The Rise and Fall of Japan's Semiconductor Industry and Samsung's Ascent: Plaza Accord, the Semiconductor Agreement, and the DRAM Defeat

Why a Security Company Is Writing About This

ShieldGuard builds security software for Windows, but every piece of hardware our products run on rests on semiconductors. Which countries and which companies make those semiconductors is a foundational premise of supply-chain security. Understanding the historical path by which the semiconductor supply structure arrived at its present shape is not wasted knowledge for an engineer.

What this article covers is one of the most thoroughly researched and documented pieces of industrial history of the past several decades: the rise and decline of Japan's DRAM industry, and the ascent of South Korea's Samsung Electronics. This is real economic history, with abundant primary sources and academic literature. Throughout, we distinguish between figures we were able to verify and figures we were not.

1. The Early 1980s: Japan's Conquest of DRAM

DRAM (Dynamic Random Access Memory) was first commercialized by Intel in 1970, and through the 1970s the market was dominated by American companies — Intel, Texas Instruments, Mostek, National Semiconductor and others. That position reversed rapidly around 1980.

The protagonists of the reversal were Japan's integrated electronics conglomerates, led by NEC, Toshiba, Hitachi, Fujitsu and Mitsubishi Electric. Japanese share overtook the US in the 64K DRAM generation around 1981, and the following 256K generation made the gap decisive. Several estimates put Japanese firms at 75–80% of the global DRAM market around 1986–1987. The US share of DRAM fell from roughly 70% in 1978 to about 20% by 1986. Looking at semiconductors as a whole (not just DRAM), Japan's world share passed the United States in 1987 (48% Japan vs. 39% US) and peaked around 1988 at approximately 51%.

What drove the reversal was primarily manufacturing quality and yield. Analyses of the period — including reports from the US Office of Technology Assessment — repeatedly note that Japanese fabs ran yields 20 to 30 points higher than their American rivals. DRAM is a textbook commodity: design originality buys little differentiation, and everything is decided by who can make the same capacity at the same specification more cheaply and at higher yield. For a manufacturing sector built on quality control and continuous process improvement, there could hardly have been a more favorable arena.

A second factor was the scale and continuity of capital investment. Because Japanese firms held DRAM as one division inside a diversified electronics conglomerate, they could keep investing through memory downturns on profits from other businesses. American companies, which were closer to pure-play memory makers, saw a downturn translate directly into an existential crisis. The memory recession of 1984–85 brought this structural asymmetry into the open.

The emblematic event is Intel's exit from DRAM in 1985. With falling prices, slumping demand and a supply glut converging, Andy Grove and his colleagues cut the memory business and redirected all resources into microprocessors. It involved plant closures and large-scale layoffs, and is recorded as one of the most painful decisions in Silicon Valley's history. The company that invented DRAM walked away from DRAM — nothing speaks more eloquently to how overwhelming the Japanese position had become.

2. The Plaza Accord (September 1985): An Exogenous Currency Shock

Almost simultaneously with the peak of Japanese semiconductors, an event occurred that changed the industry's underlying premises. On September 22, 1985, at the Plaza Hotel in New York, the finance ministers and central bank governors of the G5 (United States, Japan, West Germany, France and the United Kingdom) reached what became known as the Plaza Accord.

The backdrop was the enormous US trade deficit and the strong dollar widely blamed for part of it. In the early 1980s, capital inflows drawn by high interest rates had left the dollar substantially overvalued; US manufacturers were losing export competitiveness, and protectionist trade bills were piling up in Congress. The Plaza Accord was an attempt to relieve that pressure through exchange rate policy rather than trade policy.

The mechanism was coordinated foreign exchange intervention. The monetary authorities of the participating countries would jointly sell dollars and buy yen and Deutsche Marks. On top of that, by explicitly announcing to markets that "some further orderly appreciation of the main non-dollar currencies against the dollar is desirable," they moved market expectations themselves. The signaling effect — G5 governments are serious about a weaker dollar — is generally judged to have mattered more than the actual volume of intervention.

The result went far beyond what the participants intended. Before the accord, the dollar bought roughly ¥240. By 1986 it had moved to around ¥150–153, and by 1988 it reached roughly ¥120. In under three years, the yen's value against the dollar had roughly doubled. Accounts of the negotiations suggest the participants had in mind a dollar depreciation on the order of 10–15%; what actually happened was a currency revaluation several times that size.

The impact on semiconductors was direct. A Japanese DRAM maker's cost base was yen-denominated — domestic fabs, domestic labor, domestic capital equipment — while a large share of revenue came from dollar-denominated exports. A doubling of the yen means that the same yen cost base must now be sold at twice the dollar price to earn the same yen revenue. The overwhelming price competitiveness that Japanese firms had built on superior yield was eroding rapidly for reasons entirely outside their own control.

