APPENDIX 5: SEMINAR HELD AT THE ROYAL
SOCIETY
1. To enable the Sub-Committee to get a good
understanding of the scientific and technical issues underpinning
its Inquiry, a seminar was arranged at the Royal Society, London
on 13 March 2002.
2. Members of the Sub-Committee present were
Lord Wade of Chorlton (Chairman of the Sub-Committee), Lord Flowers,
Lord Hunt of Chesterton, Lord Lewis of Newnham, Lord Methuen,
Lord Patel and Baroness Wilcox. They were supported by the Sub-Committee's
Specialist Adviser (Professor Steve Furber) and Clerk (Mr Roger
Morgan), and by the Select Committee's Specialist Assistant (Dr
Adam Heathfield). Dr Sarah Pearce of the Parliamentary Office
for Science and Technology was also present.
Presentations
3. The day began with a series of presentations,
summarised below.
Background
4. Professor John Enderby, Physical Secretary
of the Royal Society, outlined the areas where fundamental physics
could soon begin to act as a brake on the pace of hardware improvement.
According to the ITRS, limits of the materials used in CMOS microprocessors
would start to be reached in 2008 currently a red brick
wall in the Roadmap.
5. Quoting the evidence given by the US National
Science Foundation to the House Research Committee Hearing on
"Beyond Silicon" in 2000, Professor Enderby introduced
the relevant areas of research that would be involved in finding
ways around the physical constraints computer science,
physics, mathematics, biology and engineering. A new science was
emerging from the interface of these normally separate areas.
The demand for computing power
6. Professor Tony Hey, Director of EPSRC's e-Science
Programme, spoke about the demand for computing power, with particular
reference to the prospect of handling a data deluge from large-scale
scientific work such as particle physics at CERN, genome sequencing,
and Earth observation. Different types of computing would be called
for:
a. Capability Computing for large-scale
problems requiring supercomputers consisting of many individual
processors operating in parallel; and
b. Capacity Computing for large numbers
of separate problems, each of which could be addressed by an individual
processor.
7. Companies such as British Aerospace were interested
in capability computing to run models of complex fluid dynamic
systems. Other areas of science from protein structure
investigation to weather and climate modelling would require
similar facilities. The supercomputers able to address such problems,
such as IBM's BlueGene project, would always have a market, but
never one of large numbers. There would not be significant commercial
opportunities for the United Kingdom in this sector.
8. Large particle physics experiments needed
greater capacity computing; indeed it was expected that forthcoming
projects at CERN would create such an increase in the rate of
data production that the capacity to analyse it would need to
outstrip Moore's Law growth. Professor Hey discussed the proposal
to address this capacity limitation the Grid. This would
allow researchers to interconnect scattered computing resources
and apply them to various analyses of data, essentially internet-scale
distributed computing.
9. The major challenge to make the Grid a viable
proposition was the development of middleware agreed methods
and procedures to enhance interoperability and efficient use of
processors. This was the central aim of the e-Science programme.
Professor Hey believed that significant commercial opportunities
would arise in creating this middleware. The United Kingdom should
also look to the huge market for personal mobile devices for other
opportunities to play on its strengths.
The role of architecture
10. David May, Professor of Computer Science
at Bristol University, discussed the role of architecture in improving
computer performance. Increasing processor speed and using more
transistors did not always provide a corresponding increase in
operating speed. Insufficient attention to architecture design
meant that general purpose performance had lagged behind the potential
suggested by Moore's Law. The ITRS was, in fact, explicitly based
on maintaining the industry's progress in line with Moore's Law,
enabling either function and performance to be increased at constant
cost or (of greater commercial importance) to provide the same
performance at reduced cost.
11. Economics already required very high volumes
of chip fabrication, and the current business model was becoming
increasing difficult to maintain as design and mask-making costs
escalated. Rather than making a range of different chips for different
applications, with each type being produced in relatively small
numbers, it might become necessary to have generic chips made
in larger quantities, using software to tailor them to particular
applications.
12. Improvements in architecture alone could
have significant impact in improving performance for the end user
and should be pursued energetically. The principal area where
advances were needed was in parallel computing, where many processors
were applied concurrently to a single task.
13. Other complications arose from the difficulties
in verifying that increasingly complex processors worked correctly.
Without advances in formal verification techniques, it was likely
that an ever-greater number of devices would be faulty and unreliable.
There were also limitations in wire technology: while transistors
were getting faster, the interconnecting wires were not, thus
restricting the speed at which information could be moved.
