R&D and Firm Performance in the Semiconductor Industry

Namchul Shin
Pace University
nshin@pace.edu

Kenneth L. Kraemer
University of California, Irvine
kkraemer@uci.edu

Jason Dedrick
Syracuse University
jdedrick@syr.edu

Forthcoming in Industry and Innovation

Published online at: http://www.tandfonline.com/eprint/UcjU5bzqwSWzHVtb825q/full

 R&D and Firm Performance in the Semiconductor Industry

Abstract

While the semiconductor industry is still dominated by large vertically integrated firms, fabless
firms, which outsource their manufacturing, are gaining market share. Fabless firms are
considered to have an advantage in product innovation, as they can focus their innovation efforts
on chip design and can benefit from investments in process innovation made by their
manufacturing partners. However, there is little empirical evidence of the performance of fabless
firms compared to vertically integrated firms. This research empirically examines the
relationship between R&D and the financial performance of fabless and vertically integrated
firms from 2000 to 2010. Our results show that fabless firms maintain higher gross and net
margins, earn a higher return on assets, and have greater intangible value (Tobin’s q) than
vertically integrated firms when controlling for size, capital intensity, and R&D ratio
(R&D/Sales). This supports the argument that fabless firms achieve greater performance by
focusing on one part of the innovation process. The relationship of R&D ratio to net margin is
negative for the whole sample, suggesting that the industry may be overinvesting in R&D.
Notably, the negative relationship is greater for fabless firms, which spend a higher amount of
their sales on R&D. The relationship of R&D ratio to return on assets and Tobin’s q is negative,
and there is no significant difference between fabless and integrated firms. We conclude that
fabless firms outperform integrated firms overall, but are somewhat worse in terms of increasing
profits and creating value from their R&D investments.

KEY WORDS: Innovation, firm performance, semiconductor industry, vertical disintegration,

fabless firms, vertically integrated firms

1
1. Introduction

During the last two decades, the semiconductor industry has witnessed a trend toward

decoupling of integrated circuit design from manufacturing with the emergence of “fabless”

semiconductor firms (Langlois, 2003), such as Qualcomm and Broadcom, as well as the growth

of independent contract manufacturers (foundries), such as Taiwan Semiconductor

Manufacturing Company (TSMC). Fabless firms focus on building knowledge in chip design,

development and marketing functions, and outsource manufacturing to foundries specializing in

chip fabrication (Dibiaggio, 2007). The trend toward vertical disintegration in the semiconductor

industry (i.e., specialization in either design or manufacturing) has been spurred by the extensive

use of information technologies in design activities and the effectiveness of intellectual property

rights in protecting design knowledge, along with adoption of standards to facilitate the transfer

of designs from fabless firms to foundries for fabrication (Dibiaggio, 2007; Mudambi and

Venzin, 2010).

While the semiconductor industry is still dominated by large vertically integrated firms,

fabless firms are gaining market share. However, there is little understanding of which type of

firms perform better and benefit more from investments in R&D. A key question we raise for

this research is “Do fabless firms that focus on IC (integrated circuit) design perform better and

capture higher profits from R&D than vertically integrated firms that undertake both design and

manufacturing?”

By outsourcing chip fabrication and assembly, a fabless firm does not need to spread its

R&D budget over other value chain activities and can leverage the R&D expenditures of its

suppliers. By contrast, a vertically integrated firm is obliged to spread its R&D budget over the

entire value chain, but benefits from the integration of manufacturing and design (Mudambi and

2
Venzin, 2010). In order to address the question of whether fabless firms perform better and earn

higher profits from R&D than integrated firms, we conduct an exploratory study by empirically

examining the relationship between R&D and the performance of fabless and vertically

integrated firms in the semiconductor industry. We employ the Electronic Business 300 data set

and the Hoovers database for the eleven years from 2000 to 2010.

In the next section, we discuss vertical disintegration and the emergence of fabless firms

in the semiconductor industry, and theories explaining the phenomena. Then we develop the

theoretical framework for explaining the relative performance of fabless firms and vertically

integrated firms and propose hypotheses. Section 3 describes our research methods and data

sources. We present our results in Section 4 and discussion and conclusions are provided in

Section 5.

2. Theoretical Background

2.1 Vertical Disintegration and the Emergence of Fabless Semiconductor Firms

Innovation is a key driver for creation of economic value and profitability by firms. For

successful commercialization, innovation often must be coupled with complementary assets such

as manufacturing and marketing (Teece, 1986). According to Chandler (1977), the internal

economies of the modern corporation grew because it needed to better coordinate inputs and

outputs, achieving high throughput, so it could better utilize expensive mass-production

equipment. High throughput depends on an assured supply of inputs and demand for outputs, so

complementary activities such as design and engineering tend to be kept within the boundaries of

firms, resulting in vertical integration (Sturgeon, 2002; Helper and Sako, 2010).

3
 However, when complementary assets are more standardized and not firm-specific, they

can be obtained from the competitive market through contractual relationships (Teece, 2006).

Coordination of such transactions can be facilitated by the use of new technologies, such as

electronic design automation (EDA) software, and interface standards which reduce the costs of

transactions with external partners.

In accordance with Chandler’s theory (1977), the trend towards large vertically integrated

firms was observed in much of the 20 th century. By the 1990s, however, there was a drive toward

vertical disintegration, especially in fiercely competitive and rapidly changing industries such as

electronics and semiconductors.1 Many original equipment manufacturers (OEMs) in electronics

sold their factories and outsourced manufacturing to contract manufacturers (CMs). The CMs

gained high capacity utilization and economies of scale by serving multiple OEMs, while OEMs

concentrated on R&D, product design, and marketing. OEMs could benefit from their own

product innovation and from the process innovation by CMs to improve operational efficiency

and reduce production costs.

