High Performance Computers and Export Control Policy: Issues for Congress

Congress has a strong interest in export control policy with regard to technologies that may have both commercial and military applications outside of the United States. Through its constitutionally delegated authority to regulate foreign commerce, Congress has the authority to control exports for national security or foreign policy purposes. This report examines congressional interest in the exportation of High Performance Computers, which are either single computing machines (usually called supercomputers) or a cluster of easily available, high-end workstations or personal computers.

Order Code RL31175
CRS Report for Congress
Received through the CRS Web
High Performance Computers
and Export Control Policy:
Issues for Congress
Updated January 25, 2006
Glenn J. McLoughlin
Specialist in Technology and Telecommunications Policy
Resources, Science, and Industry Division
Ian F. Fergusson
Analyst in International Trade and Finance
Foreign Affairs, Defense and Trade
Congressional Research Service ˜ The Library of Congress

High Performance Computers and Export Control
Policy: Issues for Congress
Summary
Congress has a strong interest in export control policy with regard to
technologies that may have both commercial and military applications outside of the
United States. Through its constitutionally delegated authority to regulate foreign
commerce, Congress has the authority to control exports for national security or
foreign policy purposes. The 109th Congress may examine issues of national export
control policy, including reauthorization of the Export Administration Act (EAA),
and control policies concerning key technologies. Among these technologies are High
Performance Computers — HPCs. HPCs are either single computing machines
(usually called supercomputers) or a cluster of easily available, high-end workstations
or personal computers. Congressional interest in HPCs has primarily, but not
exclusively, been focused on the dual-use applications of HPCs. That is, commercial
HPCs that may also be used to simulate nuclear weapon tests, chemical and
biological weapons production, and for military command, control, and
communications. The 109th Congress may consider the extension and revision of the
EAA through H.R. 4572 or other legislation. Both chambers last considered EAA
legislation in the 107th Congress. Legislation may also be introduced to repeal
specific controls on HPCs enacted as part of the National Defense Authorization Act
of 1998.
There are several issues that congressional policymakers may address
concerning reauthorization of the Export Administration Act. One is proliferation
— who obtains this technology and how they use it. In response to concerns about
proliferation, Congress has legislated licensing, post-shipment verification, export
notification, and licensing threshold requirements for HPCs. Some contend that
these safeguards and restrictions have not been effective. Another issue is the
notification process, in which the executive branch not only notifies the legislative
branch of HPC export control changes, but describes the national security
implications as well. Critics contend that the last several HPC export control
changes did not sufficiently provide this national security context. The U.S.
computer industry has also raised concerns about another issue — that restrictions
only hurt U.S. commercial interests. The computer industry and their supporters, as
well as others, contend that the computing benchmark for HPCs (Million Theoretical
Operations Per Second, or MTOPS), no longer reflects rapid technology innovations
in the field. Some have called for replacing MTOPS with a benchmark measuring
teraflops, and this standard has recently been adopted by the multilateral Wassenaar
Arrangement dual-use control regime.

Contents
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Congressional Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Proliferation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
U.S. Industry Concerns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Congressional Oversight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Issues for Policymakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Military Uses of HPCs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
U.S. Government and Industry Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
HPCs and Export Policy Revisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Multilateral Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
The Coordinating Committee (CoCom) . . . . . . . . . . . . . . . . . . . . . . . . 7
U.S. Japan Supercomputer Agreement . . . . . . . . . . . . . . . . . . . . . . . . . 7
Wassenaar Arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
U.S. Domestic Policy and Regulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
MTOPS and Country Tiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
The Cox Committee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
The National Defense Authorization Act of 1998 (NDAA98) . . . . . . 11
Further Decontrols: 1999-2003 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
MTOPS: Technology and Policy Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
MTOPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Eliminating MTOPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Teraflops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Other Alternatives to MTOPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
National Security, Economic Security and
Federal R&D: Issues and Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
National Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Economic Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Federal R&D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
List of Tables
Table 1. Performance Levels of Computers That Support Selected
Applications of Military Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Table 2. 1996 HPC Export Control Tiers and Policies . . . . . . . . . . . . . . . . . . . . 23

High Performance Computers and Export
Control Policy: Issues for Congress
Overview
Congressional Interest. Congress, through its constitutionally delegated
authority to regulate foreign commerce, has the authority to control exports for
national security or foreign policy purposes. The 109th Congress may examine issues
of national export control policy, including reauthorization of the Export
Administration Act (EAA), and control policies concerning several key technologies.
Among these technologies are High Performance Computers — HPCs. HPCs are
either single computing machines (usually called supercomputers) or a cluster of
widely available, high-end workstations or personal computers. Congressional
interest in HPCs has primarily, but not exclusively, been focused on the dual-use
applications of HPCs. That is, HPCs may be commercially obtained in the open
global market, but they also may have military applications, such as simulating
nuclear weapon tests, producing chemical and biological weapons production, and
for enhancing military command, control, and communications.
Legislation was last acted upon in the 107th Congress to reauthorize the Export
Administration Act, the statutory authority regulating exports of dual-use items,
including HPCs. In May 2003, the House rejected an attempt to repeal Title XII (B)
of Division A of the National Defense Authorization Act of 1998 (NDAA98), which
mandated licensing, notification, and post-shipment verification requirements of
HPC exporters by a vote of 207-217.
Proliferation. Technology is only as important as its applications. HPCs can
be used for various military applications from design and testing of weapons of mass
destruction to battlefield management. In response to concerns about proliferation,
Congress legislated licensing, post-shipment verification and Congressional
notification of exports and changes in licensing thresholds for HPCs in 1997.
Congress has focused attention on export control restrictions on certain countries.
China, India, Pakistan, the former republics of the Soviet Union, and others are
suspected of engaging or assisting in weapons proliferation. Administrative changes
in the export control system have made HPCs more available, while also attempting
to provide end-use and end-user post-shipment verification of HPCs. Many critics
contend that post-shipment verification has been ineffective, permitting HPCs to fall
into the hands of nations or entities without obtaining any clear knowledge of their
applications. Others contend that this technology is so widely available that to try to
restrict HPC sales only hurts U.S. commercial interests. There is also the issue of
rapidly evolving U.S. engagement with Pakistan, Russian, India, and China in
shaping an anti-terrorist strategy in the aftermath of the September 11 attacks.

