`Memory Market: Its Impact
`on Firm Behavior and
`Global Semiconductor
`Trade Patterns
`
`Web version:
`July 2007
`
`Author:
`Falan Yinug1
`
`Abstract
`
` This article addresses three questions about the flash memory
`market. First, will the growth of the flash memory market be a
`short- or long-term phenomenon? Second, will the growth of the
`flash memory market prompt changes in firm behavior and
`industry structure? Third, what are the implications for global
`semiconductor trade patterns of flash memory market growth?
`The analysis concludes that flash memory market growth is a
`long-term phenomenon to which producers have responded in
`four distinct ways. It also concludes that the rise in flash memory
`demand has intensified current semiconductor trade patterns but
`has not shifted them fundamentally.
`
`1 Falan Yinug (Falan.Yinug@usitc.gov) is a International Trade Analyst from the Office of
`Industries. His words are strictly his own and do not represent the opinions of the US
`International Trade Commission or of any of its Commissioners.
`
`1
`
`APPLE 1049
`Apple v. Realtime Data
`IPR2016-01737
`
`
`
`Introduction
`
`The past few years have witnessed rapid growth in a particular segment of the
`semiconductor market known as flash memory.2 In each of the past five years,
`for example, flash memory market growth has either outpaced or equaled that
`of the total integrated circuit (IC) market3 (McClean et al 2004-2007, section 5).
`One observer expects flash memory to have the third-strongest market growth
`rate over the next six years among all IC product categories (McClean et al
`2007, 5-6). As a result, the flash memory share of the total IC market has
`increased from 5.5 percent in 2002, to 8.1 percent in 2005. As a share of the
`memory market segment, flash memory has increased from 28.7 percent to 38.2
`percent during the same period. In short, the flash memory market has quickly
`become a significant part of the overall semiconductor market that cannot be
`ignored; some predict it will soon compete with the dynamic random access
`memory (DRAM) market for dominance within the memory sector in the not-
`too-distant future (McClean et al 2007, 5-4).4
`
`Given its market size and projected growth, flash memory is likely to have an
`increased impact on the global semiconductor industry, and the decisions that
`flash memory producers make are likely to have a significant influence on
`industry evolution. These decisions have already been as dynamic as the recent
`performance of the flash memory market. Some firms have shifted production
`from other products to flash memory. In addition, some other firms have
`partnered to gain flash memory market share. Also, some firms have
`aggressively moved to lock in long-term deals with certain flash memory
`consumers.
`
`This article will address three questions about the flash memory market. First,
`will the growth of the flash memory market be a short- or long-term
`phenomenon? Second, will the growth of the flash memory market prompt
`
`2 Flash memory is a type of nonvolatile memory that can be electrically erased and
`reprogrammed. Nonvolatile memory is memory that retains data when the power is turned
`off. Flash memory costs less and includes more functionality than other forms of nonvolatile
`memory.
`3 The semiconductor market is composed of two main subsets, the integrated circuit (IC)
`market and the optoelectronics, sensors, and discretes (O-S-D) market. The IC segment of the
`semiconductor market is by far the biggest (85 percent in 2006) and comprises
`semiconductors that are harder to manufacture, more advanced, and more expensive. Flash
`memory is a type of IC.
`4 DRAM is a popular type of volatile memory used mainly in computers. Compared to
`nonvolatile memory, volatile memory loses data when powered down. DRAM composes the
`largest share of the memory market, though flash memory has eroded its lead in recent years.
`
`2
`
`
`
`changes in firm behavior and industry structure? Third, what are the
`implications for global semiconductor trade patterns of flash memory market
`growth?
`
`The analysis concludes that (1) flash memory market growth is a long-term
`phenomenon; (2) flash memory producers have responded to flash memory
`market growth in four distinct ways: choosing to produce flash memory rather
`than nonvolatile memory, entering into flash memory production, increasing
`flash memory production and production capacity, and partnering with each
`other; and (3) increased demand for flash memory and the response of
`producers to meet this demand have intensified current semiconductor trade
`patterns but has not shifted them fundamentally.
`
`Flash Memory To Endure
`
`The semiconductor industry has experienced many changes since flash memory
`first appeared in the early 1980s, one of the most dramatic and long-term of
`which has been the rise of the consumer electronics market as a demand driver
`for semiconductors. This rise in the consumer electronics market has fueled
`flash memory market growth and helped to make flash memory a prominent
`segment within the semiconductor industry.