Other costs might come into play, however, and outsourcing may only pay off under

certain conditions. Based on the theory of incomplete contracts (specifically the difficulty of

including innovation efforts on the part of each party in a contract), Plambeck and Taylor (2005)

argue that outsourcing production will increase OEM profitability only if OEMs are in a strong

bargaining position relative to the CM and can capture the increased revenue that it gains from

innovation. If the OEM’s bargaining power is weak, the CM will be able to capture some of the

benefits of OEM innovation, and innovation by both parties (product innovation by OEMs,

process innovation by CMs) is likely to be less than optimum for the whole system.

1
The outsourcing of manufacturing is part of a much larger trend of vertical disintegration that occurred in other
industries (e.g., electronics manufacturing, aerospace, apparel and footwear, automotive parts, and pharmaceutical
manufacturing).

4
 Another factor influencing the impacts of outsourcing identified in Williamson’s (1975)

transaction cost theory is the extent to which OEMs are required to make asset-specific

investments to work with a partner. The need to design products that can be manufactured by the

outsourcing partner creates transaction costs for the OEM and potentially leads to a “hold-up”,

where the supplier can raise prices and cut into the OEM’s profits. An example is laptop

computers, where products are designed in collaboration with original design manufacturers

(ODMs)2 such as Quanta, Compal and Wistron with a specific factory and manufacturing process

in mind. Once the product is designed and manufacturing starts, it is costly to switch to another

manufacturer during the life of that product. As a result, laptop OEMs generally open up the

development and manufacturing of each new major product for bid to several ODMs to maintain

bargaining power (Dedrick and Kraemer, 2006).

Taking the costs and risks of incomplete contracts and asset specific investments into

account, we would expect firms that outsource production to outperform integrated

manufacturers under two conditions. The first is when the outsourcing firm maintains bargaining

power to prevent a “hold-up” of higher prices by suppliers. Bargaining power may be determined

by the relative size of the OEM and CM, and how important each is to the other’s business. If an

OEM is larger, and constitutes a large share of a CM’s business, it should retain bargaining

power and be able to capture a greater share of profits. 3 The second is asset specificity and

related switching costs. For example, if an OEM has to make asset specific investments in

information systems or design tools to work with a CM, switching costs will be higher and the

potential for a “hold-up” by CMs is greater. We look at the semiconductor industry in terms of

bargaining power and asset specific investments to posit relationships between sourcing choice
2
ODMs are a specific type of contract manufacturer that is involved in both the design and manufacturing of a
product for OEM.
3
However, Feng and Lu (2012) conducted a theoretical analysis and found that as a manufacturer’s bargaining
power decreases, its profit under outsourcing may increase.

5
and financial performance. If bargaining power is high and asset specificity is low, we would

expect firms that outsource production to do better relative to vertically integrated firms.

According to Brown and Linden (2011), one of the fundamental economic forces in the

semiconductor industry is the rising cost of fabrication. A wafer fabrication plant can cost $10

billion, a large investment that is beyond the reach of all but a few chipmakers. Facing the rising

cost of fabrication, some companies have consolidated through acquisitions to be large enough to

justify investing in new fabs. For instance, Japanese chipmaker Renesas is the product of the

merger of chip making units of NEC, Mitsubishi and Hitachi. The value of mergers and

acquisitions in the chip business reached an estimated $100 billion in 2015 alone (Pfanner and

Fukase, 2015).

Another trend has been the emergence of independent contract manufacturers (foundries)

for microchips. The largest of these is Taiwan Semiconductor Manufacturing Company (TSMC).

Others include GlobalFoundries, United Microelectronics Corporation (UMC) and

Semiconductor Manufacturing International Corporation. Even leading chipmaker Samsung

provides fabrication services for fabless chipmakers, including smart phone rival Apple. This

new business model in which companies specialize in either chip design or manufacturing is

referred to as the fabless-foundry model. The share of fabless firms in total semiconductor

revenues has grown from 7.1% in 1999 to 29.2% in 2013 (Clarke, 2014).

While fabless firms have become an integral part of the semiconductor industry, and

vertically integrated firms such as AMD and TI have sold all or part of their fabrication capacity,

the entire industry is not moving “fabless”. Companies such as Intel and Samsung, the two

largest chip makers, continue to do their own fabrication since designers can work closely with

6
process development engineers to push the technical parameters of the fabrication process

(Monteverde, 1995; Brown and Linden, 2011).

2.2 Theory Development and Hypotheses

The fabless-foundry and integrated device manufacturer models provide an opportunity

to systematically examine the relative performance of outsourced versus vertically integrated

firms in the semiconductor industry. Some industry analysts argued that fabless firms earned

higher gross margins than integrated firms in the market for core silicon 4 consisting of

application specific integrated circuits (ASICs), application specific standard products (ASSPs),

and programming logic devices (PLDs) (EE Times, 2005). However, Brown and Linden (2011)

showed that although it is volatile, on average, the performance of fabless and integrated firms in

terms of return on assets (ROA) converged over the period of 1995 to 2006.

How does the fabless model compare with vertical integration in terms of bargaining

power and asset specific investments? Beginning with bargaining power, we note that the five

largest fabless firms had revenues ranging from $17.2 billion (Qualcomm) to $4.0 billion

(Nvidia) in 2013. The largest foundry, TSMC, dominated the foundry industry with revenues of

$19.8 billion, nearly half of the industry’s revenues, with a profit margin of 32%. The fifth

largest, Powerchip, had revenues of just $1.2 billion. Overall, fabless firm revenues totaled $77

billion (Clarke, 2014), compared to total foundry revenues of $42 billion (IC Insights, 2014).

TSMC’s six largest customers accounted for 49% of its revenues, led by Qualcomm at 15%.

Meanwhile, Qualcomm relies on TSMC for most of its manufacturing, so it could be argued that

Qualcomm’s bargaining power with TSMC is limited. However, Qualcomm has engaged other

4
Core silicon refers to the semiconductors that implement specific, individual functionality in an electronic system
(iSuppli, 2010).

7
foundries in recent years, reducing TSMC’s bargaining power. In general, the sizes of fabless

firms and foundries line up pretty evenly, so bargaining power is probably balanced between the

two groups.

