Elpida Memory - Company History

DRAM is a capital-goods business with one iron rule: whoever stops paying for the next process node loses on cost. Inside a Japanese electronics conglomerate, a memory division had to win its capital budget against every other division each year — an argument it could not keep winning against Korean rivals who simply invested. That arithmetic is why NEC and Hitachi carved their DRAM operations out rather than fix them in place, incorporating NEC Hitachi Memory in December 1999 in Yaesu, Tokyo, and renaming it Elpida Memory in May 2000.

The carve-out inherited an industry in retreat. Japanese makers had held nearly 90% of the world DRAM market in the 1980s; by the late 1990s Samsung had taken it with concentrated investment, and in the years around Elpida’s founding Fujitsu and Toshiba both walked away from DRAM. But a 50-50 joint venture is a slow machine. The roughly 15% combined share the two parents brought in had fallen by more than half by early 2003, dropping Elpida to fifth in the world; consolidated revenue for the year to March 2002 was ¥78.3 billion against a net loss of ¥26.6 billion, with only 630 employees and no fabs of its own — production was contracted out to NEC’s Hiroshima plant and Hitachi’s Singapore plant.

In February 2001 Elpida began building its own 300mm wafer fab, E300Fab, at Hiroshima. Each process generation yields more chips per wafer, so whoever spends first takes the volume profit — but the plant needed roughly $1.3B (¥160bn) in equipment against ¥52 billion of paid-in capital, the rest borrowed. When the fab opened in January 2003 at only 3,000 wafers a month, it was plainly too small to survive on, and in February 2003 NEC and Hitachi told Elpida there would be no further equity from them: find the expansion money outside. What the parents supplied instead of capital was a new chief executive.

In November 2002 Sakamoto Yukio became president — an outsider courted for a year by NEC’s own president after he had turned around UMC Japan. He had joined Texas Instruments Japan out of university in 1970, run a 600-person manufacturing department at thirty-one, and reached the vice-presidency by 1993. His opening promises were specific and public: profitability within a year, a stock listing within two, a top-three world share within four. He transferred every seconded parent-company employee onto Elpida’s own payroll and abolished the practice of alternating senior posts between NEC and Hitachi appointees — the point being that a company owned half-and-half decides nothing quickly.

The money then came from outside rather than from above, which also diluted the parents and so sped up decisions further. Within a year of taking office Sakamoto raised ¥170 billion, and in March 2003 Elpida absorbed Mitsubishi Electric’s DRAM business, leaving Japan with exactly one DRAM company. In November 2004 it listed on the First Section of the Tokyo Stock Exchange, issuing 31.85 million shares. The year to March 2005 brought revenue of ¥207.0 billion, roughly 2.1 times the prior year, and the first profit in the company’s history — ¥8.2 billion. High-margin non-PC chips were made in Hiroshima; commodity PC parts were farmed out to foundries such as SMIC.

The high point came in the year to March 2007: revenue of ¥490.0 billion, operating profit of ¥68.4 billion and net profit of $449.2M (¥53bn) — the largest Elpida would ever earn — with the market capitalisation reaching ¥848.8 billion in December 2006. By then Hitachi held 11% and NEC a little over 8%. The same month Sakamoto agreed with Taiwan’s Powerchip to build Rexchip, a 50-50 fab in Taichung: ¥480 billion of combined investment, 240,000 wafers a month at full tilt, on land leased from the state at ¥900,000 per hectare per month where Hiroshima industrial land sold for some ¥200 million per hectare. He announced it without consulting his former parents. “Elpida and Powerchip used to be relatives,” he said in Taipei. “From now on we are family.”