`
`Broadly speaking, flash memory ideally suits the consumer electronics market,
`because it bestows upon electronic devices two qualities that the market
`demands: mobility and miniaturization. For example, cell phones, a major
`application for flash memory, require data storage to save and store frequently
`called numbers and perform other convenient functions for which a traditional
`hard drive would prove impractical; such information would be erased every
`time the phone were turned off. Because (1) flash memory is small, reliable,
`and (2) its memory is nonvolatile, numerous applications not practicable with
`traditional data storage technology are emerging. Flash memory brings mobility
`and miniaturization to electronics products, two defining features of most
`consumer electronics products today.
`
`Given capabilities and attractiveness of flash memory to the consumer market,
`it is clear why demand for it has rapidly grown. Flash memory allowed existing
`electronic products to adopt mobile and miniature qualities they did not have
`before and thus opened them up to new and very large consumer markets. In
`addition to cell phones, USB flash memory drives function as portable and
`smaller floppy drives. Flash memory has also prompted the growth of new
`consumer applications. Flash memory is an important component in popular
`devices such as DVD players, digital cameras, MP3 players, personal digital
`assistants (PDAs), and global positioning systems (GPS), all of which could not
`function without flash memory (McClean et al 2004, 7-2, and 2005, 7-3).
`
`3
`
`
`
`Origins and Early Growth
`
`When flash memory first appeared in the early 1980s, most industry observers
`hardly took note. The few that did most likely would not have predicted then
`that the flash memory market would become a major segment of the global
`semiconductor market (box 1). Once flash memory fully emerged in the early
`1990s, the initial industry consensus was that it had growth potential, but
`certain concerns made its growth trajectory uncertain. First, which markets
`would drive flash memory market growth? Second, how would flash memory
`compete against other types of nonvolatile memory technologies? Third, given
`its high price, how long would sluggish early sales continue?
`
`
`Box 1 Fujio Masuoka, the Inventor of Flash Memory
`
`The first flash memory device was invented in 1981 by a midlevel factory manager
`at Toshiba Corp. (Toshiba) named Fujio Masuoka. Masuoka wanted to create a
`device that would retain its memory after having been powered down. Up until then
`the main type of memory that existed was volatile memory such as DRAM, which
`lost its memory when the device was powered down. For example, any data created
`on a personal computer (PC) using such memory had to be saved to the PC's hard
`disk drive. Masuoka sought to create a chip that improved upon DRAM and hard
`disk drives. According to Masuoka,
`
`
`
`"Simply put, I wanted to make a chip that would one day replace
`all other memory technologies on the market. In the 1980s, the
`market for data storage on PCs was dominated by magnetic tape
`and disk drives0.Going after [the memory storage] market was the
`obvious thing to do for me..."
`
`The industry was initially slow to recognize Masuoka's invention and realize its
`potential. It was not until 1985, four years after patent filing, that the industry was
`introduced to the device at a conference, and some firms realized flash memory
`potential. Intel asked for a sample of the new chip and in 1987-88 announced mass
`production of its own version of flash memory. Soon thereafter, Toshiba began mass
`production of flash memory.
`
`Source: Business Week 2006a and 2006b.
`
`These concerns proved to be unfounded as the flash memory market began to
`grow in the early 1990s (table 1). First, the most significant factor in flash
`memory growth was the emergence of the portable and laptop PC market as
`a growth driver. Flash memory provided the proper benefits of size, power
`dissipation, reliability, and speed for this expanding market (ICE 1992, 6-48).
`
`4
`
`
`
`The demand for flash memory created by portable and laptop PCs in the early
`1990s hinted at a long-term trend within the semiconductor industry that would
`fuel flash memory market growth: the emergence of the consumer electronics
`market as the primary driver of end-use demand in the semiconductor industry.
`Second, within the nonvolatile memory sector, flash memory competed
`primarily against two other technologies called EPROM and EEPROM. In terms
`of price and functionality, flash memory fell somewhere in between these two
`technologies, effectively competing for space at the start of the 1990s (ICE 1992,
`6-47). Third, regarding prices, in 1992 flash memory demand received a boost
`when Intel, the leader in flash memory production at the time, effectively
`lowered flash memory price-per-megabit ratio (ICE 1992, 6-49). Understanding
`the future demand for flash memory, Intel decided in 1991 to focus its
`nonvolatile memory production on flash memory and away from EPROM (ICE
`1992, 6-49).