In terms of asset-specific investments, the importance of design for manufacturability

means that a fabless firm must work closely with the foundry to achieve needed chip

performance and cost. This requires making asset specific investments in the relationship.

(Brown and Linden, 2011). It is notable that the emergence of fabless firms has helped drive the

adoption of new technologies, such as EDA software, and standardization of the interface

between chip design and fabrication, which reduce transaction costs and the need for the transfer

of complex information.

To summarize, fabless firms have reasonable bargaining power with foundries to capture

profits generated by product and process innovation. The asset-specific investments required

create a risk of “hold-up” by foundries, which might hurt the profits of fabless firms relative to

integrated device manufacturers who internalize the design-for-manufacturability process.

However, by closely working with foundries using EDA software and interface standards,

fabless firms can lower the transaction costs incurred from asset specific investments. With

reasonable bargaining power and lowered asset specificity, we would expect fabless firms

perform better than integrated firms, given the ability of fabless firms’ to concentrate their own

innovation on product design and to tap into the process investments and the high capacity usage

of foundries. Thus we hypothesize that:

Hypothesis 1: Fabless firms achieve better financial performance than vertically

integrated firms.

8
 Next we look at the ability of fabless firms and integrated firms to capture benefits from

innovation by earning higher profits and return on their investments in R&D. We discuss this by

introducing Teece’s ‘profiting from innovation’ (PFI) framework (Teece, 1986; 2006).

Teece states that the ability of a firm to profit from innovation depends on three factors:

(1) the appropriability regime, (2) complementary assets, and (3) the dominant design paradigm.

First, in order to profit from innovation, a firm needs some mechanism to appropriate value and

prevent imitation by other firms, for instance through patents, copyright, trade secrets or a strong

brand name. Second, in order to be commercially successful, an innovation often must be

utilized in conjunction with other assets (or capabilities), such as marketing, competitive

manufacturing, and after-sales support. These are so-called complementary assets. When

complementary assets are specialized or co-specialized and cannot be obtained through

marketplace transactions, some form of vertical integration is required to obtain them. On the

other hand, when complementary assets are more standardized and are available in the

competitive market, contractual relations (or partnerships) may be a better strategy for successful

commercialization of innovation.

By outsourcing chip fabrication and assembly, fabless firms have access to

complementary assets held by their foundry partners. Unlike integrated firms, they do not have to

make huge investments on in-house capital assets while trying to keep up with rapid

technological change and volatile market demand (Sturgeon, 2002). A firm focusing on specific

activities in the value chain also does not need to spend its R&D budget on outsourced activities

(Mudambi and Venzin, 2010). Fabless firms benefit from the R&D investment of the foundries

who must invest heavily to stay near the leading edge in new fabrication processes. This should

9
enable fabless firms to leverage their foundries’ investments in valuable complementary assets,

(e.g., process knowledge), while minimizing the cost of integrating design and fabrication across

firm boundaries through the use of software tools and design standards.

In contrast, vertically integrated firms must spread their R&D budget over more activities

in the value chain, possibly reducing their return on R&D investments. However, they are more

likely to possess relevant specialized complementary assets within their boundaries, which might

give them an advantage over fabless firms, which cannot move faster on new product

development than what their foundry partners can support (e.g., larger wafers or smaller line

widths). Thus, they also capture a share of the revenue from those assets or capabilities. While

there are factors that favor either fabless or integrated chip makers, and we believe the balance

favors the fabless firms, it appears whether fabless firms achieve higher returns from R&D is an

empirical question. Thus, we propose the following hypothesis:

Hypothesis 2: Fabless firms achieve higher returns from R&D than vertically integrated

firms.

3. Methodology and Model

In order to examine the performance of fabless and integrated firms, we conducted

ordinary least-squares (OLS) and two-stage least-squares (2SLS) regression analyses of firm

performance with a “fabless firm” dummy variable, R&D spending, an interaction term for R&D

spending and the “fabless firm” dummy variable. The “fabless firm” dummy variable indicates

whether the firm is a fabless firm or an integrated firm.

10
 2SLS is employed with instrumental variables to correct potential bias caused by the

simultaneity problem (reverse causality). Although R&D spending may improve firm

performance, the opposite may also be true; firms with higher profits may spend more on R&D.

We use one-year lagged variables of R&D and capital intensity as instrumental variables. Lagged

R&D has been identified in previous research as a useful instrument (Brundell, Griffith, and Van

Reenen, 1999; Kleis et al., 2012). Capital intensity might be related to R&D spending, but not

necessarily to the variance of firm performance. We also validate our instrumental variables by

using Wooldridge chi-square statistic. These statistics are used to test the null hypothesis that the

instruments are valid and not correlated with the error term. Statistically significant test statistics

indicate that the instruments may not be valid.

Firm performance can be conceptualized and measured in different ways. We use the

following measures: (1) Gross margin and net margin, which measures profits as a percent of

revenues, (2) return on assets, which measures a firm’s ability to effectively utilize its

investments in assets such as plant and equipment, and (3) Tobin’s q, which is the difference

between the market value of a firm and the value of its physical assets. This has been interpreted

as the value of a firm’s intangible assets, such as brand name, customer knowledge and

intellectual property, as perceived by the stock market (Villalonga, 2004). Since R&D is aimed

at creating intangible assets that can contribute to future performance, it can be argued that

higher levels of R&D spending should lead to higher Tobin’s q values (Connolly and Hirschey,

2005).

