The Royal Society of Edinburgh (RSE) is pleased to respond to the request by the House of Lords Science and Technology Committee for evidence on Innovations in Microprocessing. The RSE is Scotland’s premier Learned Society, comprising Fellows elected on the basis of their distinction, from the full range of academic disciplines, and from industry, commerce and the professions. This response has been compiled by the General Secretary with the assistance of a number of Fellows with substantial experience in this technology
The specific issues identified in the consultation paper are addressed below:
What are the main drivers for increasing computer speed? Is there any reason to expect that the demand for increasing speed will abate?
The main drivers for increasing computing speed come in applications, which range from scientific number processing through telecommunications and into graphics for video and computer games. The demand for increasing speed is likely to continue, for example, as computer graphics become more realistic and model-based.
What are the physical limits to the speed of processing based on present techniques? When are these limits likely to be reached?
Processors get faster mainly because the feature size is reduced in successive generations, i.e. the transistors get smaller and the interconnections get thinner. There is a limit on size imposed by the number of electrons involved in a switching operation becoming so small that the Heisenberg Uncertainty Principle comes into play, but that this is still a long way off. The main limitation in reducing the feature size is in the photolithography, i.e. the technique used to copy the layout drawings on to the silicon during processing. As the sizes get smaller, the photolithography process must use shorter and shorter wavelengths and this becomes increasingly difficult beyond the UV region. In this context, it is believed that CMOS (complementary metal-oxide semiconductor) technology will reach its limits in the 2010-2015 timeframe.
The physical limits, however, may not be speed. Speed can, in principle, continue to increase until the MOSFET (Metal Oxide Silicon Field Effect Transistor) gate-length drops to the order of magnitude of the silicon crystal lattice spacing. However, before then, once gate-length has reached 20-30nm, (by the end of this decade), devices will become intrinsically so poorly matched (due to transistor variability) and noisy, that conventional approaches to processor design will fail.
What are the most promising alternative techniques and technologies for achieving ever greater processing speeds over the next 15 to 20 years?
The main techniques available to increase performance involve the use of parallelism, both within single processors and in systems with multiple processors. Within individual processors, however, techniques such as superscalarity will soon reach their limit. Alternative architectures may also have to be developed as the physical limits of metal-oxide semiconductor technology are reached. Asynchronous design is one such approach that may finally find its application in Deep-Sub-Micron silicon. "Neural" techniques offer another. These techniques will not, necessarily, make things intrinsically faster, but they will allow useful shrinkage of dimensions to continue, with concomitant increases in speed.
In addition, optoelectronics (in various forms and approaches) might make a major contribution to the advance of microelectronics. The concept of using 'optical interconnects' in conjunction with silicon VLSI (Very Large Scale Integration) to bypass this electronic bottleneck has been around for some time, but it has not yet made a major impact. Nevertheless, the probability that it will become important is arguably increasing – and there might well be a quite rapid (one or two year timescale) transition to a situation where optical interconnects become of major importance and are widely used. Fibre-optical communications is well-established, but communicating from one point on a silicon chip to another on the same chip (intra-chip interconnection) is the final stage in a reduction process. As an intermediate stage in that reduction process, there is already (in the USA) quite convincing, near-commercial demonstration of chip-to-chip connection (inter-chip interconnection).
What expertise does the United Kingdom have relevant to these?
There is a great deal of expertise in the UK in the effective use of multiple processor systems. Within academia, Edinburgh University (through the Edinburgh Parallel Computing Centre), Imperial College, Manchester and Southampton Universities are the main centres. An emerging form of parallelism involves integrating multiple processors (and associated components) on a single chip, a technique known as Systems Level Integration. Expertise in this area is being developed at the Institute for Systems Level Integration (a joint venture between Edinburgh, Glasgow, Heriot Watt and Strathclyde Universities and Scottish Enterprise) based in Livingston. New forms of internal parallelism are also being developed in the UK by companies such as Siroyan (based in Reading). Siroyan has developed a "soft core" which combines a RISC (Reduced Instruction-Set Computer) processor with scalable DSP (Digital Signal Processor) processing capability, using the VLIW (Very Long Instruction Word) technique. It is aimed at embedded systems markets such as mobile communications and digital TV.
There are several asynchronous groups, most notably Professor Steve Furber's Group at Manchester University and a large concentration of "neural" researchers (including Professor Alan Murray at Edinburgh University), capable of innovation at the algorithm/ architecture/ paradigm level. The UK also has many computational "mavericks", many of whose ideas are unorthodox, but who are capable of producing genuine innovation in a world where incremental movement is the order of the day.
In addition, the UK has a strong university-based R&D base in optoelectronics and one of the world's leading small- device-modelling groups exists in Glasgow University under Professor Asen Asenov.
Are there significant roles for the United Kingdom in future developments?
There are significant roles for the UK in future developments, but this is not to say that there is any guarantee of success in an endeavour to retain relevance.
What international collaborations would be beneficial?
International collaboration, particularly on a European scale, is an obvious necessity. A small number of companies like Infineon, Philips and ST Microelectronics, based in Germany, the Netherlands and Italy/France - and R&D labs such as LETI in France and IMEC in Belgium - could have major significance and linking UK activity to them could be very desirable.
What actions should be taken by the Government (through innovation policies and otherwise), publicly-funded research bodies and the private sector?
Most of the university research centres mentioned above have been set up in response to Government initiatives which were themselves prompted by academic and/or industrial pressure. Government should continue to provide on-going support for these existing centres, set up with time-constrained funding but which need longer periods of time to become self-sustaining. Providing continuation funding to successful centres and encouraging them to remain open to new ideas and collaborations is an important action which Government and others could take.
Government should try to support innovation and encourage Research Councils to fund "adventurous" and thus "risky" research. Not all projects will succeed, in that new ideas may work less well than expected, however, if the academic community does not try unconventional approaches towards genuinely new paradigms, then who will?
Greater levels of encouragement for university-based start-ups could also be an important element, but not the whole of the story. Recognising the vulnerability of major companies in the British private sector, and the need for more economic encouragement for manufacturing, in particular, would be valuable.