Mitsubishi quietly closes U.S. SoC venture

Mitsubishi quietly closes U.S. SoC venture
By Anthony Cataldo, EE Times
February 24, 2000 (5:08 p.m. EST)
URL: http://www.eetimes.com/story/OEG20000224S0042

TOKYO — VSIS Inc., a venture company backed by Mitsubishi Electronics America that was the focal point of the company's intellectual-property (IP) and chip-integration strategy in the United States, has quietly been shut down. Beyond Mitsubishi's own struggles, the move illustrates the difficulties Japan's chip giants generally have experienced in making a profit in what has proven to be a highly competitive systems-on-chip arena.

Founded in 1996, VSIS (Sunnyvale, Calif.) set out to tap Silicon Valley's budding intellectual-property and design community, acting as a quasi-startup that was to develop its own IP or serve as a venue for scouting out promising technologies to acquire. The company was also a center for R&D and product development for system-on-chip (SoC) devices.

Those ambitious plans came to a halt around mid-1999 when Mitsubishi decided to pull the plug on VSIS, an acronym for VLSI Systems Solutions. But the group's operations were not absorbed until "fairly recently," a spokesman from parent Mitsubishi Electric Corp. here said. The breakup has divided the VSIS engineering department between locations in Durham, N.C., and the Electronic Device Group in Sunnyvale, the spokesman said. The EDG will take over VSIS' SoC development efforts.

Despite several queries by EE Times, spokespeople at Mitsubishi's headquarters in Tokyo and its EDG subsidiary in Sunnyvale, Calif., declined to offer specific reasons for the closure. A spokesman in Tokyo said only that "markets and customers were moving at a really fast pace and in order to keep up with that, they merged the major functions of VSIS into the Electronics Device Group of Mitsubishi Electronics America."

Search for profits

From the beginning, VSIS' explicit goal was to be a money-making venture. But Mitsubishi and many chip companies in Japan that caught the system-on-chip fever sweeping the industry several years ago have been hard-pressed to make a pr ofit in SoC devices because of high development costs and severe price competition.

"To be honest, from a technology and manufacturing standpoint, system-on-chip is quite a difficult task," said Masamichi Ogura, group president for the Electronic Devices Group of Fujitsu Ltd. (Kawasaki, Japan). "That's especially true at the leading edge. At 0.25 and 0.18 micron it's hard to make any money. But 0.5 micron and up, that's where we have better chances of making money. So it's a matter of managing the mix."

Though sales of Mitsubishi's systems-on-chip are on the rise, the devices are not currently profitable, the company's spokesman here said. The company's sales of SoC devices are expected to rise from $636 million in 1998 to $1.2 billion by fiscal year 2001, ending March 2002.

By that time, Mitsubishi expects profits for SoC devices to be in the one to two percent range, the spokesman said. The margins are still a far cry from the 10 to 12 percent profitability that the company obtains from its microcontroller products.

With Moore's Law promising chip vendors an ability to pack ever-higher numbers of transistors on a die, semiconductor companies several years ago were quick to formulate plans to build, acquire and stitch together IP building blocks. Several industry trade groups were formed to facilitate standards and the trade of IP cores. The goal of the new groups was to create the development environment and functionality chip vendors needed to soak up all the millions of gates that new process technologies were expected to offer them.

Japanese chip makers were especially enthusiastic about SoC technology after the prices of DRAMs took a fall in late 1995. At the time, they began to emphasize the need to boost sales by selling logic devices, and later rushed headlong into SoC development as a way to add more value to their devices.

Many companies however, had underestimated the cost of development, as well as customers' resistance to paying big premiums for integration. And recent ly, some have started to make a case for better packaging technologies using multiple devices as an alternative to full-scale integration of functionality on one die.

So despite several years of developing new design environments, promoting reusable intellectual property and organizational changes, many Japanese vendors have been unable to bring down costs enough to make a profit in price-sensitive markets.

"For Japanese companies, system-on-chip is a horrible business," said Michito Kimura, a semiconductor analyst with IDC Japan. "Nothing's changed at all. All the vendors involved in system-on-chip can't make a profit."

