Forecast

2008©B.F.Dorfman

WHAT IS PREDICTABLE – WHAT IS NOT?

From: Presentation for Clarkson University, November 2004

PART 1

TECHNICAL SUCCESS IS PREDICTABLE

COMPUTER DEVELOPMENT

FORECAST FROM THE PAST AND FOR FUTURE

It is known, forecast for specific innovation is rarely successful. This is not true with regard to ‘metatechnologies’ as microelectronics and computers. The following diagrams and plots for semiconductors and all major kinds of computers are still accurate although they were built over two decades ago.

The forecast was based on correlated analysis of technological and physical parameters of elements, interconnections and systems (each constituent trend encompasses semi-linear branch that for one of them is known as ‘Moor law’ and ‘saturating’ branch). The actual development shown in colored dots is different only in that that in reality industry often missed the right time for transition to new ‘sub-paradigms’, such as multiprocessor-chips.

Top - original diagram of computers’ development with 20-years forecast from 1986 to 2006 [1,2] (Cray-2 and GF-10 – based on advanced publications describing their designed performance). Bottom - an English version published 3 years later [2]. The plots for all shown types of computers are still accurate, including exact matching for 10¹⁵ milestone reached by ‘Blue Gene’ in a simplified mode test.

Line for personal computer in bold plotted in assumption that semiconductor industry will start multi-processor chips’ production at the end 1990s.

Indeed, industry mistakenly delayed this transition for ~5 years following a single-processor line -only now fixing the situation and returning to the bold line. Similar scenario had happened with transistors: the industry was too persistent in its efforts to further increase their frequency.

The computers’ forecast was conducted based on analysis of the elements (as well as the entire hardware system) up to respective critical limits.

The forecast was based on correlated analysis of technological and physical parameters of elements, interconnections and the entire hardware system (each constituent trend encompasses semi-linear branch that for one of them is known as ‘Moor law’ and ‘saturating’ branch). The actual development shown in colored dots is different only in that that in reality industry often missed the right time for transition to new ‘sub-paradigms’, such as multiprocessor-chips.

As predicted, most major characteristics of transistors asymptotically approach the respective principle limits.
The remaining resources ~ 1 order of magnitude vs. 4 to 6 order of magnitude of the past progress.
Does the proximity of transistor limits mean the end of solid-state electronics and the urgent transition to molecular electronics, single-electron transistor and quantum computers?

Quite opposite. Stabilization of transistor structure will signify just a beginning of real evolution of solid-state systems

Transistor is a building atom of solid-state computers. Evolution may only start when the atoms reach stability; it is impossible based on radioactive, changeable atoms.

Elements’ and system’ factors in computer progress

	Elements, frequency t, ns	Elements, density cm-2	Most powerful universal computers, F, s-1	Personal computers F, s-1
1950	104=>105	10-2=>10-1	103=>104	-
2004	109=>1010	108=>109	1014=>1015	108=>109
2004/1950	105	1010	1011	105

Thus, even in supercomputer the past progress is primarily due to the progress of elements, while in personal computers it is still completely due to the elements progress.

The systems’ resources have been only slightly explored yet.

New forecast from the present into the coming a few decades also shown. Accurate forecast from the past gives a reason to believe that newly suggested forecast for successful technical progress in the coming decades is also correct. But predictable technical success obliges us to think responsibly about the possible social sequences which are not as easy predictable.

Forecast for future computer development

suggested at presentation for Clarkson University in 2004

This diagram of the year 1981 (V. F. Dorfman, On the Border of Millenniums, Moscow, 1982. In Russian) locates the brain (hatched area from the right side) vs. computers Number of elements (processors or neurons )– number of connections between each elements with other elements, frequency of elements. Shown are three major directions of computers’ development trends: 1. Maximum speed of elements vs. minimum connections in specialized computers; 2. Medium speed and connectivity in standard computers; 3. Limited speed vs. maximum connectivity in eventual provisioned trend directed to (but not reaching) the brain performance.

This diagram of the year 1981 (V. F. Dorfman, On the Border of Millenniums, Moscow, 1982. In Russian) locates the brain (hatched area from the right side) vs. computers Number of elements (processors or neurons )– number of connections between each elements with other elements, frequency of elements. Shown are three major directions of computers’ development trends: 1. Maximum speed of elements vs. minimum connections in specialized computers; 2. Medium speed and connectivity in standard computers; 3. Limited speed vs. maximum connectivity in eventual provisioned trend directed to (but not reaching) the brain performance.

Brain vs. Computer
per ‘quantum of time’

Direct comparison of performances Brain vs. Computer would be improper due to great

differences in the respective frequencies neuron vs. solid-state m-processor.

Instead, we will introduce ‘quantum of time’ – the time interval reversed to frequency and will compare the Brain vs. Computer performances during the respective quantum of time.

Best contemporary personal computer proceeds

P~ 0.1 op/quantum of time vs. brain’ P >~1 billion

Top contemporary supercomputers

P~ 1000-10,000 op/quantum of time vs. brain’ P >~1 billion

It is still far from brain performance, but basically supercomputer performance per “quantum of time” progresses very fast in this direction.

Computer technology from 1964 to 2004 was transistor technology.

Beginning from 2005, micro/nano-electronics is microprocessor technology.

Microprocessor is invariant vs. physics of elements (such as transistors).

This will create the base for eventual shift

solid-state "statistical" bit processing to single quantum phenomena in data processing.

Further development = development of system hierarchy in a single chip

Achievement of every new level of system hierarchy requires about a quarter of century (compare with 2-years stratified Moor's law).

But every new level of hierarchy in data processing will strongly change the phase of civilization.

2008©B.F.Dorfman

Source of data

1. V.F.Dorfman, L.V.Ivanov, Computer and its Elements: Development and Optimization, Moscow, 1988.
2. V.F.Dorfman,L.V.Ivanov, Problems of the System Technology of Computers, J. New Gener. Comput. Syst. Berlin, 2(1989)1, 3-23.
3. V.F.Dorfman, On the Border of Millenium, "Znanie", Moscow, 1982.
4. B.F.Dorfman, Presentation to Clarkson University, 2004

Talk presented for Clarkson University in November 2004 is available by 2 parts

http://www.clarkson.edu/camp/reports_publications/dorfman/Dorfman_Computers_Part1.pdf

http://www.clarkson.edu/camp/reports_publications/dorfman/Dorfman_Computers_Part2.pdf


"Computers: technology, performance and impact on our life: What is predictable, what is not?"