3.1 The Model

11
 Our regression model measures performance of fabless and integrated firms in the

semiconductor industry, while controlling for firm size (the logarithm of the number of

employees), capital intensity, and year-specific effects. The model includes the conditional

effects of R&D spending and being a fabless firm, as well as the interaction term of R&D

spending and being a fabless firm. The following is our regression model:

PERFit = 0 + 1Fablessi + 2R&Dit + 3Fabless*R&Dit + 4LogEMPit + CAPINTit +

YEARi + 

where for firm i in year t:

PERFit = Gross Margin, Net Margin, ROA, or Tobin’s q

Fablessi = A dummy for fabless or integrated firm
R&Dit = R&D ratio (R&D expense/sales)
LogEMPit = Ln(the number of employees)
CAPINTit = Capital Intensity (total assets/sales)
Fablessi*R&Dit = Interaction term of fabless firm and R&D ratio
YEARi = a dummy for year
= an error term

PERF stands for firm performance; its measure will be replaced in turn by gross margin,

net margin, ROA, and Tobin’s q. Fabless stands for fabless firm, which is a dummy variable

indicating whether the firm is a fabless or integrated firm. R&D represents R&D ratio (R&D

expenses divided by total sales), which is mean-centered. The mean centering reduces

multicollinearity and its negative effects, and also makes the interpretation of the conditional

effect of being a fabless firm more meaningful since zero is outside of the bounds of the

measurement of R&D. LogEMP is the number of employees is used as a control variable for

firm size. We take the log of the number of employees in order to get a normal distribution for

the value. Capital intensity (CAPINT) is also included as a control variable; since firms are likely

12
to have large sunk costs, particularly foundry firms in the semiconductor industry, capital

investment can influence performance. The interaction term of fabless and R&D is included in

order to examine if there is an interaction effect of R&D spending and being a fabless firm for

firm performance. A positive sign may suggest that fabless firms capture more profits (or

intangible value) from R&D spending.

3.2 Data Sources and Coding

This research employs two data sources: the Electronic Business (EB) 300 data set and

the Hoovers database for the eleven years from 2000 to 2010. The EB 300 data set includes the

top 300 electronics firms ranked by electronics revenue. The electronics revenue is derived from

segmentation information and Reed Research estimates. It includes revenue from the sale,

service, license or rental of electronics/computer equipment, software or components. Data items

such as sales, cost of goods sold (COGS), return on assets (ROA), R&D expense, and the

number of employees are obtained from the Hoovers database for the same firms included in the

EB 300 data set.

We select the firms operating in the semiconductor industry using the four-digit North

American Industry Classification System (NAICS) code. The NAICS code for the industry is

3344. We code these firms as fabless and integrated firms. 5 Firms that cannot be coded as pure

play fabless and integrated semiconductor firms (e.g., diversified firms, large conglomerates, and

foundry firms) are excluded. The sample includes 187 observations for 11 integrated firms and

5
Shin, Kraemer, and Dedrick (2012) coded EB 300 firms as lead firm, contract manufacturer, and component
supplier. Our coding extends the previous coding by further classifying component supplier into a fabless firm and
integrated firm. All of the fabless firms and integrated firms coded are operating in the industry of NAICS 3344.
One exception is Qualcomm (fabless firm) operating in NAICS 3342.

13
10 fabless firms.6 The full sample statistics are shown in Table 1. The list of the 21 firms is

shown in Appendix A. A correlation table is also provided in Appendix B.

= = Insert Table 1 in here = =

Our sample data (Table 1) shows that the mean values of gross margin and Tobin’s q are

higher for fabless firms than for vertically integrated firms. However, the average net margin and

ROA are negative for both fabless firms and integrated firms for the period of eleven years from

2000 to 2010. These data illustrate how fierce the competition in the semiconductor has been. It

is notable that fabless firms spend less on R&D spending, but their R&D ratio is higher,

compared to integrated firms, which means that fabless firms invest a higher portion of their

revenue into R&D, compared to integrated firms.

4. Results

Our regression results for the main effect (Models 1 and 3 in Tables 2 and 3) show that

being a fabless firm has a positive relationship with all four performance measures: gross

margin, net margin, ROA, and Tobin’s q, when we control for size, capital intensity, and R&D

ratio (R&D/Sales). The positive relationship is statically significant at a level of .001 (except for

6
These 21 firms comprised about half of the world market share in 2011. Eleven out of the 21 firms were also listed
among the top 25 semiconductor sales leaders of 2011 (IC Insights, 2012), and their market share was 40%. The
market share of the top 25 semiconductor companies was 76%. Given that the market share of most semiconductor
firms in the world is negligible, including about 100 small firms in Taiwan, our study on these 21 firms for 11 years
provides some meaningful results. We compared these sample firms (187 observations) with other semiconductor
firms in the EB300 data set (418 observations—all the firms are operating in the industry of NAICS 3344 except
Sony Corporation operating in NAICS 3343) in terms of total revenue, gross margin, net margin, and ROA. By
conducting the ANOVA, non-parametric 2 and median tests, we found that other semiconductor firms were not
systematically different from the sample firms for most of the measures (total revenue and ROA for the ANOVA
test, and total revenue, net margin, and ROA for both non-parametric 2 and median tests).

14
ROA). These findings suggest that fabless firms perform significantly better, and are perceived

to have greater intangible assets than vertically integrated firms, other things being equal. The

findings support Hypothesis 1.

When the interaction term of R&D ratio and fabless firm is included in the model

(Models 2 and 4 in Tables 2 and 3), the results also show that being a fabless firm has a positive

relationship with all four performance measures, although the magnitude of the estimate is

somewhat decreased. On the other hand, the interaction term of R&D ratio and fabless firm has a

negative relationship with all four performance measures, but the negative relationship is

statistically significant only for gross margin and net margin. These findings indicate that

although fabless firms perform better than integrated firms with the same level of R&D ratio,

they earn lower gross margins and net margins from an increased R&D ratio, compared to

integrated firms. These findings suggest that while fabless firms can better innovate by focusing

their R&D activity on chip design, such benefits are negated by the relative size of their R&D

spending (a greater portion of sales invested into R&D) 7, so the returns of R&D as measured by

net margin (bottom-line financial performance) and gross margin are lower for fabless firms than

for integrated firms. Our findings also show that there is no difference in the relationship of

R&D to return on assets or Tobin’s q between fabless and integrated firms. Thus, the estimates

do not support Hypothesis 2.

