Chip in Space:
MITI wants to fly 'em and fry 'em
by Paul Kallender

This summer, the Ministry of International Trade and Industry (MITI) presented Japan's Big Three satellite makers with an ultimatum: build these or die. MITI's June 17 Request For Proposals for its twin-bird Space Environment Reliability Verification Integrated System (SERVIS) program asked Mitsubishi Electric (Melco), Toshiba, and NEC to assemble what amounts to the ministry's last-ditch, government-funded attempt to make commercially competitive satellites. That, or "be doomed to failure," as MITI's space hitman Hirotaka Kawamoto says.

Looking at the National Space Development Agency of Japan (NASDA) - Japan's version of America's NASA - it's not hard to see why Hirotaka, President of MITI's space arm, the Institute of Unmanned Space Free-Flyer (USEF), thinks so. Of four of NASDA's last satellites, averaging $350 million a piece and taking 5-7 years to build, one, the ETS-6 buzzed off into the Van Allen Radiation Belts; the next, the ADEOS said "bye-bye" when its solar paddle unraveled and fell apart; and another, the ETS-7, catalogued 22 major faults and continually got itself lost. In addition, last year the COMETS space vehicle lived up to its name, eternally condemned to streaking through a highly elliptic orbit following a rocket failure. Meanwhile, the US Iridium satellite phone consortium, under the slogan "Build it, Ship it, Shoot it!," last year managed to loft 72 satellites in 15 launches in 12 months. At peak capacity, prime contractor Motorola was assembling the satellites every four days, the company's Wayne Daily told a flabbergasted audience assembled by Kawamoto at the Keidanren (Japan Federation of Economic Organizations) meeting this past March.

Can Japan be like Iridium?
Meanwhile, despite its technological success, the Iridium program itself is struggling with bankruptcy (nobody wants to buy $2,000 phones that look like repackaged Korean War GI field issue); but the point is clear, says Kawamoto. Of 30 years of subsidies now delivering a recent history littered with failure, Hirotaka says, "if the US is at the graduate level in building satellites, Japan is still in kindergarten ..." At the moment, Japan stands no chance of competing in next decade's $75 billion global communication satellite market, Kawamoto adds. Or as he puts it, unless Japan learns quickly, "and if we don't grasp the opportunity now, we will all go to hell."

At his SERVIS
Last November, to its credit, Melco made history when it won Japan's first-ever commercial satellite integration contract for an Australian broadcasting/military communication satellite, the Optus C-1. While much of the know-how and many of the components will come from US giant Space Systems/Loral (SS/L), the award signaled that confidence is growing in Japan's integration abilities. In the mid-90s Melco, Toshiba, and NEC made huge strides in reducing the cost of hardware components (amplifiers, solar cells, cooling systems), thus generating a small but significant space parts business for themselves. But how can economies of scale be built into entire satellites? Kawamoto thinks the answer is under your bonnet, or even on your desktop - my aging Mac 7500/100 is pepped with 126MB of memory, and I pity my poor100MHz processor and dream of a G3. Though it may be uninspiring to us, Japan's satellite makers would kill for such power and speed. While a 64MB DRAM ships at about five bucks, for NASDA, Hideo Inayoshi, deputy general manager of the company's Strategic Business Planning Division, says the cost of a space-use 1MB SRAM chip - for example Hitachi's H32 processor - "including development costs, is about ¥2 billion." Yikes.

"That's the price of reliability," argues Sumio Matsuda, senior engineer at NASDA's Electronic and Information Technology Department. After all, he says, your PC doesn't get lobbed 36,000 kilometers into space where temperatures zoom from -70 to +100 degrees Celsius in just a few seconds, nor does it get its CPU continually battered by raw solar and cosmic radiation. And when your PC crashes, you reboot it. When a satellite On-Board Computer (OBC) crashes, so might the satellite, and so did much of the US's paging services, during the Galaxy-4 failure last year.

So before any bits are lofted into orbit, NASDA, or rather HIREC, a consortium of 32 electronics companies lead by the Big Three, beats the dickens out of them with ground tests first. "It's rather a complex process," admits Matsuda, handing over a 10 (large) page schema of the basics of silicon-style S&M. Stated simply, HIREC takes a batch of, say, 145 copies of a specially designed chip and puts them through a torturing cyclical QC battery. Following microscope examination (pattern defects, assembly and bonding, channel quality), some get cooked and frozen, while others are rocked and rolled (vibration tests). Yet others get stuffed into space chambers and fried with proton beams. Survivors are tested to destruction during lifetime tests and are then prised apart for microscopic examination. Then it starts again; and again Ad Astra, Per Arduum. All this takes 18 months and, on average, a single chip will end up costing ¥2 million.

Despite the (clearly justified) safety-is-all regime, progress is being made, says Ma-tsuda. Ten years ago a Japanese satellite was lucky if it had a 10KB array 16-bit OBC. Pretty soon, however, Japan will be entering Dreamcast level. Hardware like NASDA's H2A Transfer Vehicle, a 15-ton supply tug that will have to find and automatically dock with the International Space Station (ISS), will deploy a 64-bit, 64MB DRAM and a 4MB SRAM-decked R4000 series super-duper OBC. But the R4000 series is still a symptom of the greater problem. While based on run-of-the-mill Toshiba design technology crafted onto NEC wafers, it's still a built-to-order specialist piece of ancient chip history. Satellites are still pretty dumb, and their OBC electronics costs are a key overhead that MITI can cut.

