 22(2): 101â111. ways to overcome challenges. For now, modifications in DHS (U.S. Department of Homeland Security). 2004. Characteristics of government policies affecting universities, immigration, and Specialty Occupation Workers (H-1B): Fiscal Year 2003. Washington workers affected by trade would alleviate some of the labor D.C.: DHS. market problems we have described. Freeman, R.B. 2003. Trade Wars: The Exaggerated Impact of Trade in Economic Debate. NBER Working Paper W10000. Cambridge, Mass.: National Bureau of Economic Research. ACKNOWLEDGMENTS Freeman, R.B. 2005. Does Globalization of the Scientific/Engineering Workforce Threaten U.S. Economic Leadership? NBER Working Clair Brown is professor of economics and director of the Paper 11457, June. Cambridge, Mass.: National Bureau of Economic Center for Work, Technology, and Society (IIR) at University Research. of California, Berkeley; Greg Linden is senior research fel- FSA (Fabless Semiconductor Association). 2005. Global Fabless Fundings & Financials Report: CY Q4 2004/Year End 2004. Dallas, Tex.: FSA. low at the center. Yongwook Paik provided excellent research GAO (Government Accountability Office). 2003. H-1B Foreign Workers: assistance. The authors would like to thank the Alfred P. Better Tracking Needed to Help Determine H-1B Programâs Effects on Sloan Foundation, the Institute for Industrial Relations at U.S. Workforce. GAO Report 03-883. Washington, D.C.: U.S. Gov- UC Berkeley, and the Institute for Technology, Enterprise ernment Printing Office. Available online at http://www.gao.gov/new. and Competitiveness (ITEC/COE) and Omron Fellowship items/d03883.pdf. Gereffi, G., and V. Wadhwa. 2005. Framing the Engineering Outsourcing at Doshisha University, Japan, for funding. Bob Doering Debate: Placing the United States on a Level Playing Field with China and Bill Spencer provided detailed and helpful comments and India. Engineering Management Program, Duke University. on the workshop version of this paper. We are also grateful Gwartney, J. and R. Lawson, with W. Easterly. 2006. Economic Freedom of to Ben Campbell, David Ferrell, Michael Flynn, Gartner the World: 2006 Annual Report. Vancouver, B.C.: Fraser Institute. Dataquest, Ron Hira, Dave Hodges, Rob Leachman, Daya Kumagai, J. 2003. Itâs Cold Out There. IEEE Spectrum 40(7): 13â15. Nadamuni, Elena Obukhova, Devadas Pillai, Semiconductor
178 THE OFFSHORING OF ENGINEERING Macher, J.T., D.C. Mowery, and D.A. Hodges. 1998. Back to Dominance? Garner, A.C. 2004. Offshoring in the service sector: economic impact and U.S. Resurgence in the Global Semiconductor Industry. California policy issues. Economic Review (Federal Reserve Bank of Kansas City), Management Review 41(1): 107â136. Third Quarter: 5â37. McKinsey Global Institute. 2005. The Emerging Global Labor Market. San Groshen, E.L., B. Hobijn, and M.M. McConnell. 2005. U.S. jobs gained Francisco, Calif.: McKinsey & Company. and lost through trade: a net measure. Current Issues in Economics and NRC (National Research Council). 2000. Forecasting Demand and Sup- Finance 11 (8): 1â7. ply of Doctoral Scientists and Engineers. Washington, D.C.: National Gupta, D. 2005. Ties That Bind: An Econometric Analysis of the Return Academy Press. Home of Foreign-Born U.S. Ph.D.s. Working Paper 2005-1. Amrita NRC. 2001. Building a Workforce for the Information Economy. Washing- School of Business, Ettimadai, India. ton, D.C.: National Academy Press. Ham, R.M., G. Linden, and M.M. Appleyard. 1998. The evolving role of NSF (National Science Foundation). Various years. Science and Engineering semiconductor consortia in the U.S. and Japan. California Management Doctorate Awards. Arlington, Va.: NSF Division of Science Resources Review 41(1): 137â163. Statistics. Hemani, A. 2004. Charting the EDA roadmap. IEEE Circuits & Devices PriceWaterhouseCoopers. 2004. Chinaâs Impact on the Semiconductor 20(6): 5â10. Industry, TC-02-04. San Jose, Calif.: PriceWaterhouseCoopers. Henderson, J.W. 1989. The Globalisation of High Technology Production: Saxenian, A. 2002. Transnational communities and the evolution of global Society, Space, and Semiconductors in the Restructuring of the Modern production networks: the cases of Taiwan, China and India. Industry and World. New York: Routledge. Innovation 9(3): 183â202. Henisz, W.J., and J.T. Macher. 