`
`TABLE 1 The Rise of the Flash Memory Market
`
`Flash memory
`market
`(USD Million)
`35
`135
`270
`640
`865
`1,860
`2,611
`2,702
`2,493
`4,561
`10,637
`7,595
`7,767
`11,739
`15,611
`18,569
`20,275
`
`1990
`1991
`1992
`1993
`1994
`1995
`1996
`1997
`1998
`1999
`2000
`2001
`2002
`2003
`2004
`2005
`2006
`
`Flash memory
`market annual
`percentage growth
`
`286
`130
`106
`35
`115
`40
`3
`-8
`83
`133
`-29
`2
`51
`33
`19
`9
`
`Source: WSTS and IC Insights.
`
`Flash memory as
`percentage of total
`semiconductor
`market
`
`0.1
`0.3
`0.5
`0.8
`0.9
`1.3
`2.0
`2.0
`2.0
`3.1
`5.2
`5.5
`5.5
`7.1
`7.3
`8.2
`8.1
`
`Flash memory as
`percentage of total
`memory market
`0.3
`1.0
`1.8
`3.0
`2.7
`3.5
`7.2
`9.2
`10.8
`14.1
`21.6
`30.5
`28.7
`36.1
`33.1
`38.3
`34.4
`
`5
`
`
`
`End-use Demand
`
`Scholars have noted that shifts in semiconductor end-use demand have
`historically fueled the growth and specialization of certain types of semiconduc-
`tors, thereby benefiting firms or regions or both that specialized in their
`production (Langlois and Steinmueller 1999, 68). The birth of the semiconduc-
`tor industry in the United States in the 1950s was fueled by U.S. military
`demand for high-performance semiconductors. The growth of the PC industry
`in the late 1980s and early 1990s spurred demand for microprocessors (Langlois
`and Steinmueller 1999, 23 and 52).
`
`Since the mid-1990s the importance of the consumer electronics market as a
`source of end-use demand has grown dramatically, and it is predicted to
`increase. In 1993, consumer markets accounted for a little over 20 percent of
`the overall semiconductor market (Gartner Dataquest 2004, Tully). Corporate
`and military demand were the primary market drivers of the semiconductor
`industry then, and historically in the United States these and other sources had
`always accounted for a much greater share of semiconductor end-use demand
`than the consumer market (Langlois and Steinmueller 1999, 37). However,
`since 1993 the consumer electronics market has increased its share of the
`overall semiconductor market, leading one market research firm to predict that
`by 2013, consumer markets will account for more than 50 percent of the overall
`semiconductor market, roughly a 30 percent increase in the share of the
`semiconductor market in 20 years (Gartner Dataquest 2004, Tully).
`
`Therefore, flash memory has quickly become an integral component in an end
`market of growing and sustained significance to the semiconductor industry.
`The question is how long will its importance last? Is flash memory growth truly
`a long-term phenomenon? If the prediction is correct that the consumer market
`will account for over 50 percent of the semiconductor market by 2013, then it
`is highly likely flash memory demand will continue to grow.5
`
`
`Changes in Firm Behavior and
`Industry Structure
`
`Semiconductor producers have devised various strategies to meet the increased
`demand for flash memory and obtain market share. At the beginning of flash
`
`5 Alternative nonvolatile memory solutions exist and could potentially challenge flash
`memory, though industry experts believe that these alternatives will not be widely used for
`many years. Such alternatives include FeRAM, NVRAM, PRAM, and C-RAM. IC Insights 2007,
`7-15 and 7-16; and industry official, phone interview by Commission staff, April 18, 2007.
`
`6
`
`
`
`memory growth, producers of flash memory had to decide whether to switch
`production from other nonvolatile devices to flash memory. When flash
`memory growth exploded in the late 1990s, existing firms increased production
`and firms producing nonflash memory began production. More recently, firms
`have partnered to gain a leg up on the competition. The following describes
`these four firm behaviors and considers their impact on the semiconductor
`industry.
`
`Which Nonvolatile Memory To Produce?