So, the answer, says MITI, is Commercial Off-The-Shelf components, or COTs. If Japan can put COTs on its satellites, instead of stuff rejected as too poor by the North Koreans, argues Kawamoto, it can take a big step forward to commercial competitiveness. Enter the ¥23 billion SERVIS program, a two-satellite constellation test-bed program (with the birds flying in 2002 and 05) to stuff all kinds of promising ICs, ASICs, LSIs, DRAMs, and fry them up above in a 1,000-km-high orbit for two years to see what survives. What isn't junked will be cheap and proven; then, so the reasoning goes, Japan can play ball with the Motorolas, SS/Ls, and Lockheed Martins of this world.

High frying
Far from revolutionary, Kawamoto's plan is, in fact, a distorted echo of former DoD honcho William Perry's 1994 clarion call to end the Pentagon's $1,000 spanner procurement practices. In attempting to revolutionize these, Perry de facto dumped the notorious Milspecs (military specifications) standards system, an enormous approved list of specially tested and priced (read: expensive) parts and components.

NASA immediately followed suit, with new administrator (former hard-nosed ex-TRW boss) Dan Goldin instituting a cheaper, faster, better program, which is now delivering a score of high-publicity, comparatively low cost PR successes such as the Mars Explorer. NASDA is trying to copy NASA through a mix of force and inevitability. Japan now lacks the old set of comfortable Milspec S-components (top grade, top-price). So in this light, SERVIS makes much sense.

Chip testing
However, before anything gets up there, USEF says it will have to begin its own massive chip testing program to select and identify suitable hardware. "It's too early to say what chips we will be testing and what the selection and testing process will be," says USEF Managing Director Hiroshi Kanai. According to USEF plans, MITI will begin with about 5,000 CPUs, gate arrays, and memories and - you guessed it - put them through 200 kinds of environmental, vibration, shock, thermal cycle, acceleration, and radiation tests. Mmmm, ... maybe USEF knows what it's doing. Maybe ground testing will be efficient. They'd better find out quick, say experts who attended MITI's space COTs symposium this March. Because COTs themselves can be hell. "Perhaps we sound a little negative about this," says Douglas McCormac, vice president of TRW Components International - one of the US's largest satellite components suppliers - but "COTs has large hidden costs. The Jet Propulsion laboratory says COTs systems costs more than traditional components... [in fact] COTs parts programs are not cost reducers," he warns. Daniel Guyomard of ESA's Manned Space Program recounted the reasons why.

When the European Space Agency (ESA) tried to deck out the ISS's Russian portion with a durable OBC/OS based on 250MB mass memory COTs, the problems soon multiplied. "The hard disk became obsolete before we were one year into the program," he says, as did the OS NASA used on the Mars Pathfinder and Surveyor missions. Next the CPUs, which had to fit NASA's US Space Station Honeywell computer, were geriatric 386-SXs "which were totally obsolete and totally incompatible with anything now on the market." When ESA procured SRAMs supplied from Korea, the (unnamed) company refused to work with them over debugging its chips, because it only bought 100 units. Delays in delivery and communications snafus snowballed. "Developing the processor was quite painful" he says, as was instant obsolescence, with inventory and supplier tech support melting in six months. "Twenty percent of our three-year program was eaten by parts procurement alone," he admits. "Then there was the paperwork associated with the procurement, which nobody in the end looked at and ended up in a cupboard covered in dust," he grumbles. "Obsolescence is the critical issue. We don't yet have an answer or a solution, and if anyone has a miracle to offer, we need to hear about it."

Milspec-like parts
As for the much-vaunted Iridium constellation, which was reputed to contain a staggering 74 percent COTs, only 38 percent of those were "active" and non-critical, says Daily. In fact, to its shock, Motorola found using pre-tested and expensive "Milspec-like" parts was cheaper than doing its own selection. "COTs processing costs can quickly swamp, and the volume of paperwork exceed the cost of the part very easily," he warns. NASA's next-generation COTs program involves pumping $64 million into Intel and hiring 30 designers to build a space-use Pentium I processor. Calculate that move and then apply it to MITI's ¥23 billion program; things clearly don't add up for the Japanese taxpayer.

"Yes, putting parts in test chambers is expensive," admits Kawamoto. There are trade-offs to work out, he adds. Screening will have to be rationalized. And MITI is aware that commercial chips have about a six month shelf life. But LSI is coming, he counters. The key argument is that the program will introduce economies of scale; create unified and rationalized databases; and if managed well, SERVIS will fly/fry about 1000 kinds of cheap chips in 10 categories that, he says, will save costs and make Japanese satellites competitive. We shall see.

Paul Kallender is a journalist previously based in Tokyo. In August, he moved to New York's Columbia University to pursue a Master's degree in journalism. Computing Japan wishes him success with his studies.

Back to the Table of Contents

Comments or suggestions?
Contact cjmaster@cjmag.co.jp