2004. Firm- and country-level trade-offs Sturgeon, T.J. 2006. Services Offshoring Working Group: Final report. and contingencies in the evaluation of foreign investment. Organization September. MIT Industrial Performance Center, Cambridge, Mass. Science 15(5): 537â554. Task Force on the Future of American Innovation. 2005. The Knowledge Hira, R. 2004. U.S. immigration regulations and Indiaâs information Economy: Is the United States Losing Its Competitive Edge? Available technology industry. Technological Forecasting & Social Change 71: online at www.futureofinnovation.org/ as of April 20, 2005. 837â854. Hira, R. 2005. Impacts and Trends of Offshoring Engineering Tasks and Jobs. The Bridge 35(3): 22â27. Available online at http://www.nae.edu/ ONLINE NEWS SOURCES USED NAE/bridgecom.nsf/weblinks/NAEW-6GEK5J?OpenDocument. Howell, T.R., B.L. Bartlett, W.A. Noellert, and R. Howe. 2003. Chinaâs The following list provides Internet addresses for the online publications Emerging Semiconductor Industry: The Impact of Chinaâs Preferential cited in our footnotes. In some cases, the original Internet site we used to Value-Added Tax on Current Investment Trends. San Jose, Calif.: Semi- access the article has been folded into another site. The one listed is the conductor Industry Association. current site. Some of the articles are also archived at third-party sites such International Business Strategies. 2002. Analysis of the Relationship be- as findarticles.com, so it is generally simplest to search for the article title tween EDA Expenditures and Competitive Positioning of IC Vendors: as an exact phrase. Articles older than five years may no longer be available A Custom Study for EDA Consortium. Available online at www.edac. online. org/downloads/resources/profitability/HandelJonesReport.pdf. Ac- DigiTimesâ http://digitimes.com/ cessed April 15, 2005. EDNâ http://www.edn.com Leachman, R.C., and C.H. Leachman. 2004. Globalization of Semiconduc- Electronic Businessâ http://www.edn.com (formerly separate) tors: Do Real Men Have Fabs, or Virtual Fabs? Pp. 203â231 in Locating Electronic Newsâ http://www.edn.com (formerly separate) Global Advantage: Industry Dynamics in the International Economy, Express Computer (India)â http://www.expresscomputeronline.com edited by M. Kenney with R. Florida. Stanford, Calif.: Stanford Uni- The Registerâ http://www.theregister.co.uk/ versity Press. Silicon Strategiesâ http://www.eetimes.com (formerly separate) Linden, G., C. Brown, and M. Appleyard. 2004. The Net World Orderâs Electronic Newsâ http://www.edn.com (formerly separate) Influence on Global Leadership in the Semiconductor Industry. Pp. New York Timesâ http://www.nytimes.com 232â257 in Locating Global Advantage, edited by M. Kenney with R. Nikkei Electronics Asiaâ http://techon.nikkeibp.co.jp Florida. Stanford, Calif.: Stanford University Press. Semiconductor Business Newsâ http://www.eetimes.com (formerly Linden, G., and D. Somaya. 2003. System-on-a-chip integration in the separate) semiconductor industry: industry structure and firm strategies. Industrial Semiconductor Internationalâ http://www.reed-electronics. and Corporate Change 12(3): 545â576. com/semiconductor/ MAEI (Malaysian-American Electronics Industry). 1995. Malaysian- Taipei Timesâ http://www.taipeitimes.com American Electronics Industry: Annual Survey 1994/1995. Mimeo. WSJ.comâ http://online.wsj.com Kuala Lumpur, Malaysia: MAEI. McKendrick, D.G., R.F. Doner, and S. Haggard. 2000. From Silicon Valley to Singapore : Location and Competitive Advantage in the Hard Disk ADDITIONAL READING Drive Industry. Stanford, Calif.: Stanford University Press. Semiconductor Industry Association. 2003. International Technology Borrus, M. 1988. Competing for Control: Americaâs Stake in Microelectron- Roadmap For Semiconductors: Design.â Available online at public.itrs. ics. Cambridge, Mass.: Ballinger Publishing. net/Files/2003ITRS/Home2003.htm as of April 19, 2005. Ernst, D. 2004. Internationalisation of Innovation: Why is Chip Design Tilton, J.E. 1991. International Diffusion of Technology: The Case of Semi- Moving to Asia? Working Paper 64 (rev. March 2004). Honolulu, Hi.: conductors. Washington D.C.: Brookings Institution. East-West Center. Flamm, K.S. 1985. Internationalization in the Semiconductor Industry. Pp. 38â136 in The Global Factory: Foreign Assembly in International Trade, J. Grunwald and K.S. Flamm, editors. Washington, D.C.: Brook- ings Institution.