= = Insert Tables 2 and 3 in here = =

In order to interpret the interaction effect of R&D ratio on performance for fabless firms,

we plot the estimated values of gross margin as a function of R&D ratio and being a fabless firm
7
R&D ratios of fabless firms and integrated firms are 25.59% and 17.25%, respectively (Table 1).

15
(Hayes, 2013) in Figure 1.8 The figure clearly shows that fabless firms earn higher gross margins

than integrated firms for a given R&D ratio. However, the performance gap decreases as the

R&D ratio increases; integrated firms earn higher gross margins with increased R&D ratio, while

fabless firms earn lower gross margins as they increase their R&D ratio. One reason might be

that fabless firms have decreasing returns on R&D since they have higher R&D intensity. On the

other hand, integrated firms can earn higher gross profits by investing a significant portion of

their revenue into R&D because they must spread their R&D budget over various value chain

activities. In order to corroborate these findings, we visualize the regression model using R&D

ratio as a moderator (Figure 2). The figure also shows similar illustrations 9: fabless firms earn

higher gross margins than integrated firms for a given level of R&D ratio and fabless firms earn

higher gross margins when they have a lower R&D ratio, while integrated firms earn higher

gross margins when they have a higher R&D ratio.

= = Insert Figures 1 and 2 in here = =

In order to supplement our findings, we conducted a regression analysis using patent

counts10 (the number of patents per employee), instead of R&D spending. The results are similar

to the findings above: fabless firms perform better than integrated firms. However, unlike the

results using R&D spending, the impact of patenting on the performance of fabless firms (the

estimate of the interaction term) is positive and greater for fabless firms than for integrated firms.

8
The visualization of the regression model for other performance measures, such as net margin and ROA, shows
similar illustrations.
9
Since the variable of fabless firm is a dichotomous variable, we only use the two ends of the plots for interpretation
(integrated firms are coded zero and fabless firms are coded one). It would be interesting if a continuous variable,
for example, the outsourcing percentage of chip fabrication, is used instead of a dichotomous variable. However,
such data are currently not available from any data sources we are aware of.
10
Patent data used for the analysis are derived from the COMETS database release 1.0 (Zucker and Darby, 2011).

16
The results suggest that unlike R&D spending, which is an annual expense that reduces profits,

patents do not negatively affect bottom-line financial performance, such as net profits and ROA.

The results are shown in Tables 4 and 5.

= = Insert Tables 4 and 5 in here = =

5. Discussion and Conclusions

Innovation is a key contributor to a firm’s value creation and competitiveness. However,

there has been limited theorizing or empirical research on the performance impact of R&D for

different types of firms in the semiconductor industry. This research contributes to theory by

hypothesizing and testing a relationship between innovation effort and performance for fabless

and integrated firms.

Our results show that fabless firms perform better than integrated firms in terms of all

four performance measures - gross margin, net margin, ROA, and Tobin’s q. We looked at the

impact of R&D on these performance measures and the results were different. For the full

sample of firms, the relationship between R&D ratio to gross margin is insignificant except in

Model 4 (Table 2), where it is positive, but the relationship to net margin is negative, as is the

relationship with ROA. These findings suggest that the industry as a whole may be overinvesting

in R&D, perhaps because of the intense competition in the industry and the slowness of money-

losing firms to exit during this period. The fact that the relationship of R&D ratio to Tobin’s q is

also negative in the OLS models suggests that the market does not see the industry’s R&D as

creating intangible assets that will improve future performance.

17
 While fabless firms perform better than integrated firms for a given level of R&D

spending, integrated firms perform better with increased R&D spending, compared to fabless

firms. The relationship of R&D to performance of fabless firms, as measured by gross margin

and net margin, is negative, but this negative impact is greater for fabless firms than for

integrated firms. Namely, integrated firms earn higher gross profits and net profits from their

R&D spending, compared to fabless firms. On the other hand, there is no difference in the

performance of R&D for fabless firms and integrated firms, as measured by ROA and Tobin’s q.

It appears that fabless firms are large enough to exert bargaining power over their

foundry partners and are not overly affected by the transaction costs associated with those

relationships. One interpretation is that information technologies used in design activities, such

as EDA, and interface standards between chip design and fabrication reduce transaction costs

and allow fabless firms to obtain complementary assets, such as chip fabrication and assembly,

through market transactions. Fabless firms have also grown larger and constitute a large share of

a foundry’s business, thereby gaining bargaining power over foundries. Hence, fabless firms can

focus R&D activities on chip design and development, and capture higher profits than integrated

firms. However, these benefits may be negated by the relative cost of R&D spending, so the

returns to (increased) R&D as measured by gross profits and net profits (bottom-line financial

performance) are lower for fabless firms than for integrated firms. On the other hand, integrated

firms must spread R&D budget over various activities in the value chain, and thus they are likely

to capture higher profitability from increased R&D spending, compared to fabless firms. While

this research shows that the relationship between R&D and the performance of fabless and

integrated firms is mixed, it confirms, in part, previous literature showing that fabless firms

18
earned higher gross margins than integrated firms (EE Times, 2005) and that the performance of

fabless and integrated firms in terms of ROA converged over time (Brown and Linden, 2011).

Although firms have different capabilities for managing the value chain, and some firms

are better than others, they may be approaching an equilibrium in which there would be no

significant difference in the bottom-line financial performance. Over time, large integrated firms

might raise the percentage of chips outsourced to foundries, and leading fabless firms might

become more involved with the development and mastery of process technology (Brown and

Linden, 2011). Hence, profitability of the two types of firms would converge. According to

Kapoor (2013), integrated firms have increased the proportion of manufacturing outsourced to

specialized foundries in recent years. He also argued that on average about 30% of the total sales

of foundries such as Taiwan Semiconductor Manufacturing Corporation (TSMC) and United

Microelectronics Corporation (UMC) has been generated from integrated firms in recent years.

