In the wake of the recent allegations of Chinese espionage, the release of the document that started it all, drafted by a select House committee led by California Republican Christopher Cox, is being held up over arguments about what parts should remain secret. The gist of the report, however, has already been leaked to the press: Not only did China illicitly obtain nuclear warhead technology from the U.S., but it obtained the "supercomputers" necessary to develop new warheads, courtesy of U.S. computer companies.
These computers allow the Chinese to do "virtual" testing of their weapons without violating the Comprehensive Test Ban Treaty. This situation suggests that everyone, including the government, must adjust to the realities of the computer revolution; computer speed and performance are increasing so rapidly that they have rendered meaningless current U.S. policies concerned with safeguarding national interests and security.
At the heart of this situation is the now-legendary Moore's Law, named after Gordon Moore, the former CEO of Intel Corp. As Moore put it, "The power of semiconductor technology doubles every 18 months." Moore's Law implies that even the definition of what constitutes a supercomputer is a moving target.
In the early 1980s, for instance, the fastest supercomputer in the world was the Cray X-MP, which sold for $20 million and required a special cooling system and an around-the-clock technical staff to maintain it. Today, comparable computing power can be achieved by a $1,500 PC with a Pentium II microprocessor. Today's state-of-the art supercomputers are 100,000 times faster still.
When it comes to export controls, the government relies on a measure of computer speed known as MTOPS, or millions of theoretical operations per second. The current record holder for supercomputers--built by IBM for the Lawrence Livermore National Laboratory--is capable of performing 1.6 million MTOPS. In comparison, the computers exported to China and reported to be a concern of the Cox committee are 200 machines sold last year with speeds of between 2,000 and 7,000 MTOPS.
This level of performance is currently restricted by export controls. Any manufacturer wanting to ship such computers to "Tier 3" countries--which include China, India, Israel, most of the former Soviet Union and a number of Middle Eastern nations--must first notify the Department of Commerce. That agency then has 10 days to authorize the shipment or require the manufacturer to file for an export license, a process that can take another three months.
The problem, of course, is that not only are supercomputers getting faster exponentially, but so are business and personal computers. Last year's typical PC with a Pentium II processor was capable of performing at 500 MTOPS. Machines with the new Pentium III processor, which started shipping in early March, are capable of performing more than 1,000 MTOPS, while business computers with dual Pentium III processors are already performing in the range restricted by export controls.
In 1998, the Commerce Department had to deal with 390 notifications to ship computers of 2,000 to 7,000 MTOPS to Tier 3 countries. By late this summer, the industry will be producing millions of such machines, and the Commerce Department will be faced with 390 notifications a day, rather than 390 a year. And the situation will only get worse. The Gartner Group, a technology consulting firm, predicts that by the end of next year there will be 4.3 million computers running in excess of 2,000 MTOPS. Of these, one in every four will be produced by foreign manufacturers.
Since the 1940s, Congress has repeatedly amended export control laws to address changing political, security and commercial interests.
The issue today requires rethinking a policy that has its philosophical origins in the Cold War. The questions are not whether the U.S. will be able to deny Tier 3 nations computers that are commercial commodities throughout the world, but whether U.S. or foreign companies will sell it to them and whose economy will reap the benefit. Today, anyone with a PC and a modem can be networked into a supercomputing center at numerous universities. The notion that raw computing power can and should be controlled has been made anachronistic by today's technology.
There also is a short-term imperative here. By late summer, the Commerce Department will find itself flooded by applications to export computers. But the Defense Authorization Bill requires a six-month review by Congress of any potential policy change, which means even immediate action is likely to result in the status quo still being in place when the flood gates open. In the short term, American computer manufacturers need to work with the administration to derive a solution that can be implemented by this summer.
The long-term challenge is to reform the export control system so that it is controlling truly sensitive information and not commercial commodities that can easily be purchased elsewhere. Simply raising performance levels every six months will only lead to the export control version of a fire drill every time a new chip hits the market. We need instead to develop a fundamental philosophy on export controls that is in tune with the technological, economic and security realities of the future.