14. The United Kingdom had strengths in many
areas relevant to the inquiry: high complexity microelectronic
design; simple, efficient architectures; programming languages
and tools; theory and practice of concurrency; and formal verification.
The Government should support activities in these various areas.
Particular attention should be paid to creating the clusters of
large research teams with industry links that computer architecture
research required, and to setting up more degree programmes which
spanned architecture, software and verification. More and more
people with these cross-cutting skills would be needed.
Materials options
15. Bruce Joyce, Emeritus Professor of Physics
and Senior Research Fellow at Imperial College, described the
possibilities for pursuing hardware developments in entirely new
ways. He outlined some of the research that was under way in single
electron transistors, spin resonance transistors, and quantum
dots. Some of these were still at a very early stage of development,
and none had got beyond proof of principle demonstration experiments.
There was a considerable gap between the current state of knowledge
in these technologies and any practical applications.
16. However, continuing research was needed in
these and more general areas of materials science; serendipity
would probably have a large part to play in creating a worthwhile
successor to CMOS technology. An initiative to support fairly
fundamental research whilst maintaining a focus on the overall
objective of new processor technologies would be useful. The Science
and Engineering Research Council's 1985 Low Dimensional Structures
and Devices Initiative[113]
provided a good model for such an exercise.
Metrology
17. Dr Kamal Hossain, Director of Science and
Technology at NPL, noted that CMOS chip fabrication required very
accurate measurements of position and surface conditions. For
example, oxide layers on silicon wafers had to be 1.5 nm (or about
8 atoms) thick, with tolerances of only an atom or so. Significant
advances in metrology techniques at the atomic scale were essential
for solving many of the problems identified in the industry's
Roadmap.
18. Metrology helped the semiconductor industry
to improve process control and production efficiency; to reduce
time to market; to reduce manufacturing costs; and to improve
product reliability. Metrology was also an essential tool for
innovation and development. The US and Japan had national programmes
in general nanotechnology, and the fundamental role of metrology
was emphasised in both.
19. At present, NPL were working with a variety
of techniques including atomic force microscopy, scanning tunnelling
microscopy and surface spectroscopy to measure dimensional characteristics,
physical properties and chemical composition on a close to atomic
scale. The objective was to develop reliable new techniques that
could be transferred from one laboratory to another and thus form
the basis of future measurement standards. Such developments would
provide the necessary metrology tools for effective manufacture
of new generations of microprocessors. Dr Hossain saw the United
Kingdom as well placed to lead in the development of the metrological
infrastructure required to allow the momentum of Moore's Law to
be maintained.
Nano-scale computing
20. John Pethica, Professor of Materials Science
at Oxford University, discussed the limits of existing technology
for microprocessing speed and issues arising from nano-scale computing
and quantum information processing. There was burgeoning world-wide
interest in nanotechnology, in particular regarding novel physical
properties and functionalities of ultra-small structures and their
potential synergies with biomolecules. Many areas of nanotechnology
had potential applications in future microprocessors.
21. Professor Pethica identified some factors
other than physical limits that would limit the speed of future
processors. The first was cost the semiconductor industry
was accustomed to the exponential decrease in the cost per operation
or unit of memory. However, the cost of new manufacturing plant
to make processors with smaller components was rising very rapidly,
and technology away from the leading edge might become more valuable.
Application limits would also apply the appropriate approach
to microprocessor design would ultimately depend on the applications.
22. In any case, silicon would remain the basis
of microprocessing for the next 10-15 years. The industry investment
in silicon technology was immense and would not be abandoned overnight.
New technologies for microprocessors were likely to be like silicon
but better or provide greatly changed costing or use radically
different operating principles, such as quantum information processing.
The range of options when considered in the long term was vast.
The field could be narrowed since it was improbable that technologies
that needed to operate at very low temperatures, were very bulky
or were unlikely to integrate well with silicon would take off.
23. The United Kingdom had strengths in photonics,
Silicon/Germanium and quantum computing. While the last was potentially
the most interesting, it was important to note that the many papers
published on the subject illustrated only the relevant principles;
practical applications were a very long way off.
24. Professor Pethica closed by contrasting the
structures fostering the development of innovative high-technology
companies in the United Kingdom and in the US. He saw an urgent
need for the UK Government to develop an integrated strategy to
encourage new businesses in the microprocessing sector.