From a strategic point of view, integrated firms can cope with volatile semiconductor markets by

adjusting the percentage of chips outsourced to foundries. By doing so, they can exploit external

economies of scale as fabless firms do, and focus their R&D activities more on innovation in

chip design and development.

We recognize the shortfall in our data set: its small size. This research also does not

include most successful large integrated firms, such as Samsung, Sony, and Toshiba, because

they are highly diversified firms; it is hard to obtain data for the semiconductor unit apart from

their entire business operations. Thus, it would be interesting for future research to examine the

relationship of R&D to financial performance of fabless and integrated firms, including highly

diversified firms. Another future research direction could be to examine directly the impact of

19
outsourcing of chip fabrication on the performance of fabless firms and integrated firms if the

data is available.

References

Brundell, R., R. Griffith, and J. Van Reenen. 1999. “Market Share, Market Value and Innovation
in a Panel of British Manufacturing Firms.” Review of Economic Studies, 66(3): 529-554.

Brown, C. and G. Linden. 2011. Chips and Change: How Crisis Reshapes the Semiconductor
Industry. Cambridge, MA: The MIT Press.

Chandler, A. 1977. The Visible Hand: the Managerial Revolution in American Business.
Cambridge, MA: Belknap/Harvard University Press.

Clarke, P. May 10, 2014. “Fabless Chip Companies Ranked by 2013 Sales.” EE Times.

Connolly, R. A., and M. Hirschey. 2005. “Firm size and the effect of R&D on Tobin's q.” R&D
Management, 35(2): 217-223.

Dedrick, J. and K. L. Kraemer. 2006. “Is Production Pulling Knowledge Work to China? A
Study of the Notebook PC Industry.” Computer, 39(7): 36-42.

Dibiaggio, L. 2007. “Design Complexity, Vertical Disintegration and Knowledge Organization

in the Semiconductor Industry.” Industrial and Corporate Change, 16(2): 239-267.

EE Times. March 21, 2005. “Fabless Firms Outpace IDMs in Costs, Gross Margins.”

Feng, Q. and L. X. Lu. 2012. “The Strategic Perils of Low Cost Outsourcing.” Management
Science, 58(6): 1196-1210.

Hayes, A. F. 2013. Introduction to Mediation, Moderation, and Conditional Process Analysis: A

Regression-Based Approach. New York, NY: The Guilford Press.

Helper, S. M. and M. Sako. 2010. “Management Innovation in Supply Chain: Appreciating

Chandler in the Twenty-First Century.” Industrial and Corporate Change, 19(2): 399-429.

IC Insights. April 4, 2012. “Research Bulletin: Extreme Results in Top 25 2011 Semiconductor
Sales Ranking.”

IC Insights. January 28, 2014. “Top 13 Foundries Account for 91% of Total Foundry Sales in
2013.”

20
iSuppli. September 23, 2010. “Core Silicon Semiconductor Market Set for Record Growth in
2010.”

Kapoor, R. 2013. “Persistence of Integration in the Face of Specialization: How Firms Navigated
the Winds of Disintegration and Shaped the Architecture of the Semiconductor Industry.”
Organization Science, 24(4): 1195-1213.

Kleis, L., P. Chwelos, R. V. Ramirez, and I. Cockburn. 2012. “Information Technology and
Intangible Output: The Impact of IT Investment on Innovation Productivity.” Information
Systems Research, 23(1): 42-59.

Langlois, R. N. 2003. “The Vanishing hand: the Changing Dynamics of Industrial Capitalism.”
Industrial and Corporate Change, 12(2): 351-385.

Monteverde, K. 1995. “Technical Dialogue as an Incentive for Vertical Integration in the

Semiconductor Industry.” Management Science, 41(10): 1624-1638.

Mudambi, R. and M. Venzin. 2010. “The Strategic Nexus of Offshoring and Outsourcing
Decisions.” Journal of Management Studies, 47(8): 1510-1533.

Pfanner, E. and A. Fukase. 2015. “Infineon Expressed Interest in Japanese Chip Maker
Renesas.” Wall Street Journal, November 22.

Plambeck, E. L. and T. A. Taylor. 2005. “Sell the Plant? The Impact of Contract Manufacturing
on Innovation, Capacity, and Profitability.” Management Science, 51(1): 133-150.

Shin, N., K. L. Kraemer, and J. Dedrick. 2012. “Value Capture in the Global Electronics
Industry: Empirical Evidence for the “Smiling Curve” Concept.” Industry and Innovation, 19(2):
89-107.

Sturgeon, T. J. 2002. “Modular Production Networks: a New American Model of Industrial

Organization.” Industrial and Corporate Change, 11(3): 451-496.

Teece, D. J. 1986. “Profiting from Technological Innovation: Implications for Integration,

Collaboration, Licensing and Public Policy.” Research Policy, 15(6): 285-305.

Teece, D. J. 2006. “Reflections on Profiting from Innovation.” Research Policy, 35(8): 1131-
1146.

Villalonga, B. 2004. Intangible Resources, Tobin’s q, and Sustainability of Performance

Differences.” Journal of Economic Behavior & Organization, 54(2): 205-230.

Williamson, O. E. 1975. Markets and Hierarchies. New York: Free Press.

21
Zucker, L. G. and M. R. Darby. July 1, 2011. COMETS Data Description, Release 1.0 (Los
Angeles, CA: UCLA Center for International Science, Technology, and Cultural Policy).

Figure 1. Moderation of Fabless Firm Status on the Relationship of R&D Ratio to Gross Margin

Figure 2. Moderation of R&D Ratio on the Relationship of Fabless Firm to Gross Margin

22
 Table 1. Sample Statistics (2000 to 2010)

Variables Fabless Firms Integrated firms Full Sample

Mean1 St. Dev. Obs.2 Mean St. Dev. Obs. Mean St. Dev. Obs.