`
`For firms producing nonvolatile memory in the early 1990s, uncertainty existed
`over which nonvolatile memory technology would take off—flash memory,
`EEPROM, or EPROM—thus making the decision to produce flash memory
`difficult. In addition, with flash memory accounting for less than 1 percent of
`the memory market in 1990, many firms had more immediate priorities than to
`focus on a technology with little demand. Decisions to produce flash memory
`fell into three general categories: (1) all in, (2) partially in, and (3) all out. Intel
`was one of the only nonvolatile memory producers that decided to go “all in”
`to flash memory production. In 1991, the company made the strategic decision
`to shift focus from EPROM to flash memory (ICE 1992, 6-49). More firms
`decided on the “partially in” strategy. Some were motivated by Intel's
`announced pull out of EPROM production to stay in that market (for example,
`AMD, SGS-Thomson, Fujitsu, and Texas Instruments), but they also wanted to
`maintain some flash memory production, especially at higher densities where
`some believed flash memory was superior to EPROM in terms of its functional-
`ity/cost ratio (ICE 1992, 6-49). Finally, some firms were unable to compete in
`the flash memory market and exited the market, such as Seeq Technology (ICE
`1992, 6-49).
`
`Intel's leap into the flash memory market proved critical in a technology that
`would soon dominate the nonvolatile memory market. By 1992, Intel had
`captured 75 percent market share of the flash memory market (figure 1) (ICE
`1993, 6-52). Once it was obvious that flash memory would be the dominant
`nonvolatile memory technology, many of the firms “partially in” to flash
`memory production changed strategies and increased production or jumped
`into an “all in” strategy. In 1995, AMD, Fujitsu, Atmel, and SGS-Thomson
`followed this strategy, reclaiming flash memory market share from Intel, which
`saw its share of the market drop to 42 percent (figure 1) (ICE 1996, 8-20).
`
`7
`
`
`
`Figure 1 Changing Flash Memory Market Share, selected years @ercentage)
`
`Flash Memory Market Share, 1992
`($310 million)
`
`Flash Memory Market Share, 1995
`($1.9 billion)
`
`Other
`10
`
`Other
`6
`
`AMD
`
`15
`
`SOS—Thomson
`5
`
`10
`
`Fuflsu
`10
`
` Atmel
`
`AMD
`27
`
`Intel
`42
`
`Source: ICE.
`
`Source: ICE.
`
`Flash Memory Market Share, 1999
`($4.7 billion)
`
`Flash Memory Market Share, 2005
`($20.5 billion)
`
`Other
`18
`
`Samsung
`34
`
`
`
`STMicro
`
`6 k
`
`Hynix
`8
`
`
`
`Spansion
`10
`
`Intel
`11
`
`Toshiba
`13
`
`Other
`14
`
`Intel
`26
`
`STMicro
`
`
`itsubishik
`
`5M
`
`Atmel
`
`
`
`Sharp
`
`AMD
`
`Fufisu
`
`Source: ICE.
`
`Source: Web-Feet Research.
`
`8
`
`
`
`New Producers Entering Market
`
`The fragmentation of the flash memory industry continued in the late 1990s, as
`a small number of existing flash memory producers struggled to satisfy the
`increasing demand for flash memory. Seeing the opportunity to enter a growing
`market, other semiconductor firms (e.g. Samsung, Toshiba-SanDisk) com-
`menced flash memory production. Thus, the number of flash memory
`producers went from less than 15 in 1995 to at least 28 in 2005 (ICE 1996, 8-22
`and Web-Feet Research 2006, Niebel).
`
`The entrance of new producers has had several effects on the industry. Besides
`initially helping to supply the exploding demand for flash memory at the end
`of 1998 and 1999 (though their presence and the increased production of
`existing producers still did not fully satisfy demand in 1999) and helping to
`lower Intel's market share from 42 percent in 1995 to 26 percent in 1999 (ICE
`1996, 8-20, and 2000, 68), the biggest effect of new flash memory producers has
`been the disruption of supply-demand balances in the flash memory and DRAM
`markets. This is because the recent entrants have included a host of DRAM
`producers who have shifted portions of their DRAM capacity to flash memory,
`in particular the ever popular NAND flash.6 Many DRAM producers were lured
`by the higher average selling price of flash memory from 2001-2005 (figure 2)
`and the saturation of the DRAM market. For example, Samsung, which is the
`world's leading supplier of both DRAM and flash memory, has accelerated
`production of flash memory and delayed its DRAM expansion plans (McClean
`et al 2006, 7-17).