CRS-2
U.S. Industry Concerns. While congressional policymakers are concerned
about the use of HPCs to assist in the proliferation of weapons of mass destruction,
they also have been responsive to U.S. industry concerns of competing fully in the
global economy. U.S. HPC firms tend to dominate global sales, but they are not the
only source of this technology. Therefore, several industry groups have cautioned
Congress that restricting HPCs will only mean that comparable technology will be
available from non-U.S. competitors. In the face of increasing global competition,
many in the U.S. computer industry, as well as supporters in the government and
elsewhere, have sought to expand, not restrict, HPC sales. Some have called for the
continued raising of the computing benchmark, Million Theoretical Operations Per
Second (MTOPS), which under current export control law determines which HPCs
can be sold to which users. Others have called for the complete elimination of
MTOPS as a control benchmark, contending that it is flawed and does not reflect
recent technology innovations. Some have called for replacing MTOPS with a
benchmark measuring teraflops, a standard measuring trillions of floating point
operations per second.
Congressional Oversight. Since 1998, the executive branch has raised the
MTOPS benchmark several times. The rationale for these actions has been that
HPCs are also widely available from non-U.S. sources, and continued restrictions in
the face of technological innovation only hurts U.S. firms. But many congressional
policymakers are concerned that executive branch notifications for rapidly raising the
MTOPS benchmark have not given Congress a clear assessment of the national
security implications involved. They would likely revisit the notification process and
call for greater information for assessing the national security implications for raising
export control thresholds of HPCs.
Issues for Policymakers. Among the HPC export control issues that the
109th Congress may consider include whether to maintain HPC export controls by
using MTOPS; revising HPC export control policy by using a benchmark other than
MTOPS; or to eliminate completely HPC export controls. In addition, the events of
September 11, 2001, may continue to affect the congressional debate on HPC export
controls as the United States reviews its national security concerns and foreign policy
alignments in light of the terrorist attacks.
Background
High performance computers (HPCs) are computers that can perform multiple,
complex digital operations within seconds. Sometimes called supercomputers, HPCs
are actually a wide range of technologies that also include bundled workstations,
mainframe computers, advanced microprocessors and software.1 A benchmark used
for gauging HPC computing performance is to count the millions of theoretical
operations per second, or MTOPS, that the computer can perform. The actual
MTOPS a computer can perform over time can vary, based on the operations
1 A supercomputer is usually defined as a single, complex, mainframe computer that can
undertake a series of specific computer functions. Michael S. Malone, ed., “Big Iron;
Supercomputers are Back and Changing Business, Science, and Even You,” Forbes ASAP.
February 22, 1999. 96 pages.

CRS-3
performed (some can take longer than others or can be performed while other
operations take place) and the real time cycle of the computer. In 2004, most new
desktop computers, available as off-the-shelf commercial products, ranged in
computing power from 6,000 to 12,000 MTOPS.2
HPC technologies have removed many of the technological restraints in
advanced computing by reducing long computing times and complex functions that
hindered solving mathematical, scientific, and engineering problems.3 The executive
branch has recognized that HPCs are critical in a variety of defense and security-
related areas, including nuclear and conventional weapons programs, encryption, and
military operations.4 Continual increases in the computing power of HPCs, the
extent of foreign availability of models comparable to some of those produced in the
United States, the adequacy of relying on high computing power alone as a basis for
determining the potential ability of HPCs to fulfill specific user goals, and the degree
to which foreign use of HPCs in ways that adversely affect U.S. interests can be
accurately predicted and successfully monitored, are among the factors in the export
control debate.
High performance computing can be performed in a number of ways. The scope
and effect of HPC technologies depends on several factors — how the computing
capability is structured, how the computing capability is used, and what is the desired
result of the computing capability. The first high-speed computers developed in the
late 1950s and the early 1960s, such as the IBM 704 and the IBM STRETCH, were
single computers performing a series of sequential, arithmetic functions (one
operation per one instruction in a sequential order). In the late 1960s and into the
1970s, improvements in computing memory capacity and speed led to developments
in which repetitive operations could be undertaken per instruction. An HPC which
performs one operation per one instruction uses scalar speed; an HPC which
performs repetitive operations per single instruction uses vector speed. The key
technology development was that computing capability was contained in a single
machine, or what is commonly referred to as a supercomputer.5
In the 1960s, U.S. military researchers began employing different combinations
of computers to achieve high-performance computing levels. By the 1990s,
commercial users began to more commonly link several computers with multiple
2 The growing complexity and sophistication of off-the-shelf commercial computing
products also means that pure computing power is not the only benchmark for consumer
preference; network capabilities and network interface is also an important consideration
for consumers.
3 See GAO Report GAO/NSIAD-98-196, Export Controls: Information on the Decision to
Revise High Performance Computer Controls
(September 1998), and GAO Report
GAO/NSIAD-98-200, Export Controls: National Security Issues and Foreign Availability
of High Performance Computers (
September 1998).
4 See GAO Report GAO/NSIAD-98-200, pp. 2-4; GAO Report GAO/NSIAD-98-196, at 7-
8.
5 It is important to note that scalar processing is not “better” than vector processing. The
computing application, the number of processors needed, the time, cost, and other factors
all play a role in how HPCs are used.

CRS-4
processing capability together, to alleviate delays that may occur when highly
complex data calculations reach the level of billions of operations. This permits
computations to be undertaken simultaneously, rather than sequentially, and is called
parallel processing. The most common form of parallel processing is clustering,
in which independent computers are linked, or clustered, together by a single system
of software, hardware, and network technologies. Clusters of parallel processors have
dramatically driven down the cost while increasing the scale, scope and capacity of
computing systems. Computing grids and Internet computing are extensions of
clustering — networks of computers, which may be physically separated by great
distances, but are linked (sometimes globally) together to work as a single unit.6
There are many architectures, or combinations, of computers and processing
capability, that employ parallel processing through clusters, grids, and the Internet.
Two factors that affect HPC architecture are the complexity of the operation being
performed and the response time needed for a solution. The response time may need
to be measured in microseconds (such as in a space launch) or in minutes, hours or
days. Response time determines how many computers should be used and at what
speed. Are benchmarks required at certain points of the calculation, requiring one
processor to continue to work while others remain dormant? Does the calculation
require a pre-production or post-production run to verify results?7 Regardless of the
factors considered and results sought, parallel processing requires multiple
computing processing, usually housed in multiple computers.
These advances in scalar, vector, and parallel processing have given users many
options. They have removed many of the technical constraints in advanced
computing by reducing long computing times and complex computing functions that
hindered solving mathematical, scientific, and engineering problems. Recent HPC
applications range from accurate real-time weather forecasting and climate change
modeling to simulations of nuclear weapons tests. For example, vector processing
is an integral part of Doppler radar, a computer imaging system that provides real
time weather patterns with extreme accuracy.
Military Uses of HPCs. High performance computers are used in many
military situations including weapons design, testing, battlefield modeling, and
command, control, coordination and intelligence activities (C3I). (See Appendix
Table 1 for computer performance levels used for specific military applications.)
Many experts have noted that it is possible to design nuclear weapons with
current commercially available computers. (Indeed, the first atomic weapons were
designed with slide rules). Nonetheless, it is pointed out that highly advanced
computers such as those used in nuclear weapons laboratories are necessary to
6 Nance, Scott. Clusters, Grids Moving Quickly From Lab to Market. New Technology
Week. 23 July 2001. P. 1; 4.
7 Before a long calculation is executed, it is often submitted for a shorter run to determine
if all of the inputs and factors are correct (to avoid wasting time and money). In addition,
after a long production run is made, it is often necessary to analyze the results in additional
calculations that last minutes to hours in time. National Research Council, Supercomputers:
Directions in Technology and Applications.
Washington: D.C. 1989: 32-33.