Sales (millions) 2,460.4 2,134.2 98 9,808.1 10,247.6 85 5,873.4 8,024.8 183

Total Assets (millions) 4,195.3 5,011.4 100 15,613.1 15,108.1 75 9,088.6 11,978.0 175
Gross margin (%) 49.34 13.87 82 36.31 19.04 73 43.20 17.70 155
Net margin (%) -9.73 53.66 97 -4.19 32.43 85 -7.14 44.97 182
ROA (%) -1.84 23.93 67 -3.23 29.82 78 -2.59 27.17 145
Tobin’s q 3.82 2.38 75 2.04 1.03 53 3.08 2.12 128
R&D expense (millions) 586.9 502.7 98 1,663.7 1,569.8 78 1,064.11 1,229.7 176
R&D ratio (% of sales) 25.59 13.29 98 17.25 7.08 78 21.89 11.71 176
Employees (thousands) 4.81 3.56 98 31.76 25.24 80 16.93 21.73 178
Number of Patents 150.7 173.5 95 522.6 556.2 120 358.3 468.5 215
Patents per Employee 37.26 30.40 68 24.48 28.41 84 30.19 29.90 152
Capital Intensity (total 1.75 .8814 98 1.58 .4824 75 1.68 .7385 173
assets/sales)

1
The average of all observation for 2000-2010.
2
The number of observations varies due to the availability of data for the sample firms.

23
 Table 2. OLS and 2SLS Regression Results for Gross Margin and Net Margin

Variable Gross Margin Net Margin

OLS 2SLS OLS 2SLS
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4

R&D Ratio (R&D/Sales) -.0002 .0083 .0019 .0097** -2.464*** -1.308*** -3.370*** -1.117
(.0019)1 (.0056) (.0019) (.0037) (.5136)1 (.3228) (.3316) (.8087)
Fabless Firm .2560*** .2364*** .2563*** .2457*** 34.426*** 30.590*** 40.545*** 34.252***
(.0463) (.0485) (.0438) (.0376) (8.347) (7.710) (8.112) (6.788)
R&D Ratio*Fabless Firm -.0110+ -.0107* -1.403* -2.615**
(.0058) (.0043) (.6338) (.8949)
Ln(Employee) .0731*** .0792*** .0737*** .0808*** 10.336*** 10.548*** 10.483*** 11.006***
(.0176) (.0190) (.0172) (.0180) (3.041) (2.997) (2.884) (2.794)
CAPINT .0271 .0227 .0401 .0193 -.5871 -.3318 3.705 1.555
(.0341) (.0342) (.0371) (.0340) (4.891) (4.846) (4.469) (3.941)
Controls year year year year year year year year

R2 29.96% 36.83% 32.84% 40.00% 61.41% 63.04% 65.78% 70.05%

F test 6.12*** 5.18*** NA NA 8.43*** 10.06*** NA NA
Wald 2 NA NA 95.94*** 103.13*** NA NA 254.71*** 487.47***
N 146 146 1322 132 171 171 1532 153

Instrument Test3:
Wooldridge chi-square .715 (p=.398) .009 (p=.926) 3.38 (p=.066) 1.90 (p=.168)

Key: *** (p<.001), ** (p<.01), * (p<.05) + (p<.10)

1
The values in parentheses are robust (heteroscedasticity-consistent) standard errors.
2
Each observation requires data for the current and previous periods; this eliminates observations for all of 2000 and some observations in other years
when firms do not have consecutive years of data for lagged independent and dependent variables.
3
Instrumental variables: one-year lagged R&D ratio and one-year lagged capital intensity (for 2SLS without the interaction term), and the interaction term
of one-year lagged R&D ratio and fabless firm.

24
 Table 3. OLS and 2SLS Regression Results for ROA and Tobin’s q

Variable ROA Tobin’s q

OLS 2SLS OLS 2SLS
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4

R&D Ratio (R&D/Sales) -1.074*** -.9063*** -1.469*** -1.378* -.0645** -.0429+ -.0221 -.0349
(.2232)1 (.2531) (.2682) (.6279) (.0228)1 (.0231) (.0164) (.0406)
Fabless Firm 15.969* 15.630** 19.612* 19.426** 2.388*** 2.327*** 2.363*** 2.400***
(7.535) (7.383) (7.702) (7.545) (.4931) (.4709) (.4594) (.4258)
R&D Ratio*Fabless Firm -.1990 -.0972 -.0242 .0139
(.3473) (.6514) (.0336) (.0488)
Ln(Employee) 5.109+ 5.228+ 6.002* 6.059* .0668 .0795 .2434 .2360
(2.978) (3.031) (3.000) (3.092) (.1828) (.1824) (.1508) (.1590)
CAPINT 6.222+ 6.056+ 5.740 5.441 .0154 .0214 -.4808 -.4816+
(3.602) (3.628) (3.811) (4.074) (.3096) (.3110) (.2702) (.2703)
Controls year year year year year year year year

R2 42.96% 43.08% 44.49% 44.85% 46.58% 46.70% 42.33% 42.23%

F test 6.01*** 6.48*** NA NA 5.94*** 6.31*** NA NA
Wald 2 NA NA 89.76*** 90.36*** NA NA 84.90*** 88.86***
N 134 134 1192 119 128 128 1262 126

Instrument Test3:
Wooldridge chi-square NA4 NA NA NA

Key: *** (p<.001), ** (p<.01), * (p<.05) + (p<.10)

1
The values in parentheses are robust (heteroscedasticity-consistent) standard errors.
2
Each observation requires data for the current and previous periods; this eliminates observations for all of 2000 and some observations in other years
when firms do not have consecutive years of data for lagged independent and dependent variables.
3
Instrumental variables: one-year lagged R&D ratio and one-year lagged capital intensity (for 2SLS without the interaction term), and the interaction term
of one-year lagged R&D ratio and fabless firm.
4
One year-lagged capital intensity was not used as an instrument variable because the instrument validity test showed that it might not valid as an
instrumental variable.

