Semiconductor technology for high-speed computer support

The performance of the computer is increased by an astonishing speed of about 200 times times 10 years. The central pillar of its support is the progress of semiconductor technology. This section looks at why advances in semiconductor technology can lead to advances in computers.

Moore 's law-more transistors, higher degree of parallelism

One of Intel's founders, Gordon Moore, published a paper in the 1965 Electronics magazine titled Cramming more Components onto integrated circuits, predicting an increase in integration, The chart is shown in Figure 1.9.

Figure 1.9 shows data from only 4 points from 1962 to 1965, but Moore boldly predicts that this trend will continue for 1975 years. Integrated circuit is the use of optical miniature projection lithography to make smaller parts on a silicon chip, Moore believes that the side length of components will be 0.7 times times the annual rate of decline, and the progress of this production technology will last for at least 10 years.

Fig. 1.9 Moore's Law: An improved prediction of integration of semiconductor integrated circuits

In the chart in Figure 1.9, the number of parts is doubled each year, and Moore added new data in 1975 and corrected the number of parts by 1 time times every 2 years. However, according to recent trends, it has been suggested that Moore's law should be a doubling of the number of parts every 1.5 years, but Moore himself has not said "1 time times more than 1.5 years"

Scaling Law

Moore's law is very well known, but many people may not know that the 1974 IBM Robert Dennard and other people's paper also published the Law of Scaling (Dennard scaling)

Modern CPUs, without exception, are LSI made of MOS transistors. The MOS transistor here uses a metal oxide semiconductor (METAL-OXIDE-SEMICONDUCTOR) structure that controls the current flowing to the semiconductor (Semiconductor) by applying a voltage to the metal part called the gate.

Dennard and others examined the relationship between the size of the MOS transistor and the operating speed and power consumption, and found that if the size and supply voltage were halved, the switching speed of the MOS transistor would increase by twice times and the electricity consumption would be reduced to 1/4.

In addition to these advantages, it has also been found that if you halve the size, you can reduce the area of the semiconductor chip to 1/4, or the same area to make 4 times times the transistor, this is a good thing (as shown in Figure 1.10).

Fig. 1-10 Effect of scaling law

Semiconductor miniaturization from a steady stream of development investment -the pace of miniaturization will remain long-term

As the scaling law says, reducing the size of a MOS transistor or line can achieve the following effects.

Improved performance

Cost reduction for unit transistors

Power consumption decreased

As a result, many of the best researchers in the semiconductor industry are involved, investing in large research and development costs, trying to reduce size, driving miniaturization, and building costly precision processing plants. As the level of miniaturization increases and construction costs grow proportionally, the construction costs of the cutting-edge semiconductor plant, which started in 2009, amounted to 400 billion yen (about 28 billion yuan).

These historical contents will be elaborated in the 2nd chapter. In the last 50 years, Moore's law is still in existence, and the number of transistors is still increasing exponentially over a 1-2-year period. This also benefited from the enormous benefits of miniaturization that drove the semiconductor companies to maintain a huge investment in development to maintain the pace of miniaturization. As for the next few years, the general view is that miniaturization will continue to grow at least until 2015, but it is also argued that miniaturization will be economically constrained by the continuing high cost of research and development and factory construction.

Three pillars of performance improvement-increase frequency, parallel processing, and functional expansion

Increasing the clock frequency to improve the processing speed of each step in the flow-shop is one of the pillars of improved processor performance.

Intel 4004, the earliest microprocessor, can handle only 4 bits at a time, but the microprocessor now uses a transistor that increases at the rate of Moore's law to handle 32-bit or 64-bit each time, and the performance has been greatly improved. In addition, computational circuits such as addition and multiplication are also used to achieve parallel computation by using a large number of transistors and also improve performance. This parallel processing is the second most powerful pillar for improving processor performance.

The virtualization of processors has become popular recently, but the efficiency of virtualization requires new hardware institutions. As processor usage expands, people are trying to get the processor to support new functionality, which is the extension of functionality. These extensions do not improve the performance of simple addition calculations, but can improve the overall performance of the computer, which is the third pillar.

This article is excerpted from the technology of supporting the processor-the world of endless pursuit of speed

(US) The sea-Isaac Ando;

Jian Li translation

Publishing Industry Publishing House