DRAM prices began falling in late 2007 and did not stop. A 1-gigabit chip that fetched $6 was under 60 cents by December 2008 — below cost even for Samsung — because makers starved of cash in the financial crisis dumped inventory at any price. Elpida lost ¥178.9 billion in the year to March 2009 on revenue of ¥331.0 billion; interest-bearing debt swelled to ¥495.0 billion and the equity ratio fell from 46.1% to 17.3%. The company had spent through the downturn regardless, absorbing its Hiroshima production subsidiary in April 2008 and consolidating Rexchip and Tera Probe in March 2009.

On 30 June 2009 Elpida became the first ordinary company certified under the revised Industrial Revitalisation Act: the Development Bank of Japan took $320.8M (¥30bn) of preferred shares and lent ¥10 billion, with the government guaranteeing up to 80% of the investment, alongside ¥100 billion of syndicated bank lending. Officials were explicit about the motive — one METI executive said Japan could not let Korea become the sole source of DRAM. Sakamoto was equally blunt about accepting state money: surviving and being able to invest again mattered more than pride. The catch was structural. ¥30 billion was enough to postpone a crisis and not enough to make any bank an owner of the problem, and the scheme’s assumed Taiwanese alliance never materialised.

That alliance had been Sakamoto’s real strategy. He believed only two DRAM camps would survive, and spent late 2008 flying to Taiwan fortnightly to fold its six small, patent-poor makers into Rexchip — six firms investing and pricing independently were why spot prices had fallen fifteen-fold in two and a half years. One chief executive refused to sign, and Taipei set up a state holding company instead; nothing came of it. Markets recovered enough for two profitable years, and the year to March 2011 produced Elpida’s highest-ever revenue of ¥514.3 billion — but only ¥2.1 billion of net profit, with ¥337.2 billion of debt still outstanding. From the summer of 2011 spot prices fell below $1 against a break-even nearer $1.50. “We use the most advanced technology in the world,” Sakamoto told analysts, “and it sells for the price of half a rice ball.”

Between January and April 2012 Elpida had to find over ¥120 billion for bond redemptions and loan repayments, and its Revitalisation Act certification expired that March. Having dealt with every bank at equal distance, it had no lead bank to organise a rescue — only the Development Bank of Japan, which in mid-December 2011 made refinancing conditional on finding a partner and adding ¥200 billion of capital. In January 2012 the METI official who had handled the earlier bailout was arrested for insider trading, which made further state help harder still.

A merger with Micron Technology had been negotiated since autumn 2011 and a scheme was agreed on 3 February 2012 — an eight-year courtship, since Micron’s chief executive Steve Appleton had been proposing it since 2004. Days later Appleton was killed in a light-aircraft crash and the deal died with him. The ¥200 billion condition remained. Sakamoto flew abroad weekly and could raise loans but not equity; on the morning of 27 February a company that had promised to buy half the Hiroshima plant did not call, and when he called at 10:30 the price did not work. That afternoon Elpida filed for corporate reorganisation at the Tokyo District Court with liabilities of $5.6B (¥448bn) — the largest manufacturing failure in Japanese history. The yen was under ¥80 to the dollar and 2-gigabit PC chips were selling below $1. The exchange delisted the shares in March.

Four bidders emerged — Micron, SK Hynix, Toshiba and China’s Hony Capital — and what they wanted was telling: Toshiba was interested in the Taiwanese Rexchip fab and, in one executive’s words, felt nothing at all for Hiroshima. Only Micron would add ¥80 billion of capital spending on top of the ¥200 billion, and in July 2013 Elpida became a Micron subsidiary. The brand vanished after fourteen years, though the contract protected the plants and no jobs were cut; Hiroshima ran flat out at 120,000 wafers a month on mobile DRAM. Sakamoto’s own verdict was that the fatal error had been the ¥30 billion of 2009: “the worst mistake was making the DBJ our lead bank for ¥30 billion — without it, some commercial bank would have stepped up.” His evidence was the quarter immediately after the sale, when the same business earned ¥39.0 billion on ¥130.0 billion of revenue. By March 2013 the yen had fallen to ¥97 and DRAM had recovered to $1.74.