`
`Ironically, DRAM producers' entrance into flash memory production has
`actually contributed to defeating their original purpose for entering: flash
`memory's average selling price dropped below that of DRAM in 2006 due to
`oversupply and currently DRAM is more profitable (figure 2). It is uncertain if
`these short-term supply-demand imbalances in flash memory and DRAM will
`continue (LaPedus and McGrath, 2007) and if producers will continue to shuffle
`their production in search of higher average selling prices.
`
`Estimating proper supply for the flash memory market is complicated by the
`unpredictable nature of flash memory demand – it is unclear what consumers
`will deem the next great gadget to drive the market, and when it will appear.
`
`6 NAND is a flash memory architecture that provides fast write speeds, a useful feature for
`storing large amounts of data (often used for digital photos, MP3 files, and other multimedia
`applications). The other type of popular flash memory architecture is NOR, which provides
`fast read speeds, a useful feature for quickly pulling data out of memory (cell phones are a
`major application). Currently, almost all flash memory is based on either NAND or NOR
`architectures.
`
`9
`
`
`
`One bright note in this supply-demand challenge is that a chronic oversupply
`or undersupply situation for either flash memory or DRAM is less likely given
`that now a small group of producers exists that are skilled in switching between
`flash memory and DRAM production.
`
`Flash Memory Producers Increasing Production and
`Production Capacity
`
`From 1991 to 2006, the flash memory market grew by 63 percent a year
`(calculated from various ICE and IC Insights reports) and grew from
`one-quarter of one percent to over 8 percent of the overall semiconductor
`market during this period (calculated from various ICE and IC Insights reports).
`Between 1995 and 2006 capital spending on flash memory grew from 3 percent
`to 20 percent of overall semiconductor capital spending (McClean et al 2007,
`4-15 to 4-16). Because of the long-term growth forecast of flash memory,
`
`10
`
`
`
`positive current producers are likely to continue increasing production and
`production capacity.
`
`Producers use four primary methods to increase flash memory production and
`production capacity. One of the fundamental methods firms use to increase
`production is transitioning to smaller production process geometries. Semicon-
`ductors are produced in batches on silicon wafers. Switching to smaller
`production process geometries allows firms to produce more chips per wafer,
`thus increasing chip production. Second, firms can increase their flash memory
`production by shifting existing chip production capacity from other chip
`production to flash memory production.7 Shifting existing production capacity
`allows firms to produce flash memory relatively quickly and cheaply. Recently,
`one firm has shifted existing production capacity from DRAM to flash memory
`in months instead of years and for millions of dollars instead of the billions of
`dollars required to build a new state-of-the-art semiconductor fabrication facility
`(McClean et al 2007, 8-15).8 A third option for increasing production is to buy
`existing semiconductor facilities when available and convert them to flash
`memory production. Finally, for those firms that have the financial resources,
`building new flash memory capacity from the ground up is an option, albeit a
`very expensive and high-risk venture.
`
`These methods of increasing production require different amounts of spending,
`and it is significant that flash memory producers have used the most costly
`method of increasing production. Flash memory producers have increased their
`production capacity, which is a strong indication they believe flash memory is
`a long-term phenomenon; they would not make such an investment otherwise.
`
`Indeed, a clear distinction in spending exists in the semiconductor industry
`between increasing production and increasing production capacity. Increasing
`production through R&D investment is a necessary reality in the semiconductor
`industry. Firms constantly attempt to increase production by increasing the
`number of good die per wafer, increasing the number of wafers processed per
`month, and shrinking the size of the die on wafers. The average R&D spending
`of a semiconductor firm as a percentage of sales is usually between 10 and 20
`percent. In 2006 the average was 15.5 percent (McClean et al 2007, 16-5).
`Though this investment in production is costly,9 increasing production capacity,
`by converting existing capacity, buying existing capacity, or building new
`capacity, is more costly. In 2006, the majority of semiconductor firms invested
`less than $1 billion in R&D (McClean et al 2007, 16-5). By contrast, the
`
`7 Samsung, a major producer of DRAM, employed this strategy during the late 1990s to
`enter the flash memory market. IC Insights 2000, 66.
`8 In 2006, the construction of a new state-of-the-art semiconductor fabrication facility was
`estimated at $2.5 billion. IC Insights 2007, 16-6.
`9 The only industry that spends more on R&D as a percentage of sales is the biotechnology
`industry. IC Insights 2007, 16-3.