CRS-5
produce successive generations of such weapons with greater sophistication and
capabilities.8 According to one proliferation expert, “acquiring supercomputers could
assist China in building smaller and more efficient nuclear weapons-improvements
in order to upgrade its strategic rocket forces to include multiple independent reentry
vehicles (MIRV) systems- and for cruise missiles.”9
Supercomputing capabilities can be used to simulate nuclear explosions.
However, these simulations rely on extensive data and codes obtained in part from
actual tests. Relatively few countries have tested nuclear weapons, and this data is
highly guarded classified material. Therefore, computing capacity is not enough to
successfully simulate nuclear tests. However, for countries that have conducted tests
and presumably have this data, advanced computing capability can reduce or
eliminate the number of tests needed. Not only does this contribute to greater
weapons capability, it also allows that country to conceal progress towards the
development of such weaponry.10
Battlefield management encompasses the components of command, control,
communications/computers, intelligence (C3I), and reconnaissance. In the networked
battlefield, management applications collect data from many sources simultaneously,
favoring the input-output capabilities of a mainframe HPC. The software codes used
to create battlefield management applications likewise need HPCs, but they involve
computational aspects in which clustered computers can be employed.11
Commercially available laptops such as are manufactured in numerous countries can
run these operations in the field.
U.S. Government and Industry Roles
It is important to note that the federal government has had an important role in
the development and growth of computer technology, especially HPC technology.
In the early 1950s, federal funding for computer science supported developments in
several fields: electronic components, magnetic storage, computer architectures, and
software.12 Many federally-sponsored R&D programs that supported early computer
science and technology research were undertaken for military and national security
purposes. The first two electronic digital computers, the Colossus and the ENIAC,
were developed as a consequence of the Allied war effort in the 1940s and are the
8 CSIS Commission of Technology Security in the 21st Century (CSIS), “Computer Exports
and National Security in a Global Era
,” May 2001, p. 12.
9 Bryen, Stephen D. “The Future of Strategic Export Controls,” CSIS-Stimson Working
Paper #2, April 2001, p. 22.
10 Ibid, p. 23.
11 CSIS, p. 12.
12 Flamm, Kenneth. Creating The Computer: Government, Industry, and High Technology.
The Brookings Institution. 1985: 8-28.

CRS-6
origins of the U.S. commercial computer industry.13 The Defense Advanced Projects
Research Agency (DARPA), a part of the Department of Defense that supports long-
term, high-risk, potentially high-payoff R&D, also has been instrumental in
technology developments ranging from the computer mouse to flat panel display
technology. The Internet was created as a network in which university researchers
undertaking contract work for DOD could easily exchange data. The U.S. Navy, Air
Force, and other parts of the Armed Services also provided significant funding for
early computer science research, and the National Science Foundation and what was
formerly the National Bureau of Standards have played a significant role in
supporting targeted, non-military computer science.14
As computers and computer technologies became more complex — and as
global competition for making these machines became more intense — policymakers
sought to address perceived shortcomings in technologies and applications. In 1989,
representatives from the federal government, U.S. industry, and academia called for
policymakers to address the “grand challenge” of supporting new computing and
networking applications. In FY1991, the 102nd Congress authorized the High
Performance Computing Initiative as part of a five year effort to advance high
performance computing in conjunction with the mission of federal agencies. While
authorization of this program ended at the end of FY1995, appropriations for federal
high performance computing research and development has continued. In FY2000,
support for federal HPC programs totaled $1.8 billion.15
While U.S. HPC manufacturers continue to provide technological innovations,
global competition is intense. Today’s global market leaders include IBM and Sun
Microsystems/Cray in the United States, and Japan’s NEC. Beyond the United States
and Japan, many other industrialized nations have developed high performance
capabilities, although they lag performance benchmarks of the industry leaders.
Computer software, networking, semiconductor chip manufacturing, and other
segments of the computer industry also have benefitted from, and contributed to,
HPC advances.16
HPCs and Export Policy Revisions
Before HPC technology advances in the 1990s, HPC export controls were fairly
simple. There were only a few supercomputers in the world; they were almost all
13 While the Colossus was built in England during 1942-43, the effort was considered a
joint U.S.-British accomplishment. ENIAC stands for Electronic Numerical Integrator and
Calculator. Ibid., 46-47.
14 Ibid., 29-46. The National Bureau of Standards became the National Institute of Standards
and Technology in 1995.
15 For more on the origins and history of this program, see CRS Report 97-31, Computing,
Information, and Communications R&D: Issues in High-Performance Computing
, by Glenn
J. McLoughlin.
16 Computer Industry Almanac, “Top Computer Companies,” [http://www.c-i-
a.com/statistics/prz].

CRS-7
made by U.S. firms; and computing capability ranged in the low 100 MTOPS range.
U.S. export control policy focused on controlling a single set of technologies
incorporated into specific products.
Multilateral Regimes
The Coordinating Committee (CoCom). CoCom was originally
established in 1949 to restrict the transfer of technology to the Soviet Union and its
satellite states. The organization, headquartered in Paris, met regularly to determine
which goods could be sent to the Eastern Bloc. CoCom was characterized by a
coordinated list of goods and technology, the ability to prevent the export of goods
from a member state, and frequent consultations. Its impact was realized after the
collapse of the Eastern bloc, when the woeful state of Soviet-bloc technology was
confirmed. CoCom did not long survive the end of the Cold War, as the lack of
agreement on control destinations contributed to its demise in 1994.
U.S. Japan Supercomputer Agreement. The first multilateral attempt
explicitly to control the export of supercomputers for non-proliferation purposes was
the U.S.-Japan Supercomputer Agreement of 1984. Modifications of the agreement
were adopted in 1986, 1990, and 1994. At that time, the two major supercomputer
manufacturing nations agreed to coordinate the sale of these machines to third
parties. The agreement supplemented CoCom by controlling transfers to non-
Communist countries. In 1993, the U.S. unilaterally liberalized its export controls on
supercomputers (500 MTOPS for CoCom; 2,000 for the bilateral agreement). The
Japanese protested this decision and only raised controls to 1,500 MTOPS. The
controls prohibited sales to end-users known to be engaged in weapons development,
proscribed activities, or known to be terrorist entities.
Wassenaar Arrangement. In the mid-1990s, an international agreement,
called the Wassenaar Arrangement (WA), created a multilateral framework to
control conventional arms and dual-use technology, including HPCs. The WA
resulted from negotiations culminating in July 1996 to replace the Coordinating
Committee (CoCom). The WA is a voluntary arrangement that stresses transparency,
consultation, and national restraint. WA members report on the transfers of dual-use
goods semi-annually. Sensitive List transfers are also reported semi-annually,
however, license denials are reported to the group within 60 days. WA members also
consult on the security implications of transfers of conventional arms and dual-use
goods. The WA conventional arms and dual-use lists are maintained by the
participating states and are revised as needed.
The decision to license any good or weapon remains with the Member state
based on that state’s national export control laws. Unlike CoCom, national decisions
cannot be overturned or considered by the other Member states. Neither are there
prohibitions to prevent an “undercut”, the act of selling an exact or similar good after
another country has denied such a sale. WA is not targeted at any nation, although
an informal consensus has developed against sending arms or sensitive technologies
to certain states. In December 2005, the Wassenaar Arrangement approved a new
standard for calculating computing power. This new standard, called adjusted peak
performance (APP), is the “adjusted peak rate at which digital computers perform 64-
bit or larger floating point additions and multiplications,” and is measured by a