Discussion
25. In a wide-ranging discussion involving all
the presenters together with
- Sir Robin Saxby and Mike Muller, respectively
Chairman and Chief Technology Office of ARM;
- Dr John Taylor, Director General of Research
Councils;
- Gavin Costigan, EPSRC and PPARC liaison officer
at the Office of Science and Technology; and
- Dr Tim Scragg of the Communication and Information
Industries Directorate, DTI,
the following main points were noted.
a. CMOS technology had a long life left in
it. It had received enormous cumulative investment world-wide,
and would remain the basis for producing microprocessors for many
decades even if future rates of progress became less than
predicted by Moore's Law. Any new technology seemed bound to supplement
CMOS rather than replace it.
b. The development of modern computing had been
based on the abstraction of software and applications from the
hardware on which they ran. The bulk of improvements in computer
performance to date had come from hardware developments. Future
improvements might rely to a greater extent on better architectures
and more adept software. For users of the technology, the rate
of progress in performance might equal or exceed that delivered
to date by better microprocessors.
c. General market demands were for cheaper, smaller
and faster devices normally in that priority order. Reducing
the cost of microprocessors whilst at least maintaining levels
of performance would allow more widespread use of microprocessors,
perhaps even in disposable applications again increasing
benefits that users could derive. CMOS technology was likely to
offer the only route to cheaper devices.
d. The ITRS identified many currently unsolved
technical barriers to continuing the Moore's Law trend in CMOS
technology over next ten years or so, not least in metrology.
Provided IP were safeguarded, there could be substantial financial
returns for those who found the best solutions to these problems.
However, as the United Kingdom was no longer involved in silicon
chip manufacture, the manufacturing implementation would be in
the US and Japan, or the increasingly dominant Taiwan, Korea and
China.
e. Even so, the limits of CMOS miniaturisation
were in prospect beyond about 2010. Below a certain size, the
performance of individual components would turn on atomic level
variations in composition rather than the aggregate properties
on which CMOS technology relied. In any event, continued miniaturisation
would also, at some point, expose CMOS microprocessing to unacceptable
quantum uncertainties and random thermal effects.
f. Substantially different concepts would be
needed to take miniaturisation beyond those limits. Some devices
had already been demonstrated but were a very long way from practical
application. Others were still at a mainly theoretical level.
The United Kingdom had expertise in many relevant areas. Again,
there could be substantial financial returns for those who made
practicable advances, although it would be unrealistic to think
that the United Kingdom could use any innovation in these fields
to break back into capital-intensive manufacturing capacity. (It
would, however, be important to maintain some UK or European hardware
capability to underpin necessary research and development.)
g. Blue skies expertise was spread across academic
disciplines, without any apparent co-ordination as regards possible
future microprocessing technologies. Interdisciplinary collaboration
seemed not to be a natural state, and there was much to be learnt
from previous Research Council activities that had brought different
disciplines together. The Low Dimensional Structures and Devices
Initiative and the Interdisciplinary Centres had been particularly
effective, and could provide working models for a future programme
in microprocessing.
h. Work at the interface between computer science,
physics, biology, mathematics and engineering seemed likely to
hold good prospects for developing valuable technologies. In this
and other cases, however, the time (and finance) needed for development
between any initial innovation and the market should not be underestimated.
i. Many of the largest commercial opportunities
for the United Kingdom seemed likely to arise in software, architecture,
chip design and metrology in which the country already had significant
but perhaps under-recognised skills. Such opportunities would
arise irrespective of the precise nature of the underpinning hardware.
j. Some of the opportunities that would arise
could be taken up and exploited by single UK businesses. In that
connection, the United Kingdom was slightly behind the US in producing
commercial spin-offs from academic research. This arose from a
combination of differences between general cultures, the taxation
system (especially as applied to venture capital), and the attitude
and experience of academics.
k. Other opportunities, and any requiring large-scale
technological change, could be exploited only with substantial
international collaboration. Given the American and Far East domination
of microprocessor manufacturing, securing the necessary critical
mass for UK or other European initiatives could be problematical.
The EU could have a crucial role to play in promoting UK and European
ideas.
26. Members endorsed the Chairman's thanks to
all the participants for their help in clarifying the issues for
the Inquiry in this complicated field, and to Professor Enderby
and the Specialist Adviser for their work in setting up the event.
The participants said that they too had found great value in the
sharing of knowledge throughout the day.
113 Summarised in the SERC's 1992 Review of the Initiative,
ISBN 1 870669 73 8. Back
|