`
`11
`
`
`
`construction of a new state-of-the-art semiconductor fabrication facility in 2006
`cost an estimated $2.5 billion.
`
`It is possible, however, that a continual increase in flash memory production
`capacity may lead to chronic oversupply as evidenced by the decline in flash
`memory average selling price in 2006. Downward pricing pressure may lead to
`competitor consolidation. Historical lessons from the DRAM industry, where
`regular overspending led to downward pricing pressures and consolidation, are
`a case in point (McClean et al 2007, 4-15). The nature of end-use demand for
`flash memory, however, is different than it was for DRAMS when overspending
`occurred. The future strength and stability of the flash memory market depends
`largely on development of new and diverse sources of demand from the
`consumer market coupled with producer sensitivity to creating overcapacity.
`
`Flash Memory Producers Partner
`
`Firms have also sought to increase their share of the growing flash memory
`market through partnerships (box 2). Partnering has emerged as a way for firms
`to rapidly increase production without heavily investing in new fabrication
`facility construction.10 It has also permitted firms to share R&D and manufactur-
`ing resources for mutual advantage in joint technology development, allowing
`both partners to become more competitive. Intel and Micron created IM Flash
`Technology to combine “Micron's expertise in developing NAND technology
`and operating highly efficient manufacturing facilities with Intel's multi-level
`cell technology and history of innovation in the flash memory business...” and
`to bring together “the manufacturing technology, assets, experience and scale
`necessary for Intel and Micron to successfully compete in the NAND flash
`memory business....” (Intel and Micron, joint press release, November 21,
`2005).
`
`Thus far, partnering has occurred between relatively equally matched firms
`looking to combine resources to gain market share in a rapidly growing market.
`If supply consistently exceeds demand, the nature of partnering may change
`to where stronger firms take over struggling firms. However, since most flash
`memory producers manufacture other semiconductors, the fall in prices for
`flash memory, even if persistent, will not lead quickly to consolidation.
`
`10 One industry expert estimates that a quarter to a third of current flash memory
`production comes from partnered firms. Industry official, phone interview by Commission
`staff, April 18, 2007.
`
`12
`
`
`
`Box 2 Major Partnerships among Flash Memory Producers
`IM Flash Technologies (IMFT) – joint venture between Intel and Micron
`• Began operations on January 6, 2006 to manufacture NAND flash memory
`for the exclusive benefit of its partners.
`• Key elements: Intel owns a 49 percent interest while Micron owns 51
`percent; companies share output generally in proportion to their
`investment; costs for product and process development are generally split
`evenly; product design and other research and development costs are
`shared equally. Micron contributed land and facilities in Lehi, Utah, a fully
`paid lease of a portion of its manufacturing facility in Manassas, Virginia, a
`wafer supply agreement to be supported by its operations located in Boise,
`Idaho, and $250 million in cash. Intel contributed $1.196 billion in cash and
`notes.
`Hynix and STMicroelectronics – joint venture in China
`• Signed and announced a joint venture agreement in 2004 to build a front-
`end memory manufacturing facility in Wuxi City, China. Construction began
`in 2005. The fab will employ roughly 1,500 people and will feature a 200-
`mm wafer production line planned to begin production at the end of 2006
`and a 300-mm wafer production line planned to begin production in 2007.
`Total investment planned for the project is $2 billion. STMicroelectronics
`will contribute 33 percent of the equity financing, while Hynix will
`contribute 67 percent.
`Flash Partners and FlashVision – joint ventures between Toshiba and SanDisk
`• Flash Partners formed in September 2004.
`• Key elements: SanDisk owns 49.9 percent while Toshiba owns 50.1 percent;
`purchases wafers from Toshiba and sells wafers to SanDisk and Toshiba at
`a price equal to manufacturing cost plus a markup; Toshiba operates its Fab
`3 in Japan, and SanDisk has employees assigned to work there; each firm
`is committed to take 50 percent of Flash Partners’ wafer output.
`• FlashVision formed in April 2002. Firms agreed to consolidate the NAND
`wafer fabrication manufacturing operations in Toshiba’s Fabs 1 and 2 in
`Japan.
`• Key elements: SanDisk owns 49.9 percent while Toshiba owns 50.1 percent;
`each company is committed to take 50 percent of FlashVision’s wafer
`output; each firm has a design and development team associated with
`FlashVision with each paying the cost of its design teams and 50 percent of
`the wafer processing and similar costs associated with this direct design of
`the flash memory.