CRS-8
metric know as “weighted teraflops”(TFLOPS).17 This standard, reportedly devised
with the cooperation of U.S. government and industry would need to be incorporated
into the U.S. Export Administration Regulations and may also need Congressional
approval because of the MTOPS reporting requirements in current law (see “The
National Defense Authorization Act of 1998 (NDAA98),” below).
U.S. Domestic Policy and Regulations
As both the computing power and applications of HPCs grew, so did concerns
about adjusting export controls and changing MTOPS benchmarks. In 1992, the U.S.
Commerce Department defined an HPC as 195 MTOPS; any export above this level
required an export license. This definition was revised in 1994 (1,500 MTOPS),18
to reflect new HPC technologies and expanding applications. It is important to note
that at that time, HPCs were still almost exclusively stand-alone machines.
MTOPS and Country Tiers. At the end of 1995, a study on high
performance computing and export control policy was published by the Center for
International Security and Cooperation (CISAC) and Stanford University. This
study, “Building on the Basics: An Examination of High-Performance Computing,
Export Control Policy in the 1990s” was commissioned by the Departments of
Commerce and Defense to address concerns about HPCs and export control policy.
The report analyzed then-current technology trends, applications, and export control
policy. It concluded that technological advances in computing, diffusion of HPC
technology worldwide, new applications of HPCs, and changing U.S. national
security policy in a post-Cold War era effectively made MTOPS a poor export
control benchmark. The study’s authors recommended that in the short-term, new
upper and lower export control thresholds should be based on militarily important
applications, and recognize the inability to control specific technologies.19
The current licensing process for HPCs was established by Commerce
Department rulemakings of January 25, 1996 and March 25, 1996. The former
implemented the policy of computer export reforms announced by President Clinton
on October 5, 1995; the latter implemented the reorganization of the Export
Administration Regulations (EAR) broadly, but it also announced the policy of
License Exceptions for computer exports. A License Exception is an authorization
contained in the EAR that permits exports of an item that otherwise would require
a License under certain circumstances. An exporter invoking a License Exception
certifies that all licensing criteria have been met on its export documentation.
17 “Wassenaar Arrangement Technical Note on ‘Adjusted Peak Performance’” in The Export
Practicioner
, January 2006, p. 22. Also see this report, p. 16.
18 The exception was the U.S.-Japan Supercomputer Agreement, which set the threshold at
2,000 MTOPS; however, as noted on the previous page, Japan protested the unilateral nature
of this decision and only raised controls to 1,500.
19 See Building on the Basics: An Examination of High-Performance Computing Export
Control Policy in the 1990’s
, Seymour Goodman, Peter Wolcott, and Grey Burkhart
(November 1995).

CRS-9
These regulations created the License Exception CTP under which computers
are exported today. Previously, all computers over the prevailing MTOPS threshold
required a license based on destination or type of transaction. The January 25 rule
implemented a country tiering system for computer licensing. Under the March 25
rule, a license exception could be granted based on the destination. The power or
MTOPS level eligible for a destination would be determined by the tier in which that
country was placed. The tier in which a country was placed was determined by its
adherence to proliferation control regimes, its relations with the United States, and
the potential for transhipment or diversion to particular uses (See Appendix 2).
Tier 1 countries (Canada, Western Europe, Australia, Mexico, Japan, and New
Zealand), categorized as those countries belonging to and adhering to proliferation
control regimes, would require no license for HPC exports, while Tier 2 countries
(South America, ASEAN nations, several former Eastern European bloc nations and
South Africa) had a 10,000 MTOPS threshold and reporting requirements for U.S.
sales above that level. Tier 2 represented countries in which the risk of proliferation
was relatively minimal, although greater than Tier 1.
Tier 3 countries represented countries in which there was a risk of proliferation
or diversion, either by the countries themselves or by virtue of the regions they
inhabited. These countries- China, Russia and other countries of the CIS, India, and
Pakistan- became subject to a dual control system. This system distinguishes
between civilian and military end-users and end-uses. Export licenses would be
required for HPCs above 2,000 MTOPS for military and weapons end-users and
end-uses, and above 7,000 MTOPS for any recipient or for any use. Licenses for
items in the former category would be considered on a case-by-case basis; licenses
for those in the latter would generally be approved. This end-user differentiation
ended in 2000.
Licenses were required to Tier 4 countries (Cuba, Iran, Iraq, Libya, North Korea,
Sudan, and Syria) which operationally presupposed a presumption of denial. Since
the promulgation of these regulations, the country tiers have not remained static.
Countries have been moved between tiers to reflect different geopolitical and
proliferation circumstances. MTOPS levels and destination criteria have also been
adjusted for each tier, recognizing the changes in computing power.
Under the 1996 policy change, civilian consumers in Russia, other CIS
countries, and China could buy high-performance computers without going through
the rigorous U.S. export licensing process if, for example, the computers were not to
be used at a nuclear weapons facility. After several refusals by the United States to
approve exports to Russia of computers for nuclear-related purposes, Russia
reportedly obtained 16 IBM computers through evasion of U.S. export license
requirements and installed the computers at Arzamaz-16 (now Sarov), the site where
the former Soviet Union had designed its hydrogen bomb and Russia conducts
nuclear weapons research. The United States originally insisted that the computers
be returned, but instead, after two years of negotiations, Russia agreed to remove the
computers from the nuclear weapons facility and transfer them to a new commercial
computing center which opened in Sarov in October 1999.

CRS-10
The Cox Committee. Policymakers also were concerned about exports of
HPCs to the People’s Republic of China. Reportedly, end-user and verification
problems have arisen with regard to China, since many state owned enterprises
(SOEs) have close ties to the Chinese military.20 For example, an executive branch
review reportedly discovered an HPC built by Sun Microsystems in the possession
of the Changsha Institute of Science and Technology, a weapons research facility run
by China’s People’s Liberation Army. The computer was allegedly sold to the
Institute in 1997 by one of Sun’s Hong Kong distributors without an export license.
The computer was returned to the United States in November of that year. In
addition, until 1998 China had not allowed the United States to conduct post-
shipment checks to establish end-use control; even since that time, some contend that
post-shipment verification has been erratic and incomplete.
In 1998, the House of Representatives created the Select Committee on U.S.
National Security and Military/Commercial Concerns with the People’s Republic of
China (PRC). Chaired by Representative Christopher Cox, the “Cox Committee” was
created in response to allegations that two U.S. satellite companies (Loral and
Hughes) might have transferred technology to China in the course of launching
satellites on Chinese launch vehicles. However, the committee’s mandate was
broader, and included investigations of other instances where illicit technology
transfer may have occurred — including HPCs.21 The public, declassified version of
the Committee’s report was released on May 25, 1999 (the Committee’s final report
remains classified).
The Cox Committee report was considered important (if controversial) by U.S.
policymakers, industry leaders and others because it not only assessed the amount of
HPC technology transferred to the PRC, but painted in larger brush strokes what
many consider to be the inefficacy of current U.S. HPC export control policy to Tier
3 countries. For some, the examples of HPC technology transfer to the PRC provide
concrete examples of why U.S. HPC export control policy needed substantial
revision.

The Cox Committee determined that U.S. HPC export policy has been
circumvented by PRC end users, has not been properly monitored or enforced by
U.S. officials, and that U.S. industry generally had been unaware of PRC applications
of HPCs. The major Cox Committee report findings on HPCs were: that the number
of HPCs exported to China had increased and had provided PRC end-users with
many different combinations of computing power and speed; that PRC end users
were clustering lower-end HPCs together to increase computing power and speed;
that the distinction between PRC private companies and SOEs was blurred, affecting
end-use verification and resulting in HPCs being used for military purposes and
contributing to proliferation; that PRC students visiting U.S. federal laboratories and
20 CRS Report 98-617, Technology, Trade and Security Issues Between the United States
and the People’s Republic of China: A Trip Report, August 1997,
by Glenn J. McLoughlin.
21 See archived CRS Report RL30231, Technology Transfer to China: An Overview of the
Cox Committee Investigation Regarding Satellites, Computers, and DOE Laboratory
Management
, by Marcia S. Smith, Glenn J. McLoughlin, and William C. Bowsman,
available from author.