And the timing coincided precisely with the moment a new entrant — South Korea — was getting off the ground. The won did not appreciate anything like as sharply as the yen, so Korean firms could expand production under a relatively favorable currency environment. For Japanese makers, the currency shock arrived at the worst possible moment.

3. The US-Japan Semiconductor Agreement (1986): An Unintended Price Umbrella

Exactly one year after the Plaza Accord, on September 2, 1986, the US-Japan Semiconductor Agreement was signed. This time the intervention came not through currency but directly through trade policy.

The agreement rested on two main pillars.

• Anti-dumping (price floors) — Japanese makers agreed to stop selling DRAM and EPROM below cost in the US market and in third-country markets. Operationally, the US Department of Commerce calculated a company-specific and product-specific FMV (Fair Market Value) each quarter and communicated it to Japanese firms. In effect, the US government was setting the minimum export price of Japanese memory.
• A foreign market share target inside Japan — The Government of Japan "recognized" the US semiconductor industry's expectation that sales in Japan by foreign-capital affiliated firms would grow to slightly above 20 percent of the Japanese market within five years. That 20% figure appeared not in the agreement's public text but in a confidential side letter. When the document's existence later came to light, it triggered a long-running dispute over whether Tokyo had committed to a numerical target or merely acknowledged an expectation.

Implementation went badly. In 1987 the United States imposed retaliatory tariff measures on Japanese electronics products, citing continued third-country dumping and slow progress on market access. The agreement was revised and extended in August 1991, with the revised version framing "over 20% foreign share by the end of 1992" as a more explicit target. Foreign share did reach 20.2% on the US calculation and 22.5% on the Japanese calculation in the fourth quarter of 1992, so the target was deemed met. The agreement lapsed on July 31, 1996.

The Paradox the Price Floor Created

Here is the most interesting — and most ironic — part of the agreement.

The stated purpose of the price floor was to protect American DRAM makers from Japanese price attacks. What actually happened was that the price level of the entire market was artificially propped up. Because Japanese firms supplied the great majority of world DRAM at the time, setting a minimum price on Japanese product effectively set a floor under global DRAM prices. In the immediate aftermath, 1987–88 saw a severe DRAM shortage and sharply higher prices — with the irony that Japanese makers themselves temporarily enjoyed unusually good margins.

But an artificially elevated price level is also an ideal entry environment for a newcomer with a lower cost structure. As long as Japanese firms were bound by "we cannot sell below this price," Korean and Taiwanese entrants could win customers simply by pricing a little beneath it. And because the floor kept market prices high, those entrants could capture respectable margins from the moment they entered — a condition that essentially never exists in a normal commodity market.

Under other circumstances, the barrier to entry facing a newcomer would have been exactly Japan's superb manufacturing efficiency and the relentless price offensive it made possible. The agreement had the effect of making Japanese firms dismantle that barrier with their own hands. A measure designed to protect American industry ultimately conferred its largest benefit not on American companies but on Korean and Taiwanese ones. This is the "paradox of the US-Japan Semiconductor Agreement" so often cited in trade policy history.

The numbers bear the turn out. Japan's global DRAM share fell from about 75% in 1987 to about 20% by 2001 — a collapse over fourteen years — while Korean share rose over the same period from about 5% to about 40%.

4. The 1990s: Price Collapse, Consolidation, and Elpida

Entering the 1990s, three burdens pressed on Japanese DRAM makers at once: (a) rising dollar-denominated costs from yen appreciation, (b) the loss of price as a competitive instrument under the floor agreement, and (c) the catastrophic DRAM price collapses from the mid-1990s onward.

DRAM prices peaked in late 1995 and then fell apart. One set of estimates puts the decline at roughly 51% in 1996 and a further 65% in 1997; another tabulation finds annual price declines exceeding 60% sustained from 1996 through 1998. Japanese firms fell into losses across the board in their memory businesses in the late 1990s. Reporting from 1998 describes Mitsubishi Electric forecasting a loss of roughly ¥10 billion, with Toshiba and Hitachi both guiding to sharply lower profits; US analysts at the time believed the majority of Japanese chip operations were losing money.

The strategy Japanese firms adopted in this phase was exit and consolidation. From the late 1990s into the 2000s, Fujitsu, Toshiba and Mitsubishi Electric each withdrew from commodity DRAM or spun the business out.