25
 Table 4. OLS and 2SLS Regression Results for Gross Margin and Net Margin

Variable Gross Margin Net Margin

OLS 2SLS OLS 2SLS
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4

Pat per Emp -.0004 -.0024** -.0008 -.0033** .2929** .3283* .4300** .3570*
(.0005)1 (.0009) (.0006) (.0011) (.1046)1 (.1412) (.1369) (.1592)
Fabless Firm .2772*** .2249*** .2917*** .2331*** 19.120* 19.740+ 20.903* 19.360+
(.0466) (.0517) (.0435) (.0496) (9.529) (.0517) (10.085) (11.055)
Pat per Emp*Fabless Firm .0039*** .0046*** -.0701 .1474
(.0011) (.0013) (.2140) (.2405)
Ln(Employee) .0874*** .0657** .0828*** .0573* 15.918*** 16.114*** 18.920*** 18.476***
(.0184) (.0217) (.0183) (.0224) (4.300) (4.609) (4.678) (4.985)
CAPINT .0137 .0485 .0443 .0963* -28.642** -29.077** -30.240** -29.142**
(.0362) (.0382) (.0449) (.0460) (8.846) (9.358) (10.216) (10.667)
Controls year year year year year year year year

R2 32.72% 39.90% 35.73% 43.95% 38.79% 38.82% 39.78% 39.34%

F test 5.61*** 5.05*** NA NA 2.49** 2.84*** NA NA
Wald 2 NA NA 95.63*** 90.6*** NA NA 42.08*** 45.6***
N 117 117 1052 105 140 140 1262 126

Instrument Test3:
Wooldridge chi-square .980 (p=.322) 1.38 (P=.240) NA4 NA

Key: *** (p<.001), ** (p<.01), * (p<.05) + (p<.10)

1
The values in parentheses are robust (heteroscedasticity-consistent) standard errors.
2
Each observation requires data for the current and previous periods; this eliminates observations for all of 2000 and some observations in other years
when firms do not have consecutive years of data for lagged independent and dependent variables.
3
Instrumental variables: one-year lagged patent per employee and one-year lagged capital intensity (for 2SLS without the interaction term), and the
interaction term of one-year lagged patent per employee and fabless firm.
4
One year-lagged capital intensity was not used as an instrument variable because the instrument validity test showed that it might not valid as an
instrumental variable.

26
 Table 5. OLS and 2SLS Regression Results for ROA and Tobin’s q

Variable ROA Tobin’s q

OLS 2SLS OLS 2SLS
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4

Pat per Emp .1658* .1285+ .2443** .1123 -.0019 -.0026 -.0015 -.0040
(.0639)1 (.0725) (.0795) (.0810) (.0047)1 (.0064) (.0045) (.0067)
Fabless Firm 12.100 10.828 11.881 7.039 3.202*** 3.177*** 3.227*** 3.135***
(7.932) (8.913) (8.100) (9.174) (.4688) (.5091) (.4276) (.5236)
Pat per Emp*Fabless Firm .0780 .2821 .0013 .0049
(.1470) (.1794) (.0098) (.0111)
Ln(Employee) 8.066* 7.727* 9.524** 8.397* .5025** .4958** .5664*** .5437***
(3.318) (3.594) (3.455) (3.636) (.1694) (.1641) (.1405) (.1519)
CAPINT -10.066** -9.633* -11.370* -9.667+ -1.189*** -1.175** -1.023*** -.9701**
(3.726) (3.978) (4.635) (5.025) (.3241) (.3784) (.2658) (.3188)
Controls year year year year year year year year

R2 32.56% 32.69% 32.23% 31.56% 58.06% 58.07% 56.39% 56.51%

F test 2.75** 2.66** NA NA 10.73*** 10.46*** NA NA
Wald 2 NA NA 43.10*** 41.98*** NA NA 147.64*** 153.93***
N 113 113 1002 100 100 100 992 99

Instrument Test3:
Wooldridge chi-square NA4 NA .254 (p=.614) .242 (P=.623)

Key: *** (p<.001), ** (p<.01), * (p<.05) + (p<.10)

1
The values in parentheses are robust (heteroscedasticity-consistent) standard errors.
2
Each observation requires data for the current and previous periods; this eliminates observations for all of 2000 and some observations in other years
when firms do not have consecutive years of data for lagged independent and dependent variables.
3
Instrumental variables: one-year lagged patent per employee and one-year lagged capital intensity (for 2SLS without the interaction term), and the
interaction term of one-year lagged patent per employee and fabless firm.
4
One year-lagged capital intensity was not used as an instrument variable because the instrument validity test showed that it might not valid as an
instrumental variable.

27
28
 Appendix A: The List of Sample Firms (11 Integrated Firms and 10 Fabless Firms)

Integrated Firms Fabless Firms

Advanced Micro Devices, Inc.*# Agere Systems

Freescale Semiconductor, Inc.*# Altera Corporation
Hynix Semiconductor Inc.*# ATI Technologies
Intel Corporation*# Broadcom Corporation*#
Micron Technology, Inc.*# Conexant Systems
Nanya Technology Corporation LSI Corporation (formerly LSI Logic)
Qimonda AG Marvell Technology Group Ltd.*#
Spansion Inc. NVIDIA Corporation*#
STMicroelectronics*# Qualcomm Incorporated*#
Texas Instruments Incorporated*# Xilinx Inc.#
Winbond Electronics Corp.

* Firms included in the top 25 semiconductor sales leaders of 2011 (IC Insights, 2012)
# Firms included in the top 25 semiconductor sales leaders of 2010 (iSuppli, 2010)

Appendix B: Correlation Table (Full Sample)

Variable 1 2 3 4 5 6 7

1. Gross Margin
2. Net Margin .480**
3. ROA .434** .824**
4. Tobin’s q .383** .082 .132
5. R&D Ratio .132 -.654** -.457** -.083
6. Employees .078 .178* .165 -.192* -.251**
7. Pat per Emp -.039 .082 .157 .010 .042 -.262**
8. CapInt .078 -.385** -.162 .112 .426** -.110 .134

Key: ** (p<.01), * (p<.05)

29