CRS-11
universities with HPC technologies may act on behalf of the Chinese intelligence
organizations, further blurring civilian, military, and academic lines; that the U.S.
Government’s ability to verify the location and use of HPCs in the PRC was blocked
by that country’s resistance to post-shipment, on-site verification visits; and that
these findings raised significant security issues for U.S. policymakers in formulating
export control policy and U.S.-PRC relations.22
The Cox Committee report offered four policy recommendations.
! Legislation to require testing of computers proposed for export, and
which may be potentially used for clustering and other combinations
of HPCs.
! An annual comprehensive threat assessment of HPC exports to the
PRC by the U.S. intelligence community.
! Legislation to require transparent end-use verification of PRC use of
HPCs; failure by the PRC to cooperate would result in actions by the
United States to lower the benchmark levels of HPCs sold to the
PRC, to deny certain export licenses for computers to the PRC, and
other related measures.
! Legislation to require the Executive branch to encourage other
computer-manufacturing countries, especially those countries that
manufacture HPCs, to adopt similar export policies towards the
PRC.
Shortly after the release of the report, the Clinton Administration offered a
response. The Administration agreed with the Cox Committee report that sales of
computers to the PRC should be for commercial, not military, purposes. The
Administration also stated that it was reviewing the potential national security uses
of various configurations of computers, the extent to which these computers can be
controlled, and the impact of controls on the U.S. industrial base. The
Administration agreed with the Cox Committee assessment that the United States
needs the capability to visit U.S. HPCs licensed for export to China and observe how
they are being used (although the Clinton Administration contended that it was not
possible to obtain no-notice verification visits to any country, including the PRC).
On this last point, the Administration negotiated an agreement with the PRC for
increased site visitations in 1998. However, the Clinton Administration continued
to contend that requiring the U.S. to visit every HPC site, regardless of what business
the end-user was in or how many times it was visited before, would be ineffective
and wasteful.23
22 Ibid.
23 Statement by Under Secretary for Export Administration, “Commerce Report: Growing
Demand for U.S. High Performance Computers.” Washington: U.S. Department of
Commerce. 8 January 1999.

CRS-12
The National Defense Authorization Act of 1998 (NDAA98). The106th
Congress also responded in part to the Cox Committee report when it passed the
National Defense Authorization Act for FY1998. Congress tightened Tier 3 controls
on HPC exports, and enacted new requirements for advance notification and post-
shipment verification of these items. The statute requires exporters to notify the
Commerce Department in advance of a proposed export or reexport of an HPC
greater than 2,000 MTOPS to a Tier 3 country and prohibited the export, or reexport,
of an HPC without a license if the Secretary of Commerce, Defense, State, or Energy
objected.24 The objections would have to be raised within 10 days after the agency
received the notification. The President would be authorized to raise the MTOPS
level for notification purposes, but the new threshold would not go into effect until
six months after the President justified it in a written report to Congress. Similarly,
the President had the authority to remove countries from Tier 3 (with some
exceptions) but was required to notify Congress 120 days in advance. Congress
reduced the notification period from 120 to 60 days in October 2000.25
NDAA98 also requires post-shipment verification of exports of computers of
more than 2,000 MTOPS to Tier 3 countries, though the provision will not apply to
a country moved out of Tier 3 under the authority described above. To facilitate
verification of sensitive items, the United States signed an agreement with China in
1998 setting forth terms for post-shipment visits to verify end-use of high technology
exports generally. In October 1999, Congress also directed the President to seek an
agreement with the PRC to revise the existing verification system as it applies to
NDAA-covered HPCs “so as to provide for an open and transparent system providing
for effective end-use verification for such computers.”26 It also relaxed the NDAA’s
verification requirement by making any increased MTOPS level established for
purposes of advanced notification applicable to the former requirement as well.27 On
January 24, 2002, Assistant Secretary of Commerce for export enforcement Michael
J. Garcia testified that the backlog for end-use inspections stood at approximately
700. He maintained that this number includes inspections for computers that are no
longer advanced enough to be controlled. However, he testified that access to
advanced computers with potential military applications has been denied by Chinese
authorities.28 In May 2003, the House defeated an amendment to the 2004 NDAA
that would have repealed the NDAA98 provisions. The amendment was defeated by
a vote of 207-217.
Further Decontrols: 1999-2003. In July 1999, the Clinton Administration
announced that it was notifying Congress of its intent to relax more controls on HPC
exports, including raising the export license threshold for Tier 3 countries from 7,000
to 12,300 MTOPS for civilian end-users and from 2,000 to 6,500 MTOPS for
military end-users. In addition, the President notified Congress July 26, 1999 that the
24 National Defense Authorization Act, FY1998, P.L. 105-85, § 1211.
25 National Defense Authorization Act, FY2001, P.L. 106-398, §1234.
26 NDAA, FY2000, P.L. 106-65, § 1407(a)-(b).
27 NDAA, FY1998, § 1213(e), as added by NDAA, FY2000, P.L. 106-65, § 1407(c).
28 “U.S. Unable to Monitor Computer Exports to China as Required by Congress, Aide
Says,” International Trade Reporter, January 24, 2002.

CRS-13
advance notification level would be raised from 2,000 to 6,500 MTOPS. On
February 1, 2000, at the conclusion of the six month notification period, the Clinton
Administration announced that these changes, the fourth since 1993, would be
enforced.
On February 1, 2000, the Clinton Administration raised the Tier 2 levels from
20,000 to 30,000 MTOPS and raised the notification level to Tier 3 countries from
6,500 MTOPS to 12,500 MTOPS.29 These changes went into effect on August 14,
2000. On August 30, 2000 the Clinton Administration notified Congress of
additional export control changes. The license exception limit for Tier 2 countries
was raised from 33,000 to 45,000 MTOPS; and the license exception limit for Tier
3 countries was raised from 20,000 to 28,000 MTOPS.
In January 2001, President Clinton adjusted MTOPS levels by Presidential
directive and merged country tiers 1-2. The President stated his support for ending
high-performance computer export controls, citing the lack of a meaningful or
relevant export control standard. However, he maintained the controls due to the
statutory requirements of NDAA and recommended that the incoming Administration
seek repeal of the NDAA mandates.30 Currently, a License Exception is available
to all Tier 1 countries for computers over 6,500 MTOPS. In addition, the WA
notification requirements have ended for Tier 1 countries. A Tier 3 country is eligible
for a license exception for computers with performances ranging from 6,500-85,000
MTOPS. As of this date, exports of computers with performance capabilities of over
85,000 MTOPS continue to need licenses, as do all potential sales to Tier 4
countries.
The Bush Administration implemented its first change in HPC export control
policy on January 2, 2002. It raised the license exception threshold from 6,500 to
12,000 MTOPS. Tier I countries may receive a license exception for any processors
over 12,000 MTOPS. The Administration also raised the licensing threshold from
85,000 MTOPS to 190,000 MTOPS. Thus, exporters to Tier III countries may obtain
a license exception for computers with a CTP of 12,000-190,000 MTOPS. A license
will still be required to ship computers with a CTP greater than 190,000 MTOPS to
Tier III countries. This change took effect on March 5, 2002. It has been reported that
these changes were prompted by the introduction of the Pentium Itanium Chip. A
single Itanium chip has the CTP of 6,500 MTOPS, while a popular server
configuration with 32 processors would have a CTP of 190,000 MTOPS. The new
licensing requirement would allow sales of this server configuration to Tier III
countries.31 In addition, Latvia was moved from a Tier III to a Tier I country, a
change that will become effective180 days from January 2.
29 Other changes included raising individual license levels for civilian end users in Tier 3
countries from 12,300 to 20,000 MTOPS and moving Romania from Tier 3 to Tier 2. See;
[http://www.bxa.doc.gov/press/2001].
30 “Export Controls on High Performance Computers,” White House Fact Sheet, January
10, 2001.
31 “White House Eases Computer Export Controls to China, Others,”Inside U.S. Trade,
January 4, 2002.