The culmination was Elpida Memory. At the end of 1999, NEC and Hitachi announced a joint venture combining their DRAM businesses; it launched as NEC Hitachi Memory, Inc. and was renamed Elpida Memory in 2000, headquartered in Tokyo. In 2003 it absorbed Mitsubishi Electric's DRAM business as well, consolidating Japan's commodity DRAM into a single company.

A note on Hiroshima. Elpida built its flagship 300 mm wafer fab (known as the E300 fab) in Higashihiroshima, Hiroshima Prefecture, near NEC's existing operations, breaking ground in February 2001. That Hiroshima fab remained the last major production site for commodity DRAM in Japan — for Elpida and for the country. It continues to operate today as the core of the company's Japanese DRAM production under Micron ownership, as described below.

Elpida, however, never caught up with Samsung and Hynix in the capital investment race. The post-2008 downturn, the historic yen appreciation that followed, stagnant PC demand, and the disruption to PC production caused by the 2011 Thai floods that struck hard disk drive factories all converged. On February 27, 2012, Elpida filed for corporate reorganization with total liabilities of approximately ¥448 billion (about US$5.5 billion at the time) — the largest Japanese corporate failure since Japan Airlines in 2010, and the largest ever in Japanese manufacturing.

The sponsor that stepped forward was, ironically, the American company that had first licensed DRAM technology to Samsung: Micron Technology. The Tokyo District Court approved the reorganization plan in February 2013, and on July 31, 2013, Elpida became a wholly owned subsidiary of Micron. The total consideration was approximately ¥200 billion (about US$2.5 billion), structured as ¥60 billion (about $750 million) in cash at closing plus ¥140 billion (about $1.75 billion) in annual installments through 2019. Elpida was subsequently renamed Micron Memory Japan, and the Japanese commodity DRAM maker effectively disappeared at that point.

5. Samsung's Rise: Technology Transfer, and Something More Important

1983 — The "Tokyo Declaration," and Doors That Kept Closing

Samsung Electronics committed to full-scale entry into DRAM in 1983. The starting point is usually told as the "Tokyo Declaration" — founder Lee Byung-chul's announcement, made while in Tokyo, of a serious investment push into semiconductors. At the time, Samsung was a literally unknown latecomer in DRAM. It was proposing to enter from zero at exactly the moment Japanese firms had swept the world in 64K DRAM and American firms were beginning to contemplate withdrawal.

Entry required DRAM process technology. And here lies the single most important fact of the story. According to multiple accounts, Samsung approached Hitachi, NEC, Toshiba, Texas Instruments and Motorola for a DRAM technology license, and was turned down by all of them. The industry leaders of the day saw no reason to hand their core technology to an unknown Korean firm.

The only company that agreed was a then-small American firm, Micron Technology. Around June 1983, Micron supplied Samsung with 64K DRAM design and manufacturing technology in exchange for a cash payment. Working from it, Samsung developed a 64K DRAM just six months later, in November 1983, and began mass production in mid-1984.

Chaebol Capital and Counter-Cyclical Investment

Acquiring technology only gets a company to the starting line. Most analyses agree that the decisive factor that carried Samsung to the top of the DRAM market was not technology but how it deployed capital.

DRAM is an industry with violent cycles. In a downturn, when prices collapse, the ordinary firm freezes investment, throttles output, and tries to stop the bleeding. Samsung did the opposite — expanding capital investment precisely during downturns, building new fabs and adding capacity when competitors were pulling back. This is counter-cyclical investment.

The logic is straightforward:

• Equipment and construction are cheap in a downturn; the same money buys more capacity
• While competitors halt investment, you can pull a technology generation ahead
• The moment the market recovers, your expanded capacity runs flat out and takes the demand
• If the price war drags on, weaker competitors drop out first, and the market structure itself reshapes in your favor

The catch is that this strategy presupposes the financial capacity to bleed red ink for years through a downturn. An independent company answering to shareholders every quarter simply cannot adopt it. What made it possible was South Korea's distinctive chaebol conglomerate structure: the capital of the wider Samsung Group could be channeled into the semiconductor division, backed by a South Korean government that had designated semiconductors a national strategic industry — arranging low-interest financing, organizing R&D as national projects, and so on.

In fact, Samsung's semiconductor business ran at a loss for essentially the whole of the 1980s, turning profitable around 1988. The group absorbed those losses for close to five years from entry to profitability.

A concrete example of the state's involvement is the national 4-megabit DRAM R&D project launched in October 1986. Under the management of ETRI (Electronics and Telecommunications Research Institute), Samsung, Goldstar (now LG) and Hyundai Electronics (now SK Hynix) worked jointly on 4M DRAM development, while each also pursued its own parallel program. The project is credited with narrowing Korea's technology gap with Japan to roughly one year. What matters is that this was designed not as the effort of a single firm but as a coordinated industry-government campaign.