`Spansion – joint venture between AMD and Fujitsu
`• Formed in 2003 as a manufacturing venture between AMD and Fujitsu.
`• Key elements: provides flash memory to AMD and Fujitsu, who resell it to
`customers; for fiscal 2005, AMD accounted for approximately 56 percent of
`Spansion’s net sales, and Fujitsu accounted for approximately 44 percent;
`currently, Spansion sells directly to customers previously served by AMD
`and continues relationship with Fujitsu.
`
`Source: Company annual reports and 10K and 20F filings to the SEC.
`
`13
`
`
`
`Impact on Semiconductor Trade Patterns
`
`Firm and industry changes due to the rise of the flash memory market have
`intensified current semiconductor trade patterns but has not shifted them
`fundamentally. Despite the rise in the flash memory market, major importers
`and exporters of semiconductors (HS 8542) have remained remarkably stable.11
`From 2002-2006, the top 10 semiconductor importers remained constant, and
`very little change occurred in the top 10 semiconductor exporters (Global Trade
`Atlas).12 The following section describes the nature of the change to global
`semiconductor trade patterns and briefly analyzes possible implications of this
`outcome.
`
`Changes in Current Semiconductor Trade Patterns
`
`Within the semiconductor industry major trade shifts usually occur when
`changes develop in one or more of the following three variables: the structure
`of the semiconductor manufacturing process, the location of front-end chip
`production, and/or the location of the semiconductor market. Increased flash
`memory production has not significantly changed these three variables and
`hence has not shifted current semiconductor trade patterns.
`
`Structure of the Semiconductor-Manufacturing Process
`
`Most semiconductor-manufacturing includes two distinct production processes:
`the highly capital-intensive front-end fabrication process and the less capi-
`tal-intensive (though still highly automated) back-end assembly and test
`process.13 Historically, firms have physically separated these processes, with
`the front end taking place in the firm's home country, usually the United States,
`the EU, or Asia (predominantly Japan), while the back end has occurred mostly
`in Southeast Asia. Firm response to flash memory market growth has not
`significantly altered this production process model. By and large, flash memory
`producers have increased production capacity through construction or
`conversion of facilities in their own countries while also maintaining back end
`
`11 Data on flash memory trade patterns is unavailable, because virtually no country breaks
`down its trade data by flash memory. Only South Korea maintains a subheading in its tariff
`schedule specifically for flash memory. For most countries flash memory trade data is
`aggregated into broader semiconductor groupings in their tariff schedules.
`12 Yearly changes in position among the top 10 semiconductor importers and exporters did
`occur from 2002-2006.
`13 Front-end semiconductor processing is the stage of manufacturing in which
`semiconductors are formed. To reduce semiconductor defects, this process takes place in
`ultraclean environments known as cleanrooms. Once semiconductors are formed, back-end
`processing begins in which semiconductors are assembled, tested, and packaged for final
`sale.
`
`14
`
`
`
`production in their usual locations, mainly in Southeast Asia. Therefore, the
`increase in flash memory production has actually taken place within the
`predominant production model, thus perpetuating it and the trade patterns it
`creates (box 3).
`
`It is unlikely that flash memory producers would switch to an alternative
`production model that would shift trade patterns. The most viable scenario is
`one in which flash memory producers outsource production to semiconductor
`
`Box 3 Selected Flash Memory Firms' Plant Locations
`
`With a few exceptions, firms' recent efforts to increase flash memory production
`capacity show that it is taking place in the usual areas for front-end fab
`construction (i.e. the United States, the EU, and Japan), thus reinforcing trade
`patterns.
`Current plant locations of significant flash memory firms
`Company
`Flash memory fabrication locations
`Samsung
`South Korea
`Toshiba/SanDisk
`Japan
`IMFT
`United States
`Micron
`Italy
`Spansion
`United States and Japan
`Hynix
`South Korea
`STMicro
`Italy, France and Singapore
`Qimonda
`Germany
`Hynix/STMicro joint venture
`China
`Powership
`Taiwan
`Source: Company annual reports, 10K and 20F filings to the SEC, and the McClean
`Report. 2007 ed.
`Note: Intel, which is a major producer of NOR flash memory, has fabrication
`fac