Aside from high development costs, there are many players vying for a limited number of applications which themselves have thin margins. "Almost all of them are going into digital consumer devices. DVD prices are coming down rapidly, and so are digital still cameras, MD players, set-top boxes and hard-disk drives," Kimura said.

In the mid-1990s, Mitsubishi was one of the first companies to successfully integrate embedded DRAM with a graphics accelerator logic for NeoMagic. The company later added it to other devices such as microcontrollers. But embedded DRAM has proved thus far to only be viable in a limited number of applications where it makes sense to pay extra for the integration, such as digital cameras and graphics controllers, Kimura said.

Mitsubishi is also considered to have a weak IP portfolio compared with competitors, and in the past has been reluctant to license outside intellectual property that may be considered a threat to its own proprietary products.

It was not until last year that the company licensed the ARM TDMI core, making it one of the last major semiconductor suppliers do so. Mitsubishi moved slowly despite ARM's widespread use in cellular phones and other applications that made it a standard core offering among most ASIC suppliers.

Few cores

"Mitsubishi has not been very good in the ASIC business. And they have embedded DRAM technology, and it's brilliant, but they don't have a lot of IP cores," Kimura said. "That's probably why they decided to start a business developing IP cores. Now they seem to have given that up."

When it was established, VSIS was chartered to both develop and acquire IP cores. It's still unclear how much IP, whether created in-house or acquired, came out of VSIS. The company's former vice president of engineering Kenji Baba declined to comment.

VSIS did announce in1998 that it had licensed a multimedia DSP core from Bops Inc. (Santa Clara, Calif.), but the core did not show up in a recent presentation describing Mitsubishi's IP library. Bops officials said they were unsure whether Mitsubishi would put it into silicon. Mitsubishi did recently license the TeakLite DSP core from the DSP Group, and also has its own proprietary DSP core.

The Mitsubishi spokesman said the company will continue to both develop and acquire IP. "To make use of our proprietary IP, we've set up IP projects in areas l ike telecom and information devices to make efficient use of applications. And there are cases where there's IP not within the company that we need. In that case, in order to respond to the various needs of customers, we will purchase IP," he said.

The closure of VSIS came at a time when Mitsubishi and others had set out to turn around their chip operations, which had for years been wracked by volatile DRAM prices. While nearly all Japanese chip companies have scaled back or reorganized their DRAM operations, they have been focusing more on other standard products that are more profitable, like embedded CPUs, microcontrollers and flash memory.

SoC devices, meanwhile, have hardly provided a safe haven from DRAM volatility. Blaming fierce competition, NEC said recently its ASIC revenue will fall slightly by some $45 million this year to about $5.8 billion.

But even the decline in ASIC revenue and a six percent drop in sales for memory chips were not enough to stop the company's chip group from ge nerating an expected $227 million profit this year. A spokesman said the profitable year was due to operation cost reductions and brisk sales of microprocessors, embedded CPUs and discrete devices.

In Mitsubishi's case, the company is putting more emphasis on flash memory products, which are in short supply. The company plans to double its monthly flash memory output to 10 million units by year's end, which by then will about match the company's planned production of DRAMs. Mitsubishi expects capacity increases to generate nearly $1 billion in the next fiscal year, which begins in April.

The VSIS reorganization was Mitsubishi's second in the United States since 1998, when the company decided to close its discrete IC test and assembly facility in Durham. It then reassigned 100 engineers to work in advanced SoC and microcontroller devices under the Sunnyvale-based Electronic Device Group.

The EDG is headed by president Takeo Nishimura. VSIS' former president and chief executive officer, Osamu Tomi sawa, moved to the Durham facility and now serves as vice president of the EDG.

Former VSIS executive vice president Stephen Hester is now executive vice president of Mitsubishi Electronics America Inc. in Cypress, Calif., and former engineering chief Baba now serves as director of system-level IC marketing at the EDG in Sunnyvale. Mitsubishi's spokesman here could provide no further information on staff changes.