The Stack Bet, and the 1992 Reversal

In the late 1980s the industry split over DRAM memory cell architecture: the trench approach, etching the capacitor deep into the silicon substrate, versus the stack approach, building it up above the substrate. As geometries shrank, trench structures became rapidly harder to fabricate and harder to inspect for defects. Samsung chose stack, and that technical bet is widely credited with the later difference in how fast the company could ramp volume production.

Then, across 1991 and 1992, Samsung executed capital investment on a record scale. Some reporting cites figures of 450 billion won in 1991 and 800 billion won in 1992; because the dollar conversions given in different sources do not agree, we treat these only as an indication of magnitude. What is certain is the outcome: in 1992, Samsung overtook Toshiba to become the world's largest DRAM maker. The same year, Samsung developed the world's first 64M DRAM, reportedly about six months ahead of equivalent parts from its Japanese rivals.

An unknown latecomer reached the top of the world market nine years after entry. Technology transfer alone cannot explain that pace. What was decisive was that a group and a state kept underwriting long-horizon capital deployment with losses priced in from the start.

6. Putting It Together: Where the Three Forces Overlapped

Laying the pieces out in sequence:

• 1980–1985: Japanese firms dominate DRAM on yield and investment capacity. Intel exits in 1985.
• September 1985: Plaza Accord. The yen roughly doubles against the dollar in under three years. Japanese price competitiveness is damaged by forces outside the industry.
• September 1986: US-Japan Semiconductor Agreement. Price floors deny Japanese firms their remaining weapon — the price offensive. At the same time market prices are propped up, opening room for new entrants.
• 1983–1992: Samsung enters on technology licensed from Micron, closes the gap with counter-cyclical investment backed by chaebol capital and state support, and takes the world lead in 1992.
• 1996–1998: DRAM prices collapse at 50–65% per year. Japanese makers fall into losses across the board.
• 1999–2013: Japan consolidates into Elpida, loses the capital investment race, files for reorganization in 2012, and passes to Micron in 2013.

Several causal loops emerge from that sequence. First, yen appreciation and the price floor neutralized, almost simultaneously, the two weapons Japanese firms held: a cost advantage and the freedom to set price strategically. Second, this happened at precisely the moment Korean firms were entering and trying to gain altitude. Third, the perfectly rational corporate behavior Japanese firms adopted — cut investment in a downturn and stop the losses — was structurally the losing move against a counter-cyclical investor.

And finally, the most ironic loop of all. Samsung obtained its first DRAM technology from Micron of the United States, and Micron ultimately bought Elpida, Japan's last DRAM redoubt. Meanwhile Japanese firms — Sharp above all — had contributed, if not in DRAM itself, to the early foundation of Samsung's semiconductor process capability. Japanese companies had, directly and indirectly, supplied part of the technology base of the competitor that would later displace them.

Summary

• Around 1986–1987, Japanese firms held 75–80% of the global DRAM market, driven by yields 20–30 points above US rivals and sustained capital investment backed by conglomerate balance sheets
• The Plaza Accord of September 1985 used coordinated intervention to push the yen from roughly ¥240/USD to about ¥120/USD by 1988, eroding Japanese dollar-denominated price competitiveness from outside the industry
• The US-Japan Semiconductor Agreement of September 1986 rested on price floors (FMVs calculated by the US Commerce Department) and a 20% foreign market share target recorded in a confidential side letter
• The price floor artificially propped up market prices, handing Korean and Taiwanese entrants both room to undercut from below and margins to do it with. Japan's manufacturing efficiency, its real barrier to entry, was dismantled by agreement
• Japan's global DRAM share fell from about 75% in 1987 to about 20% in 2001; Korea's rose from about 5% to about 40%
• Elpida Memory was formed in 1999 from the merged DRAM operations of NEC and Hitachi (headquartered in Tokyo, flagship 300 mm fab in Hiroshima), absorbed Mitsubishi Electric's DRAM business in 2003, filed for reorganization in February 2012 with about ¥448 billion in liabilities, and became a wholly owned Micron subsidiary in July 2013
• Samsung entered in 1983 on a 64K DRAM license from Micron after being refused by Hitachi, NEC, Toshiba and others; the clearly documented Japanese transfer is Sharp's SRAM/ROM process technology, not DRAM
• Samsung's decisive advantage was less the technology transfer than counter-cyclical capacity investment through downturns, sustained by chaebol capital and national strategic industry policy — it overtook Toshiba for the world lead in 1992
• How much each factor mattered is still debated; this is not a single-cause story