CRS-14
On January 14, 2003, the Bush Administration announced a general decontrol
of general purpose microprocessors to bring the EAR in conformance with the
February 2002 revisions of the Wassenaar Arrangement’s List of Dual Use Goods
and Technologies.32 These microprocessors are widely produced and are used for
numerous civilian applications including cellular phones, personal digital assistants,
and personal computers. However, they are also used for several military
applications such as navigation and guidance systems, communications and
information warfare systems. Based on the new regulations, microprocessors with a
CTP of over 6,500 MTOPS will only be controlled to designated terrorist supporting
countries.33 The rulemaking also establishes a license requirement for
microprocessors exported for military end-use or to military end-users in countries
such as China, Russia and newly independent states of the former Soviet Union. This
latter requirement does not require licenses for export to civilian end-use or end-users
in these countries. These regulatory changes refer only to microprocessors and,
hence, do not affect the HPC export control policy detailed above.
MTOPS: Technology and Policy Issues
Policymakers have had to face a series of technology-related policy decisions
when addressing the export policy of HPCs. The first is whether MTOPS is an
adequate and satisfactory benchmark for controlling the export of HPCs. This is a
significant policy question because if the MTOPS benchmark is removed in favor of
another technology benchmark, then a series of other policy, procedural, and
regulatory changes likely would have to be implemented as well. These would
include revising statutory authority for export control, and it is likely that export
control policy and how that policy is made, would be substantially changed as well.
MTOPS. If the United States continues to use MTOPS as a measurement for
HPC export control, it is likely that critics will continue to dispute its effectiveness.
For these critics, retaining MTOPS benchmarks (and raising export control threshold
levels when technological advances warrant) would not address the growing use of
computer clusters outside of the United States; would not differentiate among end-
users and whether these users had important ancillary technologies, such as software
and networks; would not address the applications of computers by nations and
entities not friendly to the United States; would not address the potential loss of HPC
sales, even at high MTOPS licensing thresholds, for U.S. firms; and it would not
necessarily address concerns of end-use and on-site verification raised in the Cox
Committee report and elsewhere.
Even the criteria used to determine the level of MTOPS control has been a
subject of controversy. In order to change the NDAA98 notification requirements,
the President must report to Congress on (1) the foreign availability of computers
32 “Revision of Export Controls for General Purpose Microprocessors”, 68 Federal Register
1796, January 14, 2003.
33 The control category AT (anti-terrorism) covers North Korea, Sudan, and Syria.
However, comprehensive controls on Cuba, Iran, Iraq, and Libya require licenses (with a
presumption of denial) for virtually all goods on or subject to the EAR.

CRS-15
with capabilities up to the proposed level; (2) the potential military uses that HPCs
at the new level can provide; and (3) the impact of uses on the national security. The
General Accounting Office was asked on two occasions to examine the methodology
used in the Administration’s preparation of these reports. GAO found that the
Administration had not clearly articulated specific national security interests to be
protected at various performance levels. It did not determine what military critical
applications to control or the most effective method to achieve those controls.34 The
GAO also reported that the Administration based decontrol decisions on the
availability of computing power at various levels rather than on national security
assessments. In fact, ‘widely available’ and ‘controllable’ were not (and are not)
defined in current law or regulation. Although these criticisms were directed at
actions of the Clinton Administration, the Bush administration may also need to
grapple with the issue of determining what level of technology can be controlled.
Eliminating MTOPS. Three recent reports have provided policymakers with
an extensive analysis of HPCs and export controls, as well as varying
recommendations and strategies for addressing policy questions. They are: the
General Accounting Office (GAO) report System for Controlling Exports of High
Performance Computing is Ineffective,
December 2000; the Center for Strategic and
International Studies (CSIS) report Computer Exports and National Security in a
Global Era: New Tools for a New Century,
April 24, 2001; and the Defense Science
Board (DSB) report Export Control of High Performance Computing: Analysis and
Alternative Strategies,
February 2, 2001.
The GAO, the CSIS, and the DSB all call for eliminating MTOPS as a
technology benchmark for HPC export control policy. The GAO contends that
MTOPS is outdated as a valid measurement of computing speed because it is tied to
the clock rate of the computer’s processing power, which may vary depending on the
computer application. The GAO further contends that MTOPS benchmarks do not
take into consideration other factors, such as memory retrieval times, interconnection
methods, and internal bus speeds of computers, all of which are more directly tied to
computing power and speed. The CSIS report raised similar objections to MTOPS
as a benchmark for export control policy. The CSIS report states that MTOPS is a
static measurement which does not work well when measuring computing power
from clusters, grids, or Internet-connected computers. The DSB report recommends
that current software export controls are more functional and reliable than MTOPS
as a way to restrict technology transfer to Tier 3 countries.35
Increasing computing speeds combined with networking advances have blurred
the distinction between supercomputers and mass market computers.
Microprocessors that individually comply with export regulations can be linked
together to create servers with MTOPS capabilities that breach export thresholds. If
enough processors are linked together, they can create a parallel processing system
with capabilities that approach those of a super-computer. The Defense Science
34 Harold J. Johnson, GAO, Testimony before the Senate Government Affairs Committee,
May 26,2000, p.6.
35 Defense Science Board, Report on High Performance Computing: Analysis and
Alternative Strategies.
Washington, D.C.: February 2, 2001. P. 6.

CRS-16
Board notes in its final report on Globalization and Security that the ability to cluster
commodity computers in order to multiply computing power erodes the ability to
restrict access to high-performance computing, even if high-performance stand-alone
machines can be controlled.36

