What if Moore's Law dies?

Latest update time：2015-04-18

Reads：

Click on the blue font " Semiconductor Industry Observation " under the title to follow ICBank

Recommended : Semiconductor circle WeChat platform icquan (welcome to follow)

This month marks the 50th anniversary of the publication of Moore's Law, a full half century. This month, IEEE Spectrum published four articles about Moore's Law. One of the articles is titled "Moore's Law is Dying (and That Could Be Good)". They all say that Moore's Law cannot continue. It also triggered some of my thoughts.

In 1965, the number of components on a microchip doubled, and Gordon Moore predicted that this trend would continue in the near future. In 1975, he revised it to a doubling every two years, and later said it would be every 18 months, or an exponential growth of 46% per year. This is Moore's Law. Moore's Law predicts the development of the semiconductor industry for 50 years. Moore predicted a bright future and changed the world. Such high-speed growth is not seen in other industries. The average yield of corn, the staple food in the United States, has increased by 2% per year since 1950. The efficiency of steam turbine generators that convert heat energy into electrical energy has increased by 1.5% per year in the 20th century, and the average annual growth rate of indoor lighting effectiveness (lumens per watt) from 1881 to 2014 was 2.6%, while that of outdoor lighting was 3.1%. In 1900, intercontinental travel on ocean liners traveled 35 kilometers per hour, while in 1958, the Boeing 707 traveled 885 kilometers per hour, an average annual increase of 5.6%, but this speed remained basically unchanged, even the Boeing 787 is about the same as the 707. From 1973 to 2014, the fuel conversion efficiency of automobiles increased by an average of 2.5% per year. Therefore, the exceptionally high growth of the semiconductor industry in the past 50 years is a special case and it is impossible to maintain it in the future. This is not surprising!

As components get smaller, denser, faster, and cheaper, power consumption is increased and the cost of many products and services is cut, especially computers and digital cameras, but also light-emitting diodes and photoelectric tubes. This is the revolution of the electronic, optical, and solar age. However, this revolution has both blessings and curses. Due to the highly anticipated unintended effects of technological progress, people believe that technological progress will soon give rise to self-driving electric cars, hypersonic aircraft, customized cancer treatments, 3D printing of hearts and kidneys, and the world will move from petrochemicals to renewable energy. However, this period of doubling of transistors did not lead to technological progress in human civilization. Modern life depends on many processes that need to be improved, especially the production of food and energy and the transportation of people and goods. Many historical data illustrate this reality. For example, the first commercial application of transistors was hearing aids in 1952, and microprocessors dominated the entire 20th century, or even longer. The number of transistors has doubled, but the application has not doubled. So, Moore's Law is dead, don't be depressed, there are ways to develop below.

A few years ago, China also listed a scientific research topic, how to keep Moore's Law valid? At that time, I thought, we have no relationship with Moore, so what's the big deal if his law dies? Moore's Law is not a scientific law, but a prediction of industrial development, which must be time-limited. As the integration increases, the leakage current increases, the heat dissipation problem becomes serious, and the clock frequency growth slows down and cannot be increased. The line width will be about 5 nanometers by 2020-2030, which is equivalent to the space of 10 silicon atoms. 10+ years ago, the Roadmap predicted that the line width could reach 7 nanometers, and now it says it can reach 5 nanometers. No matter what, there will always be physical limits.

In this case, what should we do? First, we should see that the number of transistors in integrated circuits is no longer a problem. Just like living in a bungalow, from one room to two rooms, you think it is a doubling of the improvement, much more comfortable. However, now you are given a 100-story building, you will think: What do I need so many houses for? There is no way to arrange it. What is the use of 29 houses for corrupt officials? In the integrated circuit chip, the number of gates is enough, you don’t know what to do with these logics. Therefore, in the past few years, multi-core processors have appeared. Putting more CPUs on a chip may improve the utilization rate. However, these CPUs cannot be parallelized, and the problem of black silicon arises. Many transistors are useless and waste power. There are still studies in this area, but they are basically minor repairs and minor modifications, and they cannot get out of the quagmire of Moore's Law. Secondly, without Moore's Law, it is beneficial to the development of small and medium-sized enterprises. In the past, integrated circuit production emphasized large quantities and monopolistic technology. It was basically impossible for small companies to compete with large monopoly companies. Now, integrated circuits must be specialized, that is, for a specific application, design and produce chips that meet their needs. Since chip production is no longer a problem, the problem lies in the product. Therefore, there will be more and more types of integrated circuits, and we need to work hard on various applications. This leads to the issue of product intelligence. If it was first-level intelligence in the past, it would be like not moving the rook towards the cannon hole in chess. This is first-level intelligence, which means that the machine can see the current step. Now we need to develop second-level intelligence, which means that we need to see that this step may allow the opponent to capture my rook, and my rook has no way to go. This is seeing two steps. Of course, higher intelligence can see many steps, and even compete with famous chess players. This is intelligence, but not artificial intelligence. Now my country has begun to promote artificial intelligence and brain science. In fact, artificial intelligence and brain science are not the same thing. Artificial intelligence has been developed for decades, but it has not achieved anything. Now we need to add intelligence to the application system. The Arduino software and hardware platform does not pursue a new model every day, and is still using hardware from 2005. Instead, it has cultivated deep roots. They appreciate the open standards of hardware accessories. Another example is China's copycat mobile phones, which are not sensitive to performance and CPU technology. Therefore, it can keep the chip unchanged and innovate in intelligence to gain benefits. This provides enough space for innovation, and more and more various intelligent application systems will appear.

A large number of chips directly facing applications require product standardization. The standardization of mobile phones can make parts interchangeable, which promotes the development of the repair industry. Small companies can spend less energy on hardware and use open source hardware just like open source software. FPGA may be more popular. As long as there is a fault in the CPU chip, the entire CPU cannot be used, but FPGA is not like that. You can avoid the fault point and it is reconfigurable. Therefore, programmable multi-core processors and reconfigurable FPGAs are more promising. The open source hardware movement will deepen further. There are relatively few companies producing open source hardware now. They only provide sufficient documentation and licenses for modification, modification, and improvement, and make their own devices. Although open source hardware has made great progress in recent years, and the United States has also established an open source research association. China's copycat mobile phones have also attracted international attention. They use the pinyin "shanzhai", which is a way for small companies to rise.

Source: Min Yinghua's ScienceNet Blog

The author, Min Yinghua, is a researcher at the Institute of Computing Technology, Chinese Academy of Sciences

Packaging, testing and manufacturing QQ group of 1,000 people 94983436 (please indicate that you are from icbank when joining the group)

Recommended Articles (Reply Number View Content)

33. The US bans chips, but China faces opportunities amid crisis

34. 2015 Most Stressful Companies Ranking

35. Introduction to chip production and packaging technology

36. 24 of the world's best semiconductor manufacturers

37. Taiwan Consulting Times: China's Semiconductor Industry

38. 50 Years of Moore's Law: Past and Future