For some proponents, while the MTOPS benchmark is not perfect, in the
absence of any viable alternatives, it brings a continued level of stability and
continuity to export control policy. Supporters contend that the HPC benchmark is
flexible enough to change as technological advances in computing occur. They
consider MTOPS a standard that, regardless of the type of computer, provides a
common technology classification with which all stakeholders can work. In the
absence of any clear technology alternative, some contend that an imperfect
benchmark is better than no benchmark.
Teraflops. Still, as computing speed gets faster, and the boundaries of high
performance computing have been broken by SGI, IBM, and NEC supercomputers,
some have advocated using a new metric for HPC export controls. The current
metric for measuring the fastest computing speed is teraflops, or trillion floating
point calculations per second. Over the last year, NEC’s record of computing speed
(41 teraflops) has been superseded by SGI (62 teraflops) and in January 2005 by IBM
(92 teraflops).37 As the boundaries of supercomputing speed are raised, some
contend that MTOPS, which now apply to more commercially available technologies,
should be completely eliminated for any barrier or export control. In its place,
teraflop computing speed could be the new threshold for export control.
Advocates contend that since many believe that controlling MTOP technology
is impractical, using a higher threshold for computing speed will keep advanced
technologies from falling into the hands of those who may use the technology against
the United States. The SGI supercomputer that set a speed and performance record
of 61 teraflops was developed for NASA; other federal agencies, particularly those
connected with defense and homeland security may find a need to both use and limit
the availability of technologies with teraflop computing speed outside of the United
States. In addition, since there are now relatively few supercomputers with teraflop
computing speed, exports of this technology may be easier to manage and control.
Critics contend that shifting the metric from MTOPS to teraflops merely
substitutes one kind of relative computing speed for another, and does not address the
larger issue of applying rapidly changing innovative technologies in a global
economy. Critics also believe that because teraflop computing technology is
relatively rare, everyone knows who has the technology; controls are not necessary.
In addition, supercomputers with teraflop computing speed tend to be very large
machines — “big iron.” Therefore, it is very unlikely that an entire supercomputer
with teraflop technology could be exported and properly used by a nation if the
recipient country does not have the requisite infrastructure, networks, and relatively
36 Defense Science Board, Final Report of Task Force on Globalization and Security,
Washington: Office of the Under Secretary of Defense for Acquisition and Technology,
December 1999, p. 27.
37 Otis Port. Holy Screaming Teraflops. BusinessWeek. January 17, 2005. P. 62.

CRS-17
well-educated workforce. Still, federal policymakers are examining this issue, and
considering teraflop computing speed as part of the overall export control debate.38
Other Alternatives to MTOPS. However, while both GAO and DSB
provide several technology options for replacing MTOPS, the CSIS contends that
there are no technology alternatives, because HPCs are no longer primarily single,
stand-alone products. According to the CSIS, “Neither MTOPS nor any other
parameter constitutes an adequate measure of system performance.”39 The CSIS calls
for the elimination of MTOPS, and with it any dual-use export control policy for
HPCs left over from the Cold War.
Among other technology options, however, two have been considered by the
GAO and others. The first is multiprocessing technology. Semiconductor chips are
made on thin wafers of material, usually silicon, in which tiny lines (linewidths) are
etched that in part determine operation, speed and memory capacity. Multiprocessors
are integrated chips which can perform multiple functions at one time. Computers
can be measured by the type of multiprocessor it contains (e.g., Intel’s 386, 486 and
Pentium processors), and can be one indicator of computing capability. Should
processing capability replace MTOPS as a metric for export control policy? The
GAO has raised two concerns about using microprocessors as a benchmark for export
control — it may be no more accurate than using MTOPS as a measurement for
computing power; and there is no consensus within the computing research or
industry on a single definition of a processor. In addition, the GAO found that
exceptions to a procedure for counting processors would be needed to differentiate
between high performance single processors and computers that have unique
hardware and software designs.40
One of the arguments used to eliminate specific metrics, such as MTOPS, as a
benchmark for export control policy is that it is computing power, not computer
hardware, that is critical for any HPC applications. One substitute control mechanism
is to restrict exports of network applications and interconnections. Two options
currently are under consideration. The first is to restrict U.S. sales or transfers of
high-end network connections that might allow the clustering of HPCs in grids or by
the Internet. Under this approach, sales of computer switches, interface cards, fiber
optics, and other connective HPC infrastructure would be restricted or prohibited,
effectively curtailing clustering of HPCs. However, critics of this approach contend
that infrastructure technology is readily available from other countries, and is
relatively inexpensive. The GAO has reported that U.S. industry considers only
some network technology controllable at the high end of HPC applications, because
38 See also: Stephen Shankand. Two Records in One Day for SGI Supercomputer. C/Net
News. [http://news.com.com/SGI+supercomputer+Two+records+in+one+day/2100-7337_3-
5427876.html]. Updated as of February 10, 2005.
39 CSIS, P.18.
40 GAO, Export Controls: System for Controlling Exports of High Performance Computing
is Ineffective.
P. 28-29.

CRS-18
of the proprietary nature of the technology, and the importance of vendor support to
make the HPC applications effective.41
A second approach is to evaluate the current status of a foreign country’s
network infrastructure, and use this as a benchmark for determining whether HPC
exports to that nation should be allowed. The rationale behind this approach is that
a nation without high-end HPC connections for clustering will have limited use of
HPC technologies. This approach may also include an assessment of software and
bandwidth capabilities; if a nation does not have the infrastructure, software or
communications pipeline for clustering HPCs, then even when it obtains an HPC, its
applications are going to be limited. However, this option does not address the
possibility that a nation can improve its networking infrastructure and HPC
capabilities over time — or that obtaining HPC technologies may be the precursor
to a more advanced computing infrastructure in a nation not friendly to the United
States.
However, policymakers may decide that neither MTOPS nor any other
technology benchmark can adequately control the transfer of high performance and
advanced computers in a global economy. If this decision is made, then the possible
ramifications would have to be further examined and the questions it raises
addressed. How would countries be classified and would country tiers still be useful;
how would the United States counter development of sophisticated computing
capabilities for warfare by nations and entities not friendly to the United States; and
how would the United States control or otherwise restrict computing technologies
that industry may develop with federal research dollars?
National Security, Economic Security and
Federal R&D: Issues and Options

National Security. The lack of a clear alternative to MTOPS as an export
control mechanism does not necessarily mean that all HPC export controls need to
be lifted. Even as MTOPS export control levels are raised — and even if they are
completely removed — other policy issues may be addressed.
U.S. policymakers usually must balance national security concerns with
economic growth and technology development interests. There is an inherent tension
in balancing those policies which restrain technology transfer to entities not friendly
with the United States, yet encourage U.S. high technology exports to those markets
that have a strong demand for these products and applications. The CSIS report
states the dynamic as:
National security can be defined at two levels. Narrowly it depends on the quality
of U.S. weapons and the readiness of U.S. forces. This focused definition of
national security historically included restrictions on exports as a way of holding
back the technical progress of U.S. opponents. . . National security also can be
defined as the vitality and strength of the U.S. economy and the dynamism of
41 Ibid, P. 36.

CRS-19
American society. . . The development of computers and information technology
was crucial to both dimensions of national security.42
The advent of the U.S. high technology industry as we know it today — widely
available computers, software, networks, and many service and manufacturing
applications — began over thirty years ago.43 In part spurred by intense global
competition from European and Asian firms, and also spurred by leaps in technology
innovation, many U.S. high technology industries became dominant in the high-end,
high-performance computing sector. This coincided with a separate but also
important development — the dissolution of the Warsaw Treaty Organization, the
breakup of the Soviet Union, and the end of the Cold War. Therefore, as U.S. high
technology became predominant and industry sought to expand into global markets,
many old restrictions fell and new markets of competition emerged.
Widespread information technology development and innovation also meant
that computers and ancillary products became less the province of a few specialized
users and more of a commodity common to everyday life. For example, it has been
reported that one common 486 semiconductor chip has more computing power than
U.S. scientists had when they developed the atomic bomb; in another frequently cited
anecdote, the F-22 (which is the most advanced U.S. fighter plane) was designed
with a 958 MTOPS Cray supercomputer — approximately one quarter the power
now found in mass-produced Pentium multi-processor semiconductor chips.44
The proliferation of high technology as a commercial commodity, along with
quantum leaps in product capabilities and the change in who the U.S. views as
potential adversaries, has had a significant impact on U.S. export control policy. As
U.S. businesses push for increased freedom and expansion of its HPC sales,
policymakers have asked: who should we keep our technology from — and can we
accomplish that goal?
Economic Security. Some contend that in the post Cold War era, national
security is really national economic security. This implies that separating military
and economic interests is not only impractical but unrealistic, and that in the absence
of a single strategic military threat, technology transfer should flow more freely in a
global economy. Traditional national security concerns based on maintaining
advances and superiority solely in military technology may represent an iceberg; it
is the 10 percent that is visible above the water while 90 percent of the base rests on
an economy and society that has grown from commercial developments in the
computer industry.45 Some question whether the 10 percent which is visible should
constrain the 90 percent that fosters economic growth and development, especially
42 CSIS, p.10.
43 Creating the Computer: Government, Industry and High Technology, by Kenneth
Flamm, The Brookings Institution, 1988.
44 CSIS, p.11. The anecdote regarding the F-22 was reported in Defense Science Board,
Final Report of the Defense Science Board Task Force on Globalization and Security,
Washington, D.C. December 1999. P. i.
45 CSIS, p.10.

CRS-20
in an increasingly global economy where technology product competition transcends
national boundaries.
The argument that national security is enhanced by robust export industries is
predicated on the changing nature of defense procurement, research and
development. During the Cold War, the formative period of the current export control
regime, the military drove much technical research and provided funds for research
and development. Now that situation is largely reversed. The military relies to a
greater extent on commercial applications and now purchases many items. Industry
argues that it is in the national security to sell current technology to generate funds
to develop future technology. If American firms are competitively hindered because
of export controls, the argument goes, foreign firms will gain market share, increase
profits, invest more in R&D, and shrink and possibly surpass our technological lead.
Thus, industry argues it needs a streamlined export process that will not needlessly
impede exports.
Non-proliferation critics reject this argument. They maintain that the United
States does not promote its national security by selling advanced technology to
potential proliferators. Debating EAA on the Senate floor, Senator Kyl remarked,
“Nations which threaten our security interests should not be armed by the United
States. The fight against proliferation and rogue regimes must include some degree
of self-discipline within our own borders.”46 Advocates of restraint point to a CIA
report released in February that claims that China interprets its non-proliferation
commitments very narrowly,47 and that China recently has used supercomputers to
improve its nuclear weapons stockpile, sold fibre optic equipment to Iraq, and sent
missiles to Pakistan.48
The argument that a robust export industry is essential to maintain
competitiveness may have merit, although at present, the reliance of the United States
on China and other controlled destinations as a market for U.S. computer products
is minimal. In 2004, the United States exported $780 million in “computers and
components” as characterized by the Harmonized Tariff Schedule to China.49 This
figure represents 2.2% of total exports to China ($34.7 billion), 3% of total exports
of computers and components to all destinations ($24.0 billion), and less than 0.1%
of all U.S. exports ($731 billion). The total value of all license applications to China
for computer equipment, technology, and software was reported to be $35 million in
FY2004.50
46 Remarks of Senator Kyl, Congressional Record, S9098, September 5, 2001.
47 Director of Central Intelligence, “Unclassified Report to Congress on the Acquisition of
Technology Relating to Weapons of Mass Destruction and Advanced Conventional
Munitions, 1 July Through 31 December 2000,” [http://www.cia.gov/cia/reports/721_
reports/july_dec2000.htm].
48 Remarks of Senator Helms, Congressional Record, S9051, September 4, 2001.
49 HTS 8471. Figures from U.S. International Trade Commission Dataweb.
50 BIS, 2004 Annual Report, [http://www.bis.doc.gov/News/2005/04AnnualRept/AppF.htm].

CRS-21
Federal R&D. The federal government supports a great deal of HPC R&D
through the federal mission agencies. In FY2000, $1.8 billion was funded for high
performance computing and communications research and development, ranging
from advanced computer software technology algorithms, information infrastructure
and applications, education and research, and high performance systems
development. In the last category, federal programs under this initiative in scalable
and parallel computing systems have advanced the use of workstations and clustered
computers in the federal government.
Although to date policymakers have not linked the federal HPC initiative to
changes in U.S. export control policy, this issue may arise during future debates.
Some may ask whether the federal government should invest a substantial amount
of federal funding in HPC research at a time when export controls on these
technologies are being relaxed. Is there a potential that federally-funded advances
in HPC technologies and applications may be commercialized by the U.S. private
sector and then sold to nations or entities not friendly to the United States? Should
U.S. HPC export control policy take into consideration the source of HPC research,
and whether it comes directly or indirectly from federal agencies? Or, returning to the
economic security argument, would curtailment of such research be self-defeating?

CRS-22
Appendix
Table 1. Performance Levels of Computers That Support
Selected Applications of Military Significance
Computer performance level
Applications
(MTOPS)
4,000 to 6,000
Joint Attack Strike Aircraft design;
nonacoustic antisubmarine warfare
sensor development; advanced synthetic
aperture radar computation
8,000 to 9,000
Bottom-contour modeling of shallow
water in submarine design; some
synthetic aperture radar applications;
algorithm development for shipboards’
infrared search and track
10,457 to 21,125
Nuclear blast simulation
15,500 to 17,500
Computational fluid dynamics
applications to model turbulence around
aircraft under extreme conditions
20,000 to 22,000
Weather forecasting; impact of blasts on
underground structures; advanced aircraft
design
21,125+
Submarine design; shallow water
acoustics analysis
24,000+
Automatic target recognition template
development
46,000-76,000
3D modeling and shock physics
simulation for nuclear weapons
applications
+120,000
Multi-line towed array signal processing
Sources: Seymor Goodman, Peter Wolcott, and Grey Burkhart. Building on the Basics: An
Examination of High Performance Computing Export Control Policy in the 1990s
(1995);
and High Performance Computing, National Security Applications, and Export Control
Policy at the Close of the 20th Century
(1998). Stanford University. Palo Alto, California.

CRS-23
Table 2. 1996 HPC Export Control Tiers and Policies
Tier/Countries
Restrictions/Controls
Record Keeping
Tier 1: Western Europe,
No licenses required for
Record keeping by U.S.
Japan, Canada, Mexico,
exports or re-exports of
exporters of HPCs above
Australia, and New
computers to/among these
2,000 MTOPS required.
Zealand.
countries; exports of
unlimited MTOPS
permitted.
Tier 2: South America,
Licenses required for
Record keeping and
South Korea, ASEAN,
exports or reexports of
reporting by U.S. exporters
Hungary, Poland, the Czech HPCs above 10,000
of HPCs up to 10,000
Republic, the Slovak
MTOPS
MTOPS required.
Republic, Slovenia, &
South Africa.
Tier 3: India, Pakistan, all
Export licenses are required Record keeping and
of the Middle
for exports or reexports of
reporting required for all
East/Maghreb, all of the
HPCs above 2,000 MTOPS
HPCs between 2,000 and
states in the former Soviet
to military end-users or
7,000 MTOPS.
Union, People’s Republic of uses, or to nuclear,
China (PRC), Vietnam, and chemical, biological, or
the rest of Eastern Europe. missile end-users or uses;
licenses required for HPCs
above 7,000 MTOPS to all
end-users or uses.
Tier 4: Cuba, Iran, Iraq,
No computers are allowed
Not applicable.
Libya, North Korea, Sudan, either through direct or
and Syria.
indirect trade.
Source: [http